LongBench v2 / 66f950acbb02136c067c5021 / In the context of the MVPF (Marginal Value of Public Funds) framework, which of…

THE
QUARTERLY JOURNAL
OF ECONOMICS
Vol. 135
2020
Issue 3
A UNIFIED WELFARE ANALYSIS OF GOVERNMENT
POLICIES∗
NATHANIEL HENDREN AND BEN SPRUNG-KEYSER
We conduct a comparative welfare analysis of 133 historical policy changes
over the past half-century in the United States, focusing on policies in social in-
surance, education and job training, taxes and cash transfers, and in-kind trans-
fers. For each policy, we use existing causal estimates to calculate the benefit
that each policy provides its recipients (measured as their willingness to pay)
and the policy’s net cost, inclusive of long-term effects on the government’s bud-
get. We divide the willingness to pay by the net cost to the government to form
each policy’s Marginal Value of Public Funds, or its “MVPF”. Comparing MVPFs
across policies provides a unified method of assessing their effect on social welfare.
Our results suggest that direct investments in low-income children’s health and
∗We first and foremost thank the several hundred researchers whose em-
pirical results form the foundation of our estimates. We are deeply indebted to
a wonderful team of research assistants: Caroline Dockes, Harris Eppsteiner,
Adriano Fernandes, Jack Hoyle, Omeed Maghzian, Kate Musen, Nicolaj Thor, and
the rest of the exceptional team of Pre-Doctoral Fellows at Opportunity Insights.
We are also grateful to Raj Chetty, David Deming, Winnie van Dijk, Amy Finkel-
stein, John Friedman, Andrew Goodman-Bacon, Jeff Grogger, Hilary Hoynes, John
Eric Humphries, Larry Katz, Sarah Miller, Evan Soltas, Larry Summers, Michael
Stepner, and Laura Wherry for helpful comments and suggestions, along with sem-
inar participants at the University of Chicago, Georgetown, IFS, the University
of Kentucky, LSE, Michigan, Minnesota, and Texas A&M, along with confer-
ence participants at the NBER and the National Tax Association meetings. This
research was funded by the National Science Foundation (#CAREER1653686
(Hendren) and #DGE1745303 (Sprung-Keyser)), the Sloan Foundation (Hendren),
the Bill & Melinda Gates Foundation (Hendren), and the Chan Zuckerberg Ini-
tiative (Hendren). Any opinions, findings, and conclusions or recommendations
expressed in this material are those of the author(s) and do not necessarily reflect
the views of the National Science Foundation.
C
⃝The Author(s) 2020. Published by Oxford University Press on behalf of President and
Fellows of Harvard College. This is an Open Access article distributed under the terms
of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/),
which permits unrestricted reuse, distribution, and reproduction in any medium, provided the
original work is properly cited.
The Quarterly Journal of Economics (2020), 1209–1318. doi:10.1093/qje/qjaa006.
Advance Access publication on March 5, 2020.
1209
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1210
THE QUARTERLY JOURNAL OF ECONOMICS
education have historically had the highest MVPFs, on average exceeding 5. Many
such policies have paid for themselves as the government recouped the cost of their
initial expenditures through additional taxes collected and reduced transfers. We
find large MVPFs for education and health policies among children of all ages,
rather than observing diminishing marginal returns throughout childhood. We
find smaller MVPFs for policies targeting adults, generally between 0.5 and 2. Ex-
penditures on adults have exceeded this MVPF range in particular if they induced
large spillovers on children. We relate our estimates to existing theories of optimal
government policy, and we discuss how the MVPF provides lessons for the design
of future research. JEL Codes: H00, I00, J24.
I. INTRODUCTION
What government expenditures are most effective at improv-
ing social well-being? Are in-kind transfers preferable to cash
transfers? Does government-provided social insurance efficiently
address market failures? Should we invest more in low-income
children? If so, at what age? Should they be direct investments
or subsidies to parents?
A large empirical literature estimates the causal effects of
historical government policies. These papers frequently conclude
with a brief welfare analysis. The method of that analysis, how-
ever, often differs from paper to paper. When reporting the ef-
fects of health insurance expansions, it is common to report cost
per life saved (e.g., Currie and Gruber 1996). Studies of tax pol-
icy changes often report the implied marginal excess burden or
the marginal cost of funds (e.g., summarized in Saez, Slemrod,
and Giertz 2012). Higher education analyses often report the cost
per enrollment (e.g., Kane 1994; Dynarski 2000). The early child-
hood education literature often reports a social benefit-cost ra-
tio (e.g., Heckman et al. 2010). These varying welfare measures
make it difficult to compare policies, especially if one wishes to
take a bird’s-eye view and perform welfare analysis across policy
categories.
This article conducts a comparative welfare analysis of
133 historical tax and expenditure policies implemented in the
United States over the past half-century. We focus on policies in
four domains: social insurance (e.g., health, unemployment, and
disability insurance), education (e.g., preschool, K–12, college, job
and vocational training), taxes and cash transfers (e.g., top tax
rates, Earned Income Tax Credit (EITC), Aid to Families with
Dependent Children (AFDC)), and in-kind transfers (e.g., housing
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1211
vouchers, food stamps). We draw on existing analyses of the
impacts of these policies to construct the benefit that each policy
provides to its recipients and the policy’s net cost to the govern-
ment. Benefits are captured by the willingness to pay of policy
recipients. The net cost combines both initial program spending
and the long-run effect of the policy on the government’s budget
(i.e., fiscal externalities). We then take the ratio of the benefits to
net government costs to generate each policy’s marginal value of
public funds (MVPF).1 Putting these components together allows
us to measure each policy’s “bang for the buck.”2
The MVPF is useful because it measures the amount of wel-
fare that can be delivered to policy beneficiaries per dollar of gov-
ernment spending on the policy. Equivalently, the MVPF mea-
sures the shadow price of raising revenue from the beneficiaries
of the policy by reducing spending on the policy. For point of refer-
ence, a simple nondistortionary transfer from the government to
an individual would have an MVPF of 1. The cost to the govern-
ment would be exactly equal to the individual beneficiary’s willing-
ness to pay. The MVPF can differ from this benchmark value of 1 if
individuals value an expenditure at more or less than its resource
cost. For instance, if the government provides insurance, willing-
ness to pay may be greater than the resource costs of provision
to individuals if the insurance provides consumption-smoothing
benefits. By contrast, willingness to pay may fall below resource
costs if individuals distort their behavior to receive higher trans-
fers.3 The MVPF may also deviate from the benchmark value of 1
if the policy induces fiscal externalities. For example, if spending
a dollar on a government policy caused individuals to work less,
government tax revenue might fall slightly and then the net cost
of the policy would rise above $1. By contrast, if spending that dol-
lar caused them to get more schooling and consequently increased
1. See Mayshar (1990), Slemrod and Yitzhaki (1996, 2001), and Kleven and
Kreiner (2006) for original definitions, and Hendren (2016) for a comparison of the
MVPF to alternative measures of welfare.
2. In several cases where authors constructed their own MVPFs, we incor-
porate those estimates directly. Where applicable, we adjust these estimates to
harmonize assumptions (e.g., discount rates). In cases where previous literature
has conducted comprehensive cost-benefit analyses of a policy, we draw on the
components of those analyses to reformulate them into their implied MVPF.
3. The intuition here comes from the envelope theorem. Willingness to pay
for a government transfer is determined by the “mechanical cost” of that transfer.
Additional costs due to behavioral responses are not valued dollar for dollar.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1212
THE QUARTERLY JOURNAL OF ECONOMICS
their income, government revenue would rise and the net cost of
the policy would fall below $1. In some cases, positive fiscal exter-
nalities may be large enough to fully offset the initial cost of the
policy. In that instance, the policy has an infinite MVPF, and conse-
quently, spending on the policy results in a Pareto improvement.4
More generally, comparisons of MVPFs correspond to precise
statements about social welfare using the intuition of Okun’s
leaky bucket experiment (Okun 1975). Given two policies, A and
B, suppose MVPFA = 2 and MVPFB = 1. Then one prefers more
spending on policy A financed by less spending on policy B if and
only if one prefers giving $2 to policy A beneficiaries over giving
$1 to policy B beneficiaries. Whether this is desirable ultimately
depends on one’s social preferences for the beneficiaries of
policies A and B. MVPFs measure the feasible trade-offs to the
government—in Okun’s metaphor, the “leaks” in the bucket. By
measuring these shadow prices of raising revenue from different
groups, the MVPF provides a unified method of welfare analysis
that can be applied both across and within diverse policy domains.
We outline the construction of the MVPF for six represen-
tative examples in Section III. At a high level, our construction
of willingness to pay often relies on intuition provided by the
envelope theorem. Our construction of net government costs
involves calculating changes in taxes paid and transfers received,
along with savings or additional costs from crowding out of other
government spending. In Online Appendices A–F we also provide
a detailed explanation of how each MVPF in our sample is
calculated. As is common with any welfare analysis, the creation
of our MVPFs requires various judgment calls. We conduct an
extensive set of robustness analyses, examining our assumptions
about interest rates, tax rates, and forecasting methods.5 In
4. To align with terminology in existing literature, we use various terms inter-
changeably to refer to the same phenomenon. Any policy with a positive willing-
ness to pay and negative net costs we define to have an infinite MVPF. Given the
negative net costs, we also say that these policies “pay for themselves” or “recoup
their initial costs.” In the taxation literature, this is also known as a Laffer effect.
We often note that spending on policies with infinite MVPFs results in a Pareto
improvement. This is because the expenditure is valued by beneficiaries and has
no net cost on the government. This final claim regarding Pareto improvement
formally assumes that all beneficiaries have positive willingness to pay, which is
natural in many of our contexts in which the policies expanded the choice sets of
all beneficiaries.
5. We also provide a Stata do-file for each program that is available on GitHub
(https://github.com/Opportunitylab/welfare analysis).
These
programs
allow
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1213
addition, many MVPF estimates for individual policies contain
considerable sampling uncertainty. We address this by construct-
ing category averages that pool across multiple policies and help
improve the precision of our conclusions. We also test and correct
for publication bias using the methods of Andrews and Kasy
(2019). Given these potential sources of uncertainty, we also focus
our results on broad patterns in the data, rather than conclusions
about individual policies.
Our analysis is inevitably constrained by the scope of existing
literature. Not all policies have been studied with the same degree
of completeness. For each policy, we incorporate all effects that
can reliably be translated into the MVPF, but an omitted impact
could affect our welfare analysis. We therefore assess the robust-
ness of our broad patterns to sample restrictions focused on more
comprehensively studied policies. In addition, we discuss how the
MVPF of each particular policy may vary with the addition (or
removal) of certain effects.6 For example, we find that our MVPF
estimates are most sensitive to changes in the estimated earnings
of beneficiaries—specifically dynamic effects within or across
generations. In the results we discuss below, we focus our primary
conclusions on the broad lessons that are robust to variations in
the availability of estimates on underlying causal estimates.
I.A. Main Results
Our estimates reveal a stark pattern: MVPFs vary sub-
stantially based on the age of each policy’s beneficiaries. We
find the highest MVPFs for direct investments in the health
and education of low-income children. This includes Medicaid
expansions, childhood education spending, and expenditures on
college. In many cases, these policies actually pay for themselves
in the long run. Children pay back the initial cost as adults
through additional tax revenue and reduced transfer payments.
For example, we examine four major health insurance expansions
to children over the past 50 years. We calculate an average across
those policies and find that for each 1ofinitialexpendituretheyrepaid1.78 back to the government in the long run. In particular,
we find that three of four policies fully repaid their initial costs.
researchers to easily modify the set of input assumptions into each MVPF
beyond the robustness we readily provide in the article and the Online Appendix.
6. We provide an extended discussion of these in the Online Appendix.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1214
THE QUARTERLY JOURNAL OF ECONOMICS
We find high MVPFs for policies targeting children through-
out childhood. We do find high MVPFs for early childhood
education programs, including an MVPF of roughly 44 for Perry
Preschool and 12 for Abecedarian.7 In addition, we find large
MVPFs for policies targeting older children, such as historical
equalizations in K–12 school financing (studied in Jackson,
Persico, and Johnson 2016) and policies increasing college
attainment. Our broad patterns contrast with the notion that
opportunities for high-return investment in children decline
rapidly with age (Heckman 2006).
Our results show lower MVPFs for policies targeted to adults.
Most of these MVPFs lie between 0.5 and 2. For example, we find
MVPFs ranging from 0.40–1.63 for health insurance expansions
to adults, 0.65–1.04 for in-kind transfers such as housing vouchers
and food stamps, and from negative values to 1.20 for tax credits
and cash welfare programs to low-income households. These lower
MVPFs reflect the fact that spending on many of these policies
reduced labor earnings. This stands in contrast to our finding that
many policies spending on children increased later-life earnings.
It is important to note that these differences in returns by age
represent general patterns but do not hold uniformly. There are a
number of exceptions. For child policies, we find large variation in
MVPFs across policies, with some estimates relatively close to 1.
In particular, we find lower MVPFs for job training programs and
for college subsidies that do not lead to increases in attainment.
We also find lower MVPFs for transfers to disabled children
and their families. This latter case illustrates that policies with
lower MVPFs are not necessarily “undesirable”—they can be
welfare enhancing depending on one’s social preferences. Unlike
expenditures with infinite MVPFs, policies with low MVPFs
involve a budgetary trade-off that should be weighed against
one’s preference for redistribution.
Among expenditures on adults, we find relatively large
MVPFs for reductions in top marginal tax rates, with estimates
7. In our baseline specifications that harmonize government revenue compo-
nents across policies, we estimate that the government recoups 92% of the up-front
cost of Perry Preschool and 78% of the cost of Abecedarian. Because the cost of
crime impacts are often difficult to quantify, they are not included in our base-
line analyses (when crime estimates are available, we we incorporate them in
alternative specifications discussed in the Online Appendix for each policy). In
this case, if one includes additional estimated effects such as the cost of crime, we
estimate that Perry Preschool does pay for itself and Abecedarian pays for 92% of
the up-front cost.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1215
from 1.16 to infinity. There is, however, substantial sampling
uncertainty in these estimates.8 We also find high MVPFs for
spending on adults that generates spillover effects on children.
For example, providing vouchers with counseling services to
families residing in high-poverty public housing (as part of the
Moving to Opportunity Experiment) helped these families move
to lower-poverty neighborhoods. This led to large increases in
children’s earnings in adulthood that generated sufficient tax rev-
enue to pay for the program cost. Our results highlight the value
of further work to uncover when such spillovers are likely to occur.
I.B. Relation to Previous Theories
The ratio of MVPFs measures the extent to which the govern-
ment can transfer welfare across individuals in society. For this
reason, it relates to the literature on optimal government policy
and redistribution (e.g., Mirrlees 1971, 1976). After presenting
our results, we interpret them in light of this theory. For example,
we tend to find tax cuts to top earners have higher MVPFs than
cuts targeted to low-income households, a result consistent with
the behavior of a progressive planner setting the tax rate in a
Mirrleesian optimal tax model (Mirrlees 1971, 1976). We also
compare the MVPFs of cash transfers to those of in-kind trans-
fers, testing the applicability of the Atkinson-Stiglitz theorem
(Atkinson and Stiglitz 1976; Hylland and Zeckhauser 1981).
I.C. Implications for Future Research
We conclude by providing three lessons for future research.
First, we show how the MVPF framework allows us to quantify
the value of such research. Because the MVPF is a shadow price,
one can use a standard decision-theoretic framework to quantify
the value of reducing uncertainty in our MVPF estimates. Just
as a consumer would be willing to pay to learn the true value of
the products he or she buys, a welfare-maximizing government
should be willing to pay to reduce uncertainty in the cost of
redistribution. Using this approach, we show that a welfare-
maximizing government deciding whether to raise taxes to spend
an additional 1ontheSupplementalNutritionAssistanceProgram(SNAP)wouldbewillingtopay0.24 to make this
decision using a more precise causal estimate of the long-run
8. For example, we estimate an infinite MVPF for the 1981 reduction in the
top marginal income tax rate from 70% to 50%. Our confidence interval, however,
includes both 1 and infinity.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1216
THE QUARTERLY JOURNAL OF ECONOMICS
impact of SNAP using administrative data (as in Bailey et al.
2019) as opposed to survey data (as in Hoynes, Schanzenbach,
and Almond 2016). This highlights the value of expanding the
access to, and use of, large administrative linked datasets for the
study of long-run policy impacts on children.
Second, we show the added insights that come using the
MVPF framework as opposed to traditional cost-benefit analy-
sis.9 It turns out that our general findings would be very similar
in a traditional cost-benefit framework, but the MVPF leads to
different conclusions in certain key instances. This is because the
MVPF and traditional benefit-cost analysis rely on similar inputs,
but the MVPF is unique in incorporating all fiscal externalities in
its denominator.10 For example, when taxes are at the top of the
Laffer curve, the social benefit of reducing taxes by 1is2,11 but
the MVPF of that policy is infinite because the benefits to the in-
dividual are 1andthenetcostofthepolicyis0. More generally,
our results suggest there is value in calculating the MVPF in other
settings, such as crime policy or tax enforcement, where the causal
effects of the policy have clear effects on the government’s budget.
Last, we discuss the implications of the MVPF framework
for future empirical designs. In particular, we highlight the im-
portance of determining whether willingness to pay is positive or
negative. In this article, we sought to analyze state-level welfare
reforms from the 1980s and 1990s. There were 27 large-scale
state-level randomized controlled trials (RCTs) analyzing welfare
reform. These studies increased our understanding of the employ-
ment and revenue impacts of welfare policy. They demonstrated
that these welfare reforms had low net costs. That said, while the
treated participants in these studies often received additional
services such as job search assistance, these policies also cut ben-
efits for those who did not comply with program requirements. As
9. The edited volume from Weimer (2009) provides a discussion of cost-benefit
analyses from different researchers in a range of different domains. The Washing-
ton State Institute for Public Policy (WSIPP 2019) conducts ongoing cost-benefit
analyses to assess policies relevant to state legislatures. See also Rea and Burton
(2020) for an application of the WSIPP data to comparative welfare analysis.
10. Traditional cost-benefit approaches include fiscal externalities in the nu-
merator (see Greenberg, Deitch, and Hamilton 2010).
11. The individual is willing to pay 1forthetaxcutandthegovernmentreceivesa1 benefit from increased tax revenue from the behavioral response to
the tax. In traditional cost-benefit analysis, increases in government tax revenue
are included in the numerator of the expression.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1217
a result, it is unclear whether willingness to pay for these reforms
was positive or negative. Despite randomizing more than 100,000
families into 27 large-scale RCTs, we are unable to reach any reli-
able estimates of the MVPFs of these policies. The evaluations of
welfare reform may have led to more valuable information if the
RCT designs had been created with a social welfare framework
in mind.
I.D. Relationship to Existing Literature
In constructing our MVPFs and presenting evidence for
high returns to investment in low-income children, we build
on a substantial line of existing research making the argument
for investment in children.12 Our work is also related to recent
research on the long-run effect of safety net protections for
children reviewed by Hoynes and Schanzenbach (2018). In light
of the evidence, they conclude that “reallocation of investments
over the life course to earlier periods can be efficiency-enhancing,”
which aligns with our conclusions.
There are also analyses—many of which we draw on in
this article—in which researchers have previously argued that
some government expenditures largely pay for themselves.
This argument is particularly prominent in discussion of early
education (e.g., Heckman et al. 2010; Garc´
ıa et al. 2017) and child
health care expenditures (e.g., Brown, Kowalski, and Lurie 2015;
Wherry et al. 2018).13 The argument also appears in the tax
literature, where some have argued that reducing top marginal
tax rates produces a “Laffer effect,” raising total revenue.14 Our
analysis builds on that work by evaluating policies at scale and
searching for the presence of high-return policies across a wide
range of policy domains. We find the most robust evidence for
Laffer effects for policies investing directly in children.
12. For example, foreshadowing many of our conclusions, Currie (1994) writes,
“Although the evidence is incomplete, it suggests that in-kind programs have
stronger effects on children than cash transfers, and that programs that target
specific benefits directly to children have the largest positive effects.”
13. Outside the scope of this article, some suggest certain macroeconomic
policies can pay for themselves, such as fiscal expansions during deep recessions
(DeLong et al. 2012). More generally, we omit many potentially relevant categories
of policies, such as macroeconomic stabilization, infrastructure investment, and
environmental policies.
14. In this sense, testing whether the MVPF of a policy change is infinite is a
generalization of Werning (2007)’s proposed test for identifying local Laffer effects
in the income tax schedule.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1218
THE QUARTERLY JOURNAL OF ECONOMICS
I.E. Roadmap
The rest of this article proceeds as follows. Section II
presents the general social welfare framework that motivates the
construction of the MVPF. Section III discusses the sample and
presents six example constructions of the MVPF. Section IV
discusses our main results and the distinction between MVPFs
of policies targeting children versus adults. Section V places the
MVPF estimates in the context of existing theories of optimal
government policy. Section VI presents lessons for future work.
Section VII concludes. As noted already, Online Appendices A–F
provide step-by-step details for constructing each MVPF, and all
Stata do-files for the construction of each MVPF are available on
GitHub.
II. MVPF FRAMEWORK
This section presents a general framework to measure the
welfare impact of changes in government policies. The frame-
work illustrates how the marginal value of public funds provides
natural guidance on the social welfare impact of economic policies.
Consider a government seeking to measure the welfare im-
pact of a government policy change under consideration. We define
social welfare, W, by the weighted sum of individual utilities,
W =

i
ψiUi,
where Ui is individual i’s utility function and ψi is their social
welfare weight. The latter measures how much a 1-unit increase
in utility corresponds to an impact on social welfare, W.15 The
utility function, Ui, measures both current and future well-being
of the individual. For example, if utility were additive over time,
one could nest uncertainty about future outcomes within this
framework, letting Ui = E[
t ⩾0βtuit] where uit is the individual’s
utility t periods from today.
Because the utility function is allowed to vary arbitrarily
across individuals, it will be helpful to normalize units across
individuals. To that aim, let λi denote individual i’s marginal
utility of income at the time the policy is under consideration.
15. For now, we do not place any assumption on these weights, and therefore
they can result from any particular social welfare function. We also assume the
weights do not change in response to the policy, but this is without loss of generality
because we focus on small policy changes below.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1219
This is equal to the effect on individual utility of providing $1
to that individual. Let ηi = ψiλi denote the individual’s social
marginal utility of income at the time of the policy. The value of
ηi measures the impact on social welfare, W, of an additional $1
placed in individual i’s budget today.
The government is considering a set of policy changes indexed
by j = 1, ..., J that change the economic environment (e.g., prices,
public goods) by a small amount. We parameterize the up-front
initial spending on policy j by dpj (which can either be an increase
or decrease). The net impact on social welfare of the policy is
(1)
dW
dpj
=

i
ψi
dUi
dpj
=

i
ηiWTP j
i = ¯
η j

i
WTP j
i ,
where 
i WTP j
i is the sum of individuals’ willingness to pay for
policy j out of their own income, WTP j
i = dUi
dpj
1
λi , and ¯
η j is the
average social marginal utility of the beneficiaries of the policy,
¯
η j =

i
ηi
WTP j
i

i WTP j
i
with weights given by the economic incidence of the policy,
WTP j
i

i WTP j
i .
The values ¯
η j measure how much social welfare increases if one
were to provide an average of $1 to the beneficiaries of policy
j. Each individual is willing to pay WTP j
i for the expansion by
dpj of policy j.16 Therefore, multiplying ¯
η j by 
i WTP j
i measures
the impact on social welfare of an expansion of the policy by
dpj. This means that the welfare effect depends on the effect of
providing $1 to a policy’s beneficiaries, ¯
η j, and the beneficiaries’
willingnesses to pay for the policy relative to cash, 
i WTP j
i .
In accounting for costs, we let R denote the present discounted
value of the government budget, and let Gj = dR
dpj denote the net
impact of the policy on the government budget.17 This net cost is
16. In the derivation of the MVPF, we remain fully general about each individ-
ual’s utility function. We abstract from any behavioral biases in the utility function
that might cause willingness to pay to be incongruent with choices that maximize
well-being. Moreover, in practice, our approaches to inferring willingness to pay
often require assumptions of rationality in individual utility that do not account
for the potential presence of behavioral biases.
17. In practice, the dpj variations that are identified in an empiricist’s re-
gressions will not, in general, correspond to budget-neutral policies. Traditional
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1220
THE QUARTERLY JOURNAL OF ECONOMICS
inclusive both of the initial cost of the program and all other effects
of behavioral responses on the government budget. For example, if
spending $1 on preschool increases wages in the future, Gj should
incorporate the effect of those increases in future tax receipts.
Crucially, both the willingness to pay measures, WTP j
i , and the
net cost, Gj, should include effects on both parents and children.
Policies that directly affect children should include willingness
to pay by parents and the impacts of their behavioral responses
on the cost of the policy. Conversely, policies that directly affect
parents should include any spillovers onto children.18
The MVPF of policy j is given by the aggregate willingness
to pay, WTP j = 
i WTP j
i , for the policy divided by the net cost to
the government, Gj:
(2)
MVPF j =

i WTP j
i
Gj
=
WTP j
Net Cost.
The MVPF is previously defined in Mayshar (1990), where it is
referred to as the marginal excess burden (MEB); in Slemrod and
Yitzhaki (1996), where it is referred to as both the marginal cost
of funds and the marginal benefit of projects, depending on the
policy in question; and in Kleven and Kreiner (2006), where it is
referred to as the marginal cost of funds (MCPF). However, the
MVPF formally differs from both the traditional definition of the
marginal excess burden in Auerbach (1985), Auerbach and Hines
(2002), and the marginal cost of funds in Stiglitz and Dasgupta
(1971), Atkinson and Stern (1974). Because of this, Hendren
(2016) defines this quantity as the MVPF to contrast it with the
MEB and MCPF.
approaches would attempt to account for government spending by modifying
the observed policy into a different policy that raised revenues via lump-sum
taxation. This would then require the researcher to observe not the causal effect of
the policy, but the “compensated effect” of the policy to identify the welfare effect.
In contrast, our approach hypothetically closes the budget constraint by comparing
two MVPFs: one that involves an increase in spending and another that involves
a reduction in spending or increase in revenue. Hence, welfare analysis can be
done with two sets of causal effects (one for the two policies under consideration)
as opposed to attempting to measure the compensated effect of a policy.
18. We sum the benefits accruing to both parents and children, but we do not
include any willingness to pay that arises because of parental altruism toward
their children (or children’s altruism toward their parents). This means that a
child’s willingness to pay for a policy is only counted once. Including willingness
to pay from parental altruism would only reinforce our central results. Similarly,
we do not incorporate individual willingness to pay for redistribution to others.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1221
Combining equations (1) and (2), the effect on social welfare
per dollar of government expenditure on policy j is
dWt
dpj
dR
dpj
= ¯
η jMVPF j.
Given the MVPF for any two policy changes, one can construct hy-
pothetical budget-neutral policy changes. For example, consider
increasing spending on policy 1 by a net amount G1, financed by
reducing spending (or increasing revenue) from policy 2 by the
same amount. Pursuing this combined policy, dp, increases social
welfare if and only if
(3)
¯
η1MVPF1 > ¯
η2MVPF2.
Welfare increases if and only if the welfare gains from increasing
spending on policy 1, ¯
η1MVPF1, exceed the welfare loss from
reducing spending on policy 2, ¯
η2MVPF2. The MVPFs of the
two policies characterize the cost of moving welfare between the
two groups of beneficiaries. One prefers the policy if and only if
¯
η1
¯
η2 > MVPF2
MVPF1 . If MVPF1 = 1 and MVPF2 = 2, then an individual
prefers spending on policy 1 financed by policy 2 if and only
if providing $1 to beneficiaries of policy 1 is valued more than
providing $2 to beneficiaries of policy 2.
As this example illustrates, welfare statements that com-
pare policies generally require comparisons of their MVPFs. The
MVPFs allows the researcher to form hypothetical budget-neutral
policies and assess their welfare implications using equation (3).
To reduce the role of social preferences in driving conclusions, one
can compare policies with the same beneficiary group. In this case,
one would expect that ¯
η1 ≈¯
η2 so that comparisons of the MVPFs
correspond to statements about social welfare. For example,
Hendren (2017a) suggests comparing the MVPF of a particular
policy to the MVPF of a tax cut with similar distributional inci-
dence. More generally, one can compare different redistributive
policies, such as food stamps and housing vouchers, among each
other to evaluate the most effective method of redistribution.
In some cases, one does not need to compare an MVPF to
another policy to reach a welfare conclusion. This occurs when
the MVPF is infinite. Mathematically, this happens when a
policy has positive willingness to pay by its beneficiaries and the
behavioral response to the policy generates fiscal externalities
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1222
THE QUARTERLY JOURNAL OF ECONOMICS
that are sufficient to cover the cost of the program, Gj < 0. The
textbook example of such a case is lowering taxes when they are
beyond the peak of the Laffer curve. In this case, lowering taxes
increases government revenue, and so these policies represent a
Pareto improvement for any positive welfare weights assigned to
the recipients.19 More generally, the MVPF framework facilitates
a search for other cases where policies have positive willingness
to pay and negative net costs, such as investment in kids.
The definition of the MVPF is theoretically motivated using
small (marginal) changes in government expenditures. Although
some empirical variation we use has marginal effects on individ-
uals’ budget constraints, one can also continue to construct the
MVPF as the ratio of willingness to pay to net government cost
for nonmarginal policy changes. This approach uses the actual
empirical variation in existing literature to estimate the return
on the observed nonmarginal expenditure. Future work could
explore how the MVPF for a given policy change varies within a
program’s size of spending. This would facilitate improved welfare
comparison for policies that were evaluated at different scales.20
II.A. Comparison to Social Cost-Benefit Analysis
The MVPF approach builds on a large literature on social
cost-benefit analysis (see the edited volume Weimer and Vining
2009 and Boardman et al. 2017, and the cost-benefit estimates
provided by WSIPP 2019). The MVPF uses many of the same un-
derlying estimates used to create benefit-cost ratios, but combines
them in a different way. A comparison with cost-benefit analysis
from Heckman et al. (2010) helps illustrate the importance of
19. In practice an expenditure policy may have been combined with a sep-
arate tax policy to raise revenue at the time the policy is implemented. In this
case, the combined expenditure and tax policy would not deliver a Pareto improve-
ment, as some current taxpayers would be made worse off. However, the infinite
MVPF corresponds to a case where the government need not raise revenue to im-
plement a policy that does not cost money in the long-run. The government could
have borrowed against the future returns on the policy and generated a Pareto
improvement.
20. Consider the case where policy 1 was a $1M government expenditure and
policy 2 was a 2Mgovernmentexpenditure.Comparingpolicy1andpolicy2wouldrequiretheMVPFforaversionofPolicy1thatisscaleduptocost2M. This same
logic would also apply if considering a large-scale expenditure on a policy that
had previously been analyzed with a narrower RCT—one would have to make the
additional assumption that the average treatment effect of this expanded policy is
given by the effect identified in the RCT.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1223
these differences. Heckman et al. (2010) compare the net social
benefits of the policy, inclusive of benefits that accrue back to
the government, against the up-front budgetary spending on the
policy, Cj. They use the following formula:
(4)
BCRj = Social Benefits
Social Costs
= WTP j + FEj
(1 + φ) C j
,
where FEj = Gj −Cj are the benefits accruing to the government
budget from the behavioral responses to the policy. The initial
program outlays in the denominator are often multiplied by 1 +
φ, where φ is the marginal deadweight loss of raising government
revenue. This is thought to translate the up-front costs into social
costs by accounting for the welfare impact of an implicit tax
policy that raises the needed funds. Often, φ is taken to be 0.3 or
0.5 (Heckman et al. 2010). Policies are then deemed to pass the
cost-benefit test if the BCR exceeds 1.
In contrast to the BCR, the MVPF is given by MVPF j =
WTP j
C j+FEj . It differs in two primary ways. First, the impact of be-
havioral responses on the government budget is counted in the
denominator, not the numerator. For example, consider a tax cut
of 1forwhichthebehavioralresponseincreasestaxrevenueby1. In this case, the policy perfectly pays for itself, and so the
MVPF is infinite. Expenditures on the policy represent a Pareto
improvement. In a BCR framework, however, that $1 in increased
tax revenue is considered social benefit and counted in the numer-
ator. That leaves a BCR estimate of (
2
1 + φ ). This illustrates why
the BCR may be a particularly misleading guide to optimal policy
when policies have strong impacts on the government budget. We
found a policy with a BCR of (
2
1 + φ ) that was a Pareto improvement,
but we could find a different policy with a BCR above 2 that does
not deliver a Pareto improvement. For example, if we compare
this hypothetical tax cut to government-provided insurance with
willingness to pay of $2 for each $1 of insurance, the traditional
cost-benefit framework cannot distinguish between these policies.
Second, the MVPF approach does not require the government
to close the budget constraint through an increase in taxation.
Therefore, one does not adjust for the “deadweight cost of tax-
ation” based on this particular assumed method of government
finance. Rather, the MVPF directly measures the amount of
welfare delivered to beneficiaries per dollar of government
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1224
THE QUARTERLY JOURNAL OF ECONOMICS
expenditure. One closes the budget constraint by comparing the
MVPF of a given policy to the MVPF of other policies. This allows
the researcher to think through the library of feasible levers
available to the government. In contrast to the cost-benefit frame-
work, this approach reinforces the idea that incidence matters: a
policy that provides benefits to the poor cannot be readily
compared to the raising of revenue on the rich without thinking
about Okun’s bucket and the social welfare weights placed on the
beneficiaries (i.e., the values of ¯
η j for the policies).
Despite our advocacy for the value of the MVPF over a tradi-
tional cost-benefit analysis, it is perhaps reassuring to note that,
in most cases, these two approaches generate similar conclusions.
So although we argue that the MVPF is more appropriate for
measuring welfare, and consequently more informative in cases
where these two welfare measures diverge, the broad pattern of
our results remain the same under either framework.
III. CALCULATING MVPFS: EXAMPLES
We estimate the MVPF for 133 policies spanning social
insurance (e.g., health, unemployment, and disability insurance),
education (e.g., preschool, K–12, college, job and vocational train-
ing), taxes and cash transfers (e.g., top tax rates, EITC, AFDC),
and in-kind transfers (e.g., housing vouchers, food stamps). Our
focus here is on policies, rather than papers. In many cases
we combine estimates from multiple different papers, putting
together the puzzle pieces to build the full picture.21
We form a sample of policies in each domain by drawing on
survey and summary articles from each field. We supplement this
initial set of estimates with recent work in each area not captured
in the survey or summary articles. We restrict our attention to
policies in which there is an experimental or quasi-experimental
identification strategy used to estimate the policy’s impact.22
Formally, such papers identify causal effects using variations dpj
in the economic environment. We form our baseline sample with
21. If multiple papers analyze the same causal effect, we generally focus on
the most recent published estimates unless otherwise noted. We provide a detailed
discussion of the alternative specifications in the Online Appendix.
22. We exclude purely cross-sectional identification using controls for ob-
servables in our baseline sample. Within the set of experimental and quasi-
experimental studies, we do not impose our own filter on the quality or validity of
these empirical designs.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1225
policies where one observes effects of the policy that are sufficient
to form a reasonably comprehensive view of both the WTP and
net cost of the policy. We discuss in Online Appendices A–F the
standard for policy inclusion in our categories and the set of
causal effects used in each case. Because this process involves
judgment calls, we also assess robustness of our conclusions to
an expanded sample (e.g., that expands the set of identification
and forecasting methods) and a more restricted sample (e.g., that
requires direct observation of causal effects on income).
Table I lists the set of policies studied, along with the
empirical papers used to form each policy’s MVPF. Column (9)
denotes the set of papers used to construct the MVPF. In many
cases, we draw from multiple papers to form a single MVPF. For
example, some publications might estimate the impact of the
policy on adults, while other papers focus on longer-run effects on
children.
In this section, we illustrate the construction of these esti-
mates using six examples spanning the domains we consider. We
attempt here to provide a diverse set of examples to demonstrate
the range of approaches used to create our estimates. Online
Appendices A–F provides a detailed step-by-step discussion of the
construction of each MVPF. In Section IV.C, we assess robustness
of our primary conclusions to alternative assumptions (e.g.,
different interest rates and tax rate imputations) and alternative
samples.
III.A. Admission to Florida International University
We begin by constructing the MVPF of admitting an addi-
tional student into Florida International University (FIU). This
example illustrates the construction of the MVPF for a policy
targeting youth with effects on later-life earnings. We use similar
methods for other child policies.
We draw on the work of Zimmerman (2014). He uses an RD
design at the school’s academic performance cutoff for applicants
to measure the effect of FIU admission on state university system
enrollment and medium-term earnings outcomes. We translate
his estimates into an MVPF, incorporating the net cost of the
policy and the beneficiaries’ willingness to pay. Throughout,
we construct confidence intervals for our estimates using a
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1226
THE QUARTERLY JOURNAL OF ECONOMICS
TABLE I
DETAILS OF ALL PROGRAMS STUDIED
Sample
Program
Label
Year
Age of
Paper
Papers
Implemented Beneficiaries Baseline Restricted Extended Estimates
Utilized
MVPF
Panel A: Education and job training
Child education
Carolina Abecedarian
Abecedarian
1975
3
x
x
x
Barnett and Masse (2007)
Study
Campbell et al. (2012)
Helburn (1995)
Masse and Barnett (2002)
Masse (2003)
Chicago Child-Parent
CPC Extended
1985
6
x
Reynolds et al. (2002)
Centers, Extended
Reynolds et al. (2011)
Program
Chicago Child-Parent
CPC
1983
4
x
Reynolds et al. (2002)
Centers, Preschool
Preschool
Reynolds et al. (2011)
Program
Chicago Child-Parent
CPC
1986
8
x
Reynolds et al. (2002)
Centers, School
School
Reynolds et al. (2011)
Age Program
(Table 1 is continued at the end of Section VII, before the Appendix.)
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1227
semiparametric bootstrap procedure discussed in detail in Online
Appendix H.23
1. Costs.
Figure I, Panel A shows how we calculate the net
cost of FIU admission. We start with initial costs of $11,403, which
represents the state university system’s educational expenditures
on each marginal admit to FIU.24 Students pay some fraction
of those educational expenses, so we subtract 3,184toaccountforprivatestudentcontributions.NextweaccountforthefactthatsomenewadmitswouldhaveattendedastatecommunitycollegeiftheyhadnotenrolledinFIU.Wesubtract5,601,
Zimmerman’s estimate of the amount the government would
have paid to support their education at those community colleges.
Taken together, that leaves us with an up-front government cost
of $2,617 per admitted student.
The remaining cost considerations all stem from earnings
changes caused by FIU admission.25 Zimmerman (2014) calcu-
lates that in the first seven years after admission, earnings fall
by $10,942.26 We use estimates from the Congressional Budget
Office to estimate that the tax and transfer rate on these earnings
is 18.6%. This suggests the earnings change reduces government
revenue by $2,035.27 Next, Zimmerman (2014) estimates that FIU
23. In particular, we conservatively account for correlations across estimates
in a given policy, and we develop a method to adjust for the uncertainty in the
denominator (with many thanks to conversations with Isaiah Andrews). We pro-
vide the intuition for the approach and Monte Carlo simulations with appropriate
coverage. In fact, the coverage is sometimes overly conservative, especially when
costs approach 0.
24. Zimmerman (2014) calculates costs and student contributions using the
data on educational expenditures from the Delta Cost Project (American Institutes
for Research 2017). We adopt this approach for other college policies analyzed in
our sample. Online Appendix B explains the details of our approach.
25. Zimmerman (2014) does not include any information on attendance of
federally supported graduate schools among marginal FIU enrollees. If that infor-
mation were available, it could be incorporated as an additional fiscal cost.
26. All earnings changes are discounted back to the time of the initial expen-
diture using a 3% discount rate. We toggle these discount rates in our robustness
discussion in Section IV.C. We also use CPI-U-RS when we need to deflate from
nominal dollar values to real ones.
27. To be conservative, we exclude payroll taxes because individuals may ben-
efit from a portion of these contributions. More detail on our calculations can be
found in Online Appendix G. The tax and transfer rate includes federal and state
income taxes along with food stamps, but excludes housing vouchers and other
welfare programs. We use the income-specific rate from the 2016 CBO estimates,
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1228
THE QUARTERLY JOURNAL OF ECONOMICS
FIGURE I
WTP and Cost Components for Admission to Florida International University
This figure illustrates the cost and willingness to pay components for admission
to Florida International University as studied in Zimmerman (2014). Panel
A breaks the total cost down into its various components, including increased
student payments on tuition, reduced government spending on community
colleges, and the changes in tax revenue from earnings. Panel B shows the
cumulative discounted cost of the policy over the lifetime of the beneficiary.
The solid line represents cumulative costs for ages up until 33, the oldest age
at which incomes are observed in Zimmerman (2014). The dotted lines provide
the 95% bootstrap (pointwise) confidence intervals with adjustments discussed
in Online Appendix H. The dashed line shows total costs inclusive of projected
costs at subsequent ages. The projection method is detailed in Section III and in
Online Appendix I. Panel C reports the components of our WTP calculations. The
point estimate measures WTP as the change in incomes after taxes and expenses
on tuition. All numbers are in 2005 dollars deflated using the CPI-U-RS and
discounted using a 3% real interest rate.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1229
admission causes earnings to rise by $36,369 in years 8–14. Once
again, we apply a tax and transfer rate and determine that the
government’s revenue rises by $7,274. At this point our net costs
are −$2,622, as shown in Figure I, Panel B. This suggests the
expenditure has paid for itself within 14 years of the initial outlay.
Finally, Zimmerman’s earnings data extend 14 years, but we
can extrapolate from the observed effects to estimate earnings
changes over the full life cycle. Online Appendix I describes this
procedure in detail, and Appendix Figure I provides a graphi-
cal illustration of the approach. We use ACS data to estimate
life cycle earnings trajectories and then map the control group in
Zimmerman (2014) onto those trajectories. In particular, we ob-
serve an average earnings for the control group of $28,964, which
we estimate to be 113% of mean earnings for this cohort in the
ACS. In contrast, the treated group earns $6,372 more during
these ages, or 22% more than the control group. We assume that
the control group earnings remain constant as a fraction of av-
erage ACS earnings throughout the life cycle. We also assume
that the percentage earnings increase for the treatment group
also remains constant throughout the life cycle. These assump-
tions mean that we assume the trajectories for the treatment and
control groups differ by a constant percentage throughout the life
cycle.28 This yields an estimated discounted earnings increase of
$117,330 through age 65. We subsequently calculate that the as-
sociated fiscal externality reduces government costs by $21,823.
When combined with our previous cost components, we find that
each marginal FIU admission has a net cost of −$24,445. The
expenditure pays for itself.
and we apply this rate uniformly across years for simplicity. With more reliable
historical information on marginal tax and transfer rates across the income distri-
bution, one could perform the analysis separately by year. We are not aware of any
comprehensive historical source on the distribution of those rates. For this reason,
we take the simpler approach of using a consistent 2016 tax and transfer rate
and then assessing the robustness of all our results to alternative rate assump-
tions. We present robustness to alternative tax and transfer rate assumptions in
Section IV.C.
28. Although this is a strong assumption, we show in the robustness analysis
that our results are actually not very sensitive to the method we use to construct
these forecasts. For example, we conduct a conservative forecast that assumes
zero income growth over the life cycle. This yields similar results (see Figure VI,
Panel B).
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1230
THE QUARTERLY JOURNAL OF ECONOMICS
2. Willingness to Pay.
Having established that the initial
costs of increasing admission at FIU leads to long-run net savings
to the government, the policy has an infinite MVPF as long as
WTP > 0. That said, constructing a measure of willingness to pay
remains useful in making our confidence intervals and evaluating
alternate specifications. The components of our baseline estimate
of WTP are illustrated in Figure I, Panel C.
Throughout, our approaches to estimating WTP rely heavily
on the logic of the envelope theorem and revealed preference. For
the baseline estimate, we assume that increases in income among
the college educated stem from returns to human capital, not from
higher levels of effort.29 In this case, the envelope theorem implies
that we can form an estimate of WTP using the policy’s impact
on net income after taxes and other expenses (and ignore the
composition of individuals’ spending).30 We begin by noting that
those who are admitted to FIU have an increase in private costs
associated with additional tuition and fee payments at the four-
year school. This leads to a negative WTP component of $2,851.
Next, the earnings fall in the first seven years after admission
leads to a further negative WTP of $8,907. The earnings gains in
years 8–14 yield a positive WTP of $29,095. Projecting through
the rest of the life cycle yields an additional WTP of 95,507.Combined,thisyieldsatotalwillingnesstopayof112,844.31
29. We refrain from incorporating general equilibrium effects in our willing-
ness to pay due to a lack of evidence on this point. If higher educational attainment
produced positive spillovers on others, aggregate willingness to pay would rise. If
the college earnings premium were driven by signaling effects, then we would
expect other individuals to have a negative willingness to pay.
30. To see this, consider the decision problem of choosing a vector of consump-
tion goods x to maximize u(x; p) subject to q · x ⩽y(p) where q is the price of goods
and y(p) is after-tax income. In principle, the government’s policy choices, p, can
directly affect utility and the budget constraint. For the baseline WTP measure
for FIU, we assume admission to FIU only affects y(p) so that ∂u
∂p = 0, which means
willingness to pay is given by dy
dp (the impact on the vector x can be ignored by
the envelope theorem). However, if effects on income of admission to FIU is the
result of higher levels of effort, that would require an adjustment for the disutility
of labor and our baseline approach would overstate WTP; conversely, if individ-
uals derive additional utility from attending college that is not captured in their
earnings, the baseline approach would understate willingness to pay.
31. We also form a “conservative WTP” of $1 that relies on the logic of revealed
preference that individuals are willing to pay a nonnegative amount for admission
into FIU.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1231
III.B. Medicaid Expansion to Pregnant Women and Infants
Now, we consider a Medicaid expansion to pregnant women
and children in the United States that occurred across states
between 1979–1992. This example illustrates a case where we
construct the MVPF using examples from several papers using the
same identification strategy but focusing on different outcomes.
We construct our MVPF using several different analyses of
these reforms, each of which use the differential timing of the re-
forms across states to measure their impacts.32 Currie and Gruber
(1996) document a significant increase in health insurance cover-
age for pregnant women, along with a corresponding reduction in
infant mortality and low birth weight. Cutler and Gruber (1996)
find significant crowd-out of private insurance policies. Dave
et al. (2015) find reductions in labor supply of eligible women.
Miller and Wherry (2019) find positive effects on children’s future
earnings and health for those whose parents obtained Medicaid
eligibility. We translate these estimates into their implied MVPF,
beginning with costs and then turning to willingness to pay.
1. Costs.
The bar chart in Figure II, Panel A illustrates
the translation of estimates from the literature into their
implied costs to the government. Currie and Gruber (1996)
estimate that the cost of insuring an additional pregnant woman
through the Medicaid expansion was $3,473.33 In addition to
the direct Medicaid costs, Dave et al. (2015) estimate that
Medicaid eligibility leads to a 21.9% reduction in female labor
force participation, which corresponds to an earnings impact
of roughly $2,834. We estimate that these individuals face a
tax-and-transfer rate of 18.9% from the CBO using our procedure
discussed in Online Appendix G. This means that the earnings
effect implies an additional cost to the government of $564 per
eligible child. As a result, a short-run analysis of the policy would
conclude that the causal effects of the policy lead to an increase
in costs.
32. Our analysis also explores other policies that expanded Medicaid to chil-
dren, such as the national expansion of Medicaid to those born after September 30,
1983. These policy changes correspond to separate MVPF constructions because
they arise from different sources of policy variation.
33. For consistency across papers analyzing the reform, we deflate all numbers
to 2012 US$ using the CPI-U-RS; as a result, they differ slightly from reported
figures in each paper.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1232
THE QUARTERLY JOURNAL OF ECONOMICS
FIGURE II
WTP and Cost Components for Medicaid Expansions to Pregnant Women and
Infants
This figure illustrates the cost to the government of providing Medicaid to
pregnant women and infants. The evidence comes from state Medicaid expansions
between 1979 and 1992. Panel A breaks the total cost down into its various
components. The savings on uncompensated care come from Currie and Gruber
(1996), who estimate rates of uninsurance, and Gold and Kenney (1985) who
estimate the quantity of uncompensated care for the uninsured. The savings
on future health costs come from Miller and Wherry (2019). The increase in
government revenue combines an effective tax rate with the estimates of earnings
gains from Miller and Wherry (2019). Panel B reports the components of our WTP
calculations. The point estimate includes the willingness to pay for reductions
in infant mortality, combined with the change in income for children over their
life cycle after taxes and educational expenses. All numbers are in 2011 dollars
deflated using the CPI-U-RS and discounted using a 3% real interest rate.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1233
Turning to the effects on children, Miller and Wherry (2019)
estimate that a 1 percentage point increase in parental eligibility
leads to a reduction in future hospitalizations of 0.237% when
children are 19 to 32 years old. With a 3% discount rate, this im-
plies government savings on Medicaid and uncompensated care
of $868 over the 14-year period from ages 19 to 32.34 Miller and
Wherry (2019) also find a 3.5% increase in college attendance and
an 11.6% increase in earnings for children made eligible. On the
one hand, to the extent to which the government subsidizes col-
lege expenses, increased enrollment raises government costs. We
estimate that effect to be $371. On the other hand, the increase in
earnings when children are 23–36 years old leads to an increase in
government revenue of $3,909. By the time children are 36 years
old, the estimates suggest that the policy has paid for itself.
As with the example in Section III.A, we forecast these
earnings gains to age 65 by assuming that the percentage impact
on earnings remains constant throughout the life cycle. This
suggests that the government recoups an additional 6,114intaxrevenueoverthisperiod,foratotalof10,024. The up-front
cost of 3,473ledtoalong−runnetgovernmentsurplusof7,014
(95% CI of [1,178, 12,971]).
Before moving on to discussing the details of willingness to
pay, it is worth noting that the MVPF of this expenditure has al-
ready been determined. For a policy to have an infinite MVPF, net
costs must be negative and willingness to pay must be any positive
value. The policy evaluated here expanded health care opportu-
nities to parents and children, so it is safe to assume willingness
to pay is positive. In fact, if the policy did not make anyone worse
off, then these expenditures resulted in a Pareto improvement.
2. WTP.
While the baseline MVPF estimate is infinite, we
calculate willingness to pay for use in constructing confidence
intervals and evaluating alternate specifications where costs are
positive. We briefly summarize this construction, which consists
of three components. (Step-by-step details of this calculation can
be found in Online Appendix D.)
First, Cutler and Gruber (1996) document that half of the
increase in Medicaid actually crowded out private coverage. As-
suming that the public and private costs of insurance were roughly
34. We forecast to age 65 by assuming a constant dollar saving and discounting
by 3%, which implies $530 of total savings, as shown in Figure II.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1234
THE QUARTERLY JOURNAL OF ECONOMICS
similar, this finding implies that beneficiaries no longer had to pay
roughly $1,737 in health insurance costs. This means that WTP
is at least 1,737.Second,CurrieandGruber(1996)estimateacausaleffectoftheMedicaidexpansiononinfantmortality.Weassumeparentshaveawillingnesstopayoutoftheirownincomeof1M to avoid an infant death (and assess robustness to alterna-
tive specifications).35 Third, we consider the WTP by the children
for improved labor market prospects in adulthood. To do so, we
assume that the increase in earnings documented by Miller and
Wherry (2019) reflects an expansion of labor market opportunities
and not an increase in costly labor effort. This means that the chil-
dren should be willing to pay the increase in their net income after
private expenses that results from increased educational attain-
ment. The increase in after-tax income is $16,775 for the observed
14-year age range (23–36) in Miller and Wherry (2019) and an
additional $26,236 in the subsequent years. Subtracting the cost
of college expenses reduces this by 111foranetWTPof47,400.
We also provide a conservative WTP estimate using solely
the transfer value of the insurance of $1,737. This would be
valid if the increase in after-tax earnings came at the expense
of increased effort as opposed to increased opportunities. To
be sure, the difference between the conservative and baseline
WTP estimate is quite large. As we discuss below, our primary
conclusions remain valid under either approach.
III.C. Introduction of Food Stamps
Third, we construct an MVPF for the impact of the intro-
duction of the Food Stamp Program, today known as the Supple-
mental Nutritional Assistance Program (SNAP). This example
illustrates how we incorporate potential spillovers of adult-
targeted policies onto children.
The Food Stamp Program provides in-kind transfers to
low-income families that can be used on food. Its introduction in
the 1970s was staggered across counties in the United States.36
35. Note we should think of this as a “private” not a “social” willingness to pay.
It assumes that parents are willing to pay $10,000 out of their own pocket to have
a 1% reduction in infant mortality. It is important to note that society may well be
willing to pay more than $1M. In the language of the social welfare function, this
suggests that the population has a high social marginal utility of income, ηj.
36. This variation was initially studied by Currie and Moretti (2006) in Cal-
ifornia and extended nationally by Almond, Hoynes, and Schanzenbach (2011),
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1235
Hoynes and Schanzenbach (2012) exploit this variation to analyze
its impact on labor income and welfare participation of adult
beneficiaries; Almond, Hoynes, and Schanzenbach (2011) study
its effect on birth outcomes. Bailey et al. (2019) use the same
variation to study its impacts on the adult earnings of children
whose parents received food stamps.
1. Costs.
The first component of our total costs is the average
yearly benefit from food stamp enrollment, equal to $2,904. To
this, we add the fiscal externality resulting from the effects on
both adults and children. For adults, Hoynes and Schanzenbach
(2012) document large yet imprecise reductions in earnings of
$3,650 that imply a fiscal externality of $471 from reductions
in tax revenue—roughly 0.16per1 of food stamps provided.
For children, Bailey et al. (2019) find increases in earnings in
adulthood corresponding to 7.1% for six full years of childhood
exposure to food stamps between the ages of zero and five. In
Online Appendix E, we show that this corresponds to an estimated
increase in tax revenue of 0.24per1 of food stamps for every
family with a child aged 0–5. We then multiply this by 0.35, the
fraction of SNAP benefits received by households with children
age 0–5. We subsequently multiply by 1.32, the average number
of children in these households. This suggests that for each 1infoodstampspending,theresultingeffectsonchildrenincreasegovernmentrevenueby0.11.37 Taken together, these estimates
imply that every 1ofspendingonfoodstampscosts1.05.38
2. WTP. We provide a willingness to pay from three compo-
nents. First, the envelope theorem suggests that individuals are
Hoynes and Schanzenbach (2012), Hoynes, Schanzenbach, and Almond (2016),
and Bailey et al. (2019).
37. We assume no impact on children at older ages, but clearly such effects
could alter the MVPF. In Section V, we discuss the implications for a policy targeted
to families with children aged 0–5; this leads to a larger MVPF.
38. Our costs estimates here are constrained by the set of observed outcomes
that we can reliably translate into effects on the government budget. For example,
Hoynes, Schanzenbach, and Almond (2016) report that the introduction of food
stamps was associated with a reduction in adult metabolic syndromes. Although
our earnings estimates likely capture the effect of those health changes on labor
supply, we lack a reliable way to measure the impact of those health changes on
health care utilization. Future work documenting long-run health impacts that
reduce (increase) government spending on medical care could lead to a higher
(lower) MVPF than we estimate here.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1236
THE QUARTERLY JOURNAL OF ECONOMICS
willing to pay for the mechanical cost of SNAP benefits, which
we estimate to be 1,809.Wearriveatthisnumberbytakingthe3,650 increase in earnings and noting that SNAP benefits
decline with earnings at a 30% phaseout rate. This means that
1,095ofthefoodstampcostistheresultofacostincreasefrombehavioralresponses.Consequently,ourpointestimatesuggestsindividualsvalue0.62 for each 1spentbythegovernmentonfoodstamps.39Second,weincorporatetheWTPforreductionsininfantmortalityandincreasesinlongevityamongtheirchildren.AsinthecaseofMedicaidinSectionIII.D,weassumethisisgivenbythereductioninchildmortalitymultipliedbyavalueofastatisticallife(VSL)of1M (2012US).Weaddtothatvaluethenumberofyearsofincreasedlongevitymultipliedbyaqualityofadjustedlife−year(QALY)of20k (2012US).ThisleadstoanadditionalWTPof0.02. Last, we incorporate an additional
willingness to pay because of increases in after-tax income among
those who received food stamps as children. These estimates of
after-tax income are based on the earnings gains we calculate
above. Combining costs with willingness to pay creates an MVPF
of 1.04 (95% CI of [−0.97, ∞]).40
It is important to note in this case that statistical uncertainty
in these estimates is quite high. The combination of substantial
earnings reductions among parents and large earnings gains
among children mean that we cannot reject MVPFs of 0 or ∞. We
return to this uncertainty in more detail in Section VI.A when
we discuss the value of additional research or data access in
reducing sampling uncertainty.
III.D. Paycheck Plus in New York City
Fourth, we measure the MVPF of the Paycheck Plus program.
This construction illustrates how we create the MVPF from RCTs.
39. It is also worth noting that this willingness to pay is nearly identical to the
value we would receive if we did not apply the envelope theorem in this context, but
rather used estimates from Whitmore (2002) suggesting that food stamps have a
trade value of at least 65%. For our “conservative” willingness to pay specification,
we make both the envelope theorem and trade value modifications and find that
the MVPF falls to 0.39.
40. In Online Appendix E, we also explore several alternate specifications
and find that these produce only small changes to the MVPF. For example, we
assume a higher VSL of 9MandaQALYof180k and find an MVPF of 1.22.
We incorporate the impact of reduced incarceration based on effects estimated in
Bailey et al. (2019) and costs of incarceration from Heckman et al. (2010). We find
that the MVPF rises from 1.04 to 1.07.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1237
It also provides guidance on the ideal set of measures future
researchers could construct to more directly estimate the MVPF
associated with RCTs.
The Paycheck Plus program is modeled after the Earned
Income Tax Credit (EITC). The EITC provides income subsidies to
low-income workers that are intended to encourage employment.
If workers face high marginal tax rates due to the benefit schedule
for means-tested transfers such as food stamps, the EITC may
offset those high rates. While the EITC generally targets adults
with children, the Paycheck Plus program in New York City
conducted an RCT to evaluate the provision of EITC-like benefits
to single adults without dependents—a group not traditionally
eligible for significant EITC benefits. The credit is worth up to
$2,000 a year and is available over three years (2014–2016 fiscal
tax years with bonuses paying out in 2015–2017).
Miller et al. (2017) estimate the effect of the policy on income,
employment, and after-tax income for the first two years of the
policy, which we translate here into their implied MVPF.41 We
begin with costs.
1. Cost.
The cost of the policy is the observed causal effect of
the policy on the government budget.42 To measure the costs, let Tj
denote the tax schedule faced by the control (j = 0) and treatment
(j = 1) groups. Let y j
i denote individual i′s earnings if they face
the j = 0, 1 tax/transfer schedule. The cost is then given by:
(5)
Cost = E

T 0 
y0
i

−E

T 1 
y1
i

.
41. As discussed in Online Appendix D, the current set of results from the third
year do not include sufficient information to form the MVPF in as precise a manner
as we do here; but we note how imposing a reasonable additional assumption
suggests that the third-year effects lead to a very similar MVPF also near 1.
42. In the context of an RCT, our approach measures the welfare impact of
randomly assigning additional people to the treatment as opposed to the control
group. As a result, one can use the reduced-form results to form our welfare
analysis (i.e., one need not separately isolate a LATE/TOT). The denominator
is the causal effect of this assignment on the budget and the numerator is the
aggregate WTP by members of the control group to be in the treatment group. As
a result, whether our welfare analysis can be externally generalized to a different
policy with different take-up of benefits would depend on how its treatment effects
vary across the population.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1238
THE QUARTERLY JOURNAL OF ECONOMICS
Because Paycheck Plus is an RCT, we compute equation (5)
using the difference in tax and transfer revenue obtained by the
government. In 2014, the causal impact on government costs was
$621; in 2015, this cost was 453.Combiningthesevalues,thecostis1,074.
2. WTP. We use the envelope theorem to estimate the WTP
for Paycheck Plus. In 2014, the average bonus paid is $1,399
among those who take it up, and 45.9% of people do so. The en-
velope theorem suggests that participants do not value the full
$1,399 subsidy dollar for dollar. This is because part of this cost
reflects the impact of behavioral responses. To the first order, those
who entered the labor force to obtain the transfer are indifferent
between working and not working. Miller et al. (2017) find a causal
effect of the program on the extensive margin labor supply of 0.9%.
Absent behavioral responses, this implies that 45% of the sample,
as opposed to 45.9%, would have received the transfer had they not
changed their behavior. Consequently, 98% of the transfer ( 45
45.9)
is valued by the beneficiaries, which implies a WTP of 630forthetransfersin2014.43Repeatingthiscalculationusingthedatafrom2015yieldsaWTPof441. This suggests a two-year WTP of
1,070.TheestimatedWTPof1,070 combined with the net cost
of 1,074impliesanMVPFof0.996(whichroundsto1inTableII).OnecanalsoconstructanMVPFseparatelyusingthe2014or2015transfersandresponses.ThisyieldssimilarMVPFsof1.014and0.973.Thisdynamicsimilaritywillbearecurringthemeamongtransferprogramstoadults.Itmeansthatastaticmodelofthelabormarketdistortionsprovidesareasonableapproxi−mationtomeasuringtheMVPFforthesepolicies.Every1 the
government spends in transfers leads to a benefit of roughly $1.44
43. This calculation assumes no intensive-margin responses. If one observed
the microdata from the RCT, one could allow for intensive-margin responses.
To the first order, the WTP is the mechanical change in the tax schedule (i.e.,
replacing T0 with T1) holding behavior fixed for each individual at y0
i :
(6)
WTP = E

T 0 
y0
i

−T 1 
y0
i

.
This means the ideal method of calculating WTP is to feed the distribution of
control group earnings into both the control and treatment group tax schedule.
In practice, this number is rarely reported, but future work conducting welfare
analyses of RCTs can directly construct this measure.
44. If the provision of work subsidies today leads to an increase in labor
earnings and thus tax revenue after the earnings subsidies have ended, then
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1239
TABLE II
MVPF, WTP, AND COST ESTIMATES WITH CONFIDENCE INTERVALS, ALL PROGRAMS
Program
MVPF
MVPF CI
WTP
WTP CI
Cost
Cost CI
Baseline
Child education
∞
[17.83, ∞]
4.82
[3.38, 6.28]
−0.21
[−0.59, 0.19]
Abecedarian
11.89
[−0.18, ∞]
2.62
[−0.24, 5.63]
0.22
[−0.76, 1.15]
x
CPC Extended
∞
[−∞, ∞]
4.15
[−21.80, 27.36]
−1.22
[−6.13, 4.36]
CPC Preschool
∞
[∞, ∞]
2.23
[0.24, 4.21]
−0.35
[−0.62, −0.10]
CPC School
1.32
[−∞, ∞]
−0.18
[−1.23, 0.83]
−0.14
[−1.27, 0.98]
Head Start
∞
[10.58, ∞]
4.42
[2.90, 6.08]
−0.11
[−0.52, 0.27]
x
Head Start RD
0.72
[−0.02, ∞]
0.72
[−12.90, 11.46]
0.99
[−0.03, 1.78]
Head Start RCT
2.41
[1.90, 3.15]
1.29
[1.10, 1.50]
0.54
[0.49, 0.59]
K12 Spend
∞
[∞, ∞]
8.78
[4.58, 13.03]
−1.03
[−2.02, −0.06]
x
K12 Spend Mich.
0.65
[0.05, 2.19]
0.62
[−0.01, 1.58]
0.95
[0.79, 1.08]
Perry Preschool
43.61
[1.83, ∞]
3.45
[1.19, 5.70]
0.08
[−0.52, 0.68]
x
College adult
−5.59
[−∞, ∞]
−2.68
[−143.04, 61.16]
0.48
[−7.74, 18.61]
AOTC (IS)
6.75
[−1.61, ∞]*
2.45
[−6.52, 13.42]*
0.36
[−6.64, 6.47]*
x
AOTC (JE)
−1.77
[−17.06, ∞]*
−7.63
[−68.99, 40.12]*
4.31
[−11.54, 24.28]*
x
AOTC (JS)
∞
[−5.96, ∞]*
9.96
[−44.67, 76.39]*
−1.37
[−16.83, 12.68]*
x
AOTC (SI)
10.05
[−18.36, ∞]
5.36
[−89.06, 104.44]
0.53
[−12.93, 13.61]
x
AOTC (SE)
−0.02
[−2.25, ∞]*
−0.02
[−7.28, 5.59]*
1.12
[−6.30, 8.17]*
x
AOTC (SS)
∞
[−8.00, ∞]*
23.39
[−112.96, 191.48]*
−1.61
[−26.75, 19.56]*
x
HOPE Cred.
12.58
[−24.72, ∞]
5.27
[−745.41, 518.28]
0.42
[−61.22, 131.34]
x
HTC (IS)
18.86
[−2.87, ∞]*
5.91
[−24.65, 43.39]*
0.31
[−11.81, 11.68]*
x
HTC (JE)
2.37
[−2.22, ∞]*
8.21
[−35.77, 61.73]*
3.47
[−21.81, 28.40]*
x
HTC (JS)
∞
[−3.51, ∞]*
18.41
[−87.39, 148.22]*
−3.15
[−69.14, 53.90]*
x
HTC (SE)
11.83
[−4.48, ∞]*
4.59
[−17.25, 31.56]*
0.39
[−4.19, 4.15]*
x
HTC (SS)
−1.64
[−13.38, ∞]*
−1.91
[−22.71, 14.11]*
1.16
[−3.87, 6.67]*
x
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1240
THE QUARTERLY JOURNAL OF ECONOMICS
TABLE II (CONTINUED)
Program
MVPF
MVPF CI
WTP
WTP CI
Cost
Cost CI
Baseline
HOPE/LLC
−8.81
[−∞, ∞]
−42.82
[−266.09, 9.92]
4.86
[−7.59, 31.13]
x
Adult Pell
2.18
[0.71, 6.11]
3.42
[1.02, 6.25]
1.57
[1.03, 2.31]
x
Tuition deduc (JE)
0.77
[−1.92, 38.88]
1.00
[−4.88, 6.49]
1.29
[0.17, 2.51]
x
Tuition deduc (JS)
−0.02
[−2.50, 5.62]
−0.03
[−5.59, 4.43]
1.38
[0.55, 2.49]
x
Tuition deduc (SE)
∞
[−∞, ∞]
1.00
[−7.76, 8.53]
−1.13
[−2.04, −0.26]
x
Tuition deduc (SS)
∞
[−∞, ∞]
5.38
[−1.58, 14.00]
−5.10
[−6.36, −3.41]
x
College child
∞
[4.18, ∞]
8.79
[3.05, 15.65]
−0.36
[−1.76, 0.73]
Cal Grant GPA
∞
[10.72, ∞]
9.41
[3.43, 16.44]
−0.57
[−1.63, 0.32]
x
Cal Grant Inc
−0.69
[−2.36, 7.41]
−1.04
[−5.37, 4.29]
1.51
[0.63, 2.21]
x
CUNY Pell
1.39
[−2.95, 12.88]
1.42
[−3.42, 7.15]
1.02
[0.48, 1.56]
x
CC Mich
29.46
[−2.33, ∞]
7.80
[−9.29, 29.19]
0.26
[−3.39, 2.72]
x
CC Texas
349.51
[1.61, ∞]
10.69
[1.73, 20.89]
0.03
[−2.08, 2.10]
x
DC Grant
22.98
7.62
0.33
x
FIU GPA
∞
[∞, ∞]
13.73
[1.40, 62.13]
−2.97
[−15.62, −0.02]
x
Florida Grant
7.42
[1.09, ∞]
7.40
[1.24, 15.61]
1.00
[−0.32, 2.42]
x
Free FAFSA (dep)
4.03
[0.65, 10.75]
33.67
[3.04, 95.17]
8.35
[2.00, 20.03]
Free FAFSA (indep)
2.12
[−0.06, 9.71]
5.32
[−0.89, 15.05]
2.51
[0.77, 4.91]
Georgia HOPE
4.00
[0.37, 20.63]
3.60
[0.73, 6.48]
0.90
[0.28, 1.57]
x
HAIL Aid
1.30
[0.24, 3.65]
0.97
[0.14, 1.78]
0.75
[0.54, 0.95]
Kalamazoo
1.93
[0.97, 5.61]
1.93
[0.93, 3.71]
1.00
[0.77, 1.24]
x
MA scholarship
0.72
[−0.92, 3.05]
1.21
[−1.81, 4.00]
1.68
[1.25, 2.23]
x
Ohio Pell
2.49
[0.80, 5.40]
2.88
[1.29, 4.40]
1.16
[0.80, 1.56]
x
TN Pell
0.84
[−1.59, 3.57]
0.78
[−1.64, 2.84]
0.93
[0.62, 1.24]
x
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1241
TABLE II (CONTINUED)
Program
MVPF
MVPF CI
WTP
WTP CI
Cost
Cost CI
Baseline
Texas Pell
∞
[∞, ∞]
85.74
[0.77, 173.60]
−17.38
[−33.15, −1.92]
x
Soc Sec College
4.86
[0.98, 52.39]
5.03
[0.82, 10.92]
1.03
[0.32, 1.95]
x
College spend
4.00
[1.25, 20.44]
3.17
[1.26, 5.48]
0.79
[0.36, 1.21]
x
TN Hope
1.86
[0.92, 5.08]
1.94
[0.94, 3.51]
1.05
[0.81, 1.36]
x
College tuition
1.02
[−1.06, 5.47]
1.02
[−1.47, 3.58]
1.00
[0.68, 1.32]
x
WI scholarship
1.43
[1.00, 2.32]
1.46
[1.04, 2.08]
1.02
[0.93, 1.13]
x
Job training
0.44
[−19.57, 0.91]
0.36
[−0.82, 1.51]
0.83
[−0.09, 1.75]
Job Corps
0.15
[−0.23, 0.58]
0.15
[−0.23, 0.55]
0.98
[0.93, 1.03]
x
JTPA adult
1.38
[−0.21, 2.13]*
1.17
[−0.17, 2.64]*
0.85
[0.08, 1.65]*
x
JTPA youth
−0.23
[−3.43, 1.27]*
−0.21
[−1.70, 1.29]*
0.91
[0.15, 1.66]*
x
JobStart
0.20
[0.04, 0.42]
0.20
[0.06, 0.34]
1.02
[0.80, 1.24]
x
NSW Women
1.48
[−∞, ∞]
0.57
[−0.50, 1.64]
0.39
[−0.09, 0.86]
x
NSW Ex-Addict
0.44
0.35
0.79
x
NSW Ex-Offender
0.64
0.53
0.82
x
NSW Youth
0.60
[−∞, ∞]
0.47
[−4.23, 5.15]
0.78
[−3.32, 4.87]
x
Work Advance
0.78
[0.26, 1.34]*
0.64
[0.21, 1.11]*
0.83
[0.83, 0.83]*
x
Year Up
0.43
[0.37, 0.48]
0.41
[0.36, 0.45]
0.96
[0.95, 0.97]
x
Disability ins.
0.85
[0.82, 0.88]
1.00
[1.00, 1.00]
1.18
[1.14, 1.22]
DI generosity
0.96
[0.95, 0.97]
1.00
1.04
[1.03, 1.05]
x
DI judge
0.78
[0.72, 0.85]
1.00
1.28
[1.18, 1.39]
x
DI examiner
0.74
[0.71, 0.78]
1.00
1.34
[1.28, 1.41]
x
DI veterans
0.95
[0.92, 0.98]
1.00
1.05
[1.02, 1.08]
x
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1242
THE QUARTERLY JOURNAL OF ECONOMICS
TABLE II (CONTINUED)
Program
MVPF
MVPF CI
WTP
WTP CI
Cost
Cost CI
Baseline
Health adult
0.89
[0.56, 1.57]
1.49
[1.00, 1.99]
1.67
[1.01, 2.39]
Mass HI (150%FPL)
0.80
1.00
1.25
x
Mass HI (200%FPL)
0.85
1.00
1.18
x
Mass HI (250%FPL)
1.09
1.00
0.92
x
Medicare intro
1.63
[0.52, 3.83]
2.00
[0.58, 3.44]
1.23
[0.48, 1.78]
x
Oregon Health
1.16
[1.08, 1.25]
1.46
[1.19, 1.83]
1.26
[1.04, 1.57]
x
Medigap tax
0.40
[0.22, 1.54]
1.00
2.53
[0.64, 4.44]
x
Health child
∞
[24.82, ∞]
6.10
[3.05, 13.17]
−0.78
[−2.52, 0.17]
MC child 83+
∞
[0.26, ∞]
0.86
[0.66, 1.44]
−0.20
[−0.47, 1.82]
x
MC pregnant & infants
∞
[∞, ∞]
13.65
[5.92, 40.80]
−2.02
[−7.85, −0.27]
x
MC child (state exp)
∞
[−0.37, ∞]
8.13
[−0.24, 14.00]
−1.08
[−2.25, 0.57]
x
MC intro
10.24
[0.93, ∞]
1.78
0.17
[−1.60, 1.93]
x
Supp. Sec. Inc.
0.75
[0.64, 0.85]
1.00
[1.00, 1.00]
1.33
[1.17, 1.56]
SSI review
0.76
[0.56, 1.00]
1.00
1.32
[1.00, 1.78]
x
SSI judge
0.74
[0.72, 0.77]
1.00
1.34
[1.30, 1.39]
x
Unemp. ins.
0.61
[0.53, 0.74]
1.20
[1.15, 1.24]
1.95
[1.63, 2.26]
UI ben (state max)
0.68
[0.48, 1.13]
1.17
[1.11, 1.22]
1.71
[0.99, 2.41]
x
UI ben (DD)
0.43
[0.28, 0.78]
1.17
[1.11, 1.22]
2.74
[1.51, 4.17]
x
UI ben (DD w UR)
0.48
[0.30, 1.69]
1.17
[1.11, 1.22]
2.43
[0.79, 3.90]
x
UI ben (GA)
1.03
[0.97, 1.09]
1.17
[1.11, 1.22]
1.14
[1.09, 1.18]
x
UI ben (MO Exp.)
0.74
[0.67, 0.81]
1.17
[1.11, 1.22]
1.59
[1.46, 1.73]
x
UI ben (MO Rec.)
0.44
[0.39, 0.50]
1.17
[1.11, 1.22]
2.68
[2.36, 3.01]
x
UI ben (NY)
0.89
[0.82, 0.97]
1.17
[1.11, 1.22]
1.31
[1.21, 1.41]
x
UI ben (RK)
0.84
[0.76, 0.92]
1.17
[1.11, 1.22]
1.40
[1.28, 1.52]
x
UI dur (DD)
0.45
[0.25, 2.12]
1.30
[1.24, 1.36]
2.89
[0.61, 5.19]
x
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1243
TABLE II (CONTINUED)
Program
MVPF
MVPF CI
WTP
WTP CI
Cost
Cost CI
Baseline
UI dur (MO)
0.83
[0.76, 0.90]
1.30
[1.24, 1.36]
1.57
[1.46, 1.69]
x
Housing vouchers
0.77
[0.74, 0.81]
0.91
[0.91, 0.91]
1.19
[1.13, 1.24]
HCV RCT to welfare
0.91
[0.86, 0.96]
1.00
1.10
[1.04, 1.17]
x
HCV Chicago lottery
0.65
[0.61, 0.70]
0.83
1.27
[1.18, 1.37]
x
Jobs+
1.42
[0.45, 2.83]∗
1.14
[0.41, 1.91]∗
0.81
[0.67, 0.93]∗
MTO
MTO
∞
[−2.80, ∞]
18.40
[−15.46, 51.85]
−2.44
[−11.35, 6.81]
x
Nutrition
WIC
1.38
[1.10, 1.66]
1.28
[1.08, 1.47]
0.93
[0.88, 0.98]
SNAP assist
0.92
[0.91, 0.96]∗
0.92
[0.91, 0.96]∗
1.00
x
SNAP info
0.89
[0.89, 0.89]∗
0.89
[0.89, 0.89]∗
1.00
x
SNAP intro
1.04
[−0.97, ∞]
1.09
[−2.45, 4.55]
1.05
[−0.38, 2.51]
x
Cash transfers
0.74
[0.36, 1.47]
0.86
[0.50, 1.37]
1.16
[0.89, 1.34]
EITC 1986
1.20
[1.05, 1.38]
1.00
0.84
[0.73, 0.95]
x
EITC 1993
1.12
[0.82, 1.21]
1.00
0.89
[0.67, 1.06]
x
AFDC generosity
0.91
[0.83, 1.00]
1.04
[0.96, 1.11]
1.14
[1.10, 1.18]
x
AFDC term limits
0.81
[0.73, 0.90]
1.00
1.23
[1.11, 1.38]
x
Alaska UBI
0.92
[0.89, 0.96]
1.00
1.09
[1.05, 1.12]
x
Paycheck+
1.00
[0.87, 1.19]
1.00
1.00
[0.85, 1.15]
x
Neg. inc. tax
−0.01
[−0.82, 9.83]
−0.02
[−2.50, 3.53]
1.96
[0.18, 3.20]
x
Top taxes
3.03
[1.35, ∞]
1.00
[1.00, 1.00]
0.33
[−0.09, 0.74]
Top tax 2013
1.16
[0.87, 1.92]
1.00
0.86
[0.54, 1.16]
x
Top tax 1993
1.85
[1.19, 4.07]
1.00
0.54
[0.25, 0.84]
x
Top tax 1986
44.27
[2.37, ∞]
1.00
0.02
[−0.37, 0.42]
x
Top tax 2001
1.37
[0.92, 2.86]
1.00
0.73
[0.36, 1.09]
x
Top tax 1981
∞
[0.94, ∞]
1.00
−0.51
[−2.13, 1.06]
x
Notes. This table presents our baseline estimates for each program in our extended sample, along with the category averages reported in the bold header rows in each category.
We exclude the welfare-to-work policies discussed in Section VI.C. For each policy, we report its MVPF, cost per dollar of programmatic spending, and willingness to pay per
dollar of programmatic spending. We also report bootstrapped 95% confidence intervals with adjustments discussed in Online Appendix H. The final column indicates whether the
program is included in the baseline estimates (and thus included in the category averages). Confidence intervals are marked with an asterisk in cases where we infer p-values
using reported interval ranges. Programs in which the confidence interval is either inferred from p-values or missing are excluded from category averages.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1244
THE QUARTERLY JOURNAL OF ECONOMICS
III.E. Job Corps
Next we construct the MVPF for an RCT of Job Corps,
one of the largest vocational education programs in the United
States. This example illustrates how not all attempts to increase
children’s human capital and earnings have high MVPFs.
Established in 1964, Job Corps is administered by the U.S.
Department of Labor and provides job training and other services
to at-risk youth between the ages of 16 and 24 via a network
of centers run by local public and private agencies (Schochet,
Burghardt, and McConnell 2008). Between 1994 and 1996, the
National Job Corps study randomized 80,000 eligible applicants
into the program. We form an MVPF for this RCT using the
recent work of Schochet (2018), who links the original RCT to tax
data; we supplement this analysis with the earlier cost-benefit
analysis of Schochet et al. (2006).
1. Cost. Schochet et al. (2006) estimates that the up-front
programmatic cost per recipient is $16,158. Schochet (2018) then
estimates the earnings impact of the program over the course of
20 years and finds minimal effects. In particular, they find that
the program increases the present discounted value of participant
earnings by $121 using a 3% discount rate. We estimate that this
corresponds to an increase in tax and transfer revenue of $52.45
To these, we add the value of the products produced by the Job
Corps participants, which Schochet, Burghardt, and McConnell
(2008) estimates to be 220.Summing,thisimpliesanetcostoftheprogramover20yearsof15,886. Given the small effects
on earnings, we use this 20-year observed period as our baseline
estimate. In Online Appendix C, we show that if one extrapolates
the MVPF would be higher. We discuss these forecasts and their implied MVPFs
in Online Appendix F. To ensure our conclusions are not biased by including
policies for adults that do not have long-run follow-ups, in Section IV.C we conduct
robustness of all our analysis to policies where long-run follow-ups have been
measured.
45. As discussed in Online Appendix C, we form this estimate by summing
the observed increase in tax revenue for years 6–20 in administrative data from
Schochet (2018) combined with an application of the CBO tax rate to the earnings
effects for the first five years. We note that a fiscal externality of 52inthiscasecorrespondstoahighimplicitmarginaltaxrate.Thisisdrivenbyalowtaxrateoninitialearningsdeclinesandacomparativelyhighertaxrateonsubsequentlysmallearningsgains.Downloadedfromhttps://academic.oup.com/qje/article/135/3/1209/5781614bygueston29September2024UNIFIEDWELFAREANALYSISOFGOVERNMENTPOLICIES1245theseearningseffectstoage65,thenetcostoftheprogramwouldfallto15,832 due to a small subsequent earnings gain.
2. WTP.
Following our approach for other policies that have
the potential to increase human capital, our baseline measure of
willingness to pay consists of the impact of the policy on after-tax
income.46 This is given by the 69increaseinafter−taxearnedincomeplusthe2,314 component of the programmatic cost that
is a transfer to participants to pay for food and clothing while
participating in the program. Summing, this yields a WTP of
2,383.Dividingbythegovernmentcostof15,886 yields an
MVPF of 0.15. If one extrapolates the earnings affects to age 65,
the resulting MVPF is 0.18.47
III.F. Top Marginal Tax Rates
Finally we turn to the MVPF of top marginal tax rate
changes. This example illustrates how we can utilize estimates
from existing literature that attempts to provide empirical
guidance on optimal government policy (e.g., optimal top tax
rates, optimal unemployment insurance benefits). Whereas those
literatures often consider the policies in isolation (e.g., optimal UI
policy), we can translate the estimates into their implied MVPF
to facilitate comparisons across policy domains.
46. A pure revealed-preference approach in this context could rely on the as-
sumption that job training is accessible in the private market at its programmatic
cost. One could then set willingness to pay equal to (or perhaps below) the up-front
cost of program enrollment. In contrast, setting willingness to pay equal to after-
tax earnings does not require the assumption that potential Job Corps enrollees
have perfect information about the returns to job training at the time of initial
enrollment. However, it does require that after-tax income is sufficient to capture
willingness to pay. This means we do not incorporate any welfare costs from opti-
mization errors in consumption decisions that stem from program participation.
47. Our analysis here focuses on the MVPF of the entire treatment group.
However, it is worth noting that Schochet (2018) finds larger effects for the sub-
sample of age 20–24 participants, including a 2.4 percentage point reduction in
disability insurance receipt and a roughly $500 a year increase in earnings. To see
how this could lead to a different MVPF, we can first take a back-of-the-envelope
calculation of a PDV of lifetime disability insurance receipt of roughly $200k con-
sistent with Von Wachter, Song, and Manchester (2011). This implies a cost saving
of 4,800.Second,wenotethatthe500 a year impact on earnings corresponds
to a PDV increase in earnings of $12.8k. Applying an approximate 20% tax and
transfer rate implies an increase in WTP by $10.2k and an increase in tax revenue
of 2.6k.Thisimpliesanetcostofroughly8,600, which implies an MVPF of 1.18.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1246
THE QUARTERLY JOURNAL OF ECONOMICS
There is a large theoretical and empirical literature dis-
cussing the optimal top marginal income tax rate, summarized
in Saez, Slemrod, and Giertz (2012). This literature notes that
a tax cut providing 1inadditionalafter−taxincomeisvaluedat1 by mechanical beneficiaries. In other words, the tax cut is
valued at cost by those who would receive it in the absence of any
behavioral response to the change in the tax code. As a result,
measuring WTP is straightforward. The cost to the government of
the tax policy is more difficult. The cost of a tax cut that provides
$1 of benefits in the absence of a behavioral response is given by
1 + FE, where FE is the impact of the behavioral response to the
tax cut on government revenue.
For top marginal tax rate reductions, Saez, Slemrod, and
Giertz (2012) and Diamond and Saez (2011) show that this FE
can be expressed as −
τ
1 −τ αϵETI, where α is the Pareto parameter
of the income distribution48 and ϵETI = 1 −τ
E[y]
dE[y]
d(1 −τ) is the elasticity
of taxable income for top earners with respect to the top marginal
“keep” rate of 1 −τ.49
The elasticity ϵETI has been estimated using various tax
reforms including the 1981 and 1986 tax decreases and 1993
increases in the top marginal income tax rate. We compute
the MVPF of the historical tax policy changes that allowed
researchers to identify ϵETI. The MVPF for each tax reform is the
ratio of WTP to cost,
1
1+FE:
(7)
MVPF =
1
1 −
τ
1 −τ αϵETI .
We translate estimates of ϵETI estimated from five major tax
reforms in 1981, 1986, 1993, 2001, and 2013, which are outlined
in Online Appendix F.
To take one example, consider the 1981 tax cut that reduced
the top marginal income tax rate from 70% to 50%. Saez (2003)
finds an estimate of ϵ = 0.311. We estimate α = 2.299 from Atkin-
son, Piketty, and Saez (2011). We plug these into equation (7).
We use marginal tax rates of τ = 75% and τ = 55% before and
after the reform, which include a 5% state tax adjustment.
48. Mathematically, α =
E[yi|yi⩾¯
y]
E[yi−¯
y|yi⩾¯
y] where ¯
y is the threshold over which the
top marginal income tax rate applies.
49. Online Appendix F provides a derivation.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1247
Combining, and averaging FE obtained using the prereform
and postreform tax rates, we obtain FE =
τ
1 −τ αϵETI = 1.51. This
means that the 70% marginal tax rate appears to have been on
the “wrong side of the Laffer curve,” so reducing tax rates may
have increased revenue. In other words, the MVPF is infinite and
the tax cut “pays for itself.” However, it is important to note the
statistical uncertainty in this estimate: we cannot reject an MVPF
of 1 or ∞.
In contrast, for later reforms we find lower MVPFs. For
example for the 1993 tax increase from 31% to 39.6% we find
an MVPF of 1.85 (95% CI of [1.19, 4.07]). This distinction is not
because of differences in ϵ, but results from the fact that τ was
much lower in 1993 than it was in 1981.
Comparison to the “Optimal” Top Tax Rate.
To compare
our results to the literature on the “optimal” top tax rates, it is
helpful to consider the case studied in Diamond and Saez (2011)
where society is assumed to place no weight on the additional
consumption of the rich. If the social welfare weights, ηi, are
equal to 0 for top earners, then the optimal tax is set to maximize
government revenue: τ is chosen to be at the peak of the Laffer
curve. This occurs when taxes are set so that the net cost to the
government of providing a tax cut is 0, or FE = −1.
This approach then makes the additional assumption that
the elasticity, ϵETI, and α do not change when the tax rate changes.
Solving for the optimal tax rate then implies τ ∗=
1
1 + αϵETI .
For α = 2.299 and ϵETI = 0.311, this implies τ ∗= 58%
inclusive of state and federal tax rates. The fact that this number
is slightly below 70% is consistent with our finding of an infinite
MVPF for the 1981 reform, in which tax rates were around
70%. In contrast to this optimal tax approach, the MVPF does
not impose an assumption that society places no weight on the
consumption of the rich.
IV. MAIN RESULTS: TARGETING KIDS VERSUS ADULTS
We construct the MVPF for each policy in our sample. Here,
we present all our baseline MVPF estimates and outline our
main results. As noted, details on our MVPF constructions are
provided in Online Appendices A–F.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1248
THE QUARTERLY JOURNAL OF ECONOMICS
FIGURE III
MVPF Estimates by Age of Policy Beneficiary
This figure presents MVPF estimates for all policies in our baseline sample. For
each MVPF, we plot them as a function of the average age of the policy’s beneficia-
ries. In cases where both parents and children potentially benefit, we assign the
age of the individuals with the highest willingness to pay. Where policies within
a category have the same age, we stagger these ages around this common value
for visual clarity. On the vertical axis, we report the MVPF estimates, capping
these estimates at 5. We separately report cases where the MVPF is infinite on
the uppermost line in green (shown in color in the online version only).
IV.A. Kids
We begin our discussion with the MVPFs of policies targeting
children. Figure III presents the MVPF for each policy on the
vertical axis plotted by the average age of the beneficiaries of the
policy on the horizontal axis.50 Each dot represents the MVPF of
a particular policy, with labels provided in Table I.
50. In cases where both parents and children are beneficiaries of the policy, we
assign the age of the “economic” beneficiary based on who has the highest WTP.
For example, when analyzing the Movement to Opportunity (MTO) experiment,
which provided housing vouchers and counseling to parents with children, the age
shown is the average age of the children in the household. This is because the
policy induced higher earnings among the children, leading them to have a higher
WTP for the policy than their parents.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1249
The figure reveals our primary result: direct investments in
children have historically had the highest MVPFs, often paying
for themselves. In addition to the evidence on the Medicaid expan-
sions and admission to FIU, we also find high MVPFs for other ed-
ucation and child health policies. For example, Wherry et al. (2018)
document that the discontinuous Medicaid coverage eligibility for
children born after September 30, 1983 led to reduced medical
costs and chronic conditions in adulthood. In Online Appendix D,
we calculate that the up-front costs are fully repaid in the long run
from reduced Medicaid and uncompensated care costs, leading to
an infinite MVPF. More generally, all four major health insurance
expansions to children studied in the past 50 years have MVPFs
in excess of 10, with three of them paying for themselves.51
In addition to health policies, we find large MVPFs for edu-
cation policies. The widely studied Perry Preschool program has
an MVPF of 43.61; the more expensive Abecedarian model has an
MVPF of 11.89 (neither of these estimates are statistically distin-
guishable from ∞).52 In contrast with the idea that the returns to
human capital investment diminish rapidly with age (Heckman
2006), we find there is potential for high MVPFs investments
throughout childhood. We find an infinite MVPF for increased
K–12 spending due to school finance equalization as studied
in Jackson, Persico, and Johnson (2016).53 We also find infinite
MVPFs for several college policies, such as admissions to FIU and
51. The only policy that does not have an infinite MVPF is the introduc-
tion of Medicaid. For this policy, we directly incorporate MVPF estimates from
Goodman-Bacon (2017). This working paper includes estimated impacts through
age 55; our back-of-the-envelope calculations suggest that it is likely that forecast-
ing these effects through 65 would lead the policy to pay for itself as well.
52. To harmonize these estimates with other programs, we do not include the
benefits to the government from reduced crime. This is both because these costs are
difficult to quantify and most papers do not estimate impacts on crime outcomes.
If we include a forecast of reduced government spending on the criminal justice
system and policing, our point estimates suggest that Perry Preschool paid for
itself. However, the standard errors of these estimates also significantly increase.
Including these costs for Abecedarian also increases its MVPF, but the policy does
not appear to pay for itself.
53. It is important to note that we only analyze one paper on K–12 education
spending because of limitations in existing evidence on long-term outcomes. While
there is a large literature looking at the effect of school spending on test scores, we
lack a reliable method to translate these effects into long-run impacts. Jackson,
Persico, and Johnson (2016) demonstrate the potential for high returns to K–12
education, but future work is needed to robustly establish the presence of high
returns to K–12 investment.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1250
THE QUARTERLY JOURNAL OF ECONOMICS
the provision of CalGrants to low-income students.54 A key insight
of our results is that many policies targeting children do not face
the classic budgetary trade-off. Instead, those expenditures pay
for themselves in the long run.
Before drawing too many conclusions about each data point
in Figure III, it is important to note there is sampling uncertainty
inherent in our estimates. Figure IV, Panel A plots each MVPF
along with its 95% confidence interval. In some cases, our
estimates are relatively precise. For example, both the Medicaid
expansion to pregnant women and infants and admissions to FIU
have confidence intervals that reject any finite MVPF. We can
rule out any positive net cost to the government. In many other
instances, however, the conclusions at the individual policy level
are less clear due to the sampling variation in the underlying
estimates. For example, the 1990 health care expansion to
children born after September 30, 1983 has a confidence interval
ranging from 0.26 to infinity. In other words, we cannot with 95%
confidence reject the hypothesis that the policy paid for itself, nor
can we reject the hypothesis that the policy provides much less
than $1 of benefits per dollar of government spending.
To reach more precise conclusions at a broader level, we pool
across policies using category averages. We imagine a new policy
that spends $1 of initial program cost on each policy j in category
J containing NJ policies. We then construct the MVPF of this
category-average policy as:
(8)
MVPFJ =
1
NJ

j∈J
WTP j
C j
1
NJ

j∈J

1 + FEj
C j
,
where the numerator is the average willingness to pay per dollar
of program cost and the denominator is the average net cost to
the government of the category-average policy.55
Figure IV, Panel B presents the category-average MVPFs.
On average, spending on child education, child health insurance,
54. It is important to be clear that although our estimates suggest high returns
to policies investing in older youth, the policies in our sample affect a range of
subpopulations. As a result, further work is needed to assess how the rate of
return on investment varies for a given child over the life cycle.
55. We construct this average measure, as opposed to a precision-weighted
average or other measure, because it corresponds to a feasible policy at the time
of initial implementation. It is straightforward for the government to construct a
policy that spends an equal amount on each of these programs.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1251
FIGURE IV
MVPF Estimates and Category Averages with Confidence Intervals
Panel A presents the MVPFs and 95% confidence intervals for each policy in our
baseline sample, plotted as a function of the average age of the policy’s beneficia-
ries. Panel B presents $1 spend domain averages and 95% confidence intervals
across categories of programs, plotted as a function of the average age of each pol-
icy’s beneficiaries within a category. Individual policy MVPFs are shown in smaller
dots, color-coded to align with their respective categories. In both panels, we report
the MVPF estimates on the vertical axis, capping these estimates at 5 and sepa-
rately reporting cases where the MVPF is infinite on the uppermost line in green
(shown in color in the online version only). All confidence intervals are 95% boot-
strapped confidence intervals with adjustments discussed in Online Appendix H.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1252
THE QUARTERLY JOURNAL OF ECONOMICS
FIGURE V
Net Government Costs per Dollar of Programmatic Spending
This figure presents estimates of costs normalized by initial programmatic for
each category-average group of policies in our baseline sample. We plot these
estimates as a function of the average age of each policy’s beneficiaries within
category. Bootstrapped 95% confidence intervals with adjustments discussed in
Online Appendix H are shown for the category averages. The normalized costs
of individual policies are shown in smaller dots, color-coded to align with their
respective categories (shown in color in the online version only).
and college policies have historically had high or infinite MVPFs.
One dollar of spending across each of the policies in each of these
categories has an MVPF of ∞in child education (95% CI of [17.8,
∞]), ∞in child health (95% CI of [24.8, ∞]), and ∞in college
policies (95% CI of [4.2, ∞]).
We can dig deeper into these category averages by focusing on
the net costs to the government of these policies (the denominator
in our formula in equation (8)). Figure V computes the average
net cost to the government per 1ofprogrammaticexpenditurespentevenlyacrossthepoliciesineachcategory.Thisallowsustoexploretheextenttowhichdifferenttypesofpolicieshavepaidforthemselves.Forexample,1 invested in the four major
Medicaid expansions to children has paid back an estimated
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1253
1.78.Inotherwords,thespendingactuallygenerated0.78 of
surplus to the government in the long run.56
Having established this primary result, it is important to
qualify that these patterns do not hold uniformly across policies.
There is considerable variation in MVPFs from policy to policy.
For example, we find lower MVPFs ranging from −0.23 to 1.48 for
job-training policies, such as an estimate of 0.15 for Job Corps—
a program targeted toward at-risk youth.57 We also analyze 14
examples of college policies where the MVPFs fall below 2.58
In most cases, this is because those policies represent trans-
fers to existing students, rather than expenditures that increase
attainment.59 In some cases, expenditures may even negatively
affect student attainment. For example, Cohodes and Goodman
(2014) analyze the impact of the Adams Scholarship in Mas-
sachusetts. They find that this merit aid program does not induce
more students to go to or complete college. Rather, it induces indi-
viduals to change colleges to attend in-state schools where they are
56. Analogously, Appendix Figure II presents willingness to pay per dollar of
programmatic spending. For our baseline WTP measures, we find very similar pat-
terns: much higher estimates of
1
NJ

j∈J
WTP j
C j
for child policies than for policies
targeting adults.
57. The one potential exception to this is the recent Year Up RCT, analyzed in
Fein and Hamadyk (2018), who document large increases in earnings in the two
years after initial implementation. As we discuss in Online Appendix C, if these
earnings gains persist for an additional 5 years, the MVPF would be 2.78, and if
they persist for 21 years, the MVPF would be infinite. In addition, in estimates
outside of our sampling frame, the nine-year follow-up results from the sectoral
training program Project Quest suggest an MVPF of 1.52, which increases to an
infinite MVPF if projected to age 65. This suggests a high value to future work
estimating the continued persistence of these more promising sectoral training
programs.
58. Our analysis also demonstrates the limitations of the traditional way that
research papers report the impact of college expenditures. It is very common for
papers to note the percentage point increase in enrollment associated with $1,000
in expenditures. The difficulty with that approach is that it doesn’t account for the
number of inframarginal students receiving the benefit. Providing $1,000 to 10% of
the school-age population to achieve a 3.6 percentage point increase in enrollment
may be a very efficient investment, while providing $1,000 to 80% of the school-
age population to achieve a 3.6 percentage point increase is mostly a transfer to
existing students. For this reason, there are cases where we find substantially
different MVPFs for policies that had similar percentage point enrollment effects.
59. In Section IV.C we discuss how our results on college expenditures vary
with the method of our MVPF calculation. Although we find persistently high
MVPFs when long-run earnings outcomes are observed, we find lower MVPFs
when we project earnings gains from attainment outcomes.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1254
THE QUARTERLY JOURNAL OF ECONOMICS
eligible to use the scholarship. The change in schooling actually
results in a fall in graduation rates arguably due to switching from
more selective schools with higher graduation rates. Incorporat-
ing these schooling declines, we calculate that the program has an
MVPF of 0.72. Job training or education polices like this one do not
substantially increase human capital and so they do not recoup
meaningful portions of their initial costs via higher tax revenue.
We also find lower MVPFs for transfers to disabled children,
such as an MVPF of 0.76 for expanded eligibility for Supplemental
Security Income (SSI) at age 18 analyzed in Deshpande (2016). It
is important to note that spending on these policies may increase
social welfare, even though they have lower MVPFs. Decisions
about optimal policy are determined by the welfare weights the
government places on policy beneficiaries. If the government
makes it a priority to provide support for disabled children, these
SSI expansions may be welfare enhancing.
IV.B. Adults
In contrast to policies targeting children, we generally find
lower MVPFs (e.g., 0.5–2) for policies targeting adults. For
example, in contrast to the nearly infinite MVPFs for child health
insurance expenditures, we find MVPFs ranging from 0.40 to
1.63 for the six health insurance policies in our baseline sample
targeted to adults.60 Along the same lines, we find MVPFs ranging
from 0.43 to 1.03 for unemployment insurance policies, 0.74–0.96
for disability insurance expansions, and 1.12–1.20 for earned
income tax credits. We find MVPFs of housing vouchers ranging
from 0.65 using assignment of vouchers in Chicago via lottery
(Jacob and Ludwig 2012) to 0.91 using an RCT of the provision of
housing vouchers to families on cash welfare (Mills et al. 2006).
The lower MVPFs reflect the fact that many of these expen-
ditures have been shown to reduce labor earnings through labor
market distortions. As depicted in Figure V, the average cost per
$1 of government spending on these adult policies is generally
60. Those adult health insurance estimates include expenditures such as the
subsidies in the Massachusetts health insurance exchange prior to the Affordable
Care Act. In that case, Finkelstein, Hendren, and Shepard (2019) exploit discon-
tinuities in the subsidy schedule to estimate both individuals’ willingness to pay
for insurance and the cost those individuals impose on the government. Translat-
ing these estimates into an MVPF suggests values ranging from 0.800 to 1.09 for
different subsidy eligibility levels.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1255
slightly above $1. This result contrasts with our findings on expen-
ditures directed toward children, for whom labor market earnings
tended to rise, leading to a decline in net costs. There are a limited
number of cases, such as the Job Training Partnership Act and Na-
tional Supported Work Experiment, where investment in adults
sought to increase earnings by increasing human capital. Those
policies, however, did not produce persistent earnings gains, so
they still yield relatively low MVPFs.61 The MVPFs of job-training
programs for adults over the age of 23 range from 0.44 to 1.48.62
As with our main results for policies targeting children, these
findings represent general patterns. They do not hold uniformly
across all policies targeting adults. In particular, there are two
types of adult policies that tend to result in higher MVPFs:
reductions of high marginal tax rates for top incomes and policies
with indirect spillovers onto children.
1. Top Tax Rates.
We find high MVPF point estimates for his-
torical reductions in the top marginal tax rate when the initial tax
rate lay at 50% or higher. In the case of the 1981 reform, the tax bill
reduced the top federal marginal tax rate on income from 70% to
50%. Using estimates of the elasticity of taxable income from Saez
(2003), we calculate that the MVPF is ∞(95% CI of [0.94, ∞]).
This implies that marginal tax rates were beyond the top of the
Laffer curve prior to 1981. Our confidence interval, however, sug-
gests this estimate contains considerable sampling uncertainty.63
61. For this reason, we calculate the MVPFs of job-training programs based
on the number of years of earnings effects observed, rather than projecting the
effects out to age 65. In Online Appendix C we discuss the sensitivity of our results
to that assumption.
62. The presence of high MVPFs for spending on children and low MVPFs for
spending on adults does not necessarily indicate that families are failing to opti-
mize their investment decisions. Even if families are fully informed of available
investment decisions, a simple model of parental investment could produce these
outcomes if parents are credit constrained. A higher MVPF for investment in chil-
dren could occur if low-income parents expect intergenerational regression to the
mean such that their children will earn more than them. That would produce lower
marginal utilities of income for those children, and therefore increase the return
on spending. In addition, this logic also suggests that when parents are given cash
transfers, they would rationally not spend all of it on their children despite high
returns—this is because their marginal utility of their own consumption is also
high.
63. As we discuss in Online Appendix F, these estimates appear to have consid-
erable uncertainty not just from sampling uncertainty but also model uncertainty:
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1256
THE QUARTERLY JOURNAL OF ECONOMICS
Along the same lines, we analyzed the 1986 reform and found an
MVPF of 44.27, with a confidence interval ranging from 2.37 to ∞.
Although this may be considered by some to be suggestive
evidence for Laffer effects in tax policy, it is important to ap-
proach that conclusion with considerable caution. In the case
of the 1981 reform our confidence intervals suggest we cannot
rule out an MVPF close to 1. In other words, we cannot rule
out the conclusion that the policy produced no positive fiscal
externality. Moreover, estimates of the impacts of recent reforms
have produced substantially smaller MVPFs (e.g., 1.16 for the
2013 top tax rate increase). Compared with these findings on
taxes, our results suggest stronger evidence for the presence of
Laffer effects when investing in young children.
2. Spillovers onto Children.
We also find that spending on
adults may have high MVPFs if those policies have spillover ef-
fects on children. For example, Chetty, Hendren, and Katz (2016)
study the long-run impact of the MTO experiment, which gave
families residing in public housing projects a voucher and coun-
seling to assist them in moving to lower-poverty neighborhoods.64
Chetty, Hendren, and Katz (2016) document that the program
significantly increased later-life earnings for young children, but
they find null or even slightly negative effects on earnings for
children who were teenagers at the time their parents obtained
the vouchers. Combining these effects across all subgroups
suggests the effects on the young children outweigh the adverse
effects on the older children, leading to an infinite MVPF.65 This
using different taxable income estimates from existing literature studying these
reforms can generate wide variation in the MVPFs of these tax reforms, preventing
precise conclusions about their MVPFs.
64. Because the program was targeted to families already in public housing
and because the cost of public housing is similar to the cost of a voucher, the
primary marginal cost of the program was the cost of the counseling (roughly
$3,783 per family).
65. Not all policies providing benefits to parents generate such large spillover
effects onto children. For example, Price and Song (2018) find that the Nega-
tive Income Tax experiment led to a reduction in children’s earnings in adult-
hood, which partially explains its low MVPF of −0.01. In other cases, such as
the provision of housing vouchers in Chicago, and the provision of housing vouch-
ers to families on AFDC and the expansion of AFDC benefits, there is sugges-
tive evidence that positive spillovers on children are small. In those cases, re-
searchers have documented that the policies have limited effects on outcomes such
as test scores, college attendance, and birthweight. Appendix Figure III, Panel A
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1257
high MVPF is driven solely by child outcomes, as the policy has
no significant effect on economic outcomes for adult beneficiaries.
One policy with a substantial degree of uncertainty about
potential
spillovers
onto
children
is
the
EITC.
Appendix
Figure III, Panel C shows how our MVPF estimates would change
if one attempted to impute effects on children using different
estimates from previous literature. In particular, we take the
MVPF for the 1993 OBRA tax reform and supplement that
estimate with spillover effects of the EITC estimated in other
contexts. Projecting earnings effects based on child test scores
produces MVPFs that range from 3.48 to ∞, while incorporating
effects on college attendance produces MVPFs from 0.84 to 1.12.66
Incorporating the work of Bastian and Michelmore (2018) on
long-term earnings would result in an infinite MVPF, suggesting
that the policy pays for itself.67
This uncertainty highlights the importance of understanding
the potential spillovers onto children. It also reinforces our
conclusion that policies raising children’s human capital often
have the highest MVPFs. We return to this issue in Section VI.A,
where we use the MVPF framework to quantify the value to
governments of more precise estimates for potential long-run
effects of policies on children.
IV.C. Robustness
Creating these MVPF estimates inevitably requires that we
make a number of judgment calls regarding the set of causal
effects included and the methodology used to translate those
effects into an MVPF. Here, we provide a short summary of the
robustness of our main conclusions to those assumptions.68
presents results for policies in our baseline sample where child effects are observed.
Panels B and C show how the MVPFs change when effects on children are incor-
porated or removed from the MVPF calculation.
66. The college effects are restricted to a small subset of recipients, so it is
unsurprising that the MVPFs remain small.
67. We exclude these results from the baseline estimates because Bastian and
Michelmore (2018) do not estimate the effect of a particular EITC expansion, but
rather pool across many state and federal policy changes. In Online Appendix F,
we note the impact of incorporating their estimates. The fact that these impacts
matter is consistent with our broader conclusions that potential spillovers onto
children can generate high MVPFs for adult-targeted policies.
68. Online Appendix J details an extensive set of robustness analyses.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1258
THE QUARTERLY JOURNAL OF ECONOMICS
Constructing the MVPF for policies with dynamic effects
requires the choice of a discount rate. Although our baseline
approach assumes 3%, Appendix Figure IV shows that higher dis-
count rates do not substantively change our conclusions. Discount
rates of 7% or 10% produce slightly lower MVPFs for child-
targeted policies (more so for young children), but we still find
those policies have higher MVPFs than policies targeting adults.
Our baseline approach uses the cross-sectional life cycle
earnings profile to forecast lifetime effects from observed earnings
changes. Our results are robust to alternative methods of fore-
casting earnings, such as assuming no income growth over the life
cycle. The baseline sample also includes some policies targeting
children for which effects on income are not directly measured.
Most notably, we include college policies where researchers have
observed a measure of attainment such as initial enrollment,
college credits, or degree receipt. In those cases we forecast
income impacts using estimates from Zimmerman (2014) on the
returns to college. Online Appendix J provides a discussion of
how our estimates vary depending on the use of intermediate
outcomes to construct long-run forecasts. In particular, Appendix
Figure III shows the effects of restricting our analysis to policies
where earnings are directly observed. We continue to find high,
often infinite MVPFs for these child-targeted policies.
In many cases, our MVPFs for policies targeting adults rely
upon estimates of short-run earnings impacts. Consequently,
one might be worried that our low MVPFs for adult policies are
driven by policies for which we do not observe long-run impacts.
In order to assess this, Appendix Figure VII, Panel B restricts
the analysis to the subset of policies for which we observe at least
five years of income estimates. We continue to find higher MVPFs
for policies targeting children.69
Our baseline willingness to pay approach often relies on
measures of a policy’s impact on after-tax income. Appendix
Figure VI, Panel A reports our MVPFs using our conservative
69. Related to this, the pattern of higher MVPFs for children could be driven by
longer payoff periods for children relative to adults, as children have their entire
lives to experience higher earnings. However, the length of the payoff period is
not what is driving our results—even restricting child benefits to accrue only up
to age 45 or 55, we find similar high returns for child-targeted policies. Rather,
the patterns are generally driven by a higher positive impact on per-year future
earnings for policies targeting children.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1259
measure of willingness to pay. Although the willingness to pay
measures are much lower, we continue to find high MVPFs for
policies targeting children, generally exceeding 5 on average. This
is to be expected as many policies analyzed have very low net costs,
leading to large MVPFs even when willingness to pay is small.
One might also be concerned that the causal effects incor-
porated in our MVPFs may vary in quality due to variation
in the underlying techniques used to produce those estimates.
Appendix Figure VII, Panel C shows our results remain the same
when restricting our sample to policies evaluated via randomized
controlled trial, lottery, or a regression discontinuity design. The
results are also robust to restricting our sample to peer-reviewed
publications. In all these robustness analyses, direct childhood
investments continue to have the highest MVPFs.
Finally, one might worry that MVPFs for child policies were
high in previous decades but have declined over time—perhaps
as the government takes advantage of high-return investments.
Appendix Figure IX assesses this by plotting the child- and adult-
average MVPFs separately by decade. We find no evidence for
that pattern of decline. Instead, we find high MVPFs for policies
targeting children throughout the past 50 years.70 The robustness
of high MVPFs for direct investments in children over time may
suggest the presence of fundamental political constraints to
enacting policies in which the benefits have a long time horizon.71
IV.D. Publication Bias
All of the robustness analyses above take the estimates from
existing literature as given. However, one might be concerned that
the research and publication process suffers from the problem
70. The one exception to this pattern is the low average MVPF among child
policies implemented in the 1970s. The child policies in that decade primarily
consisted of job-training programs that did not have significant effects on children’s
earnings.
71. There are a range of forms that these political constraints might take.
For example, it could be that governments (and politicians) apply a much higher
discount rate, requiring projects to pay off over short horizons. Alternatively, un-
derinvestment might occur because these policies require spending by state and
local governments, but much of the benefits accrue to the federal tax system.
Hence, local incentives may not be sufficient to make efficient investments. It
may also be that the high MVPF policies are undersupported because low-income
children have little political power. We leave a formal analysis of these potential
mechanisms for future work.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1260
THE QUARTERLY JOURNAL OF ECONOMICS
TABLE III
PUBLICATION BIAS ESTIMATION
Children estimates
Adult estimates
(1)
(2)
(3)
(4)
(5)
(6)
Z > 1.64
3.72
–
2.52
–
(2.46)
(1.32)
Z < −1.64
1.15
–
7.90
–
(0.44)
(1.48)
Z ∈[1.64, 1.96]
3.65
1.36
(3.46)
(1.14)
Z ∈[−1.96, −1.64]
1.02
4.19
(0.57)
(0.81)
Z > 1.96
–
3.09
3.78
–
3.27
3.59
(1.09)
(2.17)
(1.50)
(1.21)
Z < −1.96
–
1.21
1.24
–
10.39
11.52
(0.50)
(0.62)
(2.53)
(2.43)
N
237
237
237
150
150
150
Notes. The numbers shown are the estimated likelihood ratio of publication relative to an insignificant
result. Standard errors are in parentheses.
of publication bias, where studies are published only if they find
clear positive (or negative) effects. In particular, one might worry
that research on children is more likely to be published if it finds
statistically significant positive effects on children in adulthood.
Conversely, one could imagine that research on adults is more
likely to be published if it finds statistically significant evidence
of distortionary or negative effects on adult outcomes.
To address this, we implement the approach developed in
Andrews and Kasy (2019).72 They provide a method to test
and correct for the effect of publication bias on the observed
set of estimates. Online Appendix K discusses the details of
our implementation of their approach. Table III documents the
evidence of publication bias in our estimates.
The results suggest the presence of a moderate degree of
publication bias. In the baseline sample, we find studies of child
outcomes are 3.7 times more likely to be published if they find
positive effects on children with p < .10 relative to a finding of
no statistically significant effect. In contrast, we find that studies
72. We thank Isaiah Andrews and Max Kasy for their invaluable guidance in
implementing these procedures.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1261
on adult policies are 11 times more likely to be published if they
find significant distortionary effects on outcomes.
Despite evidence of publication bias in our samples, Appendix
Figure VIII, Panel A shows that correcting for the observed degree
of publication bias in this manner does not affect our conclusion
of higher MVPFs for policies targeting children. Although we find
a slight decrease in the MVPFs for child education policies, such
as preschool programs, the general patterns are quite similar
to our baseline results. Moreover, Appendix Figure VIII, Panel
B shows that even if we assumed that statistically significant
estimates of positive effects on children are 35 times73 more likely
to be published, our primary conclusions still hold.
V. MAPPING THE MVPFS TO THEORY
The MVPF provides an empirical method for evaluating the
effectiveness of different policies for improving social welfare.
Having established the key patterns of the data, it is natural
to place our empirical results into the context of theoretical
literature on optimal government policy. In this section, we
outline how our results speak to that theory.
1. Optimal Taxation.
To begin, the MVPF measures the
price of redistributing to different policy beneficiaries. In this
sense, the approach is related to a large body of theoretical and
empirical optimal tax literature in the spirit of Mirrlees (1971)
and Saez (2001).
As previously explained using the Okun’s bucket logic, the
ratio of MVPFs across two different tax changes measures the
price of moving money between the respective beneficiaries. In
general, optimal tax theory suggests that a progressive planner
should be willing to incur efficiency losses to move resources from
the affluent toward the lower regions of the income distribution.
The MVPF provides an empirical means of testing that basic
prediction: the MVPF of tax changes should increase with the
income of the beneficiaries.
Figure VI, Panel A explores the relationship between the
MVPF of each tax policy change we analyze and the income levels
of the associated beneficiaries. Consistent with this prediction,
73. A publication bias of 35x is the degree of publication bias documented in
Andrews and Kasy (2019) for small-sample experimental economics studies.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1262
THE QUARTERLY JOURNAL OF ECONOMICS
FIGURE VI
MVPF by Income of Beneficiaries
Panel A shows MVPFs for tax and transfer policies in our baseline sample
against the income of their economic beneficiaries. Panel B adds in-kind transfers
to parents and direct expenditures on children (child education, health, job train-
ing, and college policies). See Figure III for an explanation of the color scheme. The
income measures should be considered approximations, as not all papers report
consistent measures of incomes of their samples. We include all papers for which
we are able to obtain a measure of income of the beneficiaries, and we attempt
to normalize these measures to correspond to a notion of individual income per
adult in the household at age 30. All confidence intervals are 95% bootstrapped
confidence intervals with adjustments discussed in Online Appendix H.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1263
we observe an upward slope. For example, the 1993 tax reform
(OBRA93) simultaneously raised top marginal tax rates and
expanded the EITC. The MVPF of the increased top tax rates led
to an MVPF of 1.85 (95% CI of [1.19, 4.07]), and the expansion
of the EITC led to an MVPF of 1.12 (95% CI of [0.82, 1.21]).
This suggests the tax schedule created under the 1993 reform
is optimal if one is indifferent to providing 1.85totopearnersversus1.12 to those on the EITC. To the extent one’s social
preferences strictly prefer 1.12tolowearners(orstrictlyprefer1.85 to top earners), our results suggest that more progressive
(regressive) taxation than the 1993 schedule would be optimal.74
Although our MVPF estimates for tax changes are loosely
consistent with the preferences of a progressive planner, this is
no longer the case when we consider policies targeting children.
As shown in Figure VI, Panel B, there is no clear relationship
in our sample between MVPFs and the incomes of beneficiaries
when including direct investments in children. This means that,
historically, investments in the next generation have been more
efficient than transfers within generations.75
2. In-Kind versus Cash Transfers.
The low MVPFs for
policies targeting very low-income households raises the question
of whether other methods of redistribution—perhaps through
in-kind transfers—can be more effective than cash.76 Figure VII
74. Our estimate for the MVPF of the 1993 EITC is based on evidence from
Meyer and Rosenbaum (2001) on the fiscal externality associated with the labor
supply responses of single women. It is worth noting, however, that there is con-
siderable debate over the fiscal externalities associated with the EITC. On the
one hand, several recent papers have argued that reductions in transfers have
offset a substantial portion of the cost of historical EITC expansions (Hoynes and
Patel 2018; Bastian and Jones 2019). These large fiscal externalities can produce
infinite MVPFs (Bastian and Jones 2019). On the other hand, recent debates have
argued that the effects are overstated in the current literature because the impact
of the EITC expansions cannot be disentangled from the effects of contemporane-
ous welfare reforms (Kleven 2019). These conflicting estimates suggest there is a
high value to future work that reconciles these findings.
75. We develop this argument formally in Online Appendix L, where we relate
this logic to the nonexistence of a social welfare function that can rationalize our
results of high MVPFs for low-income children but low MVPFs for low-income
adults.
76. There is a large theoretical debate on this question, which largely centers
around the applicability of the Atkinson-Stiglitz theorem (Atkinson and Stiglitz
1976; Hylland and Zeckhauser 1981). When utility satisfies a “weak separability”
assumption, one would expect that the MVPF for an in-kind transfer would fall
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1264
THE QUARTERLY JOURNAL OF ECONOMICS
FIGURE VII
MVPF by Income of Beneficiaries: Cash versus In-Kind Transfers
This Figure presents MVPFs as a function of the average income of beneficiaries
for tax and transfer policies (shown in Figure IX, Panel A) combined with our
estimates for in-kind transfer policies. The income measures should be considered
approximations, as not all papers report consistent measures of incomes of their
samples. We include all papers for which we are able to obtain a measure of
income of the beneficiaries, and we attempt to normalize these measures to
correspond to a notion of individual income per adult in the household at age
30. All confidence intervals are 95% bootstrapped confidence intervals with
adjustments discussed in Online Appendix H.
adds the MVPF estimates for housing and food subsidies to the
estimates provided in Figure VI, Panel A for cash transfers and
tax credits. Broadly, we find a pattern consistent with our general
result: in-kind transfers are most effective when they induce
spillover effects onto children.
For example, the housing vouchers in Chicago (Jacob and
Ludwig 2012; Jacob, Ludwig, and Kapustin 2014) and the
provision of Welfare to Work housing vouchers (Mills et al. 2006)
find minimal spillover effects on children. This means that the
below the MVPF of a cash transfer or tax credit targeted to beneficiaries at the
same income level.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1265
distortionary impact on adults’ earnings leads them to have
MVPFs below that of distributionally equivalent tax cuts. In
contrast, the MTO experiment explained previously increased
earnings of young children by a sufficient amount to pay for the
cost of the in-kind policy (the policy had an infinite MVPF with
a 95% confidence interval of [−2.80, ∞]). Similarly, the spillover
effects onto children for the introduction of food stamps policy
leads to an MVPF of 1.04. Both point estimates suggest these
in-kind transfers are as efficient or more efficient than cash
transfers as a result of the spillovers onto children.77
3. Tagging.
There is a large literature in optimal policy
design focused on improving efficiency by targeting the right
subset of individuals. In general, this work focuses on the use of
“tags”—characteristics of program eligibility that are generally
not manipulable (Akerlof 1978).78 With that in mind, previous
literature has identified recipient age as a potentially valuable
tag for optimal government policy. Consistent with that work, we
observe that the MVPFs of certain policies differ substantially
based on the age of the recipients.
For example, our analysis of the MTO experiment finds an
infinite MVPF with a confidence interval of [−2.80, ∞]. That said,
the result masks substantial heterogeneity in the program’s ef-
fects. In families with children younger than 12, the MVPF is in-
finite with a confidence interval contained at infinity. In families
with children older than 12, the MVPF is negative, as their point
estimates imply a reduction in earnings. Along the same lines, our
analysis of the introduction of food stamps produces an MVPF of
1.04. This MVPF is partly buoyed by large positive effects on chil-
dren ages 0–5 (Bailey et al. 2019). If we excluded any effects on
children, the MVPF would fall from 1.04 to 0.54. By contrast, if
we restricted our analysis to families with young children and
assumed that causal effects of food stamp introduction remained
the same for that targeted policy, we would find an MVPF of 2.28.
77. In relation to the Atkinson-Stiglitz theorem, the violation of the weak
separability assumption for these policies comes not from a short-term change in
earnings but from the long-run indirect impact on children.
78. If the tag were manipulable, then individuals not intended as beneficiaries
of the policy could distort their behavior to obtain the benefit. To the first order,
they would not value the transfer by the envelope theorem, consequently lowering
the MVPF of the policy.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1266
THE QUARTERLY JOURNAL OF ECONOMICS
Despite this substantial variation in MVPFs by the age
of policy recipients, we do not report subgroup-specific MVPFs
in our main tables. This is a deliberate choice to restrict our
analysis to policy changes defined by explicit identification
conditions established in existing work. Reporting MVPFs for
subgroups requires the additional assumption that the observed
behavioral response to the policy among the relevant subgroup
is not impacted by the provision of the policy to other subgroups.
Although this may be plausible in certain cases, we have no
disciplined way of adjudicating its plausibility across all possible
permutations of subgroup analysis. Instead, we highlight the
potential for age-specific tagging but refrain from more definitive
statements regarding subgroup-specific welfare impacts.
VI. LESSONS FOR FUTURE WORK
In this section, we discuss three implications for future
economic research. First, we show how the MVPF framework
facilitates a straightforward method to quantify the value of
future work that reduces the statistical uncertainty in our esti-
mates. Second, we show the value-added provided by measuring
the MVPF relative to what is provided by a more traditional
cost-benefit analysis. Third, we discuss how the intuitions of the
MVPF framework might influence future empirical designs. The
key is to design experiments in a way that facilitates measuring
willingness to pay. In particular, we discuss how 27 different
welfare reform programs in the 1980s–90s randomized upwards
of 100,000 participants into RCTs, but the nature of the research
designs makes it infeasible to conduct reliable welfare analysis.
VI.A. Value of Information in Evidence-Based Policy Making
Our MVPF estimates measure the welfare impact of a
range of government policies. Although it is our hope that these
estimates can be useful for a policy maker seeking to conduct
“evidence-based” policy, it is quite clear from Figure IV, Panel A
that many of our individual policy estimates contain considerable
sampling uncertainty. Here, we show how one can use the MVPF
framework to understand the value of future research that
reduces the uncertainty in our estimates. The MVPF framework
provides a measure of the value of information because it
is a price: it measures the price faced by the government to
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1267
redistribute across beneficiaries of different types of policies.
A welfare-maximizing government should be willing to pay to
reduce the uncertainty in these prices, just as consumers would
be willing to pay to learn the true value of the products they buy.
There are many ways one could conceptualize reducing
the various sources of modeling and sampling uncertainty in
our estimates. In this section, we develop a simple approach
to measure the value of reducing sampling uncertainty. We
defer an exhaustive treatment to future work. We use this
example to illustrate the value of future research that increases
estimate precision, perhaps through improved access to larger
administrative longitudinal data sets.79
Our conceptual experiment is organized as follows: suppose
a policy maker is considering whether to raise revenue to spend
an additional $1 on policy j. The policy has a net cost to the
government of Gj and a willingness to pay of WTPj per dollar
of programmatic cost. The policy maker does not know the true
values of WTPj and Gj. Instead, we assume she only observes
the estimates,
ˆ
WTP j and ˆ
Gj, and their sampling distributions.80
We assume the policy maker has an uninformed prior about the
impact of the policy so that the estimated sampling distribution
reflects her belief about the policy’s effects.
For simplicity, we assume the policy is financed with a tax
change that targets the same beneficiaries and has an MVPF of
1. A budget-neutral policy that increases taxes to spend on policy
j has a welfare gain of
U

WTP j, Gj
	
= WTP j −Gj.
Ideally, the policy maker would wish to pursue this policy if and
only if U(WTPj,Gj) > 0 (i.e., the policy increases welfare). In
practice, the policy maker only observes estimates and sampling
distributions of these values. We assume these estimates are
unbiased but noisy estimates of the truth (e.g., E[ ˆ
Gj|Gj] = Gj).
Utility is linear in WTPj and Gj, so the policy maker will choose
the policy if and only if
ˆ
WTP j > ˆ
Gj. The expected utility of this
79. The focus here is on reducing uncertainty among the observed outcomes of
each program. Uncertainty regarding unobserved causal effects remains beyond
the scope of this exercise.
80. For simplicity, we assume programmatic costs are known and equal to
their point estimates.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1268
THE QUARTERLY JOURNAL OF ECONOMICS
strategy given the point estimates (
ˆ
WTP j, ˆ
Gj) is
EU Uninf ormed 
ˆ
WTP j, ˆ
Gj

= E

U

WTP j, Gj
	
∗1

U

ˆ
WTP j, ˆ
Gj

> 0

|W ˆ
TPj, ˆ
Gj

=

ˆ
WTP j −ˆ
Gj

∗1

ˆ
WTP j > ˆ
Gj

.
Now suppose that instead of spending $1 on the policy, the
policy maker can invest a fraction of this dollar, vj, into learning
more about the WTPj and Gj of the policy before making this
decision. We begin by considering a case where spending vj allows
the policy maker to perfectly learn WTPj and Gj before deciding
whether to invest in the policy. Once informed, the government
chooses to pursue the policy if and only if U(WTPj, Gj) > 0. Now it
can decide to pursue the policy if and only if the true WTP exceeds
the true costs. In that case, the net utility to the government is
U inf ormed 
WTP j, Gj, v j
	
=

1 −v j
	 
WTP j −Gj
	
∗1

WTP j > Gj

−v j,
where the first term is the surplus from investing the remaining
fraction 1 −vj in the policy and the second term is the cost of
paying for the information.
The value to the government of learning the true willingness
to pay and cost for policy j is the value of vinf o
j
which solves the
following equation:
E

U inf ormed 
WTP j, Gj, vinf o
j

|
ˆ
WTP j, ˆ
Gj

= EU uninf ormed 
ˆ
WTP j, ˆ
Gj

.
(9)
Here, vinf o
j
equates the government’s expected utility in the case
where it spends vinf o
j
to receive additional information and the
case where it remains uninformed. The expectation in the left side
of equation (9) is taken with respect to the distribution of the true
parameters, (WTPj, Gj), given the estimates, (
ˆ
WTP j, ˆ
Gj). Since
we assume uninformed priors, this distribution is parameterized
by the sampling distribution of the estimates. This implicitly
defines vinf o
j
as the value that makes one indifferent to remaining
uninformed versus paying for the information and making a
decision based on it.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1269
FIGURE VIII
Value of Information
This figure presents the value of information, vinfo, discussed in Section VI.A,
for each policy in our sample as a function of the average age of the policy
beneficiaries. See Figure III for an explanation of the color scheme.
1. Results.
We estimate the value of info in equation (9) both
for each individual policy and for our category averages. Figure
VIII presents the results of vinf o
j
for each policy, j, plotted relative
to the age of the policy’s beneficiaries. Broadly, we find the highest
values of future research for policies with uncertain long-run
effects on children. For example, we estimate vinf o
FS = $0.50 for
the introduction of food stamps. Moreover, we also find large
values of information for policies with potential indirect effects on
children and uncertain effects on adults. We also find large values
of information for college subsidies to parents (shown in green;
color version available online). This reflects the fact that these
policies have highly uncertain effects on college attainment, and
small increases in attainment can translate into large gains. In
contrast, we find smaller values of information for policies where
the effects have been already precisely estimated. For example,
we find the evidence-based policy maker would be willing to
pay little to remove the statistical uncertainty in the estimated
impact of disability insurance on labor earnings (e.g., we estimate
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1270
THE QUARTERLY JOURNAL OF ECONOMICS
the policy maker is willing to pay $0 to learn the precise impact
of assignment to a more lenient disability insurance judge). This
lower value of information reflects the relatively high precision
of existing estimates in those studies.
2. Administrative verus Survey Data: Long-Run Impacts of
Food Stamps.
Our estimates in Figure VIII report the value
of learning the true effect of the policy. In practice, the true
effect is never observable. That said, improved access to larger
administrative data sets can help obtain more precise effects
of government policies. For example, a policy maker can decide
whether a researcher should use a survey data set for the analysis
or obtain access to linked administrative data on the population.
To illustrate this decision, we consider the case of the intro-
duction of food stamps discussed in Section III.C. Earlier work
by Hoynes and Schanzenbach (2009) used the Panel Study of
Income Dynamics (PSID) survey data set to identify the long-run
effect of food stamps on children’s outcomes. More recently, Bailey
et al. (2019) used linked census data to estimate those effects
more precisely. Here, we imagine that a policy maker is deciding
whether to introduce food stamps based on the existing evidence.
Consider the hypothetical example that they know the PSID
estimates from Hoynes and Schanzenbach (2009),
ˆ
WTP
PSID and
ˆ
FE
PSID. Suppose that they can instead invest vCensus to learn the
estimates with the same statistical precision as those found in
Bailey et al. (2019) based on census data. Instead of learning the
true value of WTP and FE, the policy maker learns
ˆ
WTP
Census
and
ˆ
FE
Census. The policy maker will expect these estimates to
be drawn from the PSID sampling distribution but contain the
standard errors found in the census data estimates. The value of
learning the census estimates, vCensus, then solves
E
 
1 −vCensus
U

ˆ
WTP
Census, ˆ
FE
Census
× 1

ˆ
WTP
Census > 1 −
ˆ
FE
Census
−vCensus |
ˆ
WTP
PSID, ˆ
FE
PSID
= U

ˆ
WTP
PSID, ˆ
FE
PSID
1

ˆ
WTP
PSID > 1 −
ˆ
FE
PSID
(10)
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1271
The left side of equation (10) is the expected value of investing
in administrative data at a price vCensus. The right side is the
expected value of the policy if she makes her decision using the
information in the PSID.
We reconstruct the estimates of the WTP and FE for the
introduction of food stamps using the estimates from Hoynes and
Schanzenbach (2009) in place of those in Bailey et al. (2019), nor-
malizing by the mechanical program cost. This yields an infinite
point estimate for our MVPF, and we find a willingness to pay
estimate of 6.06 (95% CI of [−12.07, 23.78]) and cost of −0.19 (95%
CI of [−5.19, 4.92]). These estimates are notably less precise than
those using the results from Bailey et al. (2019) that use census
data, which generate a WTP of 1.09 (95% CI of [−2.45, 4.55]).
Plugging these estimates into equation (10) suggests the
policy maker would be willing to invest $0.24 per dollar of
investment in the food stamp program to learn the long-run
estimates from census data instead of PSID data. This exercise
illustrates that if the policy maker only knew the PSID estimates,
there would be a large value in learning additional information
before making this investment decision.
This is, of course, a stylized exercise. We are imagining a
policy maker that sees the ex post evaluation of a policy prior
to making her decision—something that is clearly not feasible.
The goal here is merely to illustrate potential value of expanding
access to administrative data sets that can generate more precise
estimates of long-run policy effects.
VI.B. Comparison to Benefit-Cost Ratios
While we focus on computing the MVPF for each policy, the
most common form of welfare analysis in previous literature is
benefit-cost analysis, as in equation (4). With that in mind, we
compare our results to the benefit-cost ratios for the same policies.
Figure IX, Panel A plots the benefit-cost ratio for a deadweight
loss of φ = 50% as in Heckman et al. (2010) as a function of the
age of the beneficiary of the policy. Our general conclusion about
the high returns to investment in children would remain true
even if one used a benefit-cost ratio instead of the MVPF. The
average benefit-cost ratio is 4.13 for child education, 5.30 for child
health, and 6.78 for college policies. In contrast, we find smaller
benefit-cost ratios for adult policies—often less than 1.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1272
THE QUARTERLY JOURNAL OF ECONOMICS
FIGURE IX
Comparison to CBA
This figure presents estimates of benefit-cost ratios for all policies evaluated
in the article and shows their relationship to the MVPF. The method for calcu-
lating these benefit-cost ratios is outlined in Section II. We assume a marginal
deadweight loss of φ = 50% for these calculations. Panel A plots the benefit-cost
ratio of each policy as a function of the age of the beneficiaries, along with cat-
egory average estimates and their confidence intervals. The capped lines show
the 95% bootstrapped confidence intervals with adjustments discussed in Online
Appendix H. Panel B plots the benefit-cost ratio of each policy as a function of the
MVPF estimate for the policy. See Figure III for an explanation of the color scheme.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1273
To directly compare the two methods of welfare analysis,
Figure IX, Panel B plots the benefit-cost ratio on the vertical
axis (again for φ = 50%) against the MVPF on the horizontal
axis. In general, we find a fairly monotonic relationship—policies
with high benefit-cost ratios also have high MVPFs. There are,
however, some notable distinctions. For example, the Medicaid
expansion to children born after September 30, 1983, has an infi-
nite MVPF but a BCR of just 1.37. Similarly, the 1981 top tax rate
reduction has an infinite MVPF but a benefit-cost ratio of 1.67. By
the standards of benefit-cost ratios, these policies may not appear
all that desirable, even though the MVPF point estimates imply
that they pay for themselves and provide a Pareto improvement.
The difference between the MVPF and benefit-cost ratio in
these cases reflects the fact that the benefit-cost ratio places all
causal effects of the program in the numerator while the MVPF
incorporates effects based on their incidence. In particular, the
numerator of the MVPF captures the effects on beneficiaries while
the denominator captures all effects on the government budget.
In measuring the welfare effects of the 1983 Medicaid expansion
and the 1981 tax cut, MVPF places all fiscal externalities in
the denominator. The results show us that these policies have
substantial benefits and limited or no net government cost. In
the benefit-cost ratio framework these reforms would have been
interpreted as high-cost policies with substantial benefits.
The second crucial distinction between the MVPF and benefit-
cost ratio is how the two approaches conceptually close the budget
constraint. While the MVPF closes the budget constraint by com-
paring MVPFs of different policies (and aggregating using Okun’s
bucket as in equation (3)), the same consistency does not exist
in the benefit-cost ratio approach. In many cases, benefit-cost
analysis includes no discussion of closing the budget constraint.
In cases where the concept is addressed, it is customary to close
the budget constraint in the same manner regardless of the policy
context. For example, BCRs in Heckman et al. (2010) and Garc´
ıa
et al. (2016) imagine that the policy was funded by an increase
in the marginal tax rate that led to a distortion in tax revenue
and a deadweight loss of φ. Consequently, the deadweight loss
parameter φ in equation (4) is not context dependent.
To see how this matters, consider the 1993 tax reform
that simultaneously raised top marginal income tax rates and
expanded the EITC. One could, in principle, use a benefit-cost
ratio to evaluate whether the EITC expansion was desirable. As
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1274
THE QUARTERLY JOURNAL OF ECONOMICS
shown in Figure IX, Panel A, the benefit-cost ratio for the 1993
EITC expansion is 0.74 after adjusting for a 50% deadweight
loss. The costs exceed the benefits and so, if the government were
applying a strict benefit-cost test, we would not expect this policy
to be implemented. This is because the hypothetical 50% cost of
raising the funds is too large to justify the expenditure.
That said, the goal of the EITC expansion was to provide
redistributive benefits to low-income workers. Its MVPF is 1.12,
near the highest among policies targeting adults. Rather than
ruling this out as a means of redistribution, we can compare
the MVPF of the EITC to the MVPF of a tax increase used to
fund this policy. Comparisons of MVPFs correspond to precise
statements of social welfare using Okun’s bucket. As noted, the
MVPF point estimate for the 1993 top tax rate change is 1.85. If
society prefers giving $1.12 to a low-income worker on EITC to
giving $1.85 to a high-income individual facing the top marginal
income tax rate, then the policy is welfare enhancing despite its
relatively low benefit-cost ratio.
VI.C. Welfare Reform: Lessons for Future RCTs
We end with a lesson of how an MVPF perspective can help
inform the design of RCTs. Throughout, we aimed to include all
possible MVPFs in the categories we considered. We included any
policy where we thought we could provide reasonable measures
of both costs and WTP. One set of notable omissions are the state-
level welfare reforms made by states that sought to increase fam-
ily self-sufficiency. Throughout the 1980s and early 1990s, states
experimented with a range of reforms to cash welfare programs
that imposed term limits, provided job training and other educa-
tional services, and provided job search and placement assistance.
The omission of these reforms is not because they were not
analyzed. Many states rigorously evaluated the effect of these
reforms. Upward of 100,000 participants were enrolled into 27
RCTs nationwide (Greenberg, Deitch, and Hamilton 2010). These
RCTs measured and provided a clear estimate of the net cost
of each reform. However, the design of the policies enacted in
each state makes it difficult to understand their welfare effects.
Generally, programs contained both a carrot and a stick.81 As
81. Welfare reform experiments were expected to place no additional costs on
the federal government and so it is natural that states bundled increases in some
types of financial support with potential decreases in others.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1275
a result, we cannot even accurately sign the WTP. As noted by
Manpower Demonstration Research Corporation (MDRC) who
implemented the evaluations of these policies,82 “all [programs]
contained a core quid pro quo arrangement in which the gov-
ernment would offer education, training, job search assistance,
and support services to people receiving cash welfare, while
most recipients—the majority of them single parents—would be
required to participate in such services in order to qualify for
benefits.” Although we can evaluate whether the government
saved money, we do not know if the people in these programs
benefited from their participation. It may be that government
revenue gains were the result of expanded job opportunities due
to program participation. In that case, willingness to pay would
be positive. By contrast, it may be that the government revenue
gains were the result of stricter attendance requirements that
drove individuals off welfare. In that case, willingness to pay
would be negative.83
This highlights the value of isolating the carrot and the stick
into separate RCTs.84 It also demonstrates the value of designing
experiments to estimate individual WTP for nonmarket goods
such as job training, job search assistance, or other educational
policies. In Appendix Figure X, we conduct a range of bounding ex-
ercises that attempt to construct lower and upper bounds on WTP
for these welfare reform programs. Unfortunately, the bounds
are very wide. In many cases, the policies are Pareto dominated,
MVPF < 0, under one set of assumptions and represent a Pareto
improvement, MVPF = ∞, under another set of assumptions.85
Despite substantial expenditures on the evaluation of these re-
forms, the designs of these reforms in each state make it difficult
82. See https://www.mdrc.org/project/evaluations-state-welfare-work-
programs#design-site-data-sources (accessed on July 7, 2019).
83. Previous work (Greenberg, Deitch, and Hamilton 2010) has conducted a
cost-benefit analysis of these reforms by assuming willingness to pay is given by
after-tax earnings. However, if the term limit is what causes individuals to choose
to move off of welfare and into the labor market (thus increasing earnings), the
envelope theorem would suggest the WTP is negative, even if after-tax earnings
increase.
84. Welfare reform RCTs have been criticized for not experimentally varying
each of the components of welfare reform (see Grogger and Karoly 2005). The
MVPF framework suggests bundling carrots and sticks into a single treatment is
particularly problematic for conducting welfare analysis, because it is difficult to
know even whether willingness to pay is positive or negative.
85. In fact, we find policies that follow this pattern in each subcategory of
welfare reform programs. These subcategories include job search assistance.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1276
THE QUARTERLY JOURNAL OF ECONOMICS
to know whether this massive shift in providing welfare benefits
to low-income families led to an increase or decrease in welfare.
VII. CONCLUSION
In this article, we examine the MVPF of 133 different histori-
cal policies over the past half-century in the United States. We find
a clear and persistent pattern that direct investments in children
have yielded the largest MVPFs. There is a large “bang for the
buck” associated with a range of expenditures on children from
early education to child health insurance to college expenditures.
We also demonstrate that in a meaningful number of cases,
these policies pay for themselves. In particular, when government
expenditures boost human capital, the resulting increase in
net government revenue can offset the policy’s up-front costs.
From a taxpayer perspective, these expenditures on children are
investments, rather than just transfers.
We find that opportunities for high-return investments in
children have persisted across policy categories for many decades.
This is, however, no guarantee that all future investment in these
categories will produce high MVPFs. Indeed, we find that MVPFs
vary substantially within policy categories. Low-return policies
exist even in high-return categories. This highlights the value of
further understanding the mechanisms behind the high MVPFs
of successful historical investments.
Even in cases where there is existing research, much remains
unknown about the welfare consequences of government policy. To
that aim, we quantify the value of future work that uses new data
to reduce estimate uncertainty. We show that in many cases, an
evidence-based policy maker seeking to maximize social welfare
should be willing to make substantial budgetary expenditures to
learn more about policy effectiveness. In particular, our results
highlight the value of expanded use of administrative data for
policy analysis.
The 133 policies included in this article are just a small subset
of those that could be analyzed using the MVPF. We do not discuss
the MVPF of crime policies, environmental policies, macroeco-
nomic stabilization policies, or infrastructure policies, among
many others. With careful tracking of willingness to pay and net
costs, the MVPF can be used in any of these contexts and can guide
cost-benefit analyses. We leave that analysis for future work.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1277
TABLE I (CONTINUED)
Sample
Program
Label
Year
Age of
Paper
Papers
Implemented Beneficiaries Baseline Restricted Extended Estimates
Utilized
MVPF
Head Start Introduction
Head Start
1965
4
x
x
x
Johnson and Jackson (2019)
Head Start Regression
Head
1970
4
x
Ludwig and Miller (2007)
Discontinuity
Start RD
Head Start
Head
2002
3
x
x
Kline and Walters (2016)
Impact Study
Start RCT
K–12 School
K12
1991
11
x
x
x
Hyman (2017)
Finance Reform
Spend
K–12 School
K12 Spend
1994
11
x
Heckman et al. (2010)
Spending in Michigan
Mich.
Perry Preschool Program Perry Preschool
1962
4
x
x
x
Heckman et al. (2011)
College adult
American Opportunity
AOTC (IS)
2011
25
x
x
Bulman and Hoxby (2015)
Tax Credit,
Independent Single
Filers at Phase Start
American Opportunity
AOTC (JE)
2011
55
x
x
Bulman and Hoxby (2015)
Tax Credit, Joint Filers
at Phase End
American Opportunity
AOTC (JS)
2011
20
x
x
Bulman and Hoxby (2015)
Tax Credit,
Joint Filers at
Phase Start
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1278
THE QUARTERLY JOURNAL OF ECONOMICS
TABLE I (CONTINUED)
Sample
Program
Label
Year
Age of
Paper
Papers
Implemented Beneficiaries Baseline Restricted Extended Estimates
Utilized
MVPF
American Opportunity
AOTC (SI)
2009
20
x
x
Bulman and Hoxby (2015)
Tax Credit,
Simulated Instrument
American Opportunity
AOTC (SE)
2011
55
x
x
Bulman and Hoxby (2015)
Tax Credit,
Single Filers at
Phase End
American Opportunity
AOTC (SS)
2011
20
x
x
Bulman and Hoxby (2015)
Tax Credit,
Single Filers at
Phase Start
Hope Tax Credit
HOPE Cred.
1999
20
x
x
Turner (2011)
Hope Tax Credit,
HTC (IS)
2007
25
x
x
Bulman and Hoxby (2015)
Independent Single
Filers at Phase Start
Hope Tax Credit,
HTC (JE)
2007
20
x
x
Bulman and Hoxby (2015)
Joint Filers at Phase End
Hope Tax Credit,
HTC (JS)
2007
20
x
x
Bulman and Hoxby (2015)
Joint Filers at
Phase Start
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1279
TABLE I (CONTINUED)
Sample
Program
Label
Year
Age of
Paper
Papers
Implemented Beneficiaries Baseline Restricted Extended Estimates
Utilized
MVPF
Hope Tax Credit,
HTC (SE)
2007
20
x
x
Bulman and Hoxby (2015)
Single Filers at
Phase End
Hope Tax Credit, Single
HTC
2007
55
x
x
Bulman and Hoxby (2015)
Filers at Phase Start
(SS)
Hope and Lifetime
HOPE/LLC
1998
55
x
x
Long (2004)
Learners Tax Credits
Pell Grants
Adult
1973
28
x
x
Seftor and Turner (2002)
Introduction to Adults
Pell
Tax Deduction for Postsecondary
Tuition
2006
55
x
x
Hoxby and Bulman (2016)
Tuition, Joint Filers at Phase End
Deduc (JE)
LaLumia (2012)
Tax Deduction for Postsecondary
Tuition
2006
55
x
x
Hoxby and Bulman (2016)
Tuition, Joint Filers at Phase Start
Deduc (JS)
LaLumia (2012)
Tax Deduction for Postsecondary
Tuition
2006
55
x
x
Hoxby and Bulman (2016)
Tuition, Single Filers at Phase End
Deduc (SE)
LaLumia (2012)
Tax Deduction for Postsecondary
Tuition
2006
20
x
x
Hoxby and Bulman (2016)
Tuition, Single Filers at Phase Start Deduc (SS)
LaLumia (2012)
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1280
THE QUARTERLY JOURNAL OF ECONOMICS
TABLE I (CONTINUED)
Sample
Program
Label
Year
Age of
Paper
Papers
Implemented Beneficiaries Baseline Restricted Extended Estimates
Utilized
MVPF
College child
Cal Grant, GPA Threshold
Cal Grant GPA
1998
20
x
x
x
Bettinger et al. (2019)
Cal Grant, Income Threshold
Cal Grant Inc
1998
20
x
x
x
Bettinger et al. (2019)
City University of New York
CUNY
2009
20
x
x
Marx and Turner (2018)
Pell Grants
Pell
Community College Tuition
CC Mich
2005
20
x
x
Acton (2018)
Changes, Michigan
Community College Tuition
CC
2005
20
x
x
Denning (2017)
Changes, Texas
Texas
District of Columbia Tuition
DC
1999
20
x
x
Abraham and Clark (2006)
Assistance Grant Program
Grant
Florida International University
Admissions at GPA Threshold
FIU GPA
1999
20
x
x
x
Zimmerman (2014)
Florida Student Access Grant
Florida Grant
2001
20
x
x
Castleman and Long (2016)
Free Application for Federal
Free FAFSA
2008
20
x
U.S. Department of Education
(Dep)
Office of Postsecondary
Student Aid, Dependent Year
Impact
Education (2010)
Bettinger et al. (2012)
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1281
TABLE I (CONTINUED)
Sample
Program
Label
Year
Age of
Paper
Papers
Implemented Beneficiaries Baseline Restricted Extended
Estimates
Utilized
MVPF
Free Application
Free FAFSA
2008
20
x
U.S. Department
for Federal Student Aid,
(Indep)
of Education Office
Independent Year
of Postsecondary
Impact
Education (2010)
Bettinger et al. (2012)
Georgia HOPE Scholarship
Georgia HOPE
1995
20
x
x
Cornwell et al. (2003)
Cornwell et al. (2006)
HAIL Michigan Aid
Awareness Letter
HAIL Aid
2016
20
x
Dynarski et al. (2018)
Hoekstra (2009)
Kalamazoo Promise
Scholarship
Kalamazoo
2006
20
x
x
Bartik et al. (2016)
Bartik et al. (forthcoming)
Massachussetts Adams
Scholarship
MA Scholarship
2005
20
x
x
Cohodes and Goodman (2014)
Goodman (2008)
Pell Grants in Ohio
Ohio Pell
2000
19
x
x
Bettinger (2004)
Pell Grants
TN Pell
2008
20
x
x
U.S. Department of Education
in Tennessee
Office of Postsecondary
Education (2010)
Carruthers and Welch (2019)
Pell Grants in Texas
Texas Pell
2008
20
x
x
x
Denning et al. (2019)
Social Security Student
Benefit Program
Soc Sec College
1982
20
x
x
Dynarski (2003)
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1282
THE QUARTERLY JOURNAL OF ECONOMICS
TABLE I (CONTINUED)
Sample
Program
Label
Year
Age of
Paper
Papers
Implemented
Beneficiaries
Baseline
Restricted
Extended
Estimates
Utilized
MVPF
Spending at Colleges from State
Appropriations
College Spend
2001
20
x
x
Deming and Walters (2017)
Tennessee HOPE Scholarships
TN Hope
2008
20
x
x
Bruce and Carruthers (2014)
Tuition Cuts at Colleges from
State Appropriations
College Tuition
2001
20
x
x
Deming and Walters (2017)
Wisconsin Scholar Grant to
Low-Income College Students
WI Scholarship
2009
20
x
x
Goldrick-Rab et al. (2016)
Job training
Job Corps
Job Corps
1995
19
x
x
x
Schochet et al. (2006, 2008)
Schochet (2018)
Job Training Partnership Act,
Adults
JTPA Adult
1988
34
x
x
x
Bloom et al. (1997)
Job Training Partnership Act,
Youth
JTPA Youth
1988
19
x
x
x
Bloom et al. (1997)
JobStart
JobStart
1986
19
x
x
x
Cave et al. (1993)
National Supported Work
Demonstration, Adult Women
NSW Women
1976
34
x
x
x
Hollister, Kemper, and Maynard (1984)
Couch (1992)
National Supported Work
Demonstration, Ex-Addicts
NSW Ex-Addict
1976
33
x
x
x
Hollister, Kemper, and Maynard (1984)
National Supported Work
Demonstration, Ex-Offenders
NSW Ex-Offender
1976
33
x
x
x
Hollister, Kemper, and Maynard (1984)
National Supported Work
Demonstration, Youth
NSW Youth
1976
18
x
x
x
Hollister, Kemper, and Maynard (1984)
Couch (1992)
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1283
TABLE I (CONTINUED)
Sample
Program
Label
Year
Age of
Paper
Papers
Implemented
Beneficiaries
Baseline
Restricted
Extended
Estimates
Utilized
MVPF
Work Advance
Work Advance
2012
34
x
x
x
Schaberg (2017)
Hendra et al. (2016)
Year Up
Year Up
2013
21
x
x
x
Fein and Hamadyk (2018)
Panel B: Social insurance
Disability ins.
Disability Insurance
Changes in Benefit
Generosity
DI Generosity
2004
50
x
x
x
Gelber, Moore, and Strand (2017)
Disability Insurance
Judge Leniency
DI Judge
2005
47
x
x
x
Maestas, Mullen, and Strand (2013)
Gelber, Moore, and Strand (2017)
Disability Insurance
DI Examiner
1995
48
x
x
x
Gelber, Moore, and Strand (2017)
Medical Examiner
French and Song (2014)
Leniency
Disability Insurance to
Veterans
DI Veterans
2001
53
x
x
x
Autor et al. (2016)
Health adult
Health Insurance
Mass HI
2011
44
x
x
x
Hendren (2017c)
Subsidies in
(150%FPL)
Finkelstein, Hendren, and
Massachusetts to Indi-
Shepard (2019)
viduals at 150% of the
Federal Poverty Line
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1284
THE QUARTERLY JOURNAL OF ECONOMICS
TABLE I (CONTINUED)
Sample
Program
Label
Year
Age of
Paper
Papers
Implemented Beneficiaries Baseline Restricted Extended Estimates
Utilized
MVPF
Health Insurance
Mass HI (200%FPL)
2011
44
x
x
x
Hendren (2017c)
Subsidies in
Finkelstein, Hendren, and
Massachusetts to
Shepard (2019)
Individuals at 200% of the
Federal Poverty Line
Health Insurance Subsidies Mass HI (250%FPL)
2011
44
x
x
x
Hendren (2017c)
in Massachusetts to
Finkelstein, Hendren, and
Individuals at 250% of the
Shepard (2019)
Federal Poverty Line
Medicare Introduction
Medicare
1965
78
x
x
x
U.S. Census Bureau (1966)
in 1965
Intro
Finkelstein and McKnight (2008)
Oregon Health Insurance
Oregon
2008
42
x
x
x
Finkelstein et al. (2012)
Experiment (Provided to
Health
Finkelstein, Hendren, and
Single Adults)
Luttmer (2019)
Taxation of Medigap
Policies
Medigap Tax
2002
75
x
x
x
Cabral and Mahoney (2019)
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1285
TABLE I (CONTINUED)
Sample
Program
Label
Year
Age of
Paper
Papers
Implemented Beneficiaries Baseline Restricted Extended
Estimates
Utilized
MVPF
Health child
Medicaid Expansion to
MC
1990
11
x
x
x
Lo Sasso and Seamster (2007)
Children Born after
Child 83+
Wherry and Meyer (2016)
September 30, 1983
Wherry et al. (2018)
Medicaid Expansions to
MC Pregnant &
1986
0
x
x
x
Dave et al. (2015)
Pregnant Women &
Infants
Currie and Gruber (1996)
Infants
Miller and Wherry (2019)
Medicaid Expansions to
MC Child
1986
9
x
x
x
Boudreaux, Golberstein, and McAlpine (2016)
Young Children
(State Exp)
Brown, Kowalski, and Lurie (2017)
Medicaid Introduction to
MC
1968
9
x
x
x
x
Goodman-Bacon (2017)
AFDC-eligible Families
Intro
Supplemental Security
Income
Supplemental Security
Income Age 18
SSI
Review
1996
18
x
x
x
x
Deshpande (2016)
Medical Review
Supplemental Security
SSI
2005
48
x
x
x
Deshpande (2016)
Income Medical
Judge
French and Song (2014)
Examiner Leniency
Gelber, Moore, and Strand (2017)
U.S. Social Security Administration (2014)
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1286
THE QUARTERLY JOURNAL OF ECONOMICS
TABLE I (CONTINUED)
Sample
Program
Label
Year
Age of
Paper
Papers
Implemented Beneficiaries Baseline Restricted Extended
Estimates
Utilized
MVPF
Unemployment insurance
Unemployment Insurance
Benefit Changes (Diff in
Diff Across States in
Chetty 2008)
UI Ben
(State Max)
1992
37
x
x
x
Chetty (2008)
Gruber (1997)
Hendren (2017b)
Schmieder and Von Wachter
(2016)
Unemployment Insurance
Benefit Changes (Diff in
Diff Across States in Katz
and Meyer 1990)
UI Ben
(DD)
1980
33
x
x
x
Gruber (1997)
Hendren (2017b)
Katz and Meyer (1990)
Schmieder and Von Wachter
(2016)
Unemployment Insurance
UI Ben
1992
37
x
x
x
Gruber (1997)
Benefit Changes (Diff in
(DD w UR)
Hendren (2017b)
Diff Across States in Kroft)
Kroft and Notowidigdo (2016)
and Notowidigdo 2016)
Schmieder and Von Wachter
(2016)
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1287
TABLE I (CONTINUED)
Sample
Program
Label
Year
Age of
Paper
Papers
Implemented Beneficiaries Baseline Restricted Extended
Estimates
Utilized
MVPF
Unemployment Insurance
Benefit Changes in Georgia
UI Ben
1979
42
x
x
x
Gruber (1997)
(GA)
Hendren (2017b)
Schmieder and Von Wachter
(2016)
Solon (1985)
Unemployment Insurance
Benefit Changes in Missouri
(Expansion Estimates)
UI Ben
2005
42
x
x
x
Card et al. (2015)
(MO Exp)
Gruber (1997)
Hendren (2017b)
Schmieder and Von Wachter
(2016)
Unemployment Insurance
UI Ben
2010
42
x
x
x
Card et al. (2015)
Benefit Changes in Missouri
(MO Rec)
Gruber (1997)
(Recession Estimates)
Hendren (2017b)
Schmieder and Von Wachter
(2016)
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1288
THE QUARTERLY JOURNAL OF ECONOMICS
TABLE I (CONTINUED)
Sample
Program
Label
Year
Age of
Paper
Papers
Implemented Beneficiaries Baseline Restricted Extended Estimates
Utilized
MVPF
Unemployment Insurance
Benefit Changes in
New York
UI Ben
(NY)
1989
42
x
x
x
Gruber (1997)
Hendren (2017b)
Meyer and Mok (2007)
Schmieder and Von Wachter (2016)
Unemployment Insurance
Benefit Changes via
Regression Kink in
Benefit Schedule
UI Ben
1980
34
x
x
x
Gruber (1997)
(RK)
Hendren (2017b)
Landais (2015)
Schmieder and Von Wachter (2016)
Unemployment Insurance
Duration Extensions (Diff
in Diff Across States in
Katz and Meyer 1990)
UI Dur
1980
33
x
x
x
Ganong and Noel (2019)
(DD)
Gruber (1997)
Hendren (2017b)
Katz and Meyer (1990)
Schmieder and Von Wachter (2016)
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1289
TABLE I (CONTINUED)
Sample
Program
Label
Year
Age of
Paper
Papers
Implemented Beneficiaries Baseline Restricted Extended
Estimates
Utilized
MVPF
Unemployment Insurance
Duration Extensions in
Missouri
UI Dur
2011
42
x
x
x
Ganong and Noel (2019)
(MO)
Gruber (1997)
Hendren (2017b)
Johnston and Mas (2018)
Schmieder and Von Wachter (2016)
Panel C: In-kind transfers
Housing vouchers
Effects of Housing Vouchers
HCV RCT
2000
31
x
x
Jacob and Ludwig (2012)
on AFDC Families Experiment
to Welfare
Mills et al. (2006)
Wood, Turnham, and Mills (2008)
Housing Vouchers
HCV
1997
31
x
x
Jacob and Ludwig (2012)
in Chicago
Chicago Lottery
Jacob, Ludwig, and Kapustin (2014)
Jobs Plus
Jobs+
1998
35
x
Bloom, Riccio, and Verma (2005)
Riccio (2006)
MTO
Moving to Opportunity
Experiment Providing
Vouchers and Counseling
MTO
1996
10
x
x
x
Chetty, Hendren, and Katz (2016)
Goering et al. (1999)
Sanbonmatsu et al. (2011)
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1290
THE QUARTERLY JOURNAL OF ECONOMICS
TABLE I (CONTINUED)
Sample
Program
Label
Year
Age of
Paper
Papers
Implemented Beneficiaries Baseline Restricted Extended
Estimates
Utilized
MVPF
Nutrition
Special Supplemental
WIC
1975
26
x
Black, Devereux, and Salvanes (2007)
Nutrition Program for
Hoynes, Page, and Stevens (2011)
Women, Infants, and
Whitmore (2002)
Children
Supplemental Nutrition
Assistance Program
Application Assistance
SNAP Assist
2016
69
x
x
x
x
Finkelstein and Notowidigdo (2019)
Supplemental Nutrition
Assistance Program
Application
Information
SNAP Info
2016
69
x
x
x
x
Finkelstein and Notowidigdo (2019)
Supplemental Nutrition
Assistance Program
Introduction
SNAP Intro
1968
32
x
x
x
Hoynes, Schanzenbach, and Almond (2016)
Almond, Hoynes, and Schanzenbach (2011)
Bailey et al. (2019)
Hoynes, Page, and Stevens (2011)
Hoynes and Schanzenbach (2012)
East (2018)
Finkelstein and Notowidigdo (2019)
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1291
TABLE I (CONTINUED)
Sample
Program
Label
Year
Age of
Paper
Papers
Implemented Beneficiaries Baseline Restricted Extended Estimates
Utilized
MVPF
Panel D: Taxes and cash transfers
1986 Earned Income
EITC 1986
1986
28
x
x
x
Ackerman, Holtzblatt, and Masken (2009)
Tax Credit Expansion
Tax Policy Center (2016)
Blank and Ruggles (1996)
Hotz and Scholz (2003)
Meyer and Rosenbaum (2001)
Moffitt (2002)
Scholz (1993)
Crouse and Waters (2014)
Eissa and Hoynes (2004)
Eissa and Liebman (1996)
1993 Earned Income
EITC 1993
1993
29
x
x
x
Tax Policy Center (2016)
Tax Credit Expansion
Hotz and Scholz (2003)
Dahl and Lochner (2012)
Meyer and Rosenbaum (2001)
Bastian and Michelmore (2018)
Chetty, Friedman, and Rockoff (2011)
Crouse and Waters (2014)
Eissa and Hoynes (2004)
Hoynes and Patel (2018)
Manoli and Turner (2018)
Maxfield (2018)
Michelmore (2013)
Meyer and Rosenbaum (2001)
Ackerman, Holtzblatt, and Masken (2009)
Currie and Cole (1993)
Moffitt (2003)
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1292
THE QUARTERLY JOURNAL OF ECONOMICS
TABLE I (CONTINUED)
Sample
Program
Label
Year
Age of
Paper
Papers
Implemented Beneficiaries Baseline Restricted Extended Estimates
Utilized
MVPF
Aid to Families with
AFDC Term Limits
1996
27
x
x
x
Grogger (2003)
Dependent Children (Term
Limit Modifications)
Pavetti (1995)
Alaska Permanent Fund
Alaska UBI
1998
34
x
x
x
Ackerman, Holtzblatt, and Masken (2009)
Dividend
Bhargava and Manoli (2015)
Blank and Ruggles (1996)
Hotz and Scholz (2003)
Jones and Marinescu (2018)
Meyer and Rosenbaum (2001)
Moffitt (2002)
Paycheck Plus Experiment
Providing EITC-benefits to
Adults without Dependents
Paycheck+
2013
35
x
x
x
Miller et al. (2017)
Seattle-Denver Income
Neg Inc Tax
1971
35
x
x
x
Price and Song (2018)
Maintenance Experiment
U.S. Social Security Administration (2018)
Tax Foundation (2013)
U.S. Department of Health
Human Services (1983)
Von Wachter, Song, and Manchester (2011)
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1293
TABLE I (CONTINUED)
Sample
Program
Label
Year
Age of
Paper
Papers
Implemented Beneficiaries Baseline Restricted Extended
Estimates
Utilized
MVPF
Top taxes
Top Tax 2013 Increases
from Affordable Care Act
Top Tax 2013
2013
49
x
x
x
Kawano, Weber, and
Whitten (2016)
Hendren (2017a)
Top Tax Rate Increase in
Top Tax 1993
1993
49
x
x
x
Atkinson, Piketty, and Saez (2011)
Omnibus Budget
Reconciliation Act 1993
Carroll (1998)
Top Tax Rate Reductions
Top Tax 1986
1986
49
x
x
x
Atkinson, Piketty, and Saez (2011)
in Tax Reform Act of 1986
Auten and Caroll (1999)
Top Taxes, Economic
Top Tax 2001
2001
49
x
x
x
Atkinson, Piketty, and Saez (2011)
Growth and Tax Relief
Reconciliation Act 2001
Heim (2009)
Top Taxes, Economic
Top Tax 1981
1981
49
x
x
x
Atkinson, Piketty, and Saez (2011)
Recovery Tax Act 1981
Saez (2003)
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1294
THE QUARTERLY JOURNAL OF ECONOMICS
TABLE I (CONTINUED)
Sample
Program
Label
Year
Age of
Paper
Papers
Implemented Beneficiaries Baseline Restricted Extended Estimates
Utilized
MVPF
Panel E: Welfare reform
Welfare to Work Alameda
GAIN Alm.
1988
31
x
Freedman et al. (1996)
Mandatory Mixed-
Initial-Activity Programs
Greenberg, Deitch, and Hamilton (2010)
Welfare to Work Atlanta
HCD NEWWS Atl.
1992
33
x
Greenberg, Deitch, and Hamilton (2010)
Mandatory Education-
First Programs
Hamilton et al. (2001)
Welfare to Work Atlanta
LFA NEWWS Atl.
1992
33
x
Hamilton et al. (2001)
Mandatory Job-Search-
First Programs
Greenberg, Deitch, and Hamilton (2010)
Welfare to Work Butte
Mandatory Mixed-Initial-
Activity Programs
GAIN Butte
1987
31
x
Hamilton et al. (2001)
Freedman et al. (1996)
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1295
TABLE I (CONTINUED)
Sample
Program
Label
Year
Age of
Paper
Papers
Implemented Beneficiaries Baseline Restricted Extended
Estimates
Utilized
MVPF
Welfare to Work Columbus
Integrated Mandatory
NEWWS Col. Int.
1992
32
x
Greenberg, Deitch, and
Hamilton (2010)
Education-First Programs
Hamilton et al. (2001)
Welfare to Work Columbus
Traditional Mandatory
Education-First Programs
NEWWS Col. Trad.
1992
32
x
Greenberg, Deitch, and
Hamilton (2010)
Hamilton et al. (2001)
Welfare to Work Connecticut
Jobs First
1996
31
x
Bloom et al. (2002)
Time-Limit-Mix Programs
Greenberg, Deitch, and
Hamilton (2010)
Welfare to Work Cook County
Mandatory Work
Experience Program
WIN Demo.
1985
32
x
Greenberg, Deitch, and
Hamilton (2010)
Brock, Butler, and Long (1993)
Welfare to Work Detroit
Mandatory Education-First
NEWWS Det.
1992
30
x
Greenberg, Deitch, and
Hamilton (2010)
Programs
Hamilton et al. (2001)
Welfare to Work Florida
Mandatory Mixed-Initial-
Proj. Ind. FL
1990
32
x
Greenberg, Deitch, and
Hamilton (2010)
Activity Programs
Kemple, Friedlander and
Fellerath (1995)
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1296
THE QUARTERLY JOURNAL OF ECONOMICS
TABLE I (CONTINUED)
Sample
Program
Label
Year
Age of
Paper
Papers
Implemented Beneficiaries Baseline Restricted Extended Estimates
Utilized
MVPF
Welfare to Work Florida
Time-Limit-Mix Programs
FTP
1994
29
x
Bloom et al. (2000)
Greenberg, Deitch, and
Hamilton (2010)
Welfare to Work Grand
Rapids Mandatory
Education-First Programs
HCD NEWWS Gr. Rap.
1991
28
x
Greenberg, Deitch, and
Hamilton (2010)
Hamilton et al. (2001)
Welfare to Work Grand
Rapids Mandatory Job-
Search-First Programs
LFA NEWWS Gr. Rap.
1991
28
x
Hamilton et al. (2001)
Greenberg, Deitch, and
Hamilton (2010)
Welfare to Work Los Angeles
Mandatory Job-Search-
First Programs
GAIN LA jobs
1996
34
x
Freedman et al. (2000)
Greenberg, Deitch, and
Hamilton (2010)
Welfare to Work Los Angeles
Mandatory Mixed-Initial-
Activity Programs
GAIN LA
1988
31
x
Greenberg, Deitch, and
Hamilton (2010)
Freedman et al. (1996)
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1297
TABLE I (CONTINUED)
Sample
Program
Label
Year
Age of
Paper
Papers
Implemented Beneficiaries Baseline Restricted Extended Estimates
Utilized
MVPF
Welfare to Work Minnesota
Earnings Supplement
Programs
MFIP
1994
29
x
Greenberg, Deitch, and
Hamilton (2010)
Miller et al. (2000)
Welfare to Work Portland
Mandatory Mixed-Initial-
Activity Programs
NEWWS Port.
1993
30
x
Greenberg, Deitch, and
Hamilton (2010)
Hamilton et al. (2001)
Welfare to Work Riverside
Mandatory Education-First
Programs
HCD NEWWS Riv.
1991
32
x
Hamilton et al. (2001)
Greenberg, Deitch, and
Hamilton (2010)
Welfare to Work Riverside
Mandatory Job-Search-First
Programs
LFA NEWWS Riv.
1991
32
x
Hamilton et al. (2001)
Greenberg, Deitch, and
Hamilton (2010)
Welfare to Work Riverside
Mandatory Mixed-Initial-
Activity Programs
GAIN Riv.
1987
31
x
Freedman et al. (1996)
Greenberg, Deitch, and
Hamilton (2010)
Welfare to Work San Diego
Mandatory Job-Search-First
Programs
SWIM
1985
31
x
Greenberg, Deitch, and
Hamilton (2010)
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1298
THE QUARTERLY JOURNAL OF ECONOMICS
TABLE I (CONTINUED)
Sample
Program
Label
Year
Age of
Paper
Papers
ImplementedBeneficiariesBaselineRestrictedExtended Estimates
Utilized
MVPF
Welfare to Work San Diego
Mandatory Mixed-
Initial-Activity Programs
GAIN SD
1987
31
x
Greenberg, Deitch, and Hamilton
(2010)
Freedman et al. (1996)
Welfare to Work San Diego
Mandatory Work
Experience Program
Work Exp. SD
1982
32
x
Greenberg, Deitch, and Hamilton
(2010)
Brock, Butler, and Long (1993)
Welfare to Work Tulane
Mandatory Mixed-Initial-
Activity Programs
GAIN Tul.
1988
31
x
Greenberg, Deitch, and Hamilton
(2010)
Freedman et al. (1996)
Welfare to Work Vermont
Earnings Supplement
Programs
WRP Earn Supp.
1994
31
x
Greenberg, Deitch, and Hamilton
(2010)
Scrivener et al. (2002)
Welfare to Work Vermont
Time-Limit-Mix Programs
WRP Time lim.
1994
31
x
Scrivener et al. (2002)
Greenberg, Deitch, and Hamilton
(2010)
Welfare to Work West
Virginia Mandatory Work
Experience Program
CWEP
1983
33
x
Brock, Butler, and Long (1993)
Greenberg, Deitch, and Hamilton
(2010)
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1299
APPENDIX
APPENDIX FIGURE I
Income Projections Using the ACS
Panels A and B present a decomposition of the elements that make up our
income projection process for the examples in Section III.A. The “Pop Avg” series
is constructed in each case from the 2015 ACS and using a 0.5% wage growth
assumption. At each age “Pop Avg” gives the mean wage level that would prevail
in the population for individuals of that age, when individuals in the treatment
group for the relevant policy were that age. This number is constructed by
assuming that the mean wage level at each age will rise (and has previously
risen) by 0.5% in each year. The “Control Forecast” series is constructed by taking
an estimate of earnings for a relevant control group at a particular age or range of
ages, then calculating the implied proportion of the “Pop Avg” series at those ages,
then projecting the series forwards (and backwards) as this constant fraction
of “Pop Avg.” The “Treatment” series is constructed by summing the observed
treatment effects in dollar terms and the “Control Forecast” series. To construct
the “Predicted” series we take the final value of the “Treatment” series, then
calculate the ratio of this value to the value of the “Pop Avg” series at that same
age, before applying this ratio to the “Pop Avg” series up to age 65. See Online
Appendix I for further details of this methodology.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1300
THE QUARTERLY JOURNAL OF ECONOMICS
APPENDIX FIGURE II
Willingness to Pay per Dollar of Programmatic Spending
This figure presents estimates of WTP normalized by initial programmatic
spending for each category-average group of policies in our baseline sample. We
plot these estimates as a function of the average age of each policy’s beneficiaries
within category. Bootstrapped 95% confidence intervals with adjustments
(discussed in Online Appendix H) are shown for the category averages. The
normalized willingness to pay of individual policies are shown in smaller dots,
color-coded to align with their respective categories.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1301
APPENDIX FIGURE III
Robustness to Child Effects
This figure assesses the impact of observing child effects on our estimates as
a function of the average age of the economic beneficiaries of the policy. Panel
A restricts our sample to the subset of policies for which we observe estimates
of the impact of the policy on children. In addition, Panel B shows projected
MVPFs for additional policies that do not observe earnings impacts but do observe
another intermediate outcome such as birthweight (AFDC), college attendance
(housing vouchers to AFDC recipients), and test scores (housing vouchers in
Chicago). Panel C reports the MVPF for the EITC under alternative methods of
incorporating indirect effects on children through test scores, college attendance,
and income of EITC more broadly.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1302
THE QUARTERLY JOURNAL OF ECONOMICS
APPENDIX FIGURE IV
Robustness to Alternative Interest Rates
This figure presents our MVPF estimates as in Figure III and the category av-
erages as in Figure IV, Panel B under alternative real interest rate assumptions,
as opposed to our baseline specification of 3%. Panel B differs slightly from our
baseline specification because we restrict to the subset of policies for which we are
able to vary the discount rate (e.g., we exclude papers where we directly import
an MVPF that relied on a particular discount rate). We omit confidence intervals
for ease of viewing, but caution the reader that the estimate for the college adult
category has a CI that includes 0 and infinity.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1303
APPENDIX FIGURE V
Robustness to Alternative Tax Rates
This figure presents our MVPF estimates as in Figure III and the category
averages as in Figure IV, Panel B under alternative tax rate assumptions. Panel
A replicates our baseline specification using the CBO estimates of the tax rates.
Panels B–D adjust the tax rate to 10%, 20%, and 30%.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1304
THE QUARTERLY JOURNAL OF ECONOMICS
APPENDIX FIGURE VI
Specification Robustness
This figure presents the category-average MVPFs from Figure III using a range
of different alternative specifications that are more conservative than our baseline
specifications. Panel A replaces our point estimate WTP measures with our
conservative measures of WTP. We report bootstrapped 95% confidence intervals
with adjustments (discussed in Online Appendix H) for each category average.
Panel B replaces our baseline income projection procedure with a procedure that
assumes zero income growth over the lifecycle. We use our restricted sample of
policies for this specification. See Online Appendix I for further details.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1305
APPENDIX FIGURE VII
Sample Restrictions
This figure presents the category-average MVPFs from Figure III using a range
of alternative sample restrictions. Panel A considers our restricted sample that
drops estimates for which we are forecasting earnings effects based on a policy’s
impact on college attendance. Panel B restricts the sample to only policies for
which earnings outcomes are estimated for at least five years of follow-up after
the policy. For this panel we show group averages even for groups with a single
policy. Panel C restricts the sample to policies whose identification strategy is a
randomized control trial, lottery, or regression discontinuity. Panel D restricts to
policies whose primary analyses have been published in a peer-reviewed journal.
We report bootstrapped 95% confidence intervals with adjustments (discussed in
Online Appendix H) for each category average.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1306
THE QUARTERLY JOURNAL OF ECONOMICS
APPENDIX FIGURE VIII
Publication Bias
This figure presents the MVPF estimates from Figure III and category averages
in Figure IV, Panel B using estimates corrected for publication bias from the
method of Andrews and Kasy (2019). Panel A reports estimates using the
corrections using the publication likelihood estimated from our model that
imposes jumps at p = .05 and p = .10, as shown in Table III, columns (3) and
(6). In Panel B we report corrected estimates under an assumption that child
policies are 35 times more likely to be published if they find a positive effect on
children’s outcomes (and we assume no publication bias for adult policies or for
child policies that find negative effects on children). This 35 times corresponds to
the estimated publication bias implied by a large-scale replication of experimental
economics papers by Camerer et al. (2016) (Table 1 of Andrews and Kasy 2019
reports that insignificant results are 0.029 times as likely to be published). We
do not report confidence intervals for these estimates (to our knowledge there is
no well-accepted method of constructing such intervals); but we refer readers to
Figure IV, Panel B to note that some of these category averages are imprecise.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1307
APPENDIX FIGURE IX
MVPFs by Decade
This figure presents MVPFs for all policies evaluated in the article based on
the year in which the policy was implemented. Policies are divided into categories
based on their decade of implementation and the average age of their economic
beneficiaries. For policies implemented in each decade there are two categories—
policies with beneficiaries over age 23 and policies with beneficiaries aged 23 or
younger. Within each decade by age category we construct the MVPF for a hypo-
thetical policy that allocates $1 of programmatic spending equally among all the
policies in the category. This is the same approach used to create MVPF estimates
for policy domains in previous figures. The capped lines show the 95% bootstrapped
confidence intervals with adjustments discussed in Online Appendix H.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1308
THE QUARTERLY JOURNAL OF ECONOMICS
APPENDIX FIGURE X
Welfare Reform
This figure presents estimates of the MVPF of 27 welfare reform policies
discussed in Section VI.C. We report MVPF estimates using three potential
measures of WTP: (i) cost, the mechanical cost of the program incurred by the
government, excluding any fiscal externalities from behavior change. Estimates
from this specification are denoted by circles. (ii) Change in transfer payments
(welfare, food stamps, and Medicaid). Estimates from this specification are
denoted by Xs. (iii) Change in post-tax income, which includes the change in
participants’ incomes due to changes in employment and the change in their
transfer payments. Estimates from this specification are denoted by triangles.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1309
HARVARD UNIVERSITY
HARVARD UNIVERSITY
SUPPLEMENTARY MATERIAL
An Online Appendix for this article can be found at The
Quarterly Journal of Economics online. Data and code replicating
tables and figures in this article can be found in Hendren and
Sprung-Keyser
(2020),
in
the
Harvard
Dataverse,
doi:
10.7910/DVN/ZHOSGC.
REFERENCES
Abraham, Katharine G., and Melissa A. Clark, “Financial Aid and Students’ Col-
lege Decisions: Evidence from the District of Columbia Tuition Assistance
Grant Program,” Journal of Human Resources, 41 (2006), 578–610.
Ackerman, Deena, Janet Holtzblatt, and Karen Masken, “The Pattern of EITC
Claims over Time: A Panel Data Analysis,” in “Conference Paper from IRS
RC 2009: Internal Revenue Service Research Conference.” Washington, DC:
Department of the Treasury.
Acton, Riley, “The Impact of Public Tuition Subsidies on College Enrollment Deci-
sions: Evidence from Michigan,” Michigan State University, 2018.
Akerlof, George A., “The Economics of ‘Tagging’ as Applied to the Optimal In-
come Tax, Welfare Programs, and Manpower Planning,” American Economic
Review, 68 (1978), 8–19.
Almond, Douglas, Hilary W. Hoynes, and Diana W. Schanzenbach, “Inside the
War on Poverty: The Impact of Food Stamps on Birth Outcomes,” Review of
Economics and Statistics, 93 (2011), 387–403.
American Institutes for Research, “Delta Cost Project,” (2017), https://www.
deltacostproject.org/delta-cost-project-database (accessed April 26, 2019).
Andrews, Isaiah, and Maximilian Kasy, “Identification of and Correction for Pub-
lication Bias,” American Economic Review, 109 (2019), 2766–2794.
Atkinson, Anthony B., Thomas Piketty, and Emmanuel Saez, “Top Incomes in the
Long Run of History,” Journal of Economic Literature, 49 (2011), 3–71.
Atkinson, Anthony B., and Nicholas H. Stern, “Pigou, Taxation and Public Goods,”
Review of Economic Studies, 41 (1974), 119–128.
Atkinson, Anthony Barnes, and Joseph E. Stiglitz, “The Design of Tax Structure:
Direct versus Indirect Taxation,” Journal of public Economics, 6 (1976), 55–75.
Auerbach, Alan, “The Theory of Excess Burden and Optimal Taxation,” in Hand-
book of Public Economics, vol. 1, A. Auerbach and M. Feldstein, eds. (Amster-
dam: Elsevier, 1985), 61–127.
Auerbach, Alan J., and James R. Hines, “Taxation and Economic Efficiency,” in
Handbook of Public Economics, vol. 3, A. Auerbach and M. Feldstein, eds.
(Amsterdam: Elsevier, 2002), 1347–1421.
Auten, Gerald, and Robert Carroll, “The Effect of Income Taxes on Household
Income,” Review of Economics and Statistics, 81 (1999), 681–693.
Autor, David H., Mark Duggan, Kyle Greenberg, and David S. Lyle, “The Impact
of Disability Benefits on Labor Supply: Evidence from the VA’s Disability
Compensation Program,” American Economic Journal: Applied Economics,
8 (2016), 31–68.
Bailey, Martha, Hilary Hoynes, Maya Rossin-Slater, and Reed Walker, “Is the
Social Safety Net a Long-Term Investment? Large-Scale Evidence from the
Food Stamps Program,” 2019. Goldman School of Public Policy Working Paper,
2019.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1310
THE QUARTERLY JOURNAL OF ECONOMICS
Barnett, W. Steven, and Leonard N. Masse, “Comparative Benefit Cost Analysis
of the Abecedarian Program and Its Policy Implications,” Economics of Edu-
cation Review, 26 (2007), 113–125.
Bartik, Timothy J., Brad Hershbein, and Marta Lachowska, “The Merits of Univer-
sal Scholarships: Benefit-Cost Evidence from the Kalamazoo Promise,” Jour-
nal of Benefit-Cost Analysis, 7 (2016), 400–433.
Bartik, Timothy J., Brad J. Hershbein, and Marta Lachowska, “The Effects of
the Kalamazoo Promise Scholarship on College Enrollment, Persistence, and
Completion,” Journal of Human Resources, forthcoming.
Bastian, Jacob, and Maggie R. Jones, “Do EITC Expansions Pay for Themselves?
Effects on Tax Revenue and Public Assistance Spending,” Rutgers University
working paper, 2019.
Bastian, Jacob, and Katherine Michelmore, “The Long-Term Impact of the Earned
Income Tax Credit on Children’s Education and Employment Outcomes,” Jour-
nal of Labor Economics, 36 (2018), 1127–1163.
Bettinger, Eric, “How Financial Aid Affects Persistence,” in College Choices: The
Economics of Where to Go, When to go, and How to Pay For It, Caroline Hoxby,
ed. (Chicago: University of Chicago Press, 2004, 207–238.
Bettinger, Eric, Oded Gurantz, Laura Kawano, Bruce Sacerdote, and Michael
Stevens, “The Long-Run Impacts of Financial Aid: Evidence from California’s
Cal Grant,” American Economic Journal: Economic Policy, 11 (2019), 64–94.
Bettinger, Eric P., Bridget Terry Long, Philip Oreopoulos, and Lisa Sanbonmatsu,
“The Role of Application Assistance and Information in College Decisions: Re-
sults from the H&R Block Fafsa Experiment,” Quarterly Journal of Economics,
127 (2012), 1205–1242.
Bhargava, Saurabh, and Dayanand Manoli, “Psychological Frictions and the In-
complete Take-Up of Social Benefits: Evidence from an IRS Field Experiment,”
American Economic Review, 105 (2015), 3489–3529.
Black, Sandra E., Paul J. Devereux, and Kjell G. Salvanes, “From the Cradle to
the Labor Market? The Effect of Birth Weight on Adult Outcomes,” Quarterly
Journal of Economics, 122 (2007), 409–439.
Blank, Rebecca M., and Patricia Ruggles, “When Do Women Use Aid to Fami-
lies with Dependent Children and Food Stamps? The Dynamics of Eligibility
versus Participation,” Journal of Human Resources, 31 (1996), 57–89.
Bloom, Dan, James J. Kemple, Pamela Morris, Susan Scrivener, Nandita Verma,
Richard Hendra, Diana Adams-Ciardullo, and David Seith, et al., “Final Re-
port on Florida’s Initial Time-Limited Welfare Program,” Manpower Demon-
stration Research Corporation, 2000.
Bloom, Dan, Susan Scrivener, Charles Michalopoulos, Pamela Morris, Richard
Hendra, Diana Adams-Ciardullo, and Johanna Walter, “Jobs First: Final Re-
port on Connecticut’s Welfare Reform Initiative,” ERIC, 2002.
Bloom, Howard S., Larry L. Orr, Stephen H. Bell, George Cave, Fred Doolittle,
Winston Lin, and Johannes M. Bos, “The Benefits and Costs of JTPA Title
II-A Programs: Key Findings from the National Job Training Partnership Act
Study,” Journal of Human Resources, 32 (1997), 549–576.
Bloom, Howard S., James A. Riccio, and Nandita Verma, “Promoting Work in
Public Housing: The Effectiveness of Jobs-Plus,” Manpower Demonstration
Research Corporation, 2005.
Boardman, Anthony E, David H. Greenberg, Aidan R. Vining, and David L.
Weimer, Cost-Benefit Analysis: Concepts and Practice (Cambridge: Cambridge
University Press, 2017).
Boudreaux, Michel H., Ezra Golberstein, and Donna D. McAlpine, “The Long-
Term Impacts of Medicaid Exposure in Early Childhood: Evidence from the
Program’s Origin,” Journal of Health Economics, 45 (2016), 161–175.
Brock, Thomas, David Butler, and David Long, “Unpaid Work Experience for Wel-
fare Recipients: Findings and Lessons from MDRC Research. MDRC Working
Papers,” Manpower Demonstration Research Corporation, 1993.
Brown, David, Amanda E. Kowalski, and Ithai Z. Lurie, “Long-Term Impacts of
Childhood Medicaid Expansions on Outcomes in Adulthood,” NBER Working
Paper no. 20835, 2017.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1311
Brown, David W., Amanda E. Kowalski, and Ithai Z. Lurie, “Medicaid as an Invest-
ment in Children: What Is the Long-Term Impact on Tax Receipts?,” NBER
Working Paper no. 20835, 2015.
Bruce, Donald J., and Celeste K. Carruthers, “Jackpot? The Impact of Lottery
Scholarships on Enrollment in Tennessee,” Journal of Urban Economics, 81
(2014), 30–44.
Bulman, George B., and Caroline M. Hoxby, “The Returns to the Federal Tax
Credits for Higher Education,” Tax Policy and the Economy, 29 (2015), 13–88.
Cabral, Marika, and Neale Mahoney, “Externalities and Taxation of Supplemental
Insurance: A Study of Medicare and Medigap,” American Economic Journal:
Applied Economics, 11 (2019), 37–73.
Camerer, Colin F., Anna Dreber, Eskil Forsell, Teck-Hua Ho, J¨
urgen Huber, Mag-
nus Johannesson, Michael Kirchler, and Johan Almenberg et al., “Evaluating
Replicability of Laboratory Experiments in Economics,” Science, 351 (2016),
1433–1436.
Campbell, Frances A., Elizabeth P. Pungello, Kirsten Kainz, Margaret Burchinal,
Yi Pan, Barbara H. Wasik, Oscar A. Barbarin, Joseph J. Sparling, and Craig
T. Ramey, “Adult Outcomes as a Function of an Early Childhood Educational
Program: An Abecedarian Project Follow-Up,” Developmental Psychology, 48
(2012), 1033–1043.
Card, David, Andrew Johnston, Pauline Leung, Alexandre Mas, and Zhuan Pei,
“The Effect of Unemployment Benefits on the Duration of Unemployment
Insurance Receipt: New Evidence from a Regression Kink Design in Missouri,
2003–2013,” American Economic Review: Papers and Proceedings, 105 (2015),
126–130.
Carroll, Robert, “Do Taxpayers Really Respond to Changes in Tax Rates? Evi-
dence from the 1993 Tax Act,” U.S. Department of the Treasury Office of Tax
Analysis, Working Paper 79, 1998.
Carruthers, Celeste K., and Jilleah G. Welch, “Not Whether, but Where? Pell
Grants and College Choices,” Journal of Public Economics, 172 (2019), 1–19.
Castleman, Benjamin L., and Bridget Terry Long, “Looking beyond Enrollment:
The Causal Effect of Need-Based Grants on College Access, Persistence, and
Graduation,” Journal of Labor Economics, 34 (2016), 1023–1073.
Cave, George, Hans Bos, Fred Doolittle, and Cyril Toussaint, “Jobstart: Final Re-
port on a Program for School Dropouts,” Manpower Demonstration Research
Corporation, 1993.
Chetty, Raj, “Moral Hazard versus Liquidity and Optimal Unemployment Insur-
ance,” Journal of Political Economy, 116 (2008), 173–234.
Chetty, Raj, John N. Friedman, and Jonah Rockoff, “New Evidence on the Long-
Term Impacts of Tax Credits,” IRS Statistics of Income White Paper, 2011.
Chetty, Raj, Nathaniel Hendren, and Lawrence F. Katz, “The Effects of Expo-
sure to Better Neighborhoods on Children: New Evidence from the Mov-
ing to Opportunity Experiment,” American Economic Review, 106 (2016),
855–902.
Cohodes, Sarah R., and Joshua S. Goodman, “Merit Aid, College Quality, and Col-
lege Completion: Massachusetts’ Adams Scholarship as an In-Kind Subsidy,”
American Economic Journal: Applied Economics, 6 (2014), 251–285.
Cornwell, Christopher, Kyung Hee Lee, and David Mustard, “The Effects of Merit-
Based Financial Aid on Course Enrollment, Withdrawal and Completion in
College,” IZA Working Paper, 2003.
Cornwell, Christopher, David B. Mustard, and Deepa J. Sridhar, “The Enrollment
Effects of Merit-Based Financial Aid: Evidence from Georgia’s HOPE Schol-
arship,” Journal of Labor Economics, 24 (2006), 761–786.
Couch, Kenneth A., “New Evidence on the Long-Term Effects of Employment
Training Programs,” Journal of Labor Economics, 10 (1992), 380–388.
Crouse, Gilbert, and Annette Waters, “Welfare Indicators and Risk Factors: Thir-
teenth Report to Congress,” US Department of Health and Human Services,
2014.
Currie, Janet, “Welfare and the Well-Being of Children: The Relative Effectiveness
of Cash and In-Kind Transfers,” Tax Policy and the Economy, 8 (1994), 1–43.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1312
THE QUARTERLY JOURNAL OF ECONOMICS
Currie, Janet, and Nancy Cole, “Welfare and Child Health: The Link between
AFDC Participation and Birth Weight,” American Economic Review, 83 (1993),
971–985.
Currie, Janet, and Jonathan Gruber, “Saving Babies: The Efficacy and Cost of
Recent Changes in the Medicaid Eligibility of Pregnant Women,” Journal of
Political Economy, 104 (1996), 1263–1296.
Currie, Janet, and Enrico Moretti, “Did the Introduction of Food Stamps Affect
Birth Outcomes in California?,” National Poverty Center Working Paper 06-
20, 2006.
Cutler, David M., and Jonathan Gruber, “Does Public Insurance Crowd out Private
Insurance?,” Quarterly Journal of Economics, 111 (1996), 391–430.
Dahl, Gordon B., and Lance Lochner, “The Impact of Family Income on Child
Achievement: Evidence from the Earned Income Tax Credit,” American Eco-
nomic Review, 102 (2012), 1927–1956.
Dave, Dhaval M., Sandra L. Decker, Robert Kaestner, and Kosali Ilayperuma Si-
mon, “The Effect of Medicaid Expansions in the Late 1980s and Early 1990s
on the Labor Supply of Pregnant Women,” American Journal of Health Eco-
nomics, 1 (2015), 195–193.
DeLong, J. Bradford, Lawrence H. Summers, Martin Feldstein, and Valerie A.
Ramey, “Fiscal Policy in a Depressed Economy,” Brookings Papers on Economic
Activity (2012), 233–297.
Deming, David J., and Christopher R. Walters, “The Impact of Price Caps and
Spending Cuts on U.S. Postsecondary Attainment,” NBER Working Paper no.
23736, 2017.
Denning, Jeffrey T., “College on the Cheap: Consequences of Community College
Tuition Reductions,” American Economic Journal: Economic Policy, 9 (2017),
155–188.
Denning, Jeffrey T., Benjamin M. Marx, and Lesley J. Turner, “ProPelled: The
Effects of Grants on Graduation, Earnings, and Welfare,” American Economic
Journal: Applied Economics, 11 (2019), 193–224.
Deshpande, Manasi, “Does Welfare Inhibit Success? The Long-Term Effects of
Removing Low-Income Youth from the Disability Rolls,” American Economic
Review, 106 (2016), 3300–3330.
Diamond, Peter, and Emmanuel Saez, “The Case for a Progressive Tax: from Basic
Research to Policy Recommendations,” Journal of Economic Perspectives, 25
(2011), 165–190.
Dynarski, Susan, “Hope for Whom? Financial Aid for the Middle Class and
Its Impact on College Attendance,” National Tax Journal, 53 (2000),
629–662.
Dynarski, Susan, C. J. Libassi, Katherine Michelmore, and Stephanie Owen, “Clos-
ing the Gap: The Effect of a Targeted, Tuition-Free Promise on College Choices
of High-Achieving, Low-Income Students,” NBER Working Paper no. 25349,
2018.
Dynarski, Susan M., “Does Aid Matter? Measuring the Effect of Student Aid on
College Attendance and Completion,” American Economic Review, 93 (2003),
279–288.
East, Chloe N., “Immigrants’ Labor Supply Response to Food Stamp Access,”
Labour Economics, 51 (2018), 202–226.
Eissa, Nada, and Hilary W. Hoynes, “Taxes and the Labor Market Participation
of Married Couples: The Earned Income Tax Credit,” Journal of Public Eco-
nomics, 88 (2004), 1931–1958.
Eissa, Nada, and Jeffrey Liebman, “Labor Supply Responses to the Earned Income
Tax Credit,” Quarterly Journal of Economics, 111 (1996), 605–637.
Fein, David, and Jill Hamadyk, “Bridging the Opportunity Divide for Low-Income
Youth: Implementation and Early Impacts of the Year Up Program,” OPRE
Report 2018-65, 2018.
Finkelstein, Amy, Nathaniel Hendren, and Erzo F. P. Luttmer, “The Value of Med-
icaid: Interpreting Results from the Oregon Health Insurance Experiment,”
Journal of Political Economy, 127 (2019).
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1313
Finkelstein, Amy, Nathaniel Hendren, and Mark Shepard, “Subsidizing Health
Insurance for Low-Income Adults: Evidence from Massachusetts,” American
Economic Review, 109 (2019), 1530–1567.
Finkelstein, Amy, and Robin McKnight, “What Did Medicare Do? The Initial Im-
pact of Medicare on Mortality and Out of Pocket Medical Spending,” Journal
of Public Economics, 92 (2008), 1644–1668.
Finkelstein, Amy, and Matthew J. Notowidigdo, “Take-Up and Targeting: Exper-
imental Evidence from SNAP,” Quarterly Journal of Economics, 134 (2019),
1505–1556.
Finkelstein, Amy, Sarah Taubman, Bill Wright, Mira Bernstein, Jonathan Gruber,
Joseph P. Newhouse, Heidi Allen, and Katherine Baicker, and Oregon Health
Study Group, “The Oregon Health Insurance Experiment: Evidence from the
First Year,” Quarterly Journal of Economics, 127 (2012), 1057–1106.
Freedman, Stephen, Daniel Friedlander, Winston Lin, and Amanda Schweder,
“The GAIN Evaluation: Five-Year Impacts on Employment, Earnings
and
AFDC
Receipt,”
Manpower
Demonstration
Research
Corporation,
1996.
Freedman, Stephen, Jean Tansey Knab, Lisa A. Gennetian, and David Navarro,
“The Los Angeles Jobs-First GAIN Evaluation: Final Report on a Work First
Program in a Major Urban Center,” Manpower Demonstration Research Cor-
poration, 2000.
French, Eric, and Jae Song, “The Effect of Disability Insurance Receipt on Labor
Supply,” American Economic Journal: Economic Policy, 6 (2014), 291–337.
Ganong, Peter, and Pascal J. Noel, “Consumer Spending during Unemployment:
Positive and Normative Implications,” American Economic Review, 109 (2019),
2383–2424.
Garc´
ıa, Jorge Luis, James J. Heckman, Andres Hojman, Yu Kyung Koh, Joshua
Shea, and Anna Ziff, “Documentation of full ABC/CARE Treatment Effects,”
Unpublished Manuscript, University of Chicago, 2016.
Garc´
ıa, Jorge Luis, James J. Heckman, Duncan Ermini Leaf, and Mar´
ıa Jos´
e
Prados, “Quantifying the Life-Cycle Benefits of a Prototypical Early Childhood
Program,” NBER Working Paper no. 23479, 2017.
Gelber, Alexander, Timothy J. Moore, and Alexander Strand, “The Effect of Dis-
ability Insurance Payments on Beneficiaries’ Earnings,” American Economic
Journal: Economic Policy, 9 (2017), 229–261.
Goering, John, Joan Kraft, Judith Feins, Debra McInnis, Mary Joel Holin, and
Huda Elhassan, “Moving to Opportunity for Fair Housing Demonstration Pro-
gram: Current Status and Initial Findings,” U.S. Department of Housing and
Urban Development, 1999.
Gold, Rachel, and Asta Kenney, “Paying for Maternity Care,” Family Planning
Perspectives, 17 (1985), 103–111.
Goldrick-Rab, Sara, Robert Kelchen, Douglas N. Harris, and James Benson, “Re-
ducing Income Inequality in Educational Attainment: Experimental Evidence
on the Impact of Financial Aid on College Completion,” American Journal of
Sociology, 121 (2016), 1762–1817.
Goodman, Joshua, “Who Merits Financial Aid?: Massachusetts’ Adams Scholar-
ship,” Journal of Public Economics, 92 (2008), 2121–2131.
Goodman-Bacon,
Andrew,
“The
Long-Run
Effects
of
Childhood
Insurance
Coverage: Medicaid Implementation, Adult Health, and Labor Market
Outcomes,” NBER Working Paper no 22899, 2017.
Greenberg, David H., Victoria Deitch, and Gayle Hamilton, “A Synthesis of Ran-
dom Assignment Benefit-Cost Studies of Welfare-to-Work Programs,” Journal
of Benefit-Cost Analysis, 1 (2010), 1–30.
Grogger, Jeff, and Lynn A. Karoly, Welfare Reform (Cambridge, MA: Harvard
University Press, 2005).
Grogger, Jeffrey, “The Effects of Time Limits, the EITC, and Other Policy Changes
on Welfare Use, Work, and Income among Female-Headed Families,” Review
of Economics and Statistics, 85 (2003), 394–408.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1314
THE QUARTERLY JOURNAL OF ECONOMICS
Gruber, Jonathan, “The Consumption Smoothing Benefits of Unemployment In-
surance,” American Economic Review, 87 (1997), 192–205.
Hamilton, Gayle, Stephen Freedman, Lisa Gennetian, Charles Michalopoulos, Jo-
hanna Walter, Diana Adams-Ciardullo, Anna Gassman-Pines, and Sharon
McGroder et al., “How Effective Are Different Welfare-to-Work Approaches?
Five-Year Adult and Child Impacts for Eleven Programs. National Evaluation
of Welfare-to-Work Strategies,” Manpower Demonstration Research Corpora-
tion, 2001.
Heckman, James J., “Skill Formation and the Economics of Investing in Disad-
vantaged Children,” Science, 312 (2006), 1900–1902.
Heckman, James J., Seong Hyeok Moon, Rodrigo Pinto, Peter A. Savelyev, and
Adam Yavitz, “The Rate of Return to the High Scope Perry Preschool Program,”
Journal of Public Economics, 94 (2010), 114–128.
Heckman, James J., Rodrigo Pinto, Azeem M. Shaikh, and Adam Yavitz, “Inference
with Imperfect Randomization: The Case of the Perry Preschool Program,”
NBER Working Paper no. 16935, 2011.
Heim, Bradley T., “The Effect of Recent Tax Changes on Taxable Income: Evidence
from a New Panel of Tax Returns,” Journal of Policy Analysis and Manage-
ment, 28 (2009), 147–163.
Helburn, Suzanne W., “Cost, Quality and Child Outcomes in Child Care Cen-
ters. Technical Report, Public Report, and Executive Summary,” Manpower
Demonstration Research Corporation, 1995.
Hendra, Richard, David H. Greenberg, Gayle Hamilton, Ari Oppenheim, Alexan-
dra Pennington, Kelsey Schaberg, and Betsy L. Tessler, “Encouraging
Evidence on a Sector-Focused Advancement Strategy: Two-Year Impacts from
the Work Advance Demonstration,” Manpower Demonstration Research Cor-
poration, 2016.
Hendren, Nathaniel, “The Policy Elasticity,” Tax Policy and the Economy, 30 (2016),
51–89.
———, “Efficient Welfare Weights,” NBER Working Paper no. 20351, 2017a.
———, “Knowledge of Future Job Loss and Implications for Unemployment In-
surance,” American Economic Review, 107 (2017b), 1778–1823.
———, “Measuring Ex-Ante Welfare in Insurance Markets,” NBER Working Paper
no. 24470, 2017c.
Hendren, Nathaniel, and Ben Sprung-Keyser, “Replication Data for: ‘A Unified
Welfare Analysis of Government Policies’,” (2020), Harvard Dataverse, doi:
10.7910/DVN/ZHOSGC.
Hoekstra, Mark, “The Effect of Attending the Flagship State University on Earn-
ings: A Discontinuity-Based Approach,” The Review of Economics and Statis-
tics, 91 (2009), 717–724.
Hollister, Robinson G., Peter Kemper, and Rebecca A. Maynard, “The National
Supported Work Demonstration,” 1984.
Hotz, V. Joseph, and John K. Scholz, “The Earned Income Tax Credit,” in Means-
Tested Transfer Programs in the United States, Robert A. Moffitt, ed. (Chicago:
University of Chicago Press, 2003), 141–198.
Hoxby, Caroline M., and George B. Bulman, “The Effects of the Tax Deduction
for Postsecondary Tuition: Implications for Structuring Tax-Based Aid,” Eco-
nomics of Education Review, 51 (2016), 23–60.
Hoynes, Hilary W., Marianne Page, and Ann Huff Stevens, “Can Targeted Trans-
fers Improve Birth Outcomes?: Evidence from the Introduction of the WIC
Program,” Journal of Public Economics, 95 (2011), 813–827.
Hoynes, Hilary W., and Ankur J. Patel, “Effective Policy for Reducing Poverty and
Inequality? The Earned Income Tax Credit and the Distribution of Income,”
Journal of Human Resources, 53 (2018), 859–890.
Hoynes, Hilary W., and Diana W. Schanzenbach, “Consumption Responses to In-
Kind Transfers: Evidence from the Introduction of the Food Stamp Program,”
American Economic Journal: Economic Policy, 1 (2009), 109–139.
———, “Work Incentives and the Food Stamp Program,” Journal of Public Eco-
nomics, 96 (2012), 151–162.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1315
———, “Safety Net Investments in Children,” Brookings Papers on Economic
Activity (2018), 89–133.
Hoynes, Hilary W., Diane Whitmore Schanzenbach, and Douglas Almond, “Long-
Run Impacts of Childhood Access to the Safety Net,” American Economic
Review, 106 (2016), 903–934.
Hylland, Aanund, and Richard Zeckhauser, “Distributional Objectives Should Af-
fect Taxes but not Program Choice or Design,” in Measurement in Public
Choice, Steinar Strøm, ed. (London: Palgrave Macmillan, 1981), 123–143.
Hyman, Joshua, “Does Money Matter in the Long Run? Effects of School Spending
on Educational Attainment,” American Economic Journal: Economic Policy, 9
(2017), 256–280.
Jackson, C. Kirabo, Claudia Persico, and Rucker C. Johnson, “The Effects of
School Spending on Educational and Economic Outcomes: Evidence from
School Finance Reforms,” Quarterly Journal of Economics, 131 (2016),
157–218.
Jacob, Brian A., and Jens Ludwig, “The Effects of Housing Assistance on Labor
Supply: Evidence from a Voucher Lottery,” American Economic Review, 102
(2012), 272–304.
Jacob, Brian A., Jens Ludwig, and Max Kapustin, “The Impact of Housing As-
sistance on Child Outcomes: Evidence from a Randomized Housing Lottery,”
Quarterly Journal of Economics, 130 (2014), 465–506.
Johnson, Rucker C., and C. Kirabo Jackson, “Reducing Inequality through Dy-
namic Complementarity: Evidence from Head Start and Public School Spend-
ing,” American Economic Journal: Economic Policy, 11 (2019), 310–349.
Johnston, Andrew C., and Alexandre Mas, “Potential Unemployment Insurance
Duration and Labor Supply: The Individual and Market-Level Response to a
Benefit Cut,” Journal of Political Economy, 126 (2018), 2480–2522.
Jones, Damon, and Ioana Marinescu, “The Labor Market Impacts of Universal
and Permanent Cash Transfers: Evidence from the Alaska Permanent Fund,”
NBER Working Paper no. 24312, 2018.
Kane, Thomas J., “College Entry by Blacks since 1970: The Role of College Costs,
Family Background, and the Returns to Education,” Journal of Political Econ-
omy, 102 (1994), 878–911.
Katz, Lawrence F., and Bruce D. Meyer, “The Impact of the Potential Duration
of Unemployment Benefits on the Duration of Unemployment,” Journal of
Public Economics, 41 (1990), 45–72.
Kawano, Laura, Caroline Weber, and Andrew Whitten, “Estimating the Elas-
ticity of Broad Income for High-Income Taxpayers,” 2016, https://papers.
ssrn.com/sol3/papers.cfm?abstract_id=2852048 (accessed July 7, 2019).
Kemple, James J., Daniel Friedlander, and Veronica Fellerath, “Florida’s Project
Independence. Benefits, Costs, and Two-Year Impacts of Florida’s JOBS Pro-
gram,” ERIC, 1995.
Kleven, Henrik, “The EITC and the Extensive Margin: A Reappraisal,” NBER
Working Paper no. 26405, 2019.
Kleven, Henrik J., and Claus Thustrup Kreiner, “The Marginal Cost of Public
Funds: Hours of Work versus Labor Force Participation,” Journal of Public
Economics, 90 (2006), 1955–1973.
Kline, Patrick, and Christopher R. Walters, “Evaluating Public Programs with
Close Substitutes: The Case of Head Start,” Quarterly Journal of Economics,
131 (2016), 1795–1848.
Kroft, Kory, and Matthew J. Notowidigdo, “Should Unemployment Insurance Vary
with the Unemployment Rate? Theory and Evidence,” Review of Economic
Studies, 83 (2016), 1092–1124.
LaLumia, Sara, “Tax Preferences for Higher Education and Adult College Enroll-
ment,” National Tax Journal, 65 (2012), 59–92.
Landais, Camille, “Assessing the Welfare Effects of Unemployment Benefits Using
the Regression Kink Design,” American Economic Journal: Economic Policy,
7 (2015), 243–278.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1316
THE QUARTERLY JOURNAL OF ECONOMICS
Lo Sasso, Anthony T., and Dorian G. Seamster, “How Federal and State Policies
Affected Hospital Uncompensated Care Provision in the 1990s,” Medical Care
Research and Review, 64 (2007), 731–744.
Long, Bridget, “The Impact of Federal Tax Credits for Higher Education Expenses,”
in College Choices: The Economics of Where to Go, When to Go, and How to Pay
for It, Caroline Hoxby, ed. (Chicago: The University of Chicago Press, 2004),
101–168.
Ludwig, Jens, and Douglas L. Miller, “Does Head Start Improve Children’s Life
Chances? Evidence from a Regression Discontinuity Design,” Quarterly Jour-
nal of Economics, 122 (2007), 159–208.
Maestas, Nicole, Kathleen J. Mullen, and Alexander Strand. “Does Disability In-
surance Receipt Discourage Work? Using Examiner Assignment to Estimate
Causal Effects of SSDI Receipt,” American Economic Review, 103 (2013),
1797–1829.
Manoli, Day, and Nicholas Turner, “Cash-on-hand and College Enrollment: Evi-
dence from Population Tax Data and the Earned Income Tax Credit,” American
Economic Journal: Economic Policy, 10 (2018), 242–271.
Marx, Benjamin M., and Lesley J. Turner, “Borrowing Trouble? Human Capital
Investment with Opt-In Costs and Implications for the Effectiveness of Grant
Aid,” American Economic Journal: Applied Economics, 10 (2018), 163–201.
Masse, Leonard N., “A Benefit Cost Analysis of the Carolina Abecedar-
ian
Preschool
Program,”
2003,
https://www.minneapolisfed.org/∼/media/
files/publications/studies/earlychild/2003conf/barnettdoc.doc?la=en.
Masse, Leonard N., and W. Steven Barnett, “A Benefit Cost Analysis of the
Abecedarian Early Childhood Intervention,” National Institute for Early Ed-
ucation Research working paper, 2002.
Maxfield, Michelle, “The Effects of the Earned Income Tax Credit on Child Achieve-
ment and Long-Term Educational Attainment,” Michigan State University
Working Paper, 2013.
Mayshar, Joram, “On Measures of Excess Burden and Their Applications,” Journal
of Public Economics, 43 (1990), 263–289.
Meyer, Bruce D., and Wallace K. C. Mok, “Quasi-Experimental Evidence on the
Effects of Unemployment Insurance from New York State,” NBER Working
Paper no. 12865, 2007.
Meyer, Bruce D., and Dan T. Rosenbaum, “Welfare, the Earned Income Tax Credit,
and the Labor Supply of Single Mothers,” Quarterly Journal of Economics, 116
(2001), 1063–1114.
Michelmore, Katherine, “The Effect of Income on Educational Attainment:
Evidence
from
State
Earned
Income
Tax
Credit
Expansions,”
2013,
https://papers.ssrn.com/sol3/papers.cfm?abstract id=2356444.
Miller, Cynthia, Lawrence F. Katz, Gilda Azurdia, Adam Isen, and Caroline B.
Schultz, “Expanding the Earned Income Tax Credit for Workers without De-
pendent Children: Interim Findings from the Paycheck Plus Demonstration
in New York City,” MDRC, 2017.
Miller, Cynthia, Virginia Knox, Lisa A. Gennetian, Martey Dodoo, Jo Anna Hunter,
and Cindy Redcross, “Reforming Welfare and Rewarding Work: Final Report
on the Minnesota Family Investment Program. Vol. 1: Effects on Adults and
Volume 2: Effects on Children,” Manpower Demonstration Research Corpora-
tion, 2000.
Miller, Sarah, and Laura R. Wherry, “The Long-Term Effects of Early Life Medi-
caid Coverage,” Journal of Human Resources, 54 (2019), 785–824.
Mills, Gregory, Daniel Gubits, Larry Orr, David Long, Judie Feins, Bulbul Kaul,
Michelle Wood, and Amy Jones et al., “Effects of Housing Vouchers on Welfare
Families,” U.S. Department of Housing and Urban Development, Office of
Policy Development and Research, 2006.
Mirrlees, James A., “An Exploration into the Theory of Optimal Income Taxation,”
Review of Economic Studies, 38 (1971), 175–208.
———, “Optimal Tax Theory: A Synthesis,” Journal of Public Economics, 6 (1976),
327–358.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1317
Moffitt, Robert A., “Welfare Programs and Labor Supply,” Handbook of Public
Economics, 4 (2002), 2393–2430.
———, Means-Tested Transfer Programs in the United States (Chicago: University
of Chicago Press, 2003).
Okun, Arthur M., Equality and Efficiency (Washington, DC: Brookings Institution
Press, 1975).
Pavetti, LaDonna, “Who is Affected by Time Limits?” in Welfare Reform: An Anal-
ysis of the Issues, Isabel V. Sawhill, ed. (Washington, DC: The Urban Institute,
1995), 31–34.
Price, David J., and Jae Song, “The Long-Term Effects of Cash Assistance,” Work-
ing Paper, 2018.
Rea, David, and Tony Burton, “New Evidence on the Heckman Curve,” Journal of
Economic Surveys, 34 (2020), 241–262.
Reynolds, Arthur J., Judy A. Temple, Dylan L. Robertson, and Emily A. Mann,
“Age 21 Cost-Benefit Analysis of the Title I Chicago Child-Parent Centers,”
Educational Evaluation and Policy Analysis, 24 (2002), 267–303.
Reynolds, Arthur J., Judy A. Temple, Suh-Ruu Ou, Irma A. Arteaga, and Barry
A.B. White, “School-Based Early Childhood Education and Age-28 Well-Being:
Effects by Timing, Dosage, and Subgroups,” Science, 333 (2011), 360–364.
Riccio, James A., “Jobs–Plus: A Promising Strategy for Increasing Employment
and Self–Sufficiency among Public Housing Residents,” Technical Report, pre-
sented before the Subcommittee on Federalism and the Census, House Com-
mittee on Government Reform, 2006.
Saez, Emmanuel, “Using Elasticities to Derive Optimal Income Tax Rates,” Review
of Economic Studies, 68 (2001), 205–229.
———, “The Effect of Marginal Tax Rates on Income: A Panel Study of ‘Bracket
Creep’,” Journal of Public Economics, 87 (2003), 1231–1258.
Saez, Emmanuel, Joel Slemrod, and Seth H. Giertz, “The Elasticity of Taxable
Income with Respect to Marginal Tax Rates: A Critical Review,” Journal of
Economic Literature, 50 (2012), 3–50.
Sanbonmatsu, Lisa, Lawrence F. Katz, Jens Ludwig, Lisa A. Gennetian, Greg J.
Duncan, Ronald C. Kessler, Emma K. Adam, and Thomas McDade et al., “Mov-
ing to Opportunity for Fair Housing Demonstration Program: Final Impacts
Evaluation,” U.S. Department of Housing and Urban Development, 2011.
Schaberg, Kelsey, “Can Sector Strategies Promote Longer-Term Effects? Three-
Year Impacts from the WorkAdvance Demonstration,” Manpower Demonstra-
tion Research Corporation, 2017.
Schmieder, Johannes F., and Till Von Wachter, “The Effects of Unemployment In-
surance Benefits: New Evidence and Interpretation,” Annual Review of Eco-
nomics, 106 (2016), 547–581.
Schochet, Peter Z., “National Job Corps Study: 20-Year Follow-Up Study Using
Tax Data,” Mathematica Policy Research Report, 2018.
Schochet, Peter Z., John Burghardt, and Sheena McConnell, “Does Job Corps
Work? Impact Findings from the National Job Corps Study,” American Eco-
nomic Review, 98 (2008), 1864–1886.
Schochet, Peter Z., John A. Burghardt, and Sheena M. McConnell et al., “National
Job Corps Study and Longer-Term Follow-Up Study: Impact and Benefit-Cost
Findings Using Survey and Summary Earnings Records Data,” U.S. Depart-
ment of Labor, Employment and Training Administration, 2006.
Scholz, John Karl, “The Earned Income Tax Credit: Participation, Compliance,
and Antipoverty Effectiveness,” Institute for Research on Poverty Discussion
Papers 1020-93, 1993.
Scrivener, Susan, Richard Hendra, Cindy Redcross, Dan Bloom, Charles
Michalopoulos, and Johanna Walter, “WRP: Final Report on Vermont’s Wel-
fare Restructuring Project, Manpower Demonstration Research Corporation,
2002.
Seftor, Neil S., and Sarah E. Turner, “Back to School: Federal Student Aid Policy
and Adult College Enrollment,” Journal of Human Resources, (2002), 336–352.
Slemrod, Joel, and Shlomo Yitzhaki, “The Social Cost of Taxation and the Marginal
Cost of Funds,” International Monetary Fund Staff Papers, 43 (1996), 172–198.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1318
THE QUARTERLY JOURNAL OF ECONOMICS
———, “Integrating Expenditure and Tax Decisions: The Marginal Cost of Funds
and the Marginal Benefit of Projects,” National Tax Journal, 54 (2001), 189–
202.
Solon, Gary, “Work Incentive Effects of Taxing Unemployment Benefits,” Econo-
metrica, 53 (1985), 295–306.
Stiglitz, Joseph E., and Parthaa Dasgupta, “Differential Taxation, Public Goods,
and Economic Efficiency,” Review of Economic Studies, 38 (1971), 151–174.
Tax Foundation, “U.S. Federal Individual Income Tax Rates History, 1862–
2013,”
2013,
https://taxfoundation.org/us-federal-individual-income-tax-
rates-history-1913-2013-nominal-and-inflation-adjusted-brackets/.
Tax Policy Center, “Earned Income Tax Credit Parameters, 1975–2016,” 2016,
https://www.taxpolicycenter.org/sites/default/files/legacy/taxfacts/content/pdf/
historical eitc parameters.pdf.
Turner, Nicholas, “The Effect of Tax-Based Federal Student Aid on College En-
rollment,” National Tax Journal, 64 (2011), 839–861.
U.S. Census Bureau, “Current Population Reports: Consumer Income,” 1996,
https://www2.census.gov/prod2/popscan/p60-049.pdf.
U.S. Department of Education Office of Postsecondary Education, “2009–
2010 Federal Pell Grant Program End-of-Year Report,” 2010, https://
www2.ed.gov/finaid/prof/resources/data/pell-2009-10/pell-eoy-09-10.pdf.
U.S. Department of Health & Human Services, “The Final Report of the
Seattle-Denver Income Maintenance Experiment,” 1983, https://aspe.hhs.gov/
report/overview-final-report-seattle-denver-income-maintenance-experiment.
U.S. Social Security Administration, “SSI Annual Statistical Report, 2013,” SSA
Publication No. 13-11827, 2014.
——–, “Annual Statistical Supplement to the Social Security Bulletin, 2017,” Pub-
lication No. 13-11700, 2018.
Von Wachter, Till, Jae Song, and Joyce Manchester, “Trends in Employment and
Earnings of Allowed and Rejected Applicants to the Social Security Disability
Insurance Program,” American Economic Review, 101 (2011), 3308–3329.
Weimer, David (ed.), Cost-Benefit Analysis and Public Policy, vol. 1 (New York:
John Wiley & Sons, 2009).
Weimer, David L., and Aidan R. Vining (eds.), Investing in the Disadvantaged
(Washington, DC: Georgetown University Press, 2009).
Werning, Ivan, “Optimal Fiscal Policy with Redistribution,” Quarterly Journal of
Economics, 122 (2007), 925–967.
Wherry, Laura R., and Bruce D. Meyer, “Saving Teens: Using a Policy Discon-
tinuity to Estimate the Effects of Medicaid Eligibility,” Journal of Human
Resources, 51 (2016), 556–588.
Wherry, Laura R., Sarah Miller, Robert Kaestner, and Bruce D. Meyer, “Child-
hood Medicaid Coverage and Later-Life Health Care Utilization,” Review of
Economics and Statistics, 100 (2018), 287–302.
Whitmore, Diane, “What Are Food Stamps Worth?,” Princeton University Indus-
trial Relations Section Working Paper no. 468, 2002.
Wood, Michelle, Jennifer Turnham, and Gregory Mills, “Housing Affordability and
Family Well-Being: Results from the Housing Voucher Evaluation,” Housing
Policy Debate, 19 (2008), 367–412.
WSIPP, “Benefit-Cost Technical Documentation,” Washington State Institute for
Public Policy Technical Report, 2019.
Zimmerman, Seth D., “The Returns to College Admission for Academically
Marginal Students,” Journal of Labor Economics, 32 (2014), 711–754.

THE
QUARTERLY JOURNAL
OF ECONOMICS
Vol. 135
2020
Issue 3
A UNIFIED WELFARE ANALYSIS OF GOVERNMENT
POLICIES∗
NATHANIEL HENDREN AND BEN SPRUNG-KEYSER
We conduct a comparative welfare analysis of 133 historical policy changes
over the past half-century in the United States, focusing on policies in social in-
surance, education and job training, taxes and cash transfers, and in-kind trans-
fers. For each policy, we use existing causal estimates to calculate the benefit
that each policy provides its recipients (measured as their willingness to pay)
and the policy’s net cost, inclusive of long-term effects on the government’s bud-
get. We divide the willingness to pay by the net cost to the government to form
each policy’s Marginal Value of Public Funds, or its “MVPF”. Comparing MVPFs
across policies provides a unified method of assessing their effect on social welfare.
Our results suggest that direct investments in low-income children’s health and
∗We first and foremost thank the several hundred researchers whose em-
pirical results form the foundation of our estimates. We are deeply indebted to
a wonderful team of research assistants: Caroline Dockes, Harris Eppsteiner,
Adriano Fernandes, Jack Hoyle, Omeed Maghzian, Kate Musen, Nicolaj Thor, and
the rest of the exceptional team of Pre-Doctoral Fellows at Opportunity Insights.
We are also grateful to Raj Chetty, David Deming, Winnie van Dijk, Amy Finkel-
stein, John Friedman, Andrew Goodman-Bacon, Jeff Grogger, Hilary Hoynes, John
Eric Humphries, Larry Katz, Sarah Miller, Evan Soltas, Larry Summers, Michael
Stepner, and Laura Wherry for helpful comments and suggestions, along with sem-
inar participants at the University of Chicago, Georgetown, IFS, the University
of Kentucky, LSE, Michigan, Minnesota, and Texas A&M, along with confer-
ence participants at the NBER and the National Tax Association meetings. This
research was funded by the National Science Foundation (#CAREER1653686
(Hendren) and #DGE1745303 (Sprung-Keyser)), the Sloan Foundation (Hendren),
the Bill & Melinda Gates Foundation (Hendren), and the Chan Zuckerberg Ini-
tiative (Hendren). Any opinions, findings, and conclusions or recommendations
expressed in this material are those of the author(s) and do not necessarily reflect
the views of the National Science Foundation.
C
⃝The Author(s) 2020. Published by Oxford University Press on behalf of President and
Fellows of Harvard College. This is an Open Access article distributed under the terms
of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/),
which permits unrestricted reuse, distribution, and reproduction in any medium, provided the
original work is properly cited.
The Quarterly Journal of Economics (2020), 1209–1318. doi:10.1093/qje/qjaa006.
Advance Access publication on March 5, 2020.
1209
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1210
THE QUARTERLY JOURNAL OF ECONOMICS
education have historically had the highest MVPFs, on average exceeding 5. Many
such policies have paid for themselves as the government recouped the cost of their
initial expenditures through additional taxes collected and reduced transfers. We
find large MVPFs for education and health policies among children of all ages,
rather than observing diminishing marginal returns throughout childhood. We
find smaller MVPFs for policies targeting adults, generally between 0.5 and 2. Ex-
penditures on adults have exceeded this MVPF range in particular if they induced
large spillovers on children. We relate our estimates to existing theories of optimal
government policy, and we discuss how the MVPF provides lessons for the design
of future research. JEL Codes: H00, I00, J24.
I. INTRODUCTION
What government expenditures are most effective at improv-
ing social well-being? Are in-kind transfers preferable to cash
transfers? Does government-provided social insurance efficiently
address market failures? Should we invest more in low-income
children? If so, at what age? Should they be direct investments
or subsidies to parents?
A large empirical literature estimates the causal effects of
historical government policies. These papers frequently conclude
with a brief welfare analysis. The method of that analysis, how-
ever, often differs from paper to paper. When reporting the ef-
fects of health insurance expansions, it is common to report cost
per life saved (e.g., Currie and Gruber 1996). Studies of tax pol-
icy changes often report the implied marginal excess burden or
the marginal cost of funds (e.g., summarized in Saez, Slemrod,
and Giertz 2012). Higher education analyses often report the cost
per enrollment (e.g., Kane 1994; Dynarski 2000). The early child-
hood education literature often reports a social benefit-cost ra-
tio (e.g., Heckman et al. 2010). These varying welfare measures
make it difficult to compare policies, especially if one wishes to
take a bird’s-eye view and perform welfare analysis across policy
categories.
This article conducts a comparative welfare analysis of
133 historical tax and expenditure policies implemented in the
United States over the past half-century. We focus on policies in
four domains: social insurance (e.g., health, unemployment, and
disability insurance), education (e.g., preschool, K–12, college, job
and vocational training), taxes and cash transfers (e.g., top tax
rates, Earned Income Tax Credit (EITC), Aid to Families with
Dependent Children (AFDC)), and in-kind transfers (e.g., housing
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1211
vouchers, food stamps). We draw on existing analyses of the
impacts of these policies to construct the benefit that each policy
provides to its recipients and the policy’s net cost to the govern-
ment. Benefits are captured by the willingness to pay of policy
recipients. The net cost combines both initial program spending
and the long-run effect of the policy on the government’s budget
(i.e., fiscal externalities). We then take the ratio of the benefits to
net government costs to generate each policy’s marginal value of
public funds (MVPF).1 Putting these components together allows
us to measure each policy’s “bang for the buck.”2
The MVPF is useful because it measures the amount of wel-
fare that can be delivered to policy beneficiaries per dollar of gov-
ernment spending on the policy. Equivalently, the MVPF mea-
sures the shadow price of raising revenue from the beneficiaries
of the policy by reducing spending on the policy. For point of refer-
ence, a simple nondistortionary transfer from the government to
an individual would have an MVPF of 1. The cost to the govern-
ment would be exactly equal to the individual beneficiary’s willing-
ness to pay. The MVPF can differ from this benchmark value of 1 if
individuals value an expenditure at more or less than its resource
cost. For instance, if the government provides insurance, willing-
ness to pay may be greater than the resource costs of provision
to individuals if the insurance provides consumption-smoothing
benefits. By contrast, willingness to pay may fall below resource
costs if individuals distort their behavior to receive higher trans-
fers.3 The MVPF may also deviate from the benchmark value of 1
if the policy induces fiscal externalities. For example, if spending
a dollar on a government policy caused individuals to work less,
government tax revenue might fall slightly and then the net cost
of the policy would rise above $1. By contrast, if spending that dol-
lar caused them to get more schooling and consequently increased
1. See Mayshar (1990), Slemrod and Yitzhaki (1996, 2001), and Kleven and
Kreiner (2006) for original definitions, and Hendren (2016) for a comparison of the
MVPF to alternative measures of welfare.
2. In several cases where authors constructed their own MVPFs, we incor-
porate those estimates directly. Where applicable, we adjust these estimates to
harmonize assumptions (e.g., discount rates). In cases where previous literature
has conducted comprehensive cost-benefit analyses of a policy, we draw on the
components of those analyses to reformulate them into their implied MVPF.
3. The intuition here comes from the envelope theorem. Willingness to pay
for a government transfer is determined by the “mechanical cost” of that transfer.
Additional costs due to behavioral responses are not valued dollar for dollar.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1212
THE QUARTERLY JOURNAL OF ECONOMICS
their income, government revenue would rise and the net cost of
the policy would fall below $1. In some cases, positive fiscal exter-
nalities may be large enough to fully offset the initial cost of the
policy. In that instance, the policy has an infinite MVPF, and conse-
quently, spending on the policy results in a Pareto improvement.4
More generally, comparisons of MVPFs correspond to precise
statements about social welfare using the intuition of Okun’s
leaky bucket experiment (Okun 1975). Given two policies, A and
B, suppose MVPFA = 2 and MVPFB = 1. Then one prefers more
spending on policy A financed by less spending on policy B if and
only if one prefers giving $2 to policy A beneficiaries over giving
$1 to policy B beneficiaries. Whether this is desirable ultimately
depends on one’s social preferences for the beneficiaries of
policies A and B. MVPFs measure the feasible trade-offs to the
government—in Okun’s metaphor, the “leaks” in the bucket. By
measuring these shadow prices of raising revenue from different
groups, the MVPF provides a unified method of welfare analysis
that can be applied both across and within diverse policy domains.
We outline the construction of the MVPF for six represen-
tative examples in Section III. At a high level, our construction
of willingness to pay often relies on intuition provided by the
envelope theorem. Our construction of net government costs
involves calculating changes in taxes paid and transfers received,
along with savings or additional costs from crowding out of other
government spending. In Online Appendices A–F we also provide
a detailed explanation of how each MVPF in our sample is
calculated. As is common with any welfare analysis, the creation
of our MVPFs requires various judgment calls. We conduct an
extensive set of robustness analyses, examining our assumptions
about interest rates, tax rates, and forecasting methods.5 In
4. To align with terminology in existing literature, we use various terms inter-
changeably to refer to the same phenomenon. Any policy with a positive willing-
ness to pay and negative net costs we define to have an infinite MVPF. Given the
negative net costs, we also say that these policies “pay for themselves” or “recoup
their initial costs.” In the taxation literature, this is also known as a Laffer effect.
We often note that spending on policies with infinite MVPFs results in a Pareto
improvement. This is because the expenditure is valued by beneficiaries and has
no net cost on the government. This final claim regarding Pareto improvement
formally assumes that all beneficiaries have positive willingness to pay, which is
natural in many of our contexts in which the policies expanded the choice sets of
all beneficiaries.
5. We also provide a Stata do-file for each program that is available on GitHub
(https://github.com/Opportunitylab/welfare analysis).
These
programs
allow
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1213
addition, many MVPF estimates for individual policies contain
considerable sampling uncertainty. We address this by construct-
ing category averages that pool across multiple policies and help
improve the precision of our conclusions. We also test and correct
for publication bias using the methods of Andrews and Kasy
(2019). Given these potential sources of uncertainty, we also focus
our results on broad patterns in the data, rather than conclusions
about individual policies.
Our analysis is inevitably constrained by the scope of existing
literature. Not all policies have been studied with the same degree
of completeness. For each policy, we incorporate all effects that
can reliably be translated into the MVPF, but an omitted impact
could affect our welfare analysis. We therefore assess the robust-
ness of our broad patterns to sample restrictions focused on more
comprehensively studied policies. In addition, we discuss how the
MVPF of each particular policy may vary with the addition (or
removal) of certain effects.6 For example, we find that our MVPF
estimates are most sensitive to changes in the estimated earnings
of beneficiaries—specifically dynamic effects within or across
generations. In the results we discuss below, we focus our primary
conclusions on the broad lessons that are robust to variations in
the availability of estimates on underlying causal estimates.
I.A. Main Results
Our estimates reveal a stark pattern: MVPFs vary sub-
stantially based on the age of each policy’s beneficiaries. We
find the highest MVPFs for direct investments in the health
and education of low-income children. This includes Medicaid
expansions, childhood education spending, and expenditures on
college. In many cases, these policies actually pay for themselves
in the long run. Children pay back the initial cost as adults
through additional tax revenue and reduced transfer payments.
For example, we examine four major health insurance expansions
to children over the past 50 years. We calculate an average across
those policies and find that for each $1 of initial expenditure they
repaid $1.78 back to the government in the long run. In particular,
we find that three of four policies fully repaid their initial costs.
researchers to easily modify the set of input assumptions into each MVPF
beyond the robustness we readily provide in the article and the Online Appendix.
6. We provide an extended discussion of these in the Online Appendix.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1214
THE QUARTERLY JOURNAL OF ECONOMICS
We find high MVPFs for policies targeting children through-
out childhood. We do find high MVPFs for early childhood
education programs, including an MVPF of roughly 44 for Perry
Preschool and 12 for Abecedarian.7 In addition, we find large
MVPFs for policies targeting older children, such as historical
equalizations in K–12 school financing (studied in Jackson,
Persico, and Johnson 2016) and policies increasing college
attainment. Our broad patterns contrast with the notion that
opportunities for high-return investment in children decline
rapidly with age (Heckman 2006).
Our results show lower MVPFs for policies targeted to adults.
Most of these MVPFs lie between 0.5 and 2. For example, we find
MVPFs ranging from 0.40–1.63 for health insurance expansions
to adults, 0.65–1.04 for in-kind transfers such as housing vouchers
and food stamps, and from negative values to 1.20 for tax credits
and cash welfare programs to low-income households. These lower
MVPFs reflect the fact that spending on many of these policies
reduced labor earnings. This stands in contrast to our finding that
many policies spending on children increased later-life earnings.
It is important to note that these differences in returns by age
represent general patterns but do not hold uniformly. There are a
number of exceptions. For child policies, we find large variation in
MVPFs across policies, with some estimates relatively close to 1.
In particular, we find lower MVPFs for job training programs and
for college subsidies that do not lead to increases in attainment.
We also find lower MVPFs for transfers to disabled children
and their families. This latter case illustrates that policies with
lower MVPFs are not necessarily “undesirable”—they can be
welfare enhancing depending on one’s social preferences. Unlike
expenditures with infinite MVPFs, policies with low MVPFs
involve a budgetary trade-off that should be weighed against
one’s preference for redistribution.
Among expenditures on adults, we find relatively large
MVPFs for reductions in top marginal tax rates, with estimates
7. In our baseline specifications that harmonize government revenue compo-
nents across policies, we estimate that the government recoups 92% of the up-front
cost of Perry Preschool and 78% of the cost of Abecedarian. Because the cost of
crime impacts are often difficult to quantify, they are not included in our base-
line analyses (when crime estimates are available, we we incorporate them in
alternative specifications discussed in the Online Appendix for each policy). In
this case, if one includes additional estimated effects such as the cost of crime, we
estimate that Perry Preschool does pay for itself and Abecedarian pays for 92% of
the up-front cost.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1215
from 1.16 to infinity. There is, however, substantial sampling
uncertainty in these estimates.8 We also find high MVPFs for
spending on adults that generates spillover effects on children.
For example, providing vouchers with counseling services to
families residing in high-poverty public housing (as part of the
Moving to Opportunity Experiment) helped these families move
to lower-poverty neighborhoods. This led to large increases in
children’s earnings in adulthood that generated sufficient tax rev-
enue to pay for the program cost. Our results highlight the value
of further work to uncover when such spillovers are likely to occur.
I.B. Relation to Previous Theories
The ratio of MVPFs measures the extent to which the govern-
ment can transfer welfare across individuals in society. For this
reason, it relates to the literature on optimal government policy
and redistribution (e.g., Mirrlees 1971, 1976). After presenting
our results, we interpret them in light of this theory. For example,
we tend to find tax cuts to top earners have higher MVPFs than
cuts targeted to low-income households, a result consistent with
the behavior of a progressive planner setting the tax rate in a
Mirrleesian optimal tax model (Mirrlees 1971, 1976). We also
compare the MVPFs of cash transfers to those of in-kind trans-
fers, testing the applicability of the Atkinson-Stiglitz theorem
(Atkinson and Stiglitz 1976; Hylland and Zeckhauser 1981).
I.C. Implications for Future Research
We conclude by providing three lessons for future research.
First, we show how the MVPF framework allows us to quantify
the value of such research. Because the MVPF is a shadow price,
one can use a standard decision-theoretic framework to quantify
the value of reducing uncertainty in our MVPF estimates. Just
as a consumer would be willing to pay to learn the true value of
the products he or she buys, a welfare-maximizing government
should be willing to pay to reduce uncertainty in the cost of
redistribution. Using this approach, we show that a welfare-
maximizing government deciding whether to raise taxes to spend
an additional $1 on the Supplemental Nutrition Assistance
Program (SNAP) would be willing to pay $0.24 to make this
decision using a more precise causal estimate of the long-run
8. For example, we estimate an infinite MVPF for the 1981 reduction in the
top marginal income tax rate from 70% to 50%. Our confidence interval, however,
includes both 1 and infinity.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1216
THE QUARTERLY JOURNAL OF ECONOMICS
impact of SNAP using administrative data (as in Bailey et al.
2019) as opposed to survey data (as in Hoynes, Schanzenbach,
and Almond 2016). This highlights the value of expanding the
access to, and use of, large administrative linked datasets for the
study of long-run policy impacts on children.
Second, we show the added insights that come using the
MVPF framework as opposed to traditional cost-benefit analy-
sis.9 It turns out that our general findings would be very similar
in a traditional cost-benefit framework, but the MVPF leads to
different conclusions in certain key instances. This is because the
MVPF and traditional benefit-cost analysis rely on similar inputs,
but the MVPF is unique in incorporating all fiscal externalities in
its denominator.10 For example, when taxes are at the top of the
Laffer curve, the social benefit of reducing taxes by $1 is $2,11 but
the MVPF of that policy is infinite because the benefits to the in-
dividual are $1 and the net cost of the policy is $0. More generally,
our results suggest there is value in calculating the MVPF in other
settings, such as crime policy or tax enforcement, where the causal
effects of the policy have clear effects on the government’s budget.
Last, we discuss the implications of the MVPF framework
for future empirical designs. In particular, we highlight the im-
portance of determining whether willingness to pay is positive or
negative. In this article, we sought to analyze state-level welfare
reforms from the 1980s and 1990s. There were 27 large-scale
state-level randomized controlled trials (RCTs) analyzing welfare
reform. These studies increased our understanding of the employ-
ment and revenue impacts of welfare policy. They demonstrated
that these welfare reforms had low net costs. That said, while the
treated participants in these studies often received additional
services such as job search assistance, these policies also cut ben-
efits for those who did not comply with program requirements. As
9. The edited volume from Weimer (2009) provides a discussion of cost-benefit
analyses from different researchers in a range of different domains. The Washing-
ton State Institute for Public Policy (WSIPP 2019) conducts ongoing cost-benefit
analyses to assess policies relevant to state legislatures. See also Rea and Burton
(2020) for an application of the WSIPP data to comparative welfare analysis.
10. Traditional cost-benefit approaches include fiscal externalities in the nu-
merator (see Greenberg, Deitch, and Hamilton 2010).
11. The individual is willing to pay $1 for the tax cut and the government
receives a $1 benefit from increased tax revenue from the behavioral response to
the tax. In traditional cost-benefit analysis, increases in government tax revenue
are included in the numerator of the expression.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1217
a result, it is unclear whether willingness to pay for these reforms
was positive or negative. Despite randomizing more than 100,000
families into 27 large-scale RCTs, we are unable to reach any reli-
able estimates of the MVPFs of these policies. The evaluations of
welfare reform may have led to more valuable information if the
RCT designs had been created with a social welfare framework
in mind.
I.D. Relationship to Existing Literature
In constructing our MVPFs and presenting evidence for
high returns to investment in low-income children, we build
on a substantial line of existing research making the argument
for investment in children.12 Our work is also related to recent
research on the long-run effect of safety net protections for
children reviewed by Hoynes and Schanzenbach (2018). In light
of the evidence, they conclude that “reallocation of investments
over the life course to earlier periods can be efficiency-enhancing,”
which aligns with our conclusions.
There are also analyses—many of which we draw on in
this article—in which researchers have previously argued that
some government expenditures largely pay for themselves.
This argument is particularly prominent in discussion of early
education (e.g., Heckman et al. 2010; Garc´
ıa et al. 2017) and child
health care expenditures (e.g., Brown, Kowalski, and Lurie 2015;
Wherry et al. 2018).13 The argument also appears in the tax
literature, where some have argued that reducing top marginal
tax rates produces a “Laffer effect,” raising total revenue.14 Our
analysis builds on that work by evaluating policies at scale and
searching for the presence of high-return policies across a wide
range of policy domains. We find the most robust evidence for
Laffer effects for policies investing directly in children.
12. For example, foreshadowing many of our conclusions, Currie (1994) writes,
“Although the evidence is incomplete, it suggests that in-kind programs have
stronger effects on children than cash transfers, and that programs that target
specific benefits directly to children have the largest positive effects.”
13. Outside the scope of this article, some suggest certain macroeconomic
policies can pay for themselves, such as fiscal expansions during deep recessions
(DeLong et al. 2012). More generally, we omit many potentially relevant categories
of policies, such as macroeconomic stabilization, infrastructure investment, and
environmental policies.
14. In this sense, testing whether the MVPF of a policy change is infinite is a
generalization of Werning (2007)’s proposed test for identifying local Laffer effects
in the income tax schedule.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1218
THE QUARTERLY JOURNAL OF ECONOMICS
I.E. Roadmap
The rest of this article proceeds as follows. Section II
presents the general social welfare framework that motivates the
construction of the MVPF. Section III discusses the sample and
presents six example constructions of the MVPF. Section IV
discusses our main results and the distinction between MVPFs
of policies targeting children versus adults. Section V places the
MVPF estimates in the context of existing theories of optimal
government policy. Section VI presents lessons for future work.
Section VII concludes. As noted already, Online Appendices A–F
provide step-by-step details for constructing each MVPF, and all
Stata do-files for the construction of each MVPF are available on
GitHub.
II. MVPF FRAMEWORK
This section presents a general framework to measure the
welfare impact of changes in government policies. The frame-
work illustrates how the marginal value of public funds provides
natural guidance on the social welfare impact of economic policies.
Consider a government seeking to measure the welfare im-
pact of a government policy change under consideration. We define
social welfare, W, by the weighted sum of individual utilities,
W =

i
ψiUi,
where Ui is individual i’s utility function and ψi is their social
welfare weight. The latter measures how much a 1-unit increase
in utility corresponds to an impact on social welfare, W.15 The
utility function, Ui, measures both current and future well-being
of the individual. For example, if utility were additive over time,
one could nest uncertainty about future outcomes within this
framework, letting Ui = E[
t ⩾0βtuit] where uit is the individual’s
utility t periods from today.
Because the utility function is allowed to vary arbitrarily
across individuals, it will be helpful to normalize units across
individuals. To that aim, let λi denote individual i’s marginal
utility of income at the time the policy is under consideration.
15. For now, we do not place any assumption on these weights, and therefore
they can result from any particular social welfare function. We also assume the
weights do not change in response to the policy, but this is without loss of generality
because we focus on small policy changes below.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1219
This is equal to the effect on individual utility of providing $1
to that individual. Let ηi = ψiλi denote the individual’s social
marginal utility of income at the time of the policy. The value of
ηi measures the impact on social welfare, W, of an additional $1
placed in individual i’s budget today.
The government is considering a set of policy changes indexed
by j = 1, ..., J that change the economic environment (e.g., prices,
public goods) by a small amount. We parameterize the up-front
initial spending on policy j by dpj (which can either be an increase
or decrease). The net impact on social welfare of the policy is
(1)
dW
dpj
=

i
ψi
dUi
dpj
=

i
ηiWTP j
i = ¯
η j

i
WTP j
i ,
where 
i WTP j
i is the sum of individuals’ willingness to pay for
policy j out of their own income, WTP j
i = dUi
dpj
1
λi , and ¯
η j is the
average social marginal utility of the beneficiaries of the policy,
¯
η j =

i
ηi
WTP j
i

i WTP j
i
with weights given by the economic incidence of the policy,
WTP j
i

i WTP j
i .
The values ¯
η j measure how much social welfare increases if one
were to provide an average of $1 to the beneficiaries of policy
j. Each individual is willing to pay WTP j
i for the expansion by
dpj of policy j.16 Therefore, multiplying ¯
η j by 
i WTP j
i measures
the impact on social welfare of an expansion of the policy by
dpj. This means that the welfare effect depends on the effect of
providing $1 to a policy’s beneficiaries, ¯
η j, and the beneficiaries’
willingnesses to pay for the policy relative to cash, 
i WTP j
i .
In accounting for costs, we let R denote the present discounted
value of the government budget, and let Gj = dR
dpj denote the net
impact of the policy on the government budget.17 This net cost is
16. In the derivation of the MVPF, we remain fully general about each individ-
ual’s utility function. We abstract from any behavioral biases in the utility function
that might cause willingness to pay to be incongruent with choices that maximize
well-being. Moreover, in practice, our approaches to inferring willingness to pay
often require assumptions of rationality in individual utility that do not account
for the potential presence of behavioral biases.
17. In practice, the dpj variations that are identified in an empiricist’s re-
gressions will not, in general, correspond to budget-neutral policies. Traditional
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1220
THE QUARTERLY JOURNAL OF ECONOMICS
inclusive both of the initial cost of the program and all other effects
of behavioral responses on the government budget. For example, if
spending $1 on preschool increases wages in the future, Gj should
incorporate the effect of those increases in future tax receipts.
Crucially, both the willingness to pay measures, WTP j
i , and the
net cost, Gj, should include effects on both parents and children.
Policies that directly affect children should include willingness
to pay by parents and the impacts of their behavioral responses
on the cost of the policy. Conversely, policies that directly affect
parents should include any spillovers onto children.18
The MVPF of policy j is given by the aggregate willingness
to pay, WTP j = 
i WTP j
i , for the policy divided by the net cost to
the government, Gj:
(2)
MVPF j =

i WTP j
i
Gj
=
WTP j
Net Cost.
The MVPF is previously defined in Mayshar (1990), where it is
referred to as the marginal excess burden (MEB); in Slemrod and
Yitzhaki (1996), where it is referred to as both the marginal cost
of funds and the marginal benefit of projects, depending on the
policy in question; and in Kleven and Kreiner (2006), where it is
referred to as the marginal cost of funds (MCPF). However, the
MVPF formally differs from both the traditional definition of the
marginal excess burden in Auerbach (1985), Auerbach and Hines
(2002), and the marginal cost of funds in Stiglitz and Dasgupta
(1971), Atkinson and Stern (1974). Because of this, Hendren
(2016) defines this quantity as the MVPF to contrast it with the
MEB and MCPF.
approaches would attempt to account for government spending by modifying
the observed policy into a different policy that raised revenues via lump-sum
taxation. This would then require the researcher to observe not the causal effect of
the policy, but the “compensated effect” of the policy to identify the welfare effect.
In contrast, our approach hypothetically closes the budget constraint by comparing
two MVPFs: one that involves an increase in spending and another that involves
a reduction in spending or increase in revenue. Hence, welfare analysis can be
done with two sets of causal effects (one for the two policies under consideration)
as opposed to attempting to measure the compensated effect of a policy.
18. We sum the benefits accruing to both parents and children, but we do not
include any willingness to pay that arises because of parental altruism toward
their children (or children’s altruism toward their parents). This means that a
child’s willingness to pay for a policy is only counted once. Including willingness
to pay from parental altruism would only reinforce our central results. Similarly,
we do not incorporate individual willingness to pay for redistribution to others.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1221
Combining equations (1) and (2), the effect on social welfare
per dollar of government expenditure on policy j is
dWt
dpj
dR
dpj
= ¯
η jMVPF j.
Given the MVPF for any two policy changes, one can construct hy-
pothetical budget-neutral policy changes. For example, consider
increasing spending on policy 1 by a net amount G1, financed by
reducing spending (or increasing revenue) from policy 2 by the
same amount. Pursuing this combined policy, dp, increases social
welfare if and only if
(3)
¯
η1MVPF1 > ¯
η2MVPF2.
Welfare increases if and only if the welfare gains from increasing
spending on policy 1, ¯
η1MVPF1, exceed the welfare loss from
reducing spending on policy 2, ¯
η2MVPF2. The MVPFs of the
two policies characterize the cost of moving welfare between the
two groups of beneficiaries. One prefers the policy if and only if
¯
η1
¯
η2 > MVPF2
MVPF1 . If MVPF1 = 1 and MVPF2 = 2, then an individual
prefers spending on policy 1 financed by policy 2 if and only
if providing $1 to beneficiaries of policy 1 is valued more than
providing $2 to beneficiaries of policy 2.
As this example illustrates, welfare statements that com-
pare policies generally require comparisons of their MVPFs. The
MVPFs allows the researcher to form hypothetical budget-neutral
policies and assess their welfare implications using equation (3).
To reduce the role of social preferences in driving conclusions, one
can compare policies with the same beneficiary group. In this case,
one would expect that ¯
η1 ≈¯
η2 so that comparisons of the MVPFs
correspond to statements about social welfare. For example,
Hendren (2017a) suggests comparing the MVPF of a particular
policy to the MVPF of a tax cut with similar distributional inci-
dence. More generally, one can compare different redistributive
policies, such as food stamps and housing vouchers, among each
other to evaluate the most effective method of redistribution.
In some cases, one does not need to compare an MVPF to
another policy to reach a welfare conclusion. This occurs when
the MVPF is infinite. Mathematically, this happens when a
policy has positive willingness to pay by its beneficiaries and the
behavioral response to the policy generates fiscal externalities
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1222
THE QUARTERLY JOURNAL OF ECONOMICS
that are sufficient to cover the cost of the program, Gj < 0. The
textbook example of such a case is lowering taxes when they are
beyond the peak of the Laffer curve. In this case, lowering taxes
increases government revenue, and so these policies represent a
Pareto improvement for any positive welfare weights assigned to
the recipients.19 More generally, the MVPF framework facilitates
a search for other cases where policies have positive willingness
to pay and negative net costs, such as investment in kids.
The definition of the MVPF is theoretically motivated using
small (marginal) changes in government expenditures. Although
some empirical variation we use has marginal effects on individ-
uals’ budget constraints, one can also continue to construct the
MVPF as the ratio of willingness to pay to net government cost
for nonmarginal policy changes. This approach uses the actual
empirical variation in existing literature to estimate the return
on the observed nonmarginal expenditure. Future work could
explore how the MVPF for a given policy change varies within a
program’s size of spending. This would facilitate improved welfare
comparison for policies that were evaluated at different scales.20
II.A. Comparison to Social Cost-Benefit Analysis
The MVPF approach builds on a large literature on social
cost-benefit analysis (see the edited volume Weimer and Vining
2009 and Boardman et al. 2017, and the cost-benefit estimates
provided by WSIPP 2019). The MVPF uses many of the same un-
derlying estimates used to create benefit-cost ratios, but combines
them in a different way. A comparison with cost-benefit analysis
from Heckman et al. (2010) helps illustrate the importance of
19. In practice an expenditure policy may have been combined with a sep-
arate tax policy to raise revenue at the time the policy is implemented. In this
case, the combined expenditure and tax policy would not deliver a Pareto improve-
ment, as some current taxpayers would be made worse off. However, the infinite
MVPF corresponds to a case where the government need not raise revenue to im-
plement a policy that does not cost money in the long-run. The government could
have borrowed against the future returns on the policy and generated a Pareto
improvement.
20. Consider the case where policy 1 was a $1M government expenditure and
policy 2 was a $2M government expenditure. Comparing policy 1 and policy 2 would
require the MVPF for a version of Policy 1 that is scaled up to cost $2M. This same
logic would also apply if considering a large-scale expenditure on a policy that
had previously been analyzed with a narrower RCT—one would have to make the
additional assumption that the average treatment effect of this expanded policy is
given by the effect identified in the RCT.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1223
these differences. Heckman et al. (2010) compare the net social
benefits of the policy, inclusive of benefits that accrue back to
the government, against the up-front budgetary spending on the
policy, Cj. They use the following formula:
(4)
BCRj = Social Benefits
Social Costs
= WTP j + FEj
(1 + φ) C j
,
where FEj = Gj −Cj are the benefits accruing to the government
budget from the behavioral responses to the policy. The initial
program outlays in the denominator are often multiplied by 1 +
φ, where φ is the marginal deadweight loss of raising government
revenue. This is thought to translate the up-front costs into social
costs by accounting for the welfare impact of an implicit tax
policy that raises the needed funds. Often, φ is taken to be 0.3 or
0.5 (Heckman et al. 2010). Policies are then deemed to pass the
cost-benefit test if the BCR exceeds 1.
In contrast to the BCR, the MVPF is given by MVPF j =
WTP j
C j+FEj . It differs in two primary ways. First, the impact of be-
havioral responses on the government budget is counted in the
denominator, not the numerator. For example, consider a tax cut
of $1 for which the behavioral response increases tax revenue by
$1. In this case, the policy perfectly pays for itself, and so the
MVPF is infinite. Expenditures on the policy represent a Pareto
improvement. In a BCR framework, however, that $1 in increased
tax revenue is considered social benefit and counted in the numer-
ator. That leaves a BCR estimate of (
2
1 + φ ). This illustrates why
the BCR may be a particularly misleading guide to optimal policy
when policies have strong impacts on the government budget. We
found a policy with a BCR of (
2
1 + φ ) that was a Pareto improvement,
but we could find a different policy with a BCR above 2 that does
not deliver a Pareto improvement. For example, if we compare
this hypothetical tax cut to government-provided insurance with
willingness to pay of $2 for each $1 of insurance, the traditional
cost-benefit framework cannot distinguish between these policies.
Second, the MVPF approach does not require the government
to close the budget constraint through an increase in taxation.
Therefore, one does not adjust for the “deadweight cost of tax-
ation” based on this particular assumed method of government
finance. Rather, the MVPF directly measures the amount of
welfare delivered to beneficiaries per dollar of government
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1224
THE QUARTERLY JOURNAL OF ECONOMICS
expenditure. One closes the budget constraint by comparing the
MVPF of a given policy to the MVPF of other policies. This allows
the researcher to think through the library of feasible levers
available to the government. In contrast to the cost-benefit frame-
work, this approach reinforces the idea that incidence matters: a
policy that provides benefits to the poor cannot be readily
compared to the raising of revenue on the rich without thinking
about Okun’s bucket and the social welfare weights placed on the
beneficiaries (i.e., the values of ¯
η j for the policies).
Despite our advocacy for the value of the MVPF over a tradi-
tional cost-benefit analysis, it is perhaps reassuring to note that,
in most cases, these two approaches generate similar conclusions.
So although we argue that the MVPF is more appropriate for
measuring welfare, and consequently more informative in cases
where these two welfare measures diverge, the broad pattern of
our results remain the same under either framework.
III. CALCULATING MVPFS: EXAMPLES
We estimate the MVPF for 133 policies spanning social
insurance (e.g., health, unemployment, and disability insurance),
education (e.g., preschool, K–12, college, job and vocational train-
ing), taxes and cash transfers (e.g., top tax rates, EITC, AFDC),
and in-kind transfers (e.g., housing vouchers, food stamps). Our
focus here is on policies, rather than papers. In many cases
we combine estimates from multiple different papers, putting
together the puzzle pieces to build the full picture.21
We form a sample of policies in each domain by drawing on
survey and summary articles from each field. We supplement this
initial set of estimates with recent work in each area not captured
in the survey or summary articles. We restrict our attention to
policies in which there is an experimental or quasi-experimental
identification strategy used to estimate the policy’s impact.22
Formally, such papers identify causal effects using variations dpj
in the economic environment. We form our baseline sample with
21. If multiple papers analyze the same causal effect, we generally focus on
the most recent published estimates unless otherwise noted. We provide a detailed
discussion of the alternative specifications in the Online Appendix.
22. We exclude purely cross-sectional identification using controls for ob-
servables in our baseline sample. Within the set of experimental and quasi-
experimental studies, we do not impose our own filter on the quality or validity of
these empirical designs.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1225
policies where one observes effects of the policy that are sufficient
to form a reasonably comprehensive view of both the WTP and
net cost of the policy. We discuss in Online Appendices A–F the
standard for policy inclusion in our categories and the set of
causal effects used in each case. Because this process involves
judgment calls, we also assess robustness of our conclusions to
an expanded sample (e.g., that expands the set of identification
and forecasting methods) and a more restricted sample (e.g., that
requires direct observation of causal effects on income).
Table I lists the set of policies studied, along with the
empirical papers used to form each policy’s MVPF. Column (9)
denotes the set of papers used to construct the MVPF. In many
cases, we draw from multiple papers to form a single MVPF. For
example, some publications might estimate the impact of the
policy on adults, while other papers focus on longer-run effects on
children.
In this section, we illustrate the construction of these esti-
mates using six examples spanning the domains we consider. We
attempt here to provide a diverse set of examples to demonstrate
the range of approaches used to create our estimates. Online
Appendices A–F provides a detailed step-by-step discussion of the
construction of each MVPF. In Section IV.C, we assess robustness
of our primary conclusions to alternative assumptions (e.g.,
different interest rates and tax rate imputations) and alternative
samples.
III.A. Admission to Florida International University
We begin by constructing the MVPF of admitting an addi-
tional student into Florida International University (FIU). This
example illustrates the construction of the MVPF for a policy
targeting youth with effects on later-life earnings. We use similar
methods for other child policies.
We draw on the work of Zimmerman (2014). He uses an RD
design at the school’s academic performance cutoff for applicants
to measure the effect of FIU admission on state university system
enrollment and medium-term earnings outcomes. We translate
his estimates into an MVPF, incorporating the net cost of the
policy and the beneficiaries’ willingness to pay. Throughout,
we construct confidence intervals for our estimates using a
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1226
THE QUARTERLY JOURNAL OF ECONOMICS
TABLE I
DETAILS OF ALL PROGRAMS STUDIED
Sample
Program
Label
Year
Age of
Paper
Papers
Implemented Beneficiaries Baseline Restricted Extended Estimates
Utilized
MVPF
Panel A: Education and job training
Child education
Carolina Abecedarian
Abecedarian
1975
3
x
x
x
Barnett and Masse (2007)
Study
Campbell et al. (2012)
Helburn (1995)
Masse and Barnett (2002)
Masse (2003)
Chicago Child-Parent
CPC Extended
1985
6
x
Reynolds et al. (2002)
Centers, Extended
Reynolds et al. (2011)
Program
Chicago Child-Parent
CPC
1983
4
x
Reynolds et al. (2002)
Centers, Preschool
Preschool
Reynolds et al. (2011)
Program
Chicago Child-Parent
CPC
1986
8
x
Reynolds et al. (2002)
Centers, School
School
Reynolds et al. (2011)
Age Program
(Table 1 is continued at the end of Section VII, before the Appendix.)
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1227
semiparametric bootstrap procedure discussed in detail in Online
Appendix H.23
1. Costs.
Figure I, Panel A shows how we calculate the net
cost of FIU admission. We start with initial costs of $11,403, which
represents the state university system’s educational expenditures
on each marginal admit to FIU.24 Students pay some fraction
of those educational expenses, so we subtract $3,184 to account
for private student contributions. Next we account for the fact
that some new admits would have attended a state community
college if they had not enrolled in FIU. We subtract $5,601,
Zimmerman’s estimate of the amount the government would
have paid to support their education at those community colleges.
Taken together, that leaves us with an up-front government cost
of $2,617 per admitted student.
The remaining cost considerations all stem from earnings
changes caused by FIU admission.25 Zimmerman (2014) calcu-
lates that in the first seven years after admission, earnings fall
by $10,942.26 We use estimates from the Congressional Budget
Office to estimate that the tax and transfer rate on these earnings
is 18.6%. This suggests the earnings change reduces government
revenue by $2,035.27 Next, Zimmerman (2014) estimates that FIU
23. In particular, we conservatively account for correlations across estimates
in a given policy, and we develop a method to adjust for the uncertainty in the
denominator (with many thanks to conversations with Isaiah Andrews). We pro-
vide the intuition for the approach and Monte Carlo simulations with appropriate
coverage. In fact, the coverage is sometimes overly conservative, especially when
costs approach 0.
24. Zimmerman (2014) calculates costs and student contributions using the
data on educational expenditures from the Delta Cost Project (American Institutes
for Research 2017). We adopt this approach for other college policies analyzed in
our sample. Online Appendix B explains the details of our approach.
25. Zimmerman (2014) does not include any information on attendance of
federally supported graduate schools among marginal FIU enrollees. If that infor-
mation were available, it could be incorporated as an additional fiscal cost.
26. All earnings changes are discounted back to the time of the initial expen-
diture using a 3% discount rate. We toggle these discount rates in our robustness
discussion in Section IV.C. We also use CPI-U-RS when we need to deflate from
nominal dollar values to real ones.
27. To be conservative, we exclude payroll taxes because individuals may ben-
efit from a portion of these contributions. More detail on our calculations can be
found in Online Appendix G. The tax and transfer rate includes federal and state
income taxes along with food stamps, but excludes housing vouchers and other
welfare programs. We use the income-specific rate from the 2016 CBO estimates,
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1228
THE QUARTERLY JOURNAL OF ECONOMICS
FIGURE I
WTP and Cost Components for Admission to Florida International University
This figure illustrates the cost and willingness to pay components for admission
to Florida International University as studied in Zimmerman (2014). Panel
A breaks the total cost down into its various components, including increased
student payments on tuition, reduced government spending on community
colleges, and the changes in tax revenue from earnings. Panel B shows the
cumulative discounted cost of the policy over the lifetime of the beneficiary.
The solid line represents cumulative costs for ages up until 33, the oldest age
at which incomes are observed in Zimmerman (2014). The dotted lines provide
the 95% bootstrap (pointwise) confidence intervals with adjustments discussed
in Online Appendix H. The dashed line shows total costs inclusive of projected
costs at subsequent ages. The projection method is detailed in Section III and in
Online Appendix I. Panel C reports the components of our WTP calculations. The
point estimate measures WTP as the change in incomes after taxes and expenses
on tuition. All numbers are in 2005 dollars deflated using the CPI-U-RS and
discounted using a 3% real interest rate.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1229
admission causes earnings to rise by $36,369 in years 8–14. Once
again, we apply a tax and transfer rate and determine that the
government’s revenue rises by $7,274. At this point our net costs
are −$2,622, as shown in Figure I, Panel B. This suggests the
expenditure has paid for itself within 14 years of the initial outlay.
Finally, Zimmerman’s earnings data extend 14 years, but we
can extrapolate from the observed effects to estimate earnings
changes over the full life cycle. Online Appendix I describes this
procedure in detail, and Appendix Figure I provides a graphi-
cal illustration of the approach. We use ACS data to estimate
life cycle earnings trajectories and then map the control group in
Zimmerman (2014) onto those trajectories. In particular, we ob-
serve an average earnings for the control group of $28,964, which
we estimate to be 113% of mean earnings for this cohort in the
ACS. In contrast, the treated group earns $6,372 more during
these ages, or 22% more than the control group. We assume that
the control group earnings remain constant as a fraction of av-
erage ACS earnings throughout the life cycle. We also assume
that the percentage earnings increase for the treatment group
also remains constant throughout the life cycle. These assump-
tions mean that we assume the trajectories for the treatment and
control groups differ by a constant percentage throughout the life
cycle.28 This yields an estimated discounted earnings increase of
$117,330 through age 65. We subsequently calculate that the as-
sociated fiscal externality reduces government costs by $21,823.
When combined with our previous cost components, we find that
each marginal FIU admission has a net cost of −$24,445. The
expenditure pays for itself.
and we apply this rate uniformly across years for simplicity. With more reliable
historical information on marginal tax and transfer rates across the income distri-
bution, one could perform the analysis separately by year. We are not aware of any
comprehensive historical source on the distribution of those rates. For this reason,
we take the simpler approach of using a consistent 2016 tax and transfer rate
and then assessing the robustness of all our results to alternative rate assump-
tions. We present robustness to alternative tax and transfer rate assumptions in
Section IV.C.
28. Although this is a strong assumption, we show in the robustness analysis
that our results are actually not very sensitive to the method we use to construct
these forecasts. For example, we conduct a conservative forecast that assumes
zero income growth over the life cycle. This yields similar results (see Figure VI,
Panel B).
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1230
THE QUARTERLY JOURNAL OF ECONOMICS
2. Willingness to Pay.
Having established that the initial
costs of increasing admission at FIU leads to long-run net savings
to the government, the policy has an infinite MVPF as long as
WTP > 0. That said, constructing a measure of willingness to pay
remains useful in making our confidence intervals and evaluating
alternate specifications. The components of our baseline estimate
of WTP are illustrated in Figure I, Panel C.
Throughout, our approaches to estimating WTP rely heavily
on the logic of the envelope theorem and revealed preference. For
the baseline estimate, we assume that increases in income among
the college educated stem from returns to human capital, not from
higher levels of effort.29 In this case, the envelope theorem implies
that we can form an estimate of WTP using the policy’s impact
on net income after taxes and other expenses (and ignore the
composition of individuals’ spending).30 We begin by noting that
those who are admitted to FIU have an increase in private costs
associated with additional tuition and fee payments at the four-
year school. This leads to a negative WTP component of $2,851.
Next, the earnings fall in the first seven years after admission
leads to a further negative WTP of $8,907. The earnings gains in
years 8–14 yield a positive WTP of $29,095. Projecting through
the rest of the life cycle yields an additional WTP of $95,507.
Combined, this yields a total willingness to pay of $112,844.31
29. We refrain from incorporating general equilibrium effects in our willing-
ness to pay due to a lack of evidence on this point. If higher educational attainment
produced positive spillovers on others, aggregate willingness to pay would rise. If
the college earnings premium were driven by signaling effects, then we would
expect other individuals to have a negative willingness to pay.
30. To see this, consider the decision problem of choosing a vector of consump-
tion goods x to maximize u(x; p) subject to q · x ⩽y(p) where q is the price of goods
and y(p) is after-tax income. In principle, the government’s policy choices, p, can
directly affect utility and the budget constraint. For the baseline WTP measure
for FIU, we assume admission to FIU only affects y(p) so that ∂u
∂p = 0, which means
willingness to pay is given by dy
dp (the impact on the vector x can be ignored by
the envelope theorem). However, if effects on income of admission to FIU is the
result of higher levels of effort, that would require an adjustment for the disutility
of labor and our baseline approach would overstate WTP; conversely, if individ-
uals derive additional utility from attending college that is not captured in their
earnings, the baseline approach would understate willingness to pay.
31. We also form a “conservative WTP” of $1 that relies on the logic of revealed
preference that individuals are willing to pay a nonnegative amount for admission
into FIU.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1231
III.B. Medicaid Expansion to Pregnant Women and Infants
Now, we consider a Medicaid expansion to pregnant women
and children in the United States that occurred across states
between 1979–1992. This example illustrates a case where we
construct the MVPF using examples from several papers using the
same identification strategy but focusing on different outcomes.
We construct our MVPF using several different analyses of
these reforms, each of which use the differential timing of the re-
forms across states to measure their impacts.32 Currie and Gruber
(1996) document a significant increase in health insurance cover-
age for pregnant women, along with a corresponding reduction in
infant mortality and low birth weight. Cutler and Gruber (1996)
find significant crowd-out of private insurance policies. Dave
et al. (2015) find reductions in labor supply of eligible women.
Miller and Wherry (2019) find positive effects on children’s future
earnings and health for those whose parents obtained Medicaid
eligibility. We translate these estimates into their implied MVPF,
beginning with costs and then turning to willingness to pay.
1. Costs.
The bar chart in Figure II, Panel A illustrates
the translation of estimates from the literature into their
implied costs to the government. Currie and Gruber (1996)
estimate that the cost of insuring an additional pregnant woman
through the Medicaid expansion was $3,473.33 In addition to
the direct Medicaid costs, Dave et al. (2015) estimate that
Medicaid eligibility leads to a 21.9% reduction in female labor
force participation, which corresponds to an earnings impact
of roughly $2,834. We estimate that these individuals face a
tax-and-transfer rate of 18.9% from the CBO using our procedure
discussed in Online Appendix G. This means that the earnings
effect implies an additional cost to the government of $564 per
eligible child. As a result, a short-run analysis of the policy would
conclude that the causal effects of the policy lead to an increase
in costs.
32. Our analysis also explores other policies that expanded Medicaid to chil-
dren, such as the national expansion of Medicaid to those born after September 30,
1983. These policy changes correspond to separate MVPF constructions because
they arise from different sources of policy variation.
33. For consistency across papers analyzing the reform, we deflate all numbers
to 2012 US$ using the CPI-U-RS; as a result, they differ slightly from reported
figures in each paper.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1232
THE QUARTERLY JOURNAL OF ECONOMICS
FIGURE II
WTP and Cost Components for Medicaid Expansions to Pregnant Women and
Infants
This figure illustrates the cost to the government of providing Medicaid to
pregnant women and infants. The evidence comes from state Medicaid expansions
between 1979 and 1992. Panel A breaks the total cost down into its various
components. The savings on uncompensated care come from Currie and Gruber
(1996), who estimate rates of uninsurance, and Gold and Kenney (1985) who
estimate the quantity of uncompensated care for the uninsured. The savings
on future health costs come from Miller and Wherry (2019). The increase in
government revenue combines an effective tax rate with the estimates of earnings
gains from Miller and Wherry (2019). Panel B reports the components of our WTP
calculations. The point estimate includes the willingness to pay for reductions
in infant mortality, combined with the change in income for children over their
life cycle after taxes and educational expenses. All numbers are in 2011 dollars
deflated using the CPI-U-RS and discounted using a 3% real interest rate.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1233
Turning to the effects on children, Miller and Wherry (2019)
estimate that a 1 percentage point increase in parental eligibility
leads to a reduction in future hospitalizations of 0.237% when
children are 19 to 32 years old. With a 3% discount rate, this im-
plies government savings on Medicaid and uncompensated care
of $868 over the 14-year period from ages 19 to 32.34 Miller and
Wherry (2019) also find a 3.5% increase in college attendance and
an 11.6% increase in earnings for children made eligible. On the
one hand, to the extent to which the government subsidizes col-
lege expenses, increased enrollment raises government costs. We
estimate that effect to be $371. On the other hand, the increase in
earnings when children are 23–36 years old leads to an increase in
government revenue of $3,909. By the time children are 36 years
old, the estimates suggest that the policy has paid for itself.
As with the example in Section III.A, we forecast these
earnings gains to age 65 by assuming that the percentage impact
on earnings remains constant throughout the life cycle. This
suggests that the government recoups an additional $6,114 in
tax revenue over this period, for a total of $10,024. The up-front
cost of $3,473 led to a long-run net government surplus of $7,014
(95% CI of [1,178, 12,971]).
Before moving on to discussing the details of willingness to
pay, it is worth noting that the MVPF of this expenditure has al-
ready been determined. For a policy to have an infinite MVPF, net
costs must be negative and willingness to pay must be any positive
value. The policy evaluated here expanded health care opportu-
nities to parents and children, so it is safe to assume willingness
to pay is positive. In fact, if the policy did not make anyone worse
off, then these expenditures resulted in a Pareto improvement.
2. WTP.
While the baseline MVPF estimate is infinite, we
calculate willingness to pay for use in constructing confidence
intervals and evaluating alternate specifications where costs are
positive. We briefly summarize this construction, which consists
of three components. (Step-by-step details of this calculation can
be found in Online Appendix D.)
First, Cutler and Gruber (1996) document that half of the
increase in Medicaid actually crowded out private coverage. As-
suming that the public and private costs of insurance were roughly
34. We forecast to age 65 by assuming a constant dollar saving and discounting
by 3%, which implies $530 of total savings, as shown in Figure II.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1234
THE QUARTERLY JOURNAL OF ECONOMICS
similar, this finding implies that beneficiaries no longer had to pay
roughly $1,737 in health insurance costs. This means that WTP
is at least $1,737. Second, Currie and Gruber (1996) estimate a
causal effect of the Medicaid expansion on infant mortality. We
assume parents have a willingness to pay out of their own income
of $1M to avoid an infant death (and assess robustness to alterna-
tive specifications).35 Third, we consider the WTP by the children
for improved labor market prospects in adulthood. To do so, we
assume that the increase in earnings documented by Miller and
Wherry (2019) reflects an expansion of labor market opportunities
and not an increase in costly labor effort. This means that the chil-
dren should be willing to pay the increase in their net income after
private expenses that results from increased educational attain-
ment. The increase in after-tax income is $16,775 for the observed
14-year age range (23–36) in Miller and Wherry (2019) and an
additional $26,236 in the subsequent years. Subtracting the cost
of college expenses reduces this by $111 for a net WTP of $47,400.
We also provide a conservative WTP estimate using solely
the transfer value of the insurance of $1,737. This would be
valid if the increase in after-tax earnings came at the expense
of increased effort as opposed to increased opportunities. To
be sure, the difference between the conservative and baseline
WTP estimate is quite large. As we discuss below, our primary
conclusions remain valid under either approach.
III.C. Introduction of Food Stamps
Third, we construct an MVPF for the impact of the intro-
duction of the Food Stamp Program, today known as the Supple-
mental Nutritional Assistance Program (SNAP). This example
illustrates how we incorporate potential spillovers of adult-
targeted policies onto children.
The Food Stamp Program provides in-kind transfers to
low-income families that can be used on food. Its introduction in
the 1970s was staggered across counties in the United States.36
35. Note we should think of this as a “private” not a “social” willingness to pay.
It assumes that parents are willing to pay $10,000 out of their own pocket to have
a 1% reduction in infant mortality. It is important to note that society may well be
willing to pay more than $1M. In the language of the social welfare function, this
suggests that the population has a high social marginal utility of income, ηj.
36. This variation was initially studied by Currie and Moretti (2006) in Cal-
ifornia and extended nationally by Almond, Hoynes, and Schanzenbach (2011),
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1235
Hoynes and Schanzenbach (2012) exploit this variation to analyze
its impact on labor income and welfare participation of adult
beneficiaries; Almond, Hoynes, and Schanzenbach (2011) study
its effect on birth outcomes. Bailey et al. (2019) use the same
variation to study its impacts on the adult earnings of children
whose parents received food stamps.
1. Costs.
The first component of our total costs is the average
yearly benefit from food stamp enrollment, equal to $2,904. To
this, we add the fiscal externality resulting from the effects on
both adults and children. For adults, Hoynes and Schanzenbach
(2012) document large yet imprecise reductions in earnings of
$3,650 that imply a fiscal externality of $471 from reductions
in tax revenue—roughly $0.16 per $1 of food stamps provided.
For children, Bailey et al. (2019) find increases in earnings in
adulthood corresponding to 7.1% for six full years of childhood
exposure to food stamps between the ages of zero and five. In
Online Appendix E, we show that this corresponds to an estimated
increase in tax revenue of $0.24 per $1 of food stamps for every
family with a child aged 0–5. We then multiply this by 0.35, the
fraction of SNAP benefits received by households with children
age 0–5. We subsequently multiply by 1.32, the average number
of children in these households. This suggests that for each $1
in food stamp spending, the resulting effects on children increase
government revenue by $0.11.37 Taken together, these estimates
imply that every $1 of spending on food stamps costs $1.05.38
2. WTP. We provide a willingness to pay from three compo-
nents. First, the envelope theorem suggests that individuals are
Hoynes and Schanzenbach (2012), Hoynes, Schanzenbach, and Almond (2016),
and Bailey et al. (2019).
37. We assume no impact on children at older ages, but clearly such effects
could alter the MVPF. In Section V, we discuss the implications for a policy targeted
to families with children aged 0–5; this leads to a larger MVPF.
38. Our costs estimates here are constrained by the set of observed outcomes
that we can reliably translate into effects on the government budget. For example,
Hoynes, Schanzenbach, and Almond (2016) report that the introduction of food
stamps was associated with a reduction in adult metabolic syndromes. Although
our earnings estimates likely capture the effect of those health changes on labor
supply, we lack a reliable way to measure the impact of those health changes on
health care utilization. Future work documenting long-run health impacts that
reduce (increase) government spending on medical care could lead to a higher
(lower) MVPF than we estimate here.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1236
THE QUARTERLY JOURNAL OF ECONOMICS
willing to pay for the mechanical cost of SNAP benefits, which
we estimate to be $1,809. We arrive at this number by taking
the $3,650 increase in earnings and noting that SNAP benefits
decline with earnings at a 30% phaseout rate. This means that
$1,095 of the food stamp cost is the result of a cost increase from
behavioral responses. Consequently, our point estimate suggests
individuals value $0.62 for each $1 spent by the government on
food stamps.39 Second, we incorporate the WTP for reductions in
infant mortality and increases in longevity among their children.
As in the case of Medicaid in Section III.D, we assume this is
given by the reduction in child mortality multiplied by a value of
a statistical life (VSL) of $1M (2012US$). We add to that value
the number of years of increased longevity multiplied by a quality
of adjusted life-year (QALY) of $20k (2012US$). This leads to
an additional WTP of $0.02. Last, we incorporate an additional
willingness to pay because of increases in after-tax income among
those who received food stamps as children. These estimates of
after-tax income are based on the earnings gains we calculate
above. Combining costs with willingness to pay creates an MVPF
of 1.04 (95% CI of [−0.97, ∞]).40
It is important to note in this case that statistical uncertainty
in these estimates is quite high. The combination of substantial
earnings reductions among parents and large earnings gains
among children mean that we cannot reject MVPFs of 0 or ∞. We
return to this uncertainty in more detail in Section VI.A when
we discuss the value of additional research or data access in
reducing sampling uncertainty.
III.D. Paycheck Plus in New York City
Fourth, we measure the MVPF of the Paycheck Plus program.
This construction illustrates how we create the MVPF from RCTs.
39. It is also worth noting that this willingness to pay is nearly identical to the
value we would receive if we did not apply the envelope theorem in this context, but
rather used estimates from Whitmore (2002) suggesting that food stamps have a
trade value of at least 65%. For our “conservative” willingness to pay specification,
we make both the envelope theorem and trade value modifications and find that
the MVPF falls to 0.39.
40. In Online Appendix E, we also explore several alternate specifications
and find that these produce only small changes to the MVPF. For example, we
assume a higher VSL of $9M and a QALY of $180k and find an MVPF of 1.22.
We incorporate the impact of reduced incarceration based on effects estimated in
Bailey et al. (2019) and costs of incarceration from Heckman et al. (2010). We find
that the MVPF rises from 1.04 to 1.07.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1237
It also provides guidance on the ideal set of measures future
researchers could construct to more directly estimate the MVPF
associated with RCTs.
The Paycheck Plus program is modeled after the Earned
Income Tax Credit (EITC). The EITC provides income subsidies to
low-income workers that are intended to encourage employment.
If workers face high marginal tax rates due to the benefit schedule
for means-tested transfers such as food stamps, the EITC may
offset those high rates. While the EITC generally targets adults
with children, the Paycheck Plus program in New York City
conducted an RCT to evaluate the provision of EITC-like benefits
to single adults without dependents—a group not traditionally
eligible for significant EITC benefits. The credit is worth up to
$2,000 a year and is available over three years (2014–2016 fiscal
tax years with bonuses paying out in 2015–2017).
Miller et al. (2017) estimate the effect of the policy on income,
employment, and after-tax income for the first two years of the
policy, which we translate here into their implied MVPF.41 We
begin with costs.
1. Cost.
The cost of the policy is the observed causal effect of
the policy on the government budget.42 To measure the costs, let Tj
denote the tax schedule faced by the control (j = 0) and treatment
(j = 1) groups. Let y j
i denote individual i′s earnings if they face
the j = 0, 1 tax/transfer schedule. The cost is then given by:
(5)
Cost = E

T 0 
y0
i

−E

T 1 
y1
i

.
41. As discussed in Online Appendix D, the current set of results from the third
year do not include sufficient information to form the MVPF in as precise a manner
as we do here; but we note how imposing a reasonable additional assumption
suggests that the third-year effects lead to a very similar MVPF also near 1.
42. In the context of an RCT, our approach measures the welfare impact of
randomly assigning additional people to the treatment as opposed to the control
group. As a result, one can use the reduced-form results to form our welfare
analysis (i.e., one need not separately isolate a LATE/TOT). The denominator
is the causal effect of this assignment on the budget and the numerator is the
aggregate WTP by members of the control group to be in the treatment group. As
a result, whether our welfare analysis can be externally generalized to a different
policy with different take-up of benefits would depend on how its treatment effects
vary across the population.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1238
THE QUARTERLY JOURNAL OF ECONOMICS
Because Paycheck Plus is an RCT, we compute equation (5)
using the difference in tax and transfer revenue obtained by the
government. In 2014, the causal impact on government costs was
$621; in 2015, this cost was $453. Combining these values, the
cost is $1,074.
2. WTP. We use the envelope theorem to estimate the WTP
for Paycheck Plus. In 2014, the average bonus paid is $1,399
among those who take it up, and 45.9% of people do so. The en-
velope theorem suggests that participants do not value the full
$1,399 subsidy dollar for dollar. This is because part of this cost
reflects the impact of behavioral responses. To the first order, those
who entered the labor force to obtain the transfer are indifferent
between working and not working. Miller et al. (2017) find a causal
effect of the program on the extensive margin labor supply of 0.9%.
Absent behavioral responses, this implies that 45% of the sample,
as opposed to 45.9%, would have received the transfer had they not
changed their behavior. Consequently, 98% of the transfer ( 45
45.9)
is valued by the beneficiaries, which implies a WTP of $630 for
the transfers in 2014.43 Repeating this calculation using the data
from 2015 yields a WTP of $441. This suggests a two-year WTP of
$1,070. The estimated WTP of $1,070 combined with the net cost
of $1,074 implies an MVPF of 0.996 (which rounds to 1 in Table II).
One can also construct an MVPF separately using the 2014 or
2015 transfers and responses. This yields similar MVPFs of 1.014
and 0.973. This dynamic similarity will be a recurring theme
among transfer programs to adults. It means that a static model
of the labor market distortions provides a reasonable approxi-
mation to measuring the MVPF for these policies. Every $1 the
government spends in transfers leads to a benefit of roughly $1.44
43. This calculation assumes no intensive-margin responses. If one observed
the microdata from the RCT, one could allow for intensive-margin responses.
To the first order, the WTP is the mechanical change in the tax schedule (i.e.,
replacing T0 with T1) holding behavior fixed for each individual at y0
i :
(6)
WTP = E

T 0 
y0
i

−T 1 
y0
i

.
This means the ideal method of calculating WTP is to feed the distribution of
control group earnings into both the control and treatment group tax schedule.
In practice, this number is rarely reported, but future work conducting welfare
analyses of RCTs can directly construct this measure.
44. If the provision of work subsidies today leads to an increase in labor
earnings and thus tax revenue after the earnings subsidies have ended, then
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1239
TABLE II
MVPF, WTP, AND COST ESTIMATES WITH CONFIDENCE INTERVALS, ALL PROGRAMS
Program
MVPF
MVPF CI
WTP
WTP CI
Cost
Cost CI
Baseline
Child education
∞
[17.83, ∞]
4.82
[3.38, 6.28]
−0.21
[−0.59, 0.19]
Abecedarian
11.89
[−0.18, ∞]
2.62
[−0.24, 5.63]
0.22
[−0.76, 1.15]
x
CPC Extended
∞
[−∞, ∞]
4.15
[−21.80, 27.36]
−1.22
[−6.13, 4.36]
CPC Preschool
∞
[∞, ∞]
2.23
[0.24, 4.21]
−0.35
[−0.62, −0.10]
CPC School
1.32
[−∞, ∞]
−0.18
[−1.23, 0.83]
−0.14
[−1.27, 0.98]
Head Start
∞
[10.58, ∞]
4.42
[2.90, 6.08]
−0.11
[−0.52, 0.27]
x
Head Start RD
0.72
[−0.02, ∞]
0.72
[−12.90, 11.46]
0.99
[−0.03, 1.78]
Head Start RCT
2.41
[1.90, 3.15]
1.29
[1.10, 1.50]
0.54
[0.49, 0.59]
K12 Spend
∞
[∞, ∞]
8.78
[4.58, 13.03]
−1.03
[−2.02, −0.06]
x
K12 Spend Mich.
0.65
[0.05, 2.19]
0.62
[−0.01, 1.58]
0.95
[0.79, 1.08]
Perry Preschool
43.61
[1.83, ∞]
3.45
[1.19, 5.70]
0.08
[−0.52, 0.68]
x
College adult
−5.59
[−∞, ∞]
−2.68
[−143.04, 61.16]
0.48
[−7.74, 18.61]
AOTC (IS)
6.75
[−1.61, ∞]*
2.45
[−6.52, 13.42]*
0.36
[−6.64, 6.47]*
x
AOTC (JE)
−1.77
[−17.06, ∞]*
−7.63
[−68.99, 40.12]*
4.31
[−11.54, 24.28]*
x
AOTC (JS)
∞
[−5.96, ∞]*
9.96
[−44.67, 76.39]*
−1.37
[−16.83, 12.68]*
x
AOTC (SI)
10.05
[−18.36, ∞]
5.36
[−89.06, 104.44]
0.53
[−12.93, 13.61]
x
AOTC (SE)
−0.02
[−2.25, ∞]*
−0.02
[−7.28, 5.59]*
1.12
[−6.30, 8.17]*
x
AOTC (SS)
∞
[−8.00, ∞]*
23.39
[−112.96, 191.48]*
−1.61
[−26.75, 19.56]*
x
HOPE Cred.
12.58
[−24.72, ∞]
5.27
[−745.41, 518.28]
0.42
[−61.22, 131.34]
x
HTC (IS)
18.86
[−2.87, ∞]*
5.91
[−24.65, 43.39]*
0.31
[−11.81, 11.68]*
x
HTC (JE)
2.37
[−2.22, ∞]*
8.21
[−35.77, 61.73]*
3.47
[−21.81, 28.40]*
x
HTC (JS)
∞
[−3.51, ∞]*
18.41
[−87.39, 148.22]*
−3.15
[−69.14, 53.90]*
x
HTC (SE)
11.83
[−4.48, ∞]*
4.59
[−17.25, 31.56]*
0.39
[−4.19, 4.15]*
x
HTC (SS)
−1.64
[−13.38, ∞]*
−1.91
[−22.71, 14.11]*
1.16
[−3.87, 6.67]*
x
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1240
THE QUARTERLY JOURNAL OF ECONOMICS
TABLE II (CONTINUED)
Program
MVPF
MVPF CI
WTP
WTP CI
Cost
Cost CI
Baseline
HOPE/LLC
−8.81
[−∞, ∞]
−42.82
[−266.09, 9.92]
4.86
[−7.59, 31.13]
x
Adult Pell
2.18
[0.71, 6.11]
3.42
[1.02, 6.25]
1.57
[1.03, 2.31]
x
Tuition deduc (JE)
0.77
[−1.92, 38.88]
1.00
[−4.88, 6.49]
1.29
[0.17, 2.51]
x
Tuition deduc (JS)
−0.02
[−2.50, 5.62]
−0.03
[−5.59, 4.43]
1.38
[0.55, 2.49]
x
Tuition deduc (SE)
∞
[−∞, ∞]
1.00
[−7.76, 8.53]
−1.13
[−2.04, −0.26]
x
Tuition deduc (SS)
∞
[−∞, ∞]
5.38
[−1.58, 14.00]
−5.10
[−6.36, −3.41]
x
College child
∞
[4.18, ∞]
8.79
[3.05, 15.65]
−0.36
[−1.76, 0.73]
Cal Grant GPA
∞
[10.72, ∞]
9.41
[3.43, 16.44]
−0.57
[−1.63, 0.32]
x
Cal Grant Inc
−0.69
[−2.36, 7.41]
−1.04
[−5.37, 4.29]
1.51
[0.63, 2.21]
x
CUNY Pell
1.39
[−2.95, 12.88]
1.42
[−3.42, 7.15]
1.02
[0.48, 1.56]
x
CC Mich
29.46
[−2.33, ∞]
7.80
[−9.29, 29.19]
0.26
[−3.39, 2.72]
x
CC Texas
349.51
[1.61, ∞]
10.69
[1.73, 20.89]
0.03
[−2.08, 2.10]
x
DC Grant
22.98
7.62
0.33
x
FIU GPA
∞
[∞, ∞]
13.73
[1.40, 62.13]
−2.97
[−15.62, −0.02]
x
Florida Grant
7.42
[1.09, ∞]
7.40
[1.24, 15.61]
1.00
[−0.32, 2.42]
x
Free FAFSA (dep)
4.03
[0.65, 10.75]
33.67
[3.04, 95.17]
8.35
[2.00, 20.03]
Free FAFSA (indep)
2.12
[−0.06, 9.71]
5.32
[−0.89, 15.05]
2.51
[0.77, 4.91]
Georgia HOPE
4.00
[0.37, 20.63]
3.60
[0.73, 6.48]
0.90
[0.28, 1.57]
x
HAIL Aid
1.30
[0.24, 3.65]
0.97
[0.14, 1.78]
0.75
[0.54, 0.95]
Kalamazoo
1.93
[0.97, 5.61]
1.93
[0.93, 3.71]
1.00
[0.77, 1.24]
x
MA scholarship
0.72
[−0.92, 3.05]
1.21
[−1.81, 4.00]
1.68
[1.25, 2.23]
x
Ohio Pell
2.49
[0.80, 5.40]
2.88
[1.29, 4.40]
1.16
[0.80, 1.56]
x
TN Pell
0.84
[−1.59, 3.57]
0.78
[−1.64, 2.84]
0.93
[0.62, 1.24]
x
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1241
TABLE II (CONTINUED)
Program
MVPF
MVPF CI
WTP
WTP CI
Cost
Cost CI
Baseline
Texas Pell
∞
[∞, ∞]
85.74
[0.77, 173.60]
−17.38
[−33.15, −1.92]
x
Soc Sec College
4.86
[0.98, 52.39]
5.03
[0.82, 10.92]
1.03
[0.32, 1.95]
x
College spend
4.00
[1.25, 20.44]
3.17
[1.26, 5.48]
0.79
[0.36, 1.21]
x
TN Hope
1.86
[0.92, 5.08]
1.94
[0.94, 3.51]
1.05
[0.81, 1.36]
x
College tuition
1.02
[−1.06, 5.47]
1.02
[−1.47, 3.58]
1.00
[0.68, 1.32]
x
WI scholarship
1.43
[1.00, 2.32]
1.46
[1.04, 2.08]
1.02
[0.93, 1.13]
x
Job training
0.44
[−19.57, 0.91]
0.36
[−0.82, 1.51]
0.83
[−0.09, 1.75]
Job Corps
0.15
[−0.23, 0.58]
0.15
[−0.23, 0.55]
0.98
[0.93, 1.03]
x
JTPA adult
1.38
[−0.21, 2.13]*
1.17
[−0.17, 2.64]*
0.85
[0.08, 1.65]*
x
JTPA youth
−0.23
[−3.43, 1.27]*
−0.21
[−1.70, 1.29]*
0.91
[0.15, 1.66]*
x
JobStart
0.20
[0.04, 0.42]
0.20
[0.06, 0.34]
1.02
[0.80, 1.24]
x
NSW Women
1.48
[−∞, ∞]
0.57
[−0.50, 1.64]
0.39
[−0.09, 0.86]
x
NSW Ex-Addict
0.44
0.35
0.79
x
NSW Ex-Offender
0.64
0.53
0.82
x
NSW Youth
0.60
[−∞, ∞]
0.47
[−4.23, 5.15]
0.78
[−3.32, 4.87]
x
Work Advance
0.78
[0.26, 1.34]*
0.64
[0.21, 1.11]*
0.83
[0.83, 0.83]*
x
Year Up
0.43
[0.37, 0.48]
0.41
[0.36, 0.45]
0.96
[0.95, 0.97]
x
Disability ins.
0.85
[0.82, 0.88]
1.00
[1.00, 1.00]
1.18
[1.14, 1.22]
DI generosity
0.96
[0.95, 0.97]
1.00
1.04
[1.03, 1.05]
x
DI judge
0.78
[0.72, 0.85]
1.00
1.28
[1.18, 1.39]
x
DI examiner
0.74
[0.71, 0.78]
1.00
1.34
[1.28, 1.41]
x
DI veterans
0.95
[0.92, 0.98]
1.00
1.05
[1.02, 1.08]
x
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1242
THE QUARTERLY JOURNAL OF ECONOMICS
TABLE II (CONTINUED)
Program
MVPF
MVPF CI
WTP
WTP CI
Cost
Cost CI
Baseline
Health adult
0.89
[0.56, 1.57]
1.49
[1.00, 1.99]
1.67
[1.01, 2.39]
Mass HI (150%FPL)
0.80
1.00
1.25
x
Mass HI (200%FPL)
0.85
1.00
1.18
x
Mass HI (250%FPL)
1.09
1.00
0.92
x
Medicare intro
1.63
[0.52, 3.83]
2.00
[0.58, 3.44]
1.23
[0.48, 1.78]
x
Oregon Health
1.16
[1.08, 1.25]
1.46
[1.19, 1.83]
1.26
[1.04, 1.57]
x
Medigap tax
0.40
[0.22, 1.54]
1.00
2.53
[0.64, 4.44]
x
Health child
∞
[24.82, ∞]
6.10
[3.05, 13.17]
−0.78
[−2.52, 0.17]
MC child 83+
∞
[0.26, ∞]
0.86
[0.66, 1.44]
−0.20
[−0.47, 1.82]
x
MC pregnant & infants
∞
[∞, ∞]
13.65
[5.92, 40.80]
−2.02
[−7.85, −0.27]
x
MC child (state exp)
∞
[−0.37, ∞]
8.13
[−0.24, 14.00]
−1.08
[−2.25, 0.57]
x
MC intro
10.24
[0.93, ∞]
1.78
0.17
[−1.60, 1.93]
x
Supp. Sec. Inc.
0.75
[0.64, 0.85]
1.00
[1.00, 1.00]
1.33
[1.17, 1.56]
SSI review
0.76
[0.56, 1.00]
1.00
1.32
[1.00, 1.78]
x
SSI judge
0.74
[0.72, 0.77]
1.00
1.34
[1.30, 1.39]
x
Unemp. ins.
0.61
[0.53, 0.74]
1.20
[1.15, 1.24]
1.95
[1.63, 2.26]
UI ben (state max)
0.68
[0.48, 1.13]
1.17
[1.11, 1.22]
1.71
[0.99, 2.41]
x
UI ben (DD)
0.43
[0.28, 0.78]
1.17
[1.11, 1.22]
2.74
[1.51, 4.17]
x
UI ben (DD w UR)
0.48
[0.30, 1.69]
1.17
[1.11, 1.22]
2.43
[0.79, 3.90]
x
UI ben (GA)
1.03
[0.97, 1.09]
1.17
[1.11, 1.22]
1.14
[1.09, 1.18]
x
UI ben (MO Exp.)
0.74
[0.67, 0.81]
1.17
[1.11, 1.22]
1.59
[1.46, 1.73]
x
UI ben (MO Rec.)
0.44
[0.39, 0.50]
1.17
[1.11, 1.22]
2.68
[2.36, 3.01]
x
UI ben (NY)
0.89
[0.82, 0.97]
1.17
[1.11, 1.22]
1.31
[1.21, 1.41]
x
UI ben (RK)
0.84
[0.76, 0.92]
1.17
[1.11, 1.22]
1.40
[1.28, 1.52]
x
UI dur (DD)
0.45
[0.25, 2.12]
1.30
[1.24, 1.36]
2.89
[0.61, 5.19]
x
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1243
TABLE II (CONTINUED)
Program
MVPF
MVPF CI
WTP
WTP CI
Cost
Cost CI
Baseline
UI dur (MO)
0.83
[0.76, 0.90]
1.30
[1.24, 1.36]
1.57
[1.46, 1.69]
x
Housing vouchers
0.77
[0.74, 0.81]
0.91
[0.91, 0.91]
1.19
[1.13, 1.24]
HCV RCT to welfare
0.91
[0.86, 0.96]
1.00
1.10
[1.04, 1.17]
x
HCV Chicago lottery
0.65
[0.61, 0.70]
0.83
1.27
[1.18, 1.37]
x
Jobs+
1.42
[0.45, 2.83]∗
1.14
[0.41, 1.91]∗
0.81
[0.67, 0.93]∗
MTO
MTO
∞
[−2.80, ∞]
18.40
[−15.46, 51.85]
−2.44
[−11.35, 6.81]
x
Nutrition
WIC
1.38
[1.10, 1.66]
1.28
[1.08, 1.47]
0.93
[0.88, 0.98]
SNAP assist
0.92
[0.91, 0.96]∗
0.92
[0.91, 0.96]∗
1.00
x
SNAP info
0.89
[0.89, 0.89]∗
0.89
[0.89, 0.89]∗
1.00
x
SNAP intro
1.04
[−0.97, ∞]
1.09
[−2.45, 4.55]
1.05
[−0.38, 2.51]
x
Cash transfers
0.74
[0.36, 1.47]
0.86
[0.50, 1.37]
1.16
[0.89, 1.34]
EITC 1986
1.20
[1.05, 1.38]
1.00
0.84
[0.73, 0.95]
x
EITC 1993
1.12
[0.82, 1.21]
1.00
0.89
[0.67, 1.06]
x
AFDC generosity
0.91
[0.83, 1.00]
1.04
[0.96, 1.11]
1.14
[1.10, 1.18]
x
AFDC term limits
0.81
[0.73, 0.90]
1.00
1.23
[1.11, 1.38]
x
Alaska UBI
0.92
[0.89, 0.96]
1.00
1.09
[1.05, 1.12]
x
Paycheck+
1.00
[0.87, 1.19]
1.00
1.00
[0.85, 1.15]
x
Neg. inc. tax
−0.01
[−0.82, 9.83]
−0.02
[−2.50, 3.53]
1.96
[0.18, 3.20]
x
Top taxes
3.03
[1.35, ∞]
1.00
[1.00, 1.00]
0.33
[−0.09, 0.74]
Top tax 2013
1.16
[0.87, 1.92]
1.00
0.86
[0.54, 1.16]
x
Top tax 1993
1.85
[1.19, 4.07]
1.00
0.54
[0.25, 0.84]
x
Top tax 1986
44.27
[2.37, ∞]
1.00
0.02
[−0.37, 0.42]
x
Top tax 2001
1.37
[0.92, 2.86]
1.00
0.73
[0.36, 1.09]
x
Top tax 1981
∞
[0.94, ∞]
1.00
−0.51
[−2.13, 1.06]
x
Notes. This table presents our baseline estimates for each program in our extended sample, along with the category averages reported in the bold header rows in each category.
We exclude the welfare-to-work policies discussed in Section VI.C. For each policy, we report its MVPF, cost per dollar of programmatic spending, and willingness to pay per
dollar of programmatic spending. We also report bootstrapped 95% confidence intervals with adjustments discussed in Online Appendix H. The final column indicates whether the
program is included in the baseline estimates (and thus included in the category averages). Confidence intervals are marked with an asterisk in cases where we infer p-values
using reported interval ranges. Programs in which the confidence interval is either inferred from p-values or missing are excluded from category averages.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1244
THE QUARTERLY JOURNAL OF ECONOMICS
III.E. Job Corps
Next we construct the MVPF for an RCT of Job Corps,
one of the largest vocational education programs in the United
States. This example illustrates how not all attempts to increase
children’s human capital and earnings have high MVPFs.
Established in 1964, Job Corps is administered by the U.S.
Department of Labor and provides job training and other services
to at-risk youth between the ages of 16 and 24 via a network
of centers run by local public and private agencies (Schochet,
Burghardt, and McConnell 2008). Between 1994 and 1996, the
National Job Corps study randomized 80,000 eligible applicants
into the program. We form an MVPF for this RCT using the
recent work of Schochet (2018), who links the original RCT to tax
data; we supplement this analysis with the earlier cost-benefit
analysis of Schochet et al. (2006).
1. Cost. Schochet et al. (2006) estimates that the up-front
programmatic cost per recipient is $16,158. Schochet (2018) then
estimates the earnings impact of the program over the course of
20 years and finds minimal effects. In particular, they find that
the program increases the present discounted value of participant
earnings by $121 using a 3% discount rate. We estimate that this
corresponds to an increase in tax and transfer revenue of $52.45
To these, we add the value of the products produced by the Job
Corps participants, which Schochet, Burghardt, and McConnell
(2008) estimates to be $220. Summing, this implies a net cost
of the program over 20 years of $15,886. Given the small effects
on earnings, we use this 20-year observed period as our baseline
estimate. In Online Appendix C, we show that if one extrapolates
the MVPF would be higher. We discuss these forecasts and their implied MVPFs
in Online Appendix F. To ensure our conclusions are not biased by including
policies for adults that do not have long-run follow-ups, in Section IV.C we conduct
robustness of all our analysis to policies where long-run follow-ups have been
measured.
45. As discussed in Online Appendix C, we form this estimate by summing
the observed increase in tax revenue for years 6–20 in administrative data from
Schochet (2018) combined with an application of the CBO tax rate to the earnings
effects for the first five years. We note that a fiscal externality of $52 in this case
corresponds to a high implicit marginal tax rate. This is driven by a low tax rate
on initial earnings declines and a comparatively higher tax rate on subsequently
small earnings gains.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1245
these earnings effects to age 65, the net cost of the program would
fall to $15,832 due to a small subsequent earnings gain.
2. WTP.
Following our approach for other policies that have
the potential to increase human capital, our baseline measure of
willingness to pay consists of the impact of the policy on after-tax
income.46 This is given by the $69 increase in after-tax earned
income plus the $2,314 component of the programmatic cost that
is a transfer to participants to pay for food and clothing while
participating in the program. Summing, this yields a WTP of
$2,383. Dividing by the government cost of $15,886 yields an
MVPF of 0.15. If one extrapolates the earnings affects to age 65,
the resulting MVPF is 0.18.47
III.F. Top Marginal Tax Rates
Finally we turn to the MVPF of top marginal tax rate
changes. This example illustrates how we can utilize estimates
from existing literature that attempts to provide empirical
guidance on optimal government policy (e.g., optimal top tax
rates, optimal unemployment insurance benefits). Whereas those
literatures often consider the policies in isolation (e.g., optimal UI
policy), we can translate the estimates into their implied MVPF
to facilitate comparisons across policy domains.
46. A pure revealed-preference approach in this context could rely on the as-
sumption that job training is accessible in the private market at its programmatic
cost. One could then set willingness to pay equal to (or perhaps below) the up-front
cost of program enrollment. In contrast, setting willingness to pay equal to after-
tax earnings does not require the assumption that potential Job Corps enrollees
have perfect information about the returns to job training at the time of initial
enrollment. However, it does require that after-tax income is sufficient to capture
willingness to pay. This means we do not incorporate any welfare costs from opti-
mization errors in consumption decisions that stem from program participation.
47. Our analysis here focuses on the MVPF of the entire treatment group.
However, it is worth noting that Schochet (2018) finds larger effects for the sub-
sample of age 20–24 participants, including a 2.4 percentage point reduction in
disability insurance receipt and a roughly $500 a year increase in earnings. To see
how this could lead to a different MVPF, we can first take a back-of-the-envelope
calculation of a PDV of lifetime disability insurance receipt of roughly $200k con-
sistent with Von Wachter, Song, and Manchester (2011). This implies a cost saving
of $4,800. Second, we note that the $500 a year impact on earnings corresponds
to a PDV increase in earnings of $12.8k. Applying an approximate 20% tax and
transfer rate implies an increase in WTP by $10.2k and an increase in tax revenue
of $2.6k. This implies a net cost of roughly $8,600, which implies an MVPF of 1.18.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1246
THE QUARTERLY JOURNAL OF ECONOMICS
There is a large theoretical and empirical literature dis-
cussing the optimal top marginal income tax rate, summarized
in Saez, Slemrod, and Giertz (2012). This literature notes that
a tax cut providing $1 in additional after-tax income is valued
at $1 by mechanical beneficiaries. In other words, the tax cut is
valued at cost by those who would receive it in the absence of any
behavioral response to the change in the tax code. As a result,
measuring WTP is straightforward. The cost to the government of
the tax policy is more difficult. The cost of a tax cut that provides
$1 of benefits in the absence of a behavioral response is given by
1 + FE, where FE is the impact of the behavioral response to the
tax cut on government revenue.
For top marginal tax rate reductions, Saez, Slemrod, and
Giertz (2012) and Diamond and Saez (2011) show that this FE
can be expressed as −
τ
1 −τ αϵETI, where α is the Pareto parameter
of the income distribution48 and ϵETI = 1 −τ
E[y]
dE[y]
d(1 −τ) is the elasticity
of taxable income for top earners with respect to the top marginal
“keep” rate of 1 −τ.49
The elasticity ϵETI has been estimated using various tax
reforms including the 1981 and 1986 tax decreases and 1993
increases in the top marginal income tax rate. We compute
the MVPF of the historical tax policy changes that allowed
researchers to identify ϵETI. The MVPF for each tax reform is the
ratio of WTP to cost,
1
1+FE:
(7)
MVPF =
1
1 −
τ
1 −τ αϵETI .
We translate estimates of ϵETI estimated from five major tax
reforms in 1981, 1986, 1993, 2001, and 2013, which are outlined
in Online Appendix F.
To take one example, consider the 1981 tax cut that reduced
the top marginal income tax rate from 70% to 50%. Saez (2003)
finds an estimate of ϵ = 0.311. We estimate α = 2.299 from Atkin-
son, Piketty, and Saez (2011). We plug these into equation (7).
We use marginal tax rates of τ = 75% and τ = 55% before and
after the reform, which include a 5% state tax adjustment.
48. Mathematically, α =
E[yi|yi⩾¯
y]
E[yi−¯
y|yi⩾¯
y] where ¯
y is the threshold over which the
top marginal income tax rate applies.
49. Online Appendix F provides a derivation.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1247
Combining, and averaging FE obtained using the prereform
and postreform tax rates, we obtain FE =
τ
1 −τ αϵETI = 1.51. This
means that the 70% marginal tax rate appears to have been on
the “wrong side of the Laffer curve,” so reducing tax rates may
have increased revenue. In other words, the MVPF is infinite and
the tax cut “pays for itself.” However, it is important to note the
statistical uncertainty in this estimate: we cannot reject an MVPF
of 1 or ∞.
In contrast, for later reforms we find lower MVPFs. For
example for the 1993 tax increase from 31% to 39.6% we find
an MVPF of 1.85 (95% CI of [1.19, 4.07]). This distinction is not
because of differences in ϵ, but results from the fact that τ was
much lower in 1993 than it was in 1981.
Comparison to the “Optimal” Top Tax Rate.
To compare
our results to the literature on the “optimal” top tax rates, it is
helpful to consider the case studied in Diamond and Saez (2011)
where society is assumed to place no weight on the additional
consumption of the rich. If the social welfare weights, ηi, are
equal to 0 for top earners, then the optimal tax is set to maximize
government revenue: τ is chosen to be at the peak of the Laffer
curve. This occurs when taxes are set so that the net cost to the
government of providing a tax cut is 0, or FE = −1.
This approach then makes the additional assumption that
the elasticity, ϵETI, and α do not change when the tax rate changes.
Solving for the optimal tax rate then implies τ ∗=
1
1 + αϵETI .
For α = 2.299 and ϵETI = 0.311, this implies τ ∗= 58%
inclusive of state and federal tax rates. The fact that this number
is slightly below 70% is consistent with our finding of an infinite
MVPF for the 1981 reform, in which tax rates were around
70%. In contrast to this optimal tax approach, the MVPF does
not impose an assumption that society places no weight on the
consumption of the rich.
IV. MAIN RESULTS: TARGETING KIDS VERSUS ADULTS
We construct the MVPF for each policy in our sample. Here,
we present all our baseline MVPF estimates and outline our
main results. As noted, details on our MVPF constructions are
provided in Online Appendices A–F.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1248
THE QUARTERLY JOURNAL OF ECONOMICS
FIGURE III
MVPF Estimates by Age of Policy Beneficiary
This figure presents MVPF estimates for all policies in our baseline sample. For
each MVPF, we plot them as a function of the average age of the policy’s beneficia-
ries. In cases where both parents and children potentially benefit, we assign the
age of the individuals with the highest willingness to pay. Where policies within
a category have the same age, we stagger these ages around this common value
for visual clarity. On the vertical axis, we report the MVPF estimates, capping
these estimates at 5. We separately report cases where the MVPF is infinite on
the uppermost line in green (shown in color in the online version only).
IV.A. Kids
We begin our discussion with the MVPFs of policies targeting
children. Figure III presents the MVPF for each policy on the
vertical axis plotted by the average age of the beneficiaries of the
policy on the horizontal axis.50 Each dot represents the MVPF of
a particular policy, with labels provided in Table I.
50. In cases where both parents and children are beneficiaries of the policy, we
assign the age of the “economic” beneficiary based on who has the highest WTP.
For example, when analyzing the Movement to Opportunity (MTO) experiment,
which provided housing vouchers and counseling to parents with children, the age
shown is the average age of the children in the household. This is because the
policy induced higher earnings among the children, leading them to have a higher
WTP for the policy than their parents.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1249
The figure reveals our primary result: direct investments in
children have historically had the highest MVPFs, often paying
for themselves. In addition to the evidence on the Medicaid expan-
sions and admission to FIU, we also find high MVPFs for other ed-
ucation and child health policies. For example, Wherry et al. (2018)
document that the discontinuous Medicaid coverage eligibility for
children born after September 30, 1983 led to reduced medical
costs and chronic conditions in adulthood. In Online Appendix D,
we calculate that the up-front costs are fully repaid in the long run
from reduced Medicaid and uncompensated care costs, leading to
an infinite MVPF. More generally, all four major health insurance
expansions to children studied in the past 50 years have MVPFs
in excess of 10, with three of them paying for themselves.51
In addition to health policies, we find large MVPFs for edu-
cation policies. The widely studied Perry Preschool program has
an MVPF of 43.61; the more expensive Abecedarian model has an
MVPF of 11.89 (neither of these estimates are statistically distin-
guishable from ∞).52 In contrast with the idea that the returns to
human capital investment diminish rapidly with age (Heckman
2006), we find there is potential for high MVPFs investments
throughout childhood. We find an infinite MVPF for increased
K–12 spending due to school finance equalization as studied
in Jackson, Persico, and Johnson (2016).53 We also find infinite
MVPFs for several college policies, such as admissions to FIU and
51. The only policy that does not have an infinite MVPF is the introduc-
tion of Medicaid. For this policy, we directly incorporate MVPF estimates from
Goodman-Bacon (2017). This working paper includes estimated impacts through
age 55; our back-of-the-envelope calculations suggest that it is likely that forecast-
ing these effects through 65 would lead the policy to pay for itself as well.
52. To harmonize these estimates with other programs, we do not include the
benefits to the government from reduced crime. This is both because these costs are
difficult to quantify and most papers do not estimate impacts on crime outcomes.
If we include a forecast of reduced government spending on the criminal justice
system and policing, our point estimates suggest that Perry Preschool paid for
itself. However, the standard errors of these estimates also significantly increase.
Including these costs for Abecedarian also increases its MVPF, but the policy does
not appear to pay for itself.
53. It is important to note that we only analyze one paper on K–12 education
spending because of limitations in existing evidence on long-term outcomes. While
there is a large literature looking at the effect of school spending on test scores, we
lack a reliable method to translate these effects into long-run impacts. Jackson,
Persico, and Johnson (2016) demonstrate the potential for high returns to K–12
education, but future work is needed to robustly establish the presence of high
returns to K–12 investment.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1250
THE QUARTERLY JOURNAL OF ECONOMICS
the provision of CalGrants to low-income students.54 A key insight
of our results is that many policies targeting children do not face
the classic budgetary trade-off. Instead, those expenditures pay
for themselves in the long run.
Before drawing too many conclusions about each data point
in Figure III, it is important to note there is sampling uncertainty
inherent in our estimates. Figure IV, Panel A plots each MVPF
along with its 95% confidence interval. In some cases, our
estimates are relatively precise. For example, both the Medicaid
expansion to pregnant women and infants and admissions to FIU
have confidence intervals that reject any finite MVPF. We can
rule out any positive net cost to the government. In many other
instances, however, the conclusions at the individual policy level
are less clear due to the sampling variation in the underlying
estimates. For example, the 1990 health care expansion to
children born after September 30, 1983 has a confidence interval
ranging from 0.26 to infinity. In other words, we cannot with 95%
confidence reject the hypothesis that the policy paid for itself, nor
can we reject the hypothesis that the policy provides much less
than $1 of benefits per dollar of government spending.
To reach more precise conclusions at a broader level, we pool
across policies using category averages. We imagine a new policy
that spends $1 of initial program cost on each policy j in category
J containing NJ policies. We then construct the MVPF of this
category-average policy as:
(8)
MVPFJ =
1
NJ

j∈J
WTP j
C j
1
NJ

j∈J

1 + FEj
C j
,
where the numerator is the average willingness to pay per dollar
of program cost and the denominator is the average net cost to
the government of the category-average policy.55
Figure IV, Panel B presents the category-average MVPFs.
On average, spending on child education, child health insurance,
54. It is important to be clear that although our estimates suggest high returns
to policies investing in older youth, the policies in our sample affect a range of
subpopulations. As a result, further work is needed to assess how the rate of
return on investment varies for a given child over the life cycle.
55. We construct this average measure, as opposed to a precision-weighted
average or other measure, because it corresponds to a feasible policy at the time
of initial implementation. It is straightforward for the government to construct a
policy that spends an equal amount on each of these programs.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1251
FIGURE IV
MVPF Estimates and Category Averages with Confidence Intervals
Panel A presents the MVPFs and 95% confidence intervals for each policy in our
baseline sample, plotted as a function of the average age of the policy’s beneficia-
ries. Panel B presents $1 spend domain averages and 95% confidence intervals
across categories of programs, plotted as a function of the average age of each pol-
icy’s beneficiaries within a category. Individual policy MVPFs are shown in smaller
dots, color-coded to align with their respective categories. In both panels, we report
the MVPF estimates on the vertical axis, capping these estimates at 5 and sepa-
rately reporting cases where the MVPF is infinite on the uppermost line in green
(shown in color in the online version only). All confidence intervals are 95% boot-
strapped confidence intervals with adjustments discussed in Online Appendix H.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1252
THE QUARTERLY JOURNAL OF ECONOMICS
FIGURE V
Net Government Costs per Dollar of Programmatic Spending
This figure presents estimates of costs normalized by initial programmatic for
each category-average group of policies in our baseline sample. We plot these
estimates as a function of the average age of each policy’s beneficiaries within
category. Bootstrapped 95% confidence intervals with adjustments discussed in
Online Appendix H are shown for the category averages. The normalized costs
of individual policies are shown in smaller dots, color-coded to align with their
respective categories (shown in color in the online version only).
and college policies have historically had high or infinite MVPFs.
One dollar of spending across each of the policies in each of these
categories has an MVPF of ∞in child education (95% CI of [17.8,
∞]), ∞in child health (95% CI of [24.8, ∞]), and ∞in college
policies (95% CI of [4.2, ∞]).
We can dig deeper into these category averages by focusing on
the net costs to the government of these policies (the denominator
in our formula in equation (8)). Figure V computes the average
net cost to the government per $1 of programmatic expenditure
spent evenly across the policies in each category. This allows us
to explore the extent to which different types of policies have
paid for themselves. For example, $1 invested in the four major
Medicaid expansions to children has paid back an estimated
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1253
$1.78. In other words, the spending actually generated $0.78 of
surplus to the government in the long run.56
Having established this primary result, it is important to
qualify that these patterns do not hold uniformly across policies.
There is considerable variation in MVPFs from policy to policy.
For example, we find lower MVPFs ranging from −0.23 to 1.48 for
job-training policies, such as an estimate of 0.15 for Job Corps—
a program targeted toward at-risk youth.57 We also analyze 14
examples of college policies where the MVPFs fall below 2.58
In most cases, this is because those policies represent trans-
fers to existing students, rather than expenditures that increase
attainment.59 In some cases, expenditures may even negatively
affect student attainment. For example, Cohodes and Goodman
(2014) analyze the impact of the Adams Scholarship in Mas-
sachusetts. They find that this merit aid program does not induce
more students to go to or complete college. Rather, it induces indi-
viduals to change colleges to attend in-state schools where they are
56. Analogously, Appendix Figure II presents willingness to pay per dollar of
programmatic spending. For our baseline WTP measures, we find very similar pat-
terns: much higher estimates of
1
NJ

j∈J
WTP j
C j
for child policies than for policies
targeting adults.
57. The one potential exception to this is the recent Year Up RCT, analyzed in
Fein and Hamadyk (2018), who document large increases in earnings in the two
years after initial implementation. As we discuss in Online Appendix C, if these
earnings gains persist for an additional 5 years, the MVPF would be 2.78, and if
they persist for 21 years, the MVPF would be infinite. In addition, in estimates
outside of our sampling frame, the nine-year follow-up results from the sectoral
training program Project Quest suggest an MVPF of 1.52, which increases to an
infinite MVPF if projected to age 65. This suggests a high value to future work
estimating the continued persistence of these more promising sectoral training
programs.
58. Our analysis also demonstrates the limitations of the traditional way that
research papers report the impact of college expenditures. It is very common for
papers to note the percentage point increase in enrollment associated with $1,000
in expenditures. The difficulty with that approach is that it doesn’t account for the
number of inframarginal students receiving the benefit. Providing $1,000 to 10% of
the school-age population to achieve a 3.6 percentage point increase in enrollment
may be a very efficient investment, while providing $1,000 to 80% of the school-
age population to achieve a 3.6 percentage point increase is mostly a transfer to
existing students. For this reason, there are cases where we find substantially
different MVPFs for policies that had similar percentage point enrollment effects.
59. In Section IV.C we discuss how our results on college expenditures vary
with the method of our MVPF calculation. Although we find persistently high
MVPFs when long-run earnings outcomes are observed, we find lower MVPFs
when we project earnings gains from attainment outcomes.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1254
THE QUARTERLY JOURNAL OF ECONOMICS
eligible to use the scholarship. The change in schooling actually
results in a fall in graduation rates arguably due to switching from
more selective schools with higher graduation rates. Incorporat-
ing these schooling declines, we calculate that the program has an
MVPF of 0.72. Job training or education polices like this one do not
substantially increase human capital and so they do not recoup
meaningful portions of their initial costs via higher tax revenue.
We also find lower MVPFs for transfers to disabled children,
such as an MVPF of 0.76 for expanded eligibility for Supplemental
Security Income (SSI) at age 18 analyzed in Deshpande (2016). It
is important to note that spending on these policies may increase
social welfare, even though they have lower MVPFs. Decisions
about optimal policy are determined by the welfare weights the
government places on policy beneficiaries. If the government
makes it a priority to provide support for disabled children, these
SSI expansions may be welfare enhancing.
IV.B. Adults
In contrast to policies targeting children, we generally find
lower MVPFs (e.g., 0.5–2) for policies targeting adults. For
example, in contrast to the nearly infinite MVPFs for child health
insurance expenditures, we find MVPFs ranging from 0.40 to
1.63 for the six health insurance policies in our baseline sample
targeted to adults.60 Along the same lines, we find MVPFs ranging
from 0.43 to 1.03 for unemployment insurance policies, 0.74–0.96
for disability insurance expansions, and 1.12–1.20 for earned
income tax credits. We find MVPFs of housing vouchers ranging
from 0.65 using assignment of vouchers in Chicago via lottery
(Jacob and Ludwig 2012) to 0.91 using an RCT of the provision of
housing vouchers to families on cash welfare (Mills et al. 2006).
The lower MVPFs reflect the fact that many of these expen-
ditures have been shown to reduce labor earnings through labor
market distortions. As depicted in Figure V, the average cost per
$1 of government spending on these adult policies is generally
60. Those adult health insurance estimates include expenditures such as the
subsidies in the Massachusetts health insurance exchange prior to the Affordable
Care Act. In that case, Finkelstein, Hendren, and Shepard (2019) exploit discon-
tinuities in the subsidy schedule to estimate both individuals’ willingness to pay
for insurance and the cost those individuals impose on the government. Translat-
ing these estimates into an MVPF suggests values ranging from 0.800 to 1.09 for
different subsidy eligibility levels.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1255
slightly above $1. This result contrasts with our findings on expen-
ditures directed toward children, for whom labor market earnings
tended to rise, leading to a decline in net costs. There are a limited
number of cases, such as the Job Training Partnership Act and Na-
tional Supported Work Experiment, where investment in adults
sought to increase earnings by increasing human capital. Those
policies, however, did not produce persistent earnings gains, so
they still yield relatively low MVPFs.61 The MVPFs of job-training
programs for adults over the age of 23 range from 0.44 to 1.48.62
As with our main results for policies targeting children, these
findings represent general patterns. They do not hold uniformly
across all policies targeting adults. In particular, there are two
types of adult policies that tend to result in higher MVPFs:
reductions of high marginal tax rates for top incomes and policies
with indirect spillovers onto children.
1. Top Tax Rates.
We find high MVPF point estimates for his-
torical reductions in the top marginal tax rate when the initial tax
rate lay at 50% or higher. In the case of the 1981 reform, the tax bill
reduced the top federal marginal tax rate on income from 70% to
50%. Using estimates of the elasticity of taxable income from Saez
(2003), we calculate that the MVPF is ∞(95% CI of [0.94, ∞]).
This implies that marginal tax rates were beyond the top of the
Laffer curve prior to 1981. Our confidence interval, however, sug-
gests this estimate contains considerable sampling uncertainty.63
61. For this reason, we calculate the MVPFs of job-training programs based
on the number of years of earnings effects observed, rather than projecting the
effects out to age 65. In Online Appendix C we discuss the sensitivity of our results
to that assumption.
62. The presence of high MVPFs for spending on children and low MVPFs for
spending on adults does not necessarily indicate that families are failing to opti-
mize their investment decisions. Even if families are fully informed of available
investment decisions, a simple model of parental investment could produce these
outcomes if parents are credit constrained. A higher MVPF for investment in chil-
dren could occur if low-income parents expect intergenerational regression to the
mean such that their children will earn more than them. That would produce lower
marginal utilities of income for those children, and therefore increase the return
on spending. In addition, this logic also suggests that when parents are given cash
transfers, they would rationally not spend all of it on their children despite high
returns—this is because their marginal utility of their own consumption is also
high.
63. As we discuss in Online Appendix F, these estimates appear to have consid-
erable uncertainty not just from sampling uncertainty but also model uncertainty:
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1256
THE QUARTERLY JOURNAL OF ECONOMICS
Along the same lines, we analyzed the 1986 reform and found an
MVPF of 44.27, with a confidence interval ranging from 2.37 to ∞.
Although this may be considered by some to be suggestive
evidence for Laffer effects in tax policy, it is important to ap-
proach that conclusion with considerable caution. In the case
of the 1981 reform our confidence intervals suggest we cannot
rule out an MVPF close to 1. In other words, we cannot rule
out the conclusion that the policy produced no positive fiscal
externality. Moreover, estimates of the impacts of recent reforms
have produced substantially smaller MVPFs (e.g., 1.16 for the
2013 top tax rate increase). Compared with these findings on
taxes, our results suggest stronger evidence for the presence of
Laffer effects when investing in young children.
2. Spillovers onto Children.
We also find that spending on
adults may have high MVPFs if those policies have spillover ef-
fects on children. For example, Chetty, Hendren, and Katz (2016)
study the long-run impact of the MTO experiment, which gave
families residing in public housing projects a voucher and coun-
seling to assist them in moving to lower-poverty neighborhoods.64
Chetty, Hendren, and Katz (2016) document that the program
significantly increased later-life earnings for young children, but
they find null or even slightly negative effects on earnings for
children who were teenagers at the time their parents obtained
the vouchers. Combining these effects across all subgroups
suggests the effects on the young children outweigh the adverse
effects on the older children, leading to an infinite MVPF.65 This
using different taxable income estimates from existing literature studying these
reforms can generate wide variation in the MVPFs of these tax reforms, preventing
precise conclusions about their MVPFs.
64. Because the program was targeted to families already in public housing
and because the cost of public housing is similar to the cost of a voucher, the
primary marginal cost of the program was the cost of the counseling (roughly
$3,783 per family).
65. Not all policies providing benefits to parents generate such large spillover
effects onto children. For example, Price and Song (2018) find that the Nega-
tive Income Tax experiment led to a reduction in children’s earnings in adult-
hood, which partially explains its low MVPF of −0.01. In other cases, such as
the provision of housing vouchers in Chicago, and the provision of housing vouch-
ers to families on AFDC and the expansion of AFDC benefits, there is sugges-
tive evidence that positive spillovers on children are small. In those cases, re-
searchers have documented that the policies have limited effects on outcomes such
as test scores, college attendance, and birthweight. Appendix Figure III, Panel A
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1257
high MVPF is driven solely by child outcomes, as the policy has
no significant effect on economic outcomes for adult beneficiaries.
One policy with a substantial degree of uncertainty about
potential
spillovers
onto
children
is
the
EITC.
Appendix
Figure III, Panel C shows how our MVPF estimates would change
if one attempted to impute effects on children using different
estimates from previous literature. In particular, we take the
MVPF for the 1993 OBRA tax reform and supplement that
estimate with spillover effects of the EITC estimated in other
contexts. Projecting earnings effects based on child test scores
produces MVPFs that range from 3.48 to ∞, while incorporating
effects on college attendance produces MVPFs from 0.84 to 1.12.66
Incorporating the work of Bastian and Michelmore (2018) on
long-term earnings would result in an infinite MVPF, suggesting
that the policy pays for itself.67
This uncertainty highlights the importance of understanding
the potential spillovers onto children. It also reinforces our
conclusion that policies raising children’s human capital often
have the highest MVPFs. We return to this issue in Section VI.A,
where we use the MVPF framework to quantify the value to
governments of more precise estimates for potential long-run
effects of policies on children.
IV.C. Robustness
Creating these MVPF estimates inevitably requires that we
make a number of judgment calls regarding the set of causal
effects included and the methodology used to translate those
effects into an MVPF. Here, we provide a short summary of the
robustness of our main conclusions to those assumptions.68
presents results for policies in our baseline sample where child effects are observed.
Panels B and C show how the MVPFs change when effects on children are incor-
porated or removed from the MVPF calculation.
66. The college effects are restricted to a small subset of recipients, so it is
unsurprising that the MVPFs remain small.
67. We exclude these results from the baseline estimates because Bastian and
Michelmore (2018) do not estimate the effect of a particular EITC expansion, but
rather pool across many state and federal policy changes. In Online Appendix F,
we note the impact of incorporating their estimates. The fact that these impacts
matter is consistent with our broader conclusions that potential spillovers onto
children can generate high MVPFs for adult-targeted policies.
68. Online Appendix J details an extensive set of robustness analyses.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1258
THE QUARTERLY JOURNAL OF ECONOMICS
Constructing the MVPF for policies with dynamic effects
requires the choice of a discount rate. Although our baseline
approach assumes 3%, Appendix Figure IV shows that higher dis-
count rates do not substantively change our conclusions. Discount
rates of 7% or 10% produce slightly lower MVPFs for child-
targeted policies (more so for young children), but we still find
those policies have higher MVPFs than policies targeting adults.
Our baseline approach uses the cross-sectional life cycle
earnings profile to forecast lifetime effects from observed earnings
changes. Our results are robust to alternative methods of fore-
casting earnings, such as assuming no income growth over the life
cycle. The baseline sample also includes some policies targeting
children for which effects on income are not directly measured.
Most notably, we include college policies where researchers have
observed a measure of attainment such as initial enrollment,
college credits, or degree receipt. In those cases we forecast
income impacts using estimates from Zimmerman (2014) on the
returns to college. Online Appendix J provides a discussion of
how our estimates vary depending on the use of intermediate
outcomes to construct long-run forecasts. In particular, Appendix
Figure III shows the effects of restricting our analysis to policies
where earnings are directly observed. We continue to find high,
often infinite MVPFs for these child-targeted policies.
In many cases, our MVPFs for policies targeting adults rely
upon estimates of short-run earnings impacts. Consequently,
one might be worried that our low MVPFs for adult policies are
driven by policies for which we do not observe long-run impacts.
In order to assess this, Appendix Figure VII, Panel B restricts
the analysis to the subset of policies for which we observe at least
five years of income estimates. We continue to find higher MVPFs
for policies targeting children.69
Our baseline willingness to pay approach often relies on
measures of a policy’s impact on after-tax income. Appendix
Figure VI, Panel A reports our MVPFs using our conservative
69. Related to this, the pattern of higher MVPFs for children could be driven by
longer payoff periods for children relative to adults, as children have their entire
lives to experience higher earnings. However, the length of the payoff period is
not what is driving our results—even restricting child benefits to accrue only up
to age 45 or 55, we find similar high returns for child-targeted policies. Rather,
the patterns are generally driven by a higher positive impact on per-year future
earnings for policies targeting children.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1259
measure of willingness to pay. Although the willingness to pay
measures are much lower, we continue to find high MVPFs for
policies targeting children, generally exceeding 5 on average. This
is to be expected as many policies analyzed have very low net costs,
leading to large MVPFs even when willingness to pay is small.
One might also be concerned that the causal effects incor-
porated in our MVPFs may vary in quality due to variation
in the underlying techniques used to produce those estimates.
Appendix Figure VII, Panel C shows our results remain the same
when restricting our sample to policies evaluated via randomized
controlled trial, lottery, or a regression discontinuity design. The
results are also robust to restricting our sample to peer-reviewed
publications. In all these robustness analyses, direct childhood
investments continue to have the highest MVPFs.
Finally, one might worry that MVPFs for child policies were
high in previous decades but have declined over time—perhaps
as the government takes advantage of high-return investments.
Appendix Figure IX assesses this by plotting the child- and adult-
average MVPFs separately by decade. We find no evidence for
that pattern of decline. Instead, we find high MVPFs for policies
targeting children throughout the past 50 years.70 The robustness
of high MVPFs for direct investments in children over time may
suggest the presence of fundamental political constraints to
enacting policies in which the benefits have a long time horizon.71
IV.D. Publication Bias
All of the robustness analyses above take the estimates from
existing literature as given. However, one might be concerned that
the research and publication process suffers from the problem
70. The one exception to this pattern is the low average MVPF among child
policies implemented in the 1970s. The child policies in that decade primarily
consisted of job-training programs that did not have significant effects on children’s
earnings.
71. There are a range of forms that these political constraints might take.
For example, it could be that governments (and politicians) apply a much higher
discount rate, requiring projects to pay off over short horizons. Alternatively, un-
derinvestment might occur because these policies require spending by state and
local governments, but much of the benefits accrue to the federal tax system.
Hence, local incentives may not be sufficient to make efficient investments. It
may also be that the high MVPF policies are undersupported because low-income
children have little political power. We leave a formal analysis of these potential
mechanisms for future work.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1260
THE QUARTERLY JOURNAL OF ECONOMICS
TABLE III
PUBLICATION BIAS ESTIMATION
Children estimates
Adult estimates
(1)
(2)
(3)
(4)
(5)
(6)
Z > 1.64
3.72
–
2.52
–
(2.46)
(1.32)
Z < −1.64
1.15
–
7.90
–
(0.44)
(1.48)
Z ∈[1.64, 1.96]
3.65
1.36
(3.46)
(1.14)
Z ∈[−1.96, −1.64]
1.02
4.19
(0.57)
(0.81)
Z > 1.96
–
3.09
3.78
–
3.27
3.59
(1.09)
(2.17)
(1.50)
(1.21)
Z < −1.96
–
1.21
1.24
–
10.39
11.52
(0.50)
(0.62)
(2.53)
(2.43)
N
237
237
237
150
150
150
Notes. The numbers shown are the estimated likelihood ratio of publication relative to an insignificant
result. Standard errors are in parentheses.
of publication bias, where studies are published only if they find
clear positive (or negative) effects. In particular, one might worry
that research on children is more likely to be published if it finds
statistically significant positive effects on children in adulthood.
Conversely, one could imagine that research on adults is more
likely to be published if it finds statistically significant evidence
of distortionary or negative effects on adult outcomes.
To address this, we implement the approach developed in
Andrews and Kasy (2019).72 They provide a method to test
and correct for the effect of publication bias on the observed
set of estimates. Online Appendix K discusses the details of
our implementation of their approach. Table III documents the
evidence of publication bias in our estimates.
The results suggest the presence of a moderate degree of
publication bias. In the baseline sample, we find studies of child
outcomes are 3.7 times more likely to be published if they find
positive effects on children with p < .10 relative to a finding of
no statistically significant effect. In contrast, we find that studies
72. We thank Isaiah Andrews and Max Kasy for their invaluable guidance in
implementing these procedures.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1261
on adult policies are 11 times more likely to be published if they
find significant distortionary effects on outcomes.
Despite evidence of publication bias in our samples, Appendix
Figure VIII, Panel A shows that correcting for the observed degree
of publication bias in this manner does not affect our conclusion
of higher MVPFs for policies targeting children. Although we find
a slight decrease in the MVPFs for child education policies, such
as preschool programs, the general patterns are quite similar
to our baseline results. Moreover, Appendix Figure VIII, Panel
B shows that even if we assumed that statistically significant
estimates of positive effects on children are 35 times73 more likely
to be published, our primary conclusions still hold.
V. MAPPING THE MVPFS TO THEORY
The MVPF provides an empirical method for evaluating the
effectiveness of different policies for improving social welfare.
Having established the key patterns of the data, it is natural
to place our empirical results into the context of theoretical
literature on optimal government policy. In this section, we
outline how our results speak to that theory.
1. Optimal Taxation.
To begin, the MVPF measures the
price of redistributing to different policy beneficiaries. In this
sense, the approach is related to a large body of theoretical and
empirical optimal tax literature in the spirit of Mirrlees (1971)
and Saez (2001).
As previously explained using the Okun’s bucket logic, the
ratio of MVPFs across two different tax changes measures the
price of moving money between the respective beneficiaries. In
general, optimal tax theory suggests that a progressive planner
should be willing to incur efficiency losses to move resources from
the affluent toward the lower regions of the income distribution.
The MVPF provides an empirical means of testing that basic
prediction: the MVPF of tax changes should increase with the
income of the beneficiaries.
Figure VI, Panel A explores the relationship between the
MVPF of each tax policy change we analyze and the income levels
of the associated beneficiaries. Consistent with this prediction,
73. A publication bias of 35x is the degree of publication bias documented in
Andrews and Kasy (2019) for small-sample experimental economics studies.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1262
THE QUARTERLY JOURNAL OF ECONOMICS
FIGURE VI
MVPF by Income of Beneficiaries
Panel A shows MVPFs for tax and transfer policies in our baseline sample
against the income of their economic beneficiaries. Panel B adds in-kind transfers
to parents and direct expenditures on children (child education, health, job train-
ing, and college policies). See Figure III for an explanation of the color scheme. The
income measures should be considered approximations, as not all papers report
consistent measures of incomes of their samples. We include all papers for which
we are able to obtain a measure of income of the beneficiaries, and we attempt
to normalize these measures to correspond to a notion of individual income per
adult in the household at age 30. All confidence intervals are 95% bootstrapped
confidence intervals with adjustments discussed in Online Appendix H.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1263
we observe an upward slope. For example, the 1993 tax reform
(OBRA93) simultaneously raised top marginal tax rates and
expanded the EITC. The MVPF of the increased top tax rates led
to an MVPF of 1.85 (95% CI of [1.19, 4.07]), and the expansion
of the EITC led to an MVPF of 1.12 (95% CI of [0.82, 1.21]).
This suggests the tax schedule created under the 1993 reform
is optimal if one is indifferent to providing $1.85 to top earners
versus $1.12 to those on the EITC. To the extent one’s social
preferences strictly prefer $1.12 to low earners (or strictly prefer
$1.85 to top earners), our results suggest that more progressive
(regressive) taxation than the 1993 schedule would be optimal.74
Although our MVPF estimates for tax changes are loosely
consistent with the preferences of a progressive planner, this is
no longer the case when we consider policies targeting children.
As shown in Figure VI, Panel B, there is no clear relationship
in our sample between MVPFs and the incomes of beneficiaries
when including direct investments in children. This means that,
historically, investments in the next generation have been more
efficient than transfers within generations.75
2. In-Kind versus Cash Transfers.
The low MVPFs for
policies targeting very low-income households raises the question
of whether other methods of redistribution—perhaps through
in-kind transfers—can be more effective than cash.76 Figure VII
74. Our estimate for the MVPF of the 1993 EITC is based on evidence from
Meyer and Rosenbaum (2001) on the fiscal externality associated with the labor
supply responses of single women. It is worth noting, however, that there is con-
siderable debate over the fiscal externalities associated with the EITC. On the
one hand, several recent papers have argued that reductions in transfers have
offset a substantial portion of the cost of historical EITC expansions (Hoynes and
Patel 2018; Bastian and Jones 2019). These large fiscal externalities can produce
infinite MVPFs (Bastian and Jones 2019). On the other hand, recent debates have
argued that the effects are overstated in the current literature because the impact
of the EITC expansions cannot be disentangled from the effects of contemporane-
ous welfare reforms (Kleven 2019). These conflicting estimates suggest there is a
high value to future work that reconciles these findings.
75. We develop this argument formally in Online Appendix L, where we relate
this logic to the nonexistence of a social welfare function that can rationalize our
results of high MVPFs for low-income children but low MVPFs for low-income
adults.
76. There is a large theoretical debate on this question, which largely centers
around the applicability of the Atkinson-Stiglitz theorem (Atkinson and Stiglitz
1976; Hylland and Zeckhauser 1981). When utility satisfies a “weak separability”
assumption, one would expect that the MVPF for an in-kind transfer would fall
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1264
THE QUARTERLY JOURNAL OF ECONOMICS
FIGURE VII
MVPF by Income of Beneficiaries: Cash versus In-Kind Transfers
This Figure presents MVPFs as a function of the average income of beneficiaries
for tax and transfer policies (shown in Figure IX, Panel A) combined with our
estimates for in-kind transfer policies. The income measures should be considered
approximations, as not all papers report consistent measures of incomes of their
samples. We include all papers for which we are able to obtain a measure of
income of the beneficiaries, and we attempt to normalize these measures to
correspond to a notion of individual income per adult in the household at age
30. All confidence intervals are 95% bootstrapped confidence intervals with
adjustments discussed in Online Appendix H.
adds the MVPF estimates for housing and food subsidies to the
estimates provided in Figure VI, Panel A for cash transfers and
tax credits. Broadly, we find a pattern consistent with our general
result: in-kind transfers are most effective when they induce
spillover effects onto children.
For example, the housing vouchers in Chicago (Jacob and
Ludwig 2012; Jacob, Ludwig, and Kapustin 2014) and the
provision of Welfare to Work housing vouchers (Mills et al. 2006)
find minimal spillover effects on children. This means that the
below the MVPF of a cash transfer or tax credit targeted to beneficiaries at the
same income level.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1265
distortionary impact on adults’ earnings leads them to have
MVPFs below that of distributionally equivalent tax cuts. In
contrast, the MTO experiment explained previously increased
earnings of young children by a sufficient amount to pay for the
cost of the in-kind policy (the policy had an infinite MVPF with
a 95% confidence interval of [−2.80, ∞]). Similarly, the spillover
effects onto children for the introduction of food stamps policy
leads to an MVPF of 1.04. Both point estimates suggest these
in-kind transfers are as efficient or more efficient than cash
transfers as a result of the spillovers onto children.77
3. Tagging.
There is a large literature in optimal policy
design focused on improving efficiency by targeting the right
subset of individuals. In general, this work focuses on the use of
“tags”—characteristics of program eligibility that are generally
not manipulable (Akerlof 1978).78 With that in mind, previous
literature has identified recipient age as a potentially valuable
tag for optimal government policy. Consistent with that work, we
observe that the MVPFs of certain policies differ substantially
based on the age of the recipients.
For example, our analysis of the MTO experiment finds an
infinite MVPF with a confidence interval of [−2.80, ∞]. That said,
the result masks substantial heterogeneity in the program’s ef-
fects. In families with children younger than 12, the MVPF is in-
finite with a confidence interval contained at infinity. In families
with children older than 12, the MVPF is negative, as their point
estimates imply a reduction in earnings. Along the same lines, our
analysis of the introduction of food stamps produces an MVPF of
1.04. This MVPF is partly buoyed by large positive effects on chil-
dren ages 0–5 (Bailey et al. 2019). If we excluded any effects on
children, the MVPF would fall from 1.04 to 0.54. By contrast, if
we restricted our analysis to families with young children and
assumed that causal effects of food stamp introduction remained
the same for that targeted policy, we would find an MVPF of 2.28.
77. In relation to the Atkinson-Stiglitz theorem, the violation of the weak
separability assumption for these policies comes not from a short-term change in
earnings but from the long-run indirect impact on children.
78. If the tag were manipulable, then individuals not intended as beneficiaries
of the policy could distort their behavior to obtain the benefit. To the first order,
they would not value the transfer by the envelope theorem, consequently lowering
the MVPF of the policy.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1266
THE QUARTERLY JOURNAL OF ECONOMICS
Despite this substantial variation in MVPFs by the age
of policy recipients, we do not report subgroup-specific MVPFs
in our main tables. This is a deliberate choice to restrict our
analysis to policy changes defined by explicit identification
conditions established in existing work. Reporting MVPFs for
subgroups requires the additional assumption that the observed
behavioral response to the policy among the relevant subgroup
is not impacted by the provision of the policy to other subgroups.
Although this may be plausible in certain cases, we have no
disciplined way of adjudicating its plausibility across all possible
permutations of subgroup analysis. Instead, we highlight the
potential for age-specific tagging but refrain from more definitive
statements regarding subgroup-specific welfare impacts.
VI. LESSONS FOR FUTURE WORK
In this section, we discuss three implications for future
economic research. First, we show how the MVPF framework
facilitates a straightforward method to quantify the value of
future work that reduces the statistical uncertainty in our esti-
mates. Second, we show the value-added provided by measuring
the MVPF relative to what is provided by a more traditional
cost-benefit analysis. Third, we discuss how the intuitions of the
MVPF framework might influence future empirical designs. The
key is to design experiments in a way that facilitates measuring
willingness to pay. In particular, we discuss how 27 different
welfare reform programs in the 1980s–90s randomized upwards
of 100,000 participants into RCTs, but the nature of the research
designs makes it infeasible to conduct reliable welfare analysis.
VI.A. Value of Information in Evidence-Based Policy Making
Our MVPF estimates measure the welfare impact of a
range of government policies. Although it is our hope that these
estimates can be useful for a policy maker seeking to conduct
“evidence-based” policy, it is quite clear from Figure IV, Panel A
that many of our individual policy estimates contain considerable
sampling uncertainty. Here, we show how one can use the MVPF
framework to understand the value of future research that
reduces the uncertainty in our estimates. The MVPF framework
provides a measure of the value of information because it
is a price: it measures the price faced by the government to
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1267
redistribute across beneficiaries of different types of policies.
A welfare-maximizing government should be willing to pay to
reduce the uncertainty in these prices, just as consumers would
be willing to pay to learn the true value of the products they buy.
There are many ways one could conceptualize reducing
the various sources of modeling and sampling uncertainty in
our estimates. In this section, we develop a simple approach
to measure the value of reducing sampling uncertainty. We
defer an exhaustive treatment to future work. We use this
example to illustrate the value of future research that increases
estimate precision, perhaps through improved access to larger
administrative longitudinal data sets.79
Our conceptual experiment is organized as follows: suppose
a policy maker is considering whether to raise revenue to spend
an additional $1 on policy j. The policy has a net cost to the
government of Gj and a willingness to pay of WTPj per dollar
of programmatic cost. The policy maker does not know the true
values of WTPj and Gj. Instead, we assume she only observes
the estimates,
ˆ
WTP j and ˆ
Gj, and their sampling distributions.80
We assume the policy maker has an uninformed prior about the
impact of the policy so that the estimated sampling distribution
reflects her belief about the policy’s effects.
For simplicity, we assume the policy is financed with a tax
change that targets the same beneficiaries and has an MVPF of
1. A budget-neutral policy that increases taxes to spend on policy
j has a welfare gain of
U

WTP j, Gj
	
= WTP j −Gj.
Ideally, the policy maker would wish to pursue this policy if and
only if U(WTPj,Gj) > 0 (i.e., the policy increases welfare). In
practice, the policy maker only observes estimates and sampling
distributions of these values. We assume these estimates are
unbiased but noisy estimates of the truth (e.g., E[ ˆ
Gj|Gj] = Gj).
Utility is linear in WTPj and Gj, so the policy maker will choose
the policy if and only if
ˆ
WTP j > ˆ
Gj. The expected utility of this
79. The focus here is on reducing uncertainty among the observed outcomes of
each program. Uncertainty regarding unobserved causal effects remains beyond
the scope of this exercise.
80. For simplicity, we assume programmatic costs are known and equal to
their point estimates.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1268
THE QUARTERLY JOURNAL OF ECONOMICS
strategy given the point estimates (
ˆ
WTP j, ˆ
Gj) is
EU Uninf ormed 
ˆ
WTP j, ˆ
Gj

= E

U

WTP j, Gj
	
∗1

U

ˆ
WTP j, ˆ
Gj

> 0

|W ˆ
TPj, ˆ
Gj

=

ˆ
WTP j −ˆ
Gj

∗1

ˆ
WTP j > ˆ
Gj

.
Now suppose that instead of spending $1 on the policy, the
policy maker can invest a fraction of this dollar, vj, into learning
more about the WTPj and Gj of the policy before making this
decision. We begin by considering a case where spending vj allows
the policy maker to perfectly learn WTPj and Gj before deciding
whether to invest in the policy. Once informed, the government
chooses to pursue the policy if and only if U(WTPj, Gj) > 0. Now it
can decide to pursue the policy if and only if the true WTP exceeds
the true costs. In that case, the net utility to the government is
U inf ormed 
WTP j, Gj, v j
	
=

1 −v j
	 
WTP j −Gj
	
∗1

WTP j > Gj

−v j,
where the first term is the surplus from investing the remaining
fraction 1 −vj in the policy and the second term is the cost of
paying for the information.
The value to the government of learning the true willingness
to pay and cost for policy j is the value of vinf o
j
which solves the
following equation:
E

U inf ormed 
WTP j, Gj, vinf o
j

|
ˆ
WTP j, ˆ
Gj

= EU uninf ormed 
ˆ
WTP j, ˆ
Gj

.
(9)
Here, vinf o
j
equates the government’s expected utility in the case
where it spends vinf o
j
to receive additional information and the
case where it remains uninformed. The expectation in the left side
of equation (9) is taken with respect to the distribution of the true
parameters, (WTPj, Gj), given the estimates, (
ˆ
WTP j, ˆ
Gj). Since
we assume uninformed priors, this distribution is parameterized
by the sampling distribution of the estimates. This implicitly
defines vinf o
j
as the value that makes one indifferent to remaining
uninformed versus paying for the information and making a
decision based on it.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1269
FIGURE VIII
Value of Information
This figure presents the value of information, vinfo, discussed in Section VI.A,
for each policy in our sample as a function of the average age of the policy
beneficiaries. See Figure III for an explanation of the color scheme.
1. Results.
We estimate the value of info in equation (9) both
for each individual policy and for our category averages. Figure
VIII presents the results of vinf o
j
for each policy, j, plotted relative
to the age of the policy’s beneficiaries. Broadly, we find the highest
values of future research for policies with uncertain long-run
effects on children. For example, we estimate vinf o
FS = $0.50 for
the introduction of food stamps. Moreover, we also find large
values of information for policies with potential indirect effects on
children and uncertain effects on adults. We also find large values
of information for college subsidies to parents (shown in green;
color version available online). This reflects the fact that these
policies have highly uncertain effects on college attainment, and
small increases in attainment can translate into large gains. In
contrast, we find smaller values of information for policies where
the effects have been already precisely estimated. For example,
we find the evidence-based policy maker would be willing to
pay little to remove the statistical uncertainty in the estimated
impact of disability insurance on labor earnings (e.g., we estimate
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1270
THE QUARTERLY JOURNAL OF ECONOMICS
the policy maker is willing to pay $0 to learn the precise impact
of assignment to a more lenient disability insurance judge). This
lower value of information reflects the relatively high precision
of existing estimates in those studies.
2. Administrative verus Survey Data: Long-Run Impacts of
Food Stamps.
Our estimates in Figure VIII report the value
of learning the true effect of the policy. In practice, the true
effect is never observable. That said, improved access to larger
administrative data sets can help obtain more precise effects
of government policies. For example, a policy maker can decide
whether a researcher should use a survey data set for the analysis
or obtain access to linked administrative data on the population.
To illustrate this decision, we consider the case of the intro-
duction of food stamps discussed in Section III.C. Earlier work
by Hoynes and Schanzenbach (2009) used the Panel Study of
Income Dynamics (PSID) survey data set to identify the long-run
effect of food stamps on children’s outcomes. More recently, Bailey
et al. (2019) used linked census data to estimate those effects
more precisely. Here, we imagine that a policy maker is deciding
whether to introduce food stamps based on the existing evidence.
Consider the hypothetical example that they know the PSID
estimates from Hoynes and Schanzenbach (2009),
ˆ
WTP
PSID and
ˆ
FE
PSID. Suppose that they can instead invest vCensus to learn the
estimates with the same statistical precision as those found in
Bailey et al. (2019) based on census data. Instead of learning the
true value of WTP and FE, the policy maker learns
ˆ
WTP
Census
and
ˆ
FE
Census. The policy maker will expect these estimates to
be drawn from the PSID sampling distribution but contain the
standard errors found in the census data estimates. The value of
learning the census estimates, vCensus, then solves
E
 
1 −vCensus
U

ˆ
WTP
Census, ˆ
FE
Census
× 1

ˆ
WTP
Census > 1 −
ˆ
FE
Census
−vCensus |
ˆ
WTP
PSID, ˆ
FE
PSID
= U

ˆ
WTP
PSID, ˆ
FE
PSID
1

ˆ
WTP
PSID > 1 −
ˆ
FE
PSID
(10)
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1271
The left side of equation (10) is the expected value of investing
in administrative data at a price vCensus. The right side is the
expected value of the policy if she makes her decision using the
information in the PSID.
We reconstruct the estimates of the WTP and FE for the
introduction of food stamps using the estimates from Hoynes and
Schanzenbach (2009) in place of those in Bailey et al. (2019), nor-
malizing by the mechanical program cost. This yields an infinite
point estimate for our MVPF, and we find a willingness to pay
estimate of 6.06 (95% CI of [−12.07, 23.78]) and cost of −0.19 (95%
CI of [−5.19, 4.92]). These estimates are notably less precise than
those using the results from Bailey et al. (2019) that use census
data, which generate a WTP of 1.09 (95% CI of [−2.45, 4.55]).
Plugging these estimates into equation (10) suggests the
policy maker would be willing to invest $0.24 per dollar of
investment in the food stamp program to learn the long-run
estimates from census data instead of PSID data. This exercise
illustrates that if the policy maker only knew the PSID estimates,
there would be a large value in learning additional information
before making this investment decision.
This is, of course, a stylized exercise. We are imagining a
policy maker that sees the ex post evaluation of a policy prior
to making her decision—something that is clearly not feasible.
The goal here is merely to illustrate potential value of expanding
access to administrative data sets that can generate more precise
estimates of long-run policy effects.
VI.B. Comparison to Benefit-Cost Ratios
While we focus on computing the MVPF for each policy, the
most common form of welfare analysis in previous literature is
benefit-cost analysis, as in equation (4). With that in mind, we
compare our results to the benefit-cost ratios for the same policies.
Figure IX, Panel A plots the benefit-cost ratio for a deadweight
loss of φ = 50% as in Heckman et al. (2010) as a function of the
age of the beneficiary of the policy. Our general conclusion about
the high returns to investment in children would remain true
even if one used a benefit-cost ratio instead of the MVPF. The
average benefit-cost ratio is 4.13 for child education, 5.30 for child
health, and 6.78 for college policies. In contrast, we find smaller
benefit-cost ratios for adult policies—often less than 1.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1272
THE QUARTERLY JOURNAL OF ECONOMICS
FIGURE IX
Comparison to CBA
This figure presents estimates of benefit-cost ratios for all policies evaluated
in the article and shows their relationship to the MVPF. The method for calcu-
lating these benefit-cost ratios is outlined in Section II. We assume a marginal
deadweight loss of φ = 50% for these calculations. Panel A plots the benefit-cost
ratio of each policy as a function of the age of the beneficiaries, along with cat-
egory average estimates and their confidence intervals. The capped lines show
the 95% bootstrapped confidence intervals with adjustments discussed in Online
Appendix H. Panel B plots the benefit-cost ratio of each policy as a function of the
MVPF estimate for the policy. See Figure III for an explanation of the color scheme.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1273
To directly compare the two methods of welfare analysis,
Figure IX, Panel B plots the benefit-cost ratio on the vertical
axis (again for φ = 50%) against the MVPF on the horizontal
axis. In general, we find a fairly monotonic relationship—policies
with high benefit-cost ratios also have high MVPFs. There are,
however, some notable distinctions. For example, the Medicaid
expansion to children born after September 30, 1983, has an infi-
nite MVPF but a BCR of just 1.37. Similarly, the 1981 top tax rate
reduction has an infinite MVPF but a benefit-cost ratio of 1.67. By
the standards of benefit-cost ratios, these policies may not appear
all that desirable, even though the MVPF point estimates imply
that they pay for themselves and provide a Pareto improvement.
The difference between the MVPF and benefit-cost ratio in
these cases reflects the fact that the benefit-cost ratio places all
causal effects of the program in the numerator while the MVPF
incorporates effects based on their incidence. In particular, the
numerator of the MVPF captures the effects on beneficiaries while
the denominator captures all effects on the government budget.
In measuring the welfare effects of the 1983 Medicaid expansion
and the 1981 tax cut, MVPF places all fiscal externalities in
the denominator. The results show us that these policies have
substantial benefits and limited or no net government cost. In
the benefit-cost ratio framework these reforms would have been
interpreted as high-cost policies with substantial benefits.
The second crucial distinction between the MVPF and benefit-
cost ratio is how the two approaches conceptually close the budget
constraint. While the MVPF closes the budget constraint by com-
paring MVPFs of different policies (and aggregating using Okun’s
bucket as in equation (3)), the same consistency does not exist
in the benefit-cost ratio approach. In many cases, benefit-cost
analysis includes no discussion of closing the budget constraint.
In cases where the concept is addressed, it is customary to close
the budget constraint in the same manner regardless of the policy
context. For example, BCRs in Heckman et al. (2010) and Garc´
ıa
et al. (2016) imagine that the policy was funded by an increase
in the marginal tax rate that led to a distortion in tax revenue
and a deadweight loss of φ. Consequently, the deadweight loss
parameter φ in equation (4) is not context dependent.
To see how this matters, consider the 1993 tax reform
that simultaneously raised top marginal income tax rates and
expanded the EITC. One could, in principle, use a benefit-cost
ratio to evaluate whether the EITC expansion was desirable. As
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1274
THE QUARTERLY JOURNAL OF ECONOMICS
shown in Figure IX, Panel A, the benefit-cost ratio for the 1993
EITC expansion is 0.74 after adjusting for a 50% deadweight
loss. The costs exceed the benefits and so, if the government were
applying a strict benefit-cost test, we would not expect this policy
to be implemented. This is because the hypothetical 50% cost of
raising the funds is too large to justify the expenditure.
That said, the goal of the EITC expansion was to provide
redistributive benefits to low-income workers. Its MVPF is 1.12,
near the highest among policies targeting adults. Rather than
ruling this out as a means of redistribution, we can compare
the MVPF of the EITC to the MVPF of a tax increase used to
fund this policy. Comparisons of MVPFs correspond to precise
statements of social welfare using Okun’s bucket. As noted, the
MVPF point estimate for the 1993 top tax rate change is 1.85. If
society prefers giving $1.12 to a low-income worker on EITC to
giving $1.85 to a high-income individual facing the top marginal
income tax rate, then the policy is welfare enhancing despite its
relatively low benefit-cost ratio.
VI.C. Welfare Reform: Lessons for Future RCTs
We end with a lesson of how an MVPF perspective can help
inform the design of RCTs. Throughout, we aimed to include all
possible MVPFs in the categories we considered. We included any
policy where we thought we could provide reasonable measures
of both costs and WTP. One set of notable omissions are the state-
level welfare reforms made by states that sought to increase fam-
ily self-sufficiency. Throughout the 1980s and early 1990s, states
experimented with a range of reforms to cash welfare programs
that imposed term limits, provided job training and other educa-
tional services, and provided job search and placement assistance.
The omission of these reforms is not because they were not
analyzed. Many states rigorously evaluated the effect of these
reforms. Upward of 100,000 participants were enrolled into 27
RCTs nationwide (Greenberg, Deitch, and Hamilton 2010). These
RCTs measured and provided a clear estimate of the net cost
of each reform. However, the design of the policies enacted in
each state makes it difficult to understand their welfare effects.
Generally, programs contained both a carrot and a stick.81 As
81. Welfare reform experiments were expected to place no additional costs on
the federal government and so it is natural that states bundled increases in some
types of financial support with potential decreases in others.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1275
a result, we cannot even accurately sign the WTP. As noted by
Manpower Demonstration Research Corporation (MDRC) who
implemented the evaluations of these policies,82 “all [programs]
contained a core quid pro quo arrangement in which the gov-
ernment would offer education, training, job search assistance,
and support services to people receiving cash welfare, while
most recipients—the majority of them single parents—would be
required to participate in such services in order to qualify for
benefits.” Although we can evaluate whether the government
saved money, we do not know if the people in these programs
benefited from their participation. It may be that government
revenue gains were the result of expanded job opportunities due
to program participation. In that case, willingness to pay would
be positive. By contrast, it may be that the government revenue
gains were the result of stricter attendance requirements that
drove individuals off welfare. In that case, willingness to pay
would be negative.83
This highlights the value of isolating the carrot and the stick
into separate RCTs.84 It also demonstrates the value of designing
experiments to estimate individual WTP for nonmarket goods
such as job training, job search assistance, or other educational
policies. In Appendix Figure X, we conduct a range of bounding ex-
ercises that attempt to construct lower and upper bounds on WTP
for these welfare reform programs. Unfortunately, the bounds
are very wide. In many cases, the policies are Pareto dominated,
MVPF < 0, under one set of assumptions and represent a Pareto
improvement, MVPF = ∞, under another set of assumptions.85
Despite substantial expenditures on the evaluation of these re-
forms, the designs of these reforms in each state make it difficult
82. See https://www.mdrc.org/project/evaluations-state-welfare-work-
programs#design-site-data-sources (accessed on July 7, 2019).
83. Previous work (Greenberg, Deitch, and Hamilton 2010) has conducted a
cost-benefit analysis of these reforms by assuming willingness to pay is given by
after-tax earnings. However, if the term limit is what causes individuals to choose
to move off of welfare and into the labor market (thus increasing earnings), the
envelope theorem would suggest the WTP is negative, even if after-tax earnings
increase.
84. Welfare reform RCTs have been criticized for not experimentally varying
each of the components of welfare reform (see Grogger and Karoly 2005). The
MVPF framework suggests bundling carrots and sticks into a single treatment is
particularly problematic for conducting welfare analysis, because it is difficult to
know even whether willingness to pay is positive or negative.
85. In fact, we find policies that follow this pattern in each subcategory of
welfare reform programs. These subcategories include job search assistance.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1276
THE QUARTERLY JOURNAL OF ECONOMICS
to know whether this massive shift in providing welfare benefits
to low-income families led to an increase or decrease in welfare.
VII. CONCLUSION
In this article, we examine the MVPF of 133 different histori-
cal policies over the past half-century in the United States. We find
a clear and persistent pattern that direct investments in children
have yielded the largest MVPFs. There is a large “bang for the
buck” associated with a range of expenditures on children from
early education to child health insurance to college expenditures.
We also demonstrate that in a meaningful number of cases,
these policies pay for themselves. In particular, when government
expenditures boost human capital, the resulting increase in
net government revenue can offset the policy’s up-front costs.
From a taxpayer perspective, these expenditures on children are
investments, rather than just transfers.
We find that opportunities for high-return investments in
children have persisted across policy categories for many decades.
This is, however, no guarantee that all future investment in these
categories will produce high MVPFs. Indeed, we find that MVPFs
vary substantially within policy categories. Low-return policies
exist even in high-return categories. This highlights the value of
further understanding the mechanisms behind the high MVPFs
of successful historical investments.
Even in cases where there is existing research, much remains
unknown about the welfare consequences of government policy. To
that aim, we quantify the value of future work that uses new data
to reduce estimate uncertainty. We show that in many cases, an
evidence-based policy maker seeking to maximize social welfare
should be willing to make substantial budgetary expenditures to
learn more about policy effectiveness. In particular, our results
highlight the value of expanded use of administrative data for
policy analysis.
The 133 policies included in this article are just a small subset
of those that could be analyzed using the MVPF. We do not discuss
the MVPF of crime policies, environmental policies, macroeco-
nomic stabilization policies, or infrastructure policies, among
many others. With careful tracking of willingness to pay and net
costs, the MVPF can be used in any of these contexts and can guide
cost-benefit analyses. We leave that analysis for future work.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1277
TABLE I (CONTINUED)
Sample
Program
Label
Year
Age of
Paper
Papers
Implemented Beneficiaries Baseline Restricted Extended Estimates
Utilized
MVPF
Head Start Introduction
Head Start
1965
4
x
x
x
Johnson and Jackson (2019)
Head Start Regression
Head
1970
4
x
Ludwig and Miller (2007)
Discontinuity
Start RD
Head Start
Head
2002
3
x
x
Kline and Walters (2016)
Impact Study
Start RCT
K–12 School
K12
1991
11
x
x
x
Hyman (2017)
Finance Reform
Spend
K–12 School
K12 Spend
1994
11
x
Heckman et al. (2010)
Spending in Michigan
Mich.
Perry Preschool Program Perry Preschool
1962
4
x
x
x
Heckman et al. (2011)
College adult
American Opportunity
AOTC (IS)
2011
25
x
x
Bulman and Hoxby (2015)
Tax Credit,
Independent Single
Filers at Phase Start
American Opportunity
AOTC (JE)
2011
55
x
x
Bulman and Hoxby (2015)
Tax Credit, Joint Filers
at Phase End
American Opportunity
AOTC (JS)
2011
20
x
x
Bulman and Hoxby (2015)
Tax Credit,
Joint Filers at
Phase Start
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1278
THE QUARTERLY JOURNAL OF ECONOMICS
TABLE I (CONTINUED)
Sample
Program
Label
Year
Age of
Paper
Papers
Implemented Beneficiaries Baseline Restricted Extended Estimates
Utilized
MVPF
American Opportunity
AOTC (SI)
2009
20
x
x
Bulman and Hoxby (2015)
Tax Credit,
Simulated Instrument
American Opportunity
AOTC (SE)
2011
55
x
x
Bulman and Hoxby (2015)
Tax Credit,
Single Filers at
Phase End
American Opportunity
AOTC (SS)
2011
20
x
x
Bulman and Hoxby (2015)
Tax Credit,
Single Filers at
Phase Start
Hope Tax Credit
HOPE Cred.
1999
20
x
x
Turner (2011)
Hope Tax Credit,
HTC (IS)
2007
25
x
x
Bulman and Hoxby (2015)
Independent Single
Filers at Phase Start
Hope Tax Credit,
HTC (JE)
2007
20
x
x
Bulman and Hoxby (2015)
Joint Filers at Phase End
Hope Tax Credit,
HTC (JS)
2007
20
x
x
Bulman and Hoxby (2015)
Joint Filers at
Phase Start
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1279
TABLE I (CONTINUED)
Sample
Program
Label
Year
Age of
Paper
Papers
Implemented Beneficiaries Baseline Restricted Extended Estimates
Utilized
MVPF
Hope Tax Credit,
HTC (SE)
2007
20
x
x
Bulman and Hoxby (2015)
Single Filers at
Phase End
Hope Tax Credit, Single
HTC
2007
55
x
x
Bulman and Hoxby (2015)
Filers at Phase Start
(SS)
Hope and Lifetime
HOPE/LLC
1998
55
x
x
Long (2004)
Learners Tax Credits
Pell Grants
Adult
1973
28
x
x
Seftor and Turner (2002)
Introduction to Adults
Pell
Tax Deduction for Postsecondary
Tuition
2006
55
x
x
Hoxby and Bulman (2016)
Tuition, Joint Filers at Phase End
Deduc (JE)
LaLumia (2012)
Tax Deduction for Postsecondary
Tuition
2006
55
x
x
Hoxby and Bulman (2016)
Tuition, Joint Filers at Phase Start
Deduc (JS)
LaLumia (2012)
Tax Deduction for Postsecondary
Tuition
2006
55
x
x
Hoxby and Bulman (2016)
Tuition, Single Filers at Phase End
Deduc (SE)
LaLumia (2012)
Tax Deduction for Postsecondary
Tuition
2006
20
x
x
Hoxby and Bulman (2016)
Tuition, Single Filers at Phase Start Deduc (SS)
LaLumia (2012)
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1280
THE QUARTERLY JOURNAL OF ECONOMICS
TABLE I (CONTINUED)
Sample
Program
Label
Year
Age of
Paper
Papers
Implemented Beneficiaries Baseline Restricted Extended Estimates
Utilized
MVPF
College child
Cal Grant, GPA Threshold
Cal Grant GPA
1998
20
x
x
x
Bettinger et al. (2019)
Cal Grant, Income Threshold
Cal Grant Inc
1998
20
x
x
x
Bettinger et al. (2019)
City University of New York
CUNY
2009
20
x
x
Marx and Turner (2018)
Pell Grants
Pell
Community College Tuition
CC Mich
2005
20
x
x
Acton (2018)
Changes, Michigan
Community College Tuition
CC
2005
20
x
x
Denning (2017)
Changes, Texas
Texas
District of Columbia Tuition
DC
1999
20
x
x
Abraham and Clark (2006)
Assistance Grant Program
Grant
Florida International University
Admissions at GPA Threshold
FIU GPA
1999
20
x
x
x
Zimmerman (2014)
Florida Student Access Grant
Florida Grant
2001
20
x
x
Castleman and Long (2016)
Free Application for Federal
Free FAFSA
2008
20
x
U.S. Department of Education
(Dep)
Office of Postsecondary
Student Aid, Dependent Year
Impact
Education (2010)
Bettinger et al. (2012)
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1281
TABLE I (CONTINUED)
Sample
Program
Label
Year
Age of
Paper
Papers
Implemented Beneficiaries Baseline Restricted Extended
Estimates
Utilized
MVPF
Free Application
Free FAFSA
2008
20
x
U.S. Department
for Federal Student Aid,
(Indep)
of Education Office
Independent Year
of Postsecondary
Impact
Education (2010)
Bettinger et al. (2012)
Georgia HOPE Scholarship
Georgia HOPE
1995
20
x
x
Cornwell et al. (2003)
Cornwell et al. (2006)
HAIL Michigan Aid
Awareness Letter
HAIL Aid
2016
20
x
Dynarski et al. (2018)
Hoekstra (2009)
Kalamazoo Promise
Scholarship
Kalamazoo
2006
20
x
x
Bartik et al. (2016)
Bartik et al. (forthcoming)
Massachussetts Adams
Scholarship
MA Scholarship
2005
20
x
x
Cohodes and Goodman (2014)
Goodman (2008)
Pell Grants in Ohio
Ohio Pell
2000
19
x
x
Bettinger (2004)
Pell Grants
TN Pell
2008
20
x
x
U.S. Department of Education
in Tennessee
Office of Postsecondary
Education (2010)
Carruthers and Welch (2019)
Pell Grants in Texas
Texas Pell
2008
20
x
x
x
Denning et al. (2019)
Social Security Student
Benefit Program
Soc Sec College
1982
20
x
x
Dynarski (2003)
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1282
THE QUARTERLY JOURNAL OF ECONOMICS
TABLE I (CONTINUED)
Sample
Program
Label
Year
Age of
Paper
Papers
Implemented
Beneficiaries
Baseline
Restricted
Extended
Estimates
Utilized
MVPF
Spending at Colleges from State
Appropriations
College Spend
2001
20
x
x
Deming and Walters (2017)
Tennessee HOPE Scholarships
TN Hope
2008
20
x
x
Bruce and Carruthers (2014)
Tuition Cuts at Colleges from
State Appropriations
College Tuition
2001
20
x
x
Deming and Walters (2017)
Wisconsin Scholar Grant to
Low-Income College Students
WI Scholarship
2009
20
x
x
Goldrick-Rab et al. (2016)
Job training
Job Corps
Job Corps
1995
19
x
x
x
Schochet et al. (2006, 2008)
Schochet (2018)
Job Training Partnership Act,
Adults
JTPA Adult
1988
34
x
x
x
Bloom et al. (1997)
Job Training Partnership Act,
Youth
JTPA Youth
1988
19
x
x
x
Bloom et al. (1997)
JobStart
JobStart
1986
19
x
x
x
Cave et al. (1993)
National Supported Work
Demonstration, Adult Women
NSW Women
1976
34
x
x
x
Hollister, Kemper, and Maynard (1984)
Couch (1992)
National Supported Work
Demonstration, Ex-Addicts
NSW Ex-Addict
1976
33
x
x
x
Hollister, Kemper, and Maynard (1984)
National Supported Work
Demonstration, Ex-Offenders
NSW Ex-Offender
1976
33
x
x
x
Hollister, Kemper, and Maynard (1984)
National Supported Work
Demonstration, Youth
NSW Youth
1976
18
x
x
x
Hollister, Kemper, and Maynard (1984)
Couch (1992)
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1283
TABLE I (CONTINUED)
Sample
Program
Label
Year
Age of
Paper
Papers
Implemented
Beneficiaries
Baseline
Restricted
Extended
Estimates
Utilized
MVPF
Work Advance
Work Advance
2012
34
x
x
x
Schaberg (2017)
Hendra et al. (2016)
Year Up
Year Up
2013
21
x
x
x
Fein and Hamadyk (2018)
Panel B: Social insurance
Disability ins.
Disability Insurance
Changes in Benefit
Generosity
DI Generosity
2004
50
x
x
x
Gelber, Moore, and Strand (2017)
Disability Insurance
Judge Leniency
DI Judge
2005
47
x
x
x
Maestas, Mullen, and Strand (2013)
Gelber, Moore, and Strand (2017)
Disability Insurance
DI Examiner
1995
48
x
x
x
Gelber, Moore, and Strand (2017)
Medical Examiner
French and Song (2014)
Leniency
Disability Insurance to
Veterans
DI Veterans
2001
53
x
x
x
Autor et al. (2016)
Health adult
Health Insurance
Mass HI
2011
44
x
x
x
Hendren (2017c)
Subsidies in
(150%FPL)
Finkelstein, Hendren, and
Massachusetts to Indi-
Shepard (2019)
viduals at 150% of the
Federal Poverty Line
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1284
THE QUARTERLY JOURNAL OF ECONOMICS
TABLE I (CONTINUED)
Sample
Program
Label
Year
Age of
Paper
Papers
Implemented Beneficiaries Baseline Restricted Extended Estimates
Utilized
MVPF
Health Insurance
Mass HI (200%FPL)
2011
44
x
x
x
Hendren (2017c)
Subsidies in
Finkelstein, Hendren, and
Massachusetts to
Shepard (2019)
Individuals at 200% of the
Federal Poverty Line
Health Insurance Subsidies Mass HI (250%FPL)
2011
44
x
x
x
Hendren (2017c)
in Massachusetts to
Finkelstein, Hendren, and
Individuals at 250% of the
Shepard (2019)
Federal Poverty Line
Medicare Introduction
Medicare
1965
78
x
x
x
U.S. Census Bureau (1966)
in 1965
Intro
Finkelstein and McKnight (2008)
Oregon Health Insurance
Oregon
2008
42
x
x
x
Finkelstein et al. (2012)
Experiment (Provided to
Health
Finkelstein, Hendren, and
Single Adults)
Luttmer (2019)
Taxation of Medigap
Policies
Medigap Tax
2002
75
x
x
x
Cabral and Mahoney (2019)
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1285
TABLE I (CONTINUED)
Sample
Program
Label
Year
Age of
Paper
Papers
Implemented Beneficiaries Baseline Restricted Extended
Estimates
Utilized
MVPF
Health child
Medicaid Expansion to
MC
1990
11
x
x
x
Lo Sasso and Seamster (2007)
Children Born after
Child 83+
Wherry and Meyer (2016)
September 30, 1983
Wherry et al. (2018)
Medicaid Expansions to
MC Pregnant &
1986
0
x
x
x
Dave et al. (2015)
Pregnant Women &
Infants
Currie and Gruber (1996)
Infants
Miller and Wherry (2019)
Medicaid Expansions to
MC Child
1986
9
x
x
x
Boudreaux, Golberstein, and McAlpine (2016)
Young Children
(State Exp)
Brown, Kowalski, and Lurie (2017)
Medicaid Introduction to
MC
1968
9
x
x
x
x
Goodman-Bacon (2017)
AFDC-eligible Families
Intro
Supplemental Security
Income
Supplemental Security
Income Age 18
SSI
Review
1996
18
x
x
x
x
Deshpande (2016)
Medical Review
Supplemental Security
SSI
2005
48
x
x
x
Deshpande (2016)
Income Medical
Judge
French and Song (2014)
Examiner Leniency
Gelber, Moore, and Strand (2017)
U.S. Social Security Administration (2014)
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1286
THE QUARTERLY JOURNAL OF ECONOMICS
TABLE I (CONTINUED)
Sample
Program
Label
Year
Age of
Paper
Papers
Implemented Beneficiaries Baseline Restricted Extended
Estimates
Utilized
MVPF
Unemployment insurance
Unemployment Insurance
Benefit Changes (Diff in
Diff Across States in
Chetty 2008)
UI Ben
(State Max)
1992
37
x
x
x
Chetty (2008)
Gruber (1997)
Hendren (2017b)
Schmieder and Von Wachter
(2016)
Unemployment Insurance
Benefit Changes (Diff in
Diff Across States in Katz
and Meyer 1990)
UI Ben
(DD)
1980
33
x
x
x
Gruber (1997)
Hendren (2017b)
Katz and Meyer (1990)
Schmieder and Von Wachter
(2016)
Unemployment Insurance
UI Ben
1992
37
x
x
x
Gruber (1997)
Benefit Changes (Diff in
(DD w UR)
Hendren (2017b)
Diff Across States in Kroft)
Kroft and Notowidigdo (2016)
and Notowidigdo 2016)
Schmieder and Von Wachter
(2016)
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1287
TABLE I (CONTINUED)
Sample
Program
Label
Year
Age of
Paper
Papers
Implemented Beneficiaries Baseline Restricted Extended
Estimates
Utilized
MVPF
Unemployment Insurance
Benefit Changes in Georgia
UI Ben
1979
42
x
x
x
Gruber (1997)
(GA)
Hendren (2017b)
Schmieder and Von Wachter
(2016)
Solon (1985)
Unemployment Insurance
Benefit Changes in Missouri
(Expansion Estimates)
UI Ben
2005
42
x
x
x
Card et al. (2015)
(MO Exp)
Gruber (1997)
Hendren (2017b)
Schmieder and Von Wachter
(2016)
Unemployment Insurance
UI Ben
2010
42
x
x
x
Card et al. (2015)
Benefit Changes in Missouri
(MO Rec)
Gruber (1997)
(Recession Estimates)
Hendren (2017b)
Schmieder and Von Wachter
(2016)
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1288
THE QUARTERLY JOURNAL OF ECONOMICS
TABLE I (CONTINUED)
Sample
Program
Label
Year
Age of
Paper
Papers
Implemented Beneficiaries Baseline Restricted Extended Estimates
Utilized
MVPF
Unemployment Insurance
Benefit Changes in
New York
UI Ben
(NY)
1989
42
x
x
x
Gruber (1997)
Hendren (2017b)
Meyer and Mok (2007)
Schmieder and Von Wachter (2016)
Unemployment Insurance
Benefit Changes via
Regression Kink in
Benefit Schedule
UI Ben
1980
34
x
x
x
Gruber (1997)
(RK)
Hendren (2017b)
Landais (2015)
Schmieder and Von Wachter (2016)
Unemployment Insurance
Duration Extensions (Diff
in Diff Across States in
Katz and Meyer 1990)
UI Dur
1980
33
x
x
x
Ganong and Noel (2019)
(DD)
Gruber (1997)
Hendren (2017b)
Katz and Meyer (1990)
Schmieder and Von Wachter (2016)
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1289
TABLE I (CONTINUED)
Sample
Program
Label
Year
Age of
Paper
Papers
Implemented Beneficiaries Baseline Restricted Extended
Estimates
Utilized
MVPF
Unemployment Insurance
Duration Extensions in
Missouri
UI Dur
2011
42
x
x
x
Ganong and Noel (2019)
(MO)
Gruber (1997)
Hendren (2017b)
Johnston and Mas (2018)
Schmieder and Von Wachter (2016)
Panel C: In-kind transfers
Housing vouchers
Effects of Housing Vouchers
HCV RCT
2000
31
x
x
Jacob and Ludwig (2012)
on AFDC Families Experiment
to Welfare
Mills et al. (2006)
Wood, Turnham, and Mills (2008)
Housing Vouchers
HCV
1997
31
x
x
Jacob and Ludwig (2012)
in Chicago
Chicago Lottery
Jacob, Ludwig, and Kapustin (2014)
Jobs Plus
Jobs+
1998
35
x
Bloom, Riccio, and Verma (2005)
Riccio (2006)
MTO
Moving to Opportunity
Experiment Providing
Vouchers and Counseling
MTO
1996
10
x
x
x
Chetty, Hendren, and Katz (2016)
Goering et al. (1999)
Sanbonmatsu et al. (2011)
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1290
THE QUARTERLY JOURNAL OF ECONOMICS
TABLE I (CONTINUED)
Sample
Program
Label
Year
Age of
Paper
Papers
Implemented Beneficiaries Baseline Restricted Extended
Estimates
Utilized
MVPF
Nutrition
Special Supplemental
WIC
1975
26
x
Black, Devereux, and Salvanes (2007)
Nutrition Program for
Hoynes, Page, and Stevens (2011)
Women, Infants, and
Whitmore (2002)
Children
Supplemental Nutrition
Assistance Program
Application Assistance
SNAP Assist
2016
69
x
x
x
x
Finkelstein and Notowidigdo (2019)
Supplemental Nutrition
Assistance Program
Application
Information
SNAP Info
2016
69
x
x
x
x
Finkelstein and Notowidigdo (2019)
Supplemental Nutrition
Assistance Program
Introduction
SNAP Intro
1968
32
x
x
x
Hoynes, Schanzenbach, and Almond (2016)
Almond, Hoynes, and Schanzenbach (2011)
Bailey et al. (2019)
Hoynes, Page, and Stevens (2011)
Hoynes and Schanzenbach (2012)
East (2018)
Finkelstein and Notowidigdo (2019)
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1291
TABLE I (CONTINUED)
Sample
Program
Label
Year
Age of
Paper
Papers
Implemented Beneficiaries Baseline Restricted Extended Estimates
Utilized
MVPF
Panel D: Taxes and cash transfers
1986 Earned Income
EITC 1986
1986
28
x
x
x
Ackerman, Holtzblatt, and Masken (2009)
Tax Credit Expansion
Tax Policy Center (2016)
Blank and Ruggles (1996)
Hotz and Scholz (2003)
Meyer and Rosenbaum (2001)
Moffitt (2002)
Scholz (1993)
Crouse and Waters (2014)
Eissa and Hoynes (2004)
Eissa and Liebman (1996)
1993 Earned Income
EITC 1993
1993
29
x
x
x
Tax Policy Center (2016)
Tax Credit Expansion
Hotz and Scholz (2003)
Dahl and Lochner (2012)
Meyer and Rosenbaum (2001)
Bastian and Michelmore (2018)
Chetty, Friedman, and Rockoff (2011)
Crouse and Waters (2014)
Eissa and Hoynes (2004)
Hoynes and Patel (2018)
Manoli and Turner (2018)
Maxfield (2018)
Michelmore (2013)
Meyer and Rosenbaum (2001)
Ackerman, Holtzblatt, and Masken (2009)
Currie and Cole (1993)
Moffitt (2003)
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1292
THE QUARTERLY JOURNAL OF ECONOMICS
TABLE I (CONTINUED)
Sample
Program
Label
Year
Age of
Paper
Papers
Implemented Beneficiaries Baseline Restricted Extended Estimates
Utilized
MVPF
Aid to Families with
AFDC Term Limits
1996
27
x
x
x
Grogger (2003)
Dependent Children (Term
Limit Modifications)
Pavetti (1995)
Alaska Permanent Fund
Alaska UBI
1998
34
x
x
x
Ackerman, Holtzblatt, and Masken (2009)
Dividend
Bhargava and Manoli (2015)
Blank and Ruggles (1996)
Hotz and Scholz (2003)
Jones and Marinescu (2018)
Meyer and Rosenbaum (2001)
Moffitt (2002)
Paycheck Plus Experiment
Providing EITC-benefits to
Adults without Dependents
Paycheck+
2013
35
x
x
x
Miller et al. (2017)
Seattle-Denver Income
Neg Inc Tax
1971
35
x
x
x
Price and Song (2018)
Maintenance Experiment
U.S. Social Security Administration (2018)
Tax Foundation (2013)
U.S. Department of Health
Human Services (1983)
Von Wachter, Song, and Manchester (2011)
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1293
TABLE I (CONTINUED)
Sample
Program
Label
Year
Age of
Paper
Papers
Implemented Beneficiaries Baseline Restricted Extended
Estimates
Utilized
MVPF
Top taxes
Top Tax 2013 Increases
from Affordable Care Act
Top Tax 2013
2013
49
x
x
x
Kawano, Weber, and
Whitten (2016)
Hendren (2017a)
Top Tax Rate Increase in
Top Tax 1993
1993
49
x
x
x
Atkinson, Piketty, and Saez (2011)
Omnibus Budget
Reconciliation Act 1993
Carroll (1998)
Top Tax Rate Reductions
Top Tax 1986
1986
49
x
x
x
Atkinson, Piketty, and Saez (2011)
in Tax Reform Act of 1986
Auten and Caroll (1999)
Top Taxes, Economic
Top Tax 2001
2001
49
x
x
x
Atkinson, Piketty, and Saez (2011)
Growth and Tax Relief
Reconciliation Act 2001
Heim (2009)
Top Taxes, Economic
Top Tax 1981
1981
49
x
x
x
Atkinson, Piketty, and Saez (2011)
Recovery Tax Act 1981
Saez (2003)
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1294
THE QUARTERLY JOURNAL OF ECONOMICS
TABLE I (CONTINUED)
Sample
Program
Label
Year
Age of
Paper
Papers
Implemented Beneficiaries Baseline Restricted Extended Estimates
Utilized
MVPF
Panel E: Welfare reform
Welfare to Work Alameda
GAIN Alm.
1988
31
x
Freedman et al. (1996)
Mandatory Mixed-
Initial-Activity Programs
Greenberg, Deitch, and Hamilton (2010)
Welfare to Work Atlanta
HCD NEWWS Atl.
1992
33
x
Greenberg, Deitch, and Hamilton (2010)
Mandatory Education-
First Programs
Hamilton et al. (2001)
Welfare to Work Atlanta
LFA NEWWS Atl.
1992
33
x
Hamilton et al. (2001)
Mandatory Job-Search-
First Programs
Greenberg, Deitch, and Hamilton (2010)
Welfare to Work Butte
Mandatory Mixed-Initial-
Activity Programs
GAIN Butte
1987
31
x
Hamilton et al. (2001)
Freedman et al. (1996)
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1295
TABLE I (CONTINUED)
Sample
Program
Label
Year
Age of
Paper
Papers
Implemented Beneficiaries Baseline Restricted Extended
Estimates
Utilized
MVPF
Welfare to Work Columbus
Integrated Mandatory
NEWWS Col. Int.
1992
32
x
Greenberg, Deitch, and
Hamilton (2010)
Education-First Programs
Hamilton et al. (2001)
Welfare to Work Columbus
Traditional Mandatory
Education-First Programs
NEWWS Col. Trad.
1992
32
x
Greenberg, Deitch, and
Hamilton (2010)
Hamilton et al. (2001)
Welfare to Work Connecticut
Jobs First
1996
31
x
Bloom et al. (2002)
Time-Limit-Mix Programs
Greenberg, Deitch, and
Hamilton (2010)
Welfare to Work Cook County
Mandatory Work
Experience Program
WIN Demo.
1985
32
x
Greenberg, Deitch, and
Hamilton (2010)
Brock, Butler, and Long (1993)
Welfare to Work Detroit
Mandatory Education-First
NEWWS Det.
1992
30
x
Greenberg, Deitch, and
Hamilton (2010)
Programs
Hamilton et al. (2001)
Welfare to Work Florida
Mandatory Mixed-Initial-
Proj. Ind. FL
1990
32
x
Greenberg, Deitch, and
Hamilton (2010)
Activity Programs
Kemple, Friedlander and
Fellerath (1995)
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1296
THE QUARTERLY JOURNAL OF ECONOMICS
TABLE I (CONTINUED)
Sample
Program
Label
Year
Age of
Paper
Papers
Implemented Beneficiaries Baseline Restricted Extended Estimates
Utilized
MVPF
Welfare to Work Florida
Time-Limit-Mix Programs
FTP
1994
29
x
Bloom et al. (2000)
Greenberg, Deitch, and
Hamilton (2010)
Welfare to Work Grand
Rapids Mandatory
Education-First Programs
HCD NEWWS Gr. Rap.
1991
28
x
Greenberg, Deitch, and
Hamilton (2010)
Hamilton et al. (2001)
Welfare to Work Grand
Rapids Mandatory Job-
Search-First Programs
LFA NEWWS Gr. Rap.
1991
28
x
Hamilton et al. (2001)
Greenberg, Deitch, and
Hamilton (2010)
Welfare to Work Los Angeles
Mandatory Job-Search-
First Programs
GAIN LA jobs
1996
34
x
Freedman et al. (2000)
Greenberg, Deitch, and
Hamilton (2010)
Welfare to Work Los Angeles
Mandatory Mixed-Initial-
Activity Programs
GAIN LA
1988
31
x
Greenberg, Deitch, and
Hamilton (2010)
Freedman et al. (1996)
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1297
TABLE I (CONTINUED)
Sample
Program
Label
Year
Age of
Paper
Papers
Implemented Beneficiaries Baseline Restricted Extended Estimates
Utilized
MVPF
Welfare to Work Minnesota
Earnings Supplement
Programs
MFIP
1994
29
x
Greenberg, Deitch, and
Hamilton (2010)
Miller et al. (2000)
Welfare to Work Portland
Mandatory Mixed-Initial-
Activity Programs
NEWWS Port.
1993
30
x
Greenberg, Deitch, and
Hamilton (2010)
Hamilton et al. (2001)
Welfare to Work Riverside
Mandatory Education-First
Programs
HCD NEWWS Riv.
1991
32
x
Hamilton et al. (2001)
Greenberg, Deitch, and
Hamilton (2010)
Welfare to Work Riverside
Mandatory Job-Search-First
Programs
LFA NEWWS Riv.
1991
32
x
Hamilton et al. (2001)
Greenberg, Deitch, and
Hamilton (2010)
Welfare to Work Riverside
Mandatory Mixed-Initial-
Activity Programs
GAIN Riv.
1987
31
x
Freedman et al. (1996)
Greenberg, Deitch, and
Hamilton (2010)
Welfare to Work San Diego
Mandatory Job-Search-First
Programs
SWIM
1985
31
x
Greenberg, Deitch, and
Hamilton (2010)
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1298
THE QUARTERLY JOURNAL OF ECONOMICS
TABLE I (CONTINUED)
Sample
Program
Label
Year
Age of
Paper
Papers
ImplementedBeneficiariesBaselineRestrictedExtended Estimates
Utilized
MVPF
Welfare to Work San Diego
Mandatory Mixed-
Initial-Activity Programs
GAIN SD
1987
31
x
Greenberg, Deitch, and Hamilton
(2010)
Freedman et al. (1996)
Welfare to Work San Diego
Mandatory Work
Experience Program
Work Exp. SD
1982
32
x
Greenberg, Deitch, and Hamilton
(2010)
Brock, Butler, and Long (1993)
Welfare to Work Tulane
Mandatory Mixed-Initial-
Activity Programs
GAIN Tul.
1988
31
x
Greenberg, Deitch, and Hamilton
(2010)
Freedman et al. (1996)
Welfare to Work Vermont
Earnings Supplement
Programs
WRP Earn Supp.
1994
31
x
Greenberg, Deitch, and Hamilton
(2010)
Scrivener et al. (2002)
Welfare to Work Vermont
Time-Limit-Mix Programs
WRP Time lim.
1994
31
x
Scrivener et al. (2002)
Greenberg, Deitch, and Hamilton
(2010)
Welfare to Work West
Virginia Mandatory Work
Experience Program
CWEP
1983
33
x
Brock, Butler, and Long (1993)
Greenberg, Deitch, and Hamilton
(2010)
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1299
APPENDIX
APPENDIX FIGURE I
Income Projections Using the ACS
Panels A and B present a decomposition of the elements that make up our
income projection process for the examples in Section III.A. The “Pop Avg” series
is constructed in each case from the 2015 ACS and using a 0.5% wage growth
assumption. At each age “Pop Avg” gives the mean wage level that would prevail
in the population for individuals of that age, when individuals in the treatment
group for the relevant policy were that age. This number is constructed by
assuming that the mean wage level at each age will rise (and has previously
risen) by 0.5% in each year. The “Control Forecast” series is constructed by taking
an estimate of earnings for a relevant control group at a particular age or range of
ages, then calculating the implied proportion of the “Pop Avg” series at those ages,
then projecting the series forwards (and backwards) as this constant fraction
of “Pop Avg.” The “Treatment” series is constructed by summing the observed
treatment effects in dollar terms and the “Control Forecast” series. To construct
the “Predicted” series we take the final value of the “Treatment” series, then
calculate the ratio of this value to the value of the “Pop Avg” series at that same
age, before applying this ratio to the “Pop Avg” series up to age 65. See Online
Appendix I for further details of this methodology.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1300
THE QUARTERLY JOURNAL OF ECONOMICS
APPENDIX FIGURE II
Willingness to Pay per Dollar of Programmatic Spending
This figure presents estimates of WTP normalized by initial programmatic
spending for each category-average group of policies in our baseline sample. We
plot these estimates as a function of the average age of each policy’s beneficiaries
within category. Bootstrapped 95% confidence intervals with adjustments
(discussed in Online Appendix H) are shown for the category averages. The
normalized willingness to pay of individual policies are shown in smaller dots,
color-coded to align with their respective categories.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1301
APPENDIX FIGURE III
Robustness to Child Effects
This figure assesses the impact of observing child effects on our estimates as
a function of the average age of the economic beneficiaries of the policy. Panel
A restricts our sample to the subset of policies for which we observe estimates
of the impact of the policy on children. In addition, Panel B shows projected
MVPFs for additional policies that do not observe earnings impacts but do observe
another intermediate outcome such as birthweight (AFDC), college attendance
(housing vouchers to AFDC recipients), and test scores (housing vouchers in
Chicago). Panel C reports the MVPF for the EITC under alternative methods of
incorporating indirect effects on children through test scores, college attendance,
and income of EITC more broadly.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1302
THE QUARTERLY JOURNAL OF ECONOMICS
APPENDIX FIGURE IV
Robustness to Alternative Interest Rates
This figure presents our MVPF estimates as in Figure III and the category av-
erages as in Figure IV, Panel B under alternative real interest rate assumptions,
as opposed to our baseline specification of 3%. Panel B differs slightly from our
baseline specification because we restrict to the subset of policies for which we are
able to vary the discount rate (e.g., we exclude papers where we directly import
an MVPF that relied on a particular discount rate). We omit confidence intervals
for ease of viewing, but caution the reader that the estimate for the college adult
category has a CI that includes 0 and infinity.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1303
APPENDIX FIGURE V
Robustness to Alternative Tax Rates
This figure presents our MVPF estimates as in Figure III and the category
averages as in Figure IV, Panel B under alternative tax rate assumptions. Panel
A replicates our baseline specification using the CBO estimates of the tax rates.
Panels B–D adjust the tax rate to 10%, 20%, and 30%.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1304
THE QUARTERLY JOURNAL OF ECONOMICS
APPENDIX FIGURE VI
Specification Robustness
This figure presents the category-average MVPFs from Figure III using a range
of different alternative specifications that are more conservative than our baseline
specifications. Panel A replaces our point estimate WTP measures with our
conservative measures of WTP. We report bootstrapped 95% confidence intervals
with adjustments (discussed in Online Appendix H) for each category average.
Panel B replaces our baseline income projection procedure with a procedure that
assumes zero income growth over the lifecycle. We use our restricted sample of
policies for this specification. See Online Appendix I for further details.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1305
APPENDIX FIGURE VII
Sample Restrictions
This figure presents the category-average MVPFs from Figure III using a range
of alternative sample restrictions. Panel A considers our restricted sample that
drops estimates for which we are forecasting earnings effects based on a policy’s
impact on college attendance. Panel B restricts the sample to only policies for
which earnings outcomes are estimated for at least five years of follow-up after
the policy. For this panel we show group averages even for groups with a single
policy. Panel C restricts the sample to policies whose identification strategy is a
randomized control trial, lottery, or regression discontinuity. Panel D restricts to
policies whose primary analyses have been published in a peer-reviewed journal.
We report bootstrapped 95% confidence intervals with adjustments (discussed in
Online Appendix H) for each category average.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1306
THE QUARTERLY JOURNAL OF ECONOMICS
APPENDIX FIGURE VIII
Publication Bias
This figure presents the MVPF estimates from Figure III and category averages
in Figure IV, Panel B using estimates corrected for publication bias from the
method of Andrews and Kasy (2019). Panel A reports estimates using the
corrections using the publication likelihood estimated from our model that
imposes jumps at p = .05 and p = .10, as shown in Table III, columns (3) and
(6). In Panel B we report corrected estimates under an assumption that child
policies are 35 times more likely to be published if they find a positive effect on
children’s outcomes (and we assume no publication bias for adult policies or for
child policies that find negative effects on children). This 35 times corresponds to
the estimated publication bias implied by a large-scale replication of experimental
economics papers by Camerer et al. (2016) (Table 1 of Andrews and Kasy 2019
reports that insignificant results are 0.029 times as likely to be published). We
do not report confidence intervals for these estimates (to our knowledge there is
no well-accepted method of constructing such intervals); but we refer readers to
Figure IV, Panel B to note that some of these category averages are imprecise.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1307
APPENDIX FIGURE IX
MVPFs by Decade
This figure presents MVPFs for all policies evaluated in the article based on
the year in which the policy was implemented. Policies are divided into categories
based on their decade of implementation and the average age of their economic
beneficiaries. For policies implemented in each decade there are two categories—
policies with beneficiaries over age 23 and policies with beneficiaries aged 23 or
younger. Within each decade by age category we construct the MVPF for a hypo-
thetical policy that allocates $1 of programmatic spending equally among all the
policies in the category. This is the same approach used to create MVPF estimates
for policy domains in previous figures. The capped lines show the 95% bootstrapped
confidence intervals with adjustments discussed in Online Appendix H.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1308
THE QUARTERLY JOURNAL OF ECONOMICS
APPENDIX FIGURE X
Welfare Reform
This figure presents estimates of the MVPF of 27 welfare reform policies
discussed in Section VI.C. We report MVPF estimates using three potential
measures of WTP: (i) cost, the mechanical cost of the program incurred by the
government, excluding any fiscal externalities from behavior change. Estimates
from this specification are denoted by circles. (ii) Change in transfer payments
(welfare, food stamps, and Medicaid). Estimates from this specification are
denoted by Xs. (iii) Change in post-tax income, which includes the change in
participants’ incomes due to changes in employment and the change in their
transfer payments. Estimates from this specification are denoted by triangles.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1309
HARVARD UNIVERSITY
HARVARD UNIVERSITY
SUPPLEMENTARY MATERIAL
An Online Appendix for this article can be found at The
Quarterly Journal of Economics online. Data and code replicating
tables and figures in this article can be found in Hendren and
Sprung-Keyser
(2020),
in
the
Harvard
Dataverse,
doi:
10.7910/DVN/ZHOSGC.
REFERENCES
Abraham, Katharine G., and Melissa A. Clark, “Financial Aid and Students’ Col-
lege Decisions: Evidence from the District of Columbia Tuition Assistance
Grant Program,” Journal of Human Resources, 41 (2006), 578–610.
Ackerman, Deena, Janet Holtzblatt, and Karen Masken, “The Pattern of EITC
Claims over Time: A Panel Data Analysis,” in “Conference Paper from IRS
RC 2009: Internal Revenue Service Research Conference.” Washington, DC:
Department of the Treasury.
Acton, Riley, “The Impact of Public Tuition Subsidies on College Enrollment Deci-
sions: Evidence from Michigan,” Michigan State University, 2018.
Akerlof, George A., “The Economics of ‘Tagging’ as Applied to the Optimal In-
come Tax, Welfare Programs, and Manpower Planning,” American Economic
Review, 68 (1978), 8–19.
Almond, Douglas, Hilary W. Hoynes, and Diana W. Schanzenbach, “Inside the
War on Poverty: The Impact of Food Stamps on Birth Outcomes,” Review of
Economics and Statistics, 93 (2011), 387–403.
American Institutes for Research, “Delta Cost Project,” (2017), https://www.
deltacostproject.org/delta-cost-project-database (accessed April 26, 2019).
Andrews, Isaiah, and Maximilian Kasy, “Identification of and Correction for Pub-
lication Bias,” American Economic Review, 109 (2019), 2766–2794.
Atkinson, Anthony B., Thomas Piketty, and Emmanuel Saez, “Top Incomes in the
Long Run of History,” Journal of Economic Literature, 49 (2011), 3–71.
Atkinson, Anthony B., and Nicholas H. Stern, “Pigou, Taxation and Public Goods,”
Review of Economic Studies, 41 (1974), 119–128.
Atkinson, Anthony Barnes, and Joseph E. Stiglitz, “The Design of Tax Structure:
Direct versus Indirect Taxation,” Journal of public Economics, 6 (1976), 55–75.
Auerbach, Alan, “The Theory of Excess Burden and Optimal Taxation,” in Hand-
book of Public Economics, vol. 1, A. Auerbach and M. Feldstein, eds. (Amster-
dam: Elsevier, 1985), 61–127.
Auerbach, Alan J., and James R. Hines, “Taxation and Economic Efficiency,” in
Handbook of Public Economics, vol. 3, A. Auerbach and M. Feldstein, eds.
(Amsterdam: Elsevier, 2002), 1347–1421.
Auten, Gerald, and Robert Carroll, “The Effect of Income Taxes on Household
Income,” Review of Economics and Statistics, 81 (1999), 681–693.
Autor, David H., Mark Duggan, Kyle Greenberg, and David S. Lyle, “The Impact
of Disability Benefits on Labor Supply: Evidence from the VA’s Disability
Compensation Program,” American Economic Journal: Applied Economics,
8 (2016), 31–68.
Bailey, Martha, Hilary Hoynes, Maya Rossin-Slater, and Reed Walker, “Is the
Social Safety Net a Long-Term Investment? Large-Scale Evidence from the
Food Stamps Program,” 2019. Goldman School of Public Policy Working Paper,
2019.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1310
THE QUARTERLY JOURNAL OF ECONOMICS
Barnett, W. Steven, and Leonard N. Masse, “Comparative Benefit Cost Analysis
of the Abecedarian Program and Its Policy Implications,” Economics of Edu-
cation Review, 26 (2007), 113–125.
Bartik, Timothy J., Brad Hershbein, and Marta Lachowska, “The Merits of Univer-
sal Scholarships: Benefit-Cost Evidence from the Kalamazoo Promise,” Jour-
nal of Benefit-Cost Analysis, 7 (2016), 400–433.
Bartik, Timothy J., Brad J. Hershbein, and Marta Lachowska, “The Effects of
the Kalamazoo Promise Scholarship on College Enrollment, Persistence, and
Completion,” Journal of Human Resources, forthcoming.
Bastian, Jacob, and Maggie R. Jones, “Do EITC Expansions Pay for Themselves?
Effects on Tax Revenue and Public Assistance Spending,” Rutgers University
working paper, 2019.
Bastian, Jacob, and Katherine Michelmore, “The Long-Term Impact of the Earned
Income Tax Credit on Children’s Education and Employment Outcomes,” Jour-
nal of Labor Economics, 36 (2018), 1127–1163.
Bettinger, Eric, “How Financial Aid Affects Persistence,” in College Choices: The
Economics of Where to Go, When to go, and How to Pay For It, Caroline Hoxby,
ed. (Chicago: University of Chicago Press, 2004, 207–238.
Bettinger, Eric, Oded Gurantz, Laura Kawano, Bruce Sacerdote, and Michael
Stevens, “The Long-Run Impacts of Financial Aid: Evidence from California’s
Cal Grant,” American Economic Journal: Economic Policy, 11 (2019), 64–94.
Bettinger, Eric P., Bridget Terry Long, Philip Oreopoulos, and Lisa Sanbonmatsu,
“The Role of Application Assistance and Information in College Decisions: Re-
sults from the H&R Block Fafsa Experiment,” Quarterly Journal of Economics,
127 (2012), 1205–1242.
Bhargava, Saurabh, and Dayanand Manoli, “Psychological Frictions and the In-
complete Take-Up of Social Benefits: Evidence from an IRS Field Experiment,”
American Economic Review, 105 (2015), 3489–3529.
Black, Sandra E., Paul J. Devereux, and Kjell G. Salvanes, “From the Cradle to
the Labor Market? The Effect of Birth Weight on Adult Outcomes,” Quarterly
Journal of Economics, 122 (2007), 409–439.
Blank, Rebecca M., and Patricia Ruggles, “When Do Women Use Aid to Fami-
lies with Dependent Children and Food Stamps? The Dynamics of Eligibility
versus Participation,” Journal of Human Resources, 31 (1996), 57–89.
Bloom, Dan, James J. Kemple, Pamela Morris, Susan Scrivener, Nandita Verma,
Richard Hendra, Diana Adams-Ciardullo, and David Seith, et al., “Final Re-
port on Florida’s Initial Time-Limited Welfare Program,” Manpower Demon-
stration Research Corporation, 2000.
Bloom, Dan, Susan Scrivener, Charles Michalopoulos, Pamela Morris, Richard
Hendra, Diana Adams-Ciardullo, and Johanna Walter, “Jobs First: Final Re-
port on Connecticut’s Welfare Reform Initiative,” ERIC, 2002.
Bloom, Howard S., Larry L. Orr, Stephen H. Bell, George Cave, Fred Doolittle,
Winston Lin, and Johannes M. Bos, “The Benefits and Costs of JTPA Title
II-A Programs: Key Findings from the National Job Training Partnership Act
Study,” Journal of Human Resources, 32 (1997), 549–576.
Bloom, Howard S., James A. Riccio, and Nandita Verma, “Promoting Work in
Public Housing: The Effectiveness of Jobs-Plus,” Manpower Demonstration
Research Corporation, 2005.
Boardman, Anthony E, David H. Greenberg, Aidan R. Vining, and David L.
Weimer, Cost-Benefit Analysis: Concepts and Practice (Cambridge: Cambridge
University Press, 2017).
Boudreaux, Michel H., Ezra Golberstein, and Donna D. McAlpine, “The Long-
Term Impacts of Medicaid Exposure in Early Childhood: Evidence from the
Program’s Origin,” Journal of Health Economics, 45 (2016), 161–175.
Brock, Thomas, David Butler, and David Long, “Unpaid Work Experience for Wel-
fare Recipients: Findings and Lessons from MDRC Research. MDRC Working
Papers,” Manpower Demonstration Research Corporation, 1993.
Brown, David, Amanda E. Kowalski, and Ithai Z. Lurie, “Long-Term Impacts of
Childhood Medicaid Expansions on Outcomes in Adulthood,” NBER Working
Paper no. 20835, 2017.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1311
Brown, David W., Amanda E. Kowalski, and Ithai Z. Lurie, “Medicaid as an Invest-
ment in Children: What Is the Long-Term Impact on Tax Receipts?,” NBER
Working Paper no. 20835, 2015.
Bruce, Donald J., and Celeste K. Carruthers, “Jackpot? The Impact of Lottery
Scholarships on Enrollment in Tennessee,” Journal of Urban Economics, 81
(2014), 30–44.
Bulman, George B., and Caroline M. Hoxby, “The Returns to the Federal Tax
Credits for Higher Education,” Tax Policy and the Economy, 29 (2015), 13–88.
Cabral, Marika, and Neale Mahoney, “Externalities and Taxation of Supplemental
Insurance: A Study of Medicare and Medigap,” American Economic Journal:
Applied Economics, 11 (2019), 37–73.
Camerer, Colin F., Anna Dreber, Eskil Forsell, Teck-Hua Ho, J¨
urgen Huber, Mag-
nus Johannesson, Michael Kirchler, and Johan Almenberg et al., “Evaluating
Replicability of Laboratory Experiments in Economics,” Science, 351 (2016),
1433–1436.
Campbell, Frances A., Elizabeth P. Pungello, Kirsten Kainz, Margaret Burchinal,
Yi Pan, Barbara H. Wasik, Oscar A. Barbarin, Joseph J. Sparling, and Craig
T. Ramey, “Adult Outcomes as a Function of an Early Childhood Educational
Program: An Abecedarian Project Follow-Up,” Developmental Psychology, 48
(2012), 1033–1043.
Card, David, Andrew Johnston, Pauline Leung, Alexandre Mas, and Zhuan Pei,
“The Effect of Unemployment Benefits on the Duration of Unemployment
Insurance Receipt: New Evidence from a Regression Kink Design in Missouri,
2003–2013,” American Economic Review: Papers and Proceedings, 105 (2015),
126–130.
Carroll, Robert, “Do Taxpayers Really Respond to Changes in Tax Rates? Evi-
dence from the 1993 Tax Act,” U.S. Department of the Treasury Office of Tax
Analysis, Working Paper 79, 1998.
Carruthers, Celeste K., and Jilleah G. Welch, “Not Whether, but Where? Pell
Grants and College Choices,” Journal of Public Economics, 172 (2019), 1–19.
Castleman, Benjamin L., and Bridget Terry Long, “Looking beyond Enrollment:
The Causal Effect of Need-Based Grants on College Access, Persistence, and
Graduation,” Journal of Labor Economics, 34 (2016), 1023–1073.
Cave, George, Hans Bos, Fred Doolittle, and Cyril Toussaint, “Jobstart: Final Re-
port on a Program for School Dropouts,” Manpower Demonstration Research
Corporation, 1993.
Chetty, Raj, “Moral Hazard versus Liquidity and Optimal Unemployment Insur-
ance,” Journal of Political Economy, 116 (2008), 173–234.
Chetty, Raj, John N. Friedman, and Jonah Rockoff, “New Evidence on the Long-
Term Impacts of Tax Credits,” IRS Statistics of Income White Paper, 2011.
Chetty, Raj, Nathaniel Hendren, and Lawrence F. Katz, “The Effects of Expo-
sure to Better Neighborhoods on Children: New Evidence from the Mov-
ing to Opportunity Experiment,” American Economic Review, 106 (2016),
855–902.
Cohodes, Sarah R., and Joshua S. Goodman, “Merit Aid, College Quality, and Col-
lege Completion: Massachusetts’ Adams Scholarship as an In-Kind Subsidy,”
American Economic Journal: Applied Economics, 6 (2014), 251–285.
Cornwell, Christopher, Kyung Hee Lee, and David Mustard, “The Effects of Merit-
Based Financial Aid on Course Enrollment, Withdrawal and Completion in
College,” IZA Working Paper, 2003.
Cornwell, Christopher, David B. Mustard, and Deepa J. Sridhar, “The Enrollment
Effects of Merit-Based Financial Aid: Evidence from Georgia’s HOPE Schol-
arship,” Journal of Labor Economics, 24 (2006), 761–786.
Couch, Kenneth A., “New Evidence on the Long-Term Effects of Employment
Training Programs,” Journal of Labor Economics, 10 (1992), 380–388.
Crouse, Gilbert, and Annette Waters, “Welfare Indicators and Risk Factors: Thir-
teenth Report to Congress,” US Department of Health and Human Services,
2014.
Currie, Janet, “Welfare and the Well-Being of Children: The Relative Effectiveness
of Cash and In-Kind Transfers,” Tax Policy and the Economy, 8 (1994), 1–43.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1312
THE QUARTERLY JOURNAL OF ECONOMICS
Currie, Janet, and Nancy Cole, “Welfare and Child Health: The Link between
AFDC Participation and Birth Weight,” American Economic Review, 83 (1993),
971–985.
Currie, Janet, and Jonathan Gruber, “Saving Babies: The Efficacy and Cost of
Recent Changes in the Medicaid Eligibility of Pregnant Women,” Journal of
Political Economy, 104 (1996), 1263–1296.
Currie, Janet, and Enrico Moretti, “Did the Introduction of Food Stamps Affect
Birth Outcomes in California?,” National Poverty Center Working Paper 06-
20, 2006.
Cutler, David M., and Jonathan Gruber, “Does Public Insurance Crowd out Private
Insurance?,” Quarterly Journal of Economics, 111 (1996), 391–430.
Dahl, Gordon B., and Lance Lochner, “The Impact of Family Income on Child
Achievement: Evidence from the Earned Income Tax Credit,” American Eco-
nomic Review, 102 (2012), 1927–1956.
Dave, Dhaval M., Sandra L. Decker, Robert Kaestner, and Kosali Ilayperuma Si-
mon, “The Effect of Medicaid Expansions in the Late 1980s and Early 1990s
on the Labor Supply of Pregnant Women,” American Journal of Health Eco-
nomics, 1 (2015), 195–193.
DeLong, J. Bradford, Lawrence H. Summers, Martin Feldstein, and Valerie A.
Ramey, “Fiscal Policy in a Depressed Economy,” Brookings Papers on Economic
Activity (2012), 233–297.
Deming, David J., and Christopher R. Walters, “The Impact of Price Caps and
Spending Cuts on U.S. Postsecondary Attainment,” NBER Working Paper no.
23736, 2017.
Denning, Jeffrey T., “College on the Cheap: Consequences of Community College
Tuition Reductions,” American Economic Journal: Economic Policy, 9 (2017),
155–188.
Denning, Jeffrey T., Benjamin M. Marx, and Lesley J. Turner, “ProPelled: The
Effects of Grants on Graduation, Earnings, and Welfare,” American Economic
Journal: Applied Economics, 11 (2019), 193–224.
Deshpande, Manasi, “Does Welfare Inhibit Success? The Long-Term Effects of
Removing Low-Income Youth from the Disability Rolls,” American Economic
Review, 106 (2016), 3300–3330.
Diamond, Peter, and Emmanuel Saez, “The Case for a Progressive Tax: from Basic
Research to Policy Recommendations,” Journal of Economic Perspectives, 25
(2011), 165–190.
Dynarski, Susan, “Hope for Whom? Financial Aid for the Middle Class and
Its Impact on College Attendance,” National Tax Journal, 53 (2000),
629–662.
Dynarski, Susan, C. J. Libassi, Katherine Michelmore, and Stephanie Owen, “Clos-
ing the Gap: The Effect of a Targeted, Tuition-Free Promise on College Choices
of High-Achieving, Low-Income Students,” NBER Working Paper no. 25349,
2018.
Dynarski, Susan M., “Does Aid Matter? Measuring the Effect of Student Aid on
College Attendance and Completion,” American Economic Review, 93 (2003),
279–288.
East, Chloe N., “Immigrants’ Labor Supply Response to Food Stamp Access,”
Labour Economics, 51 (2018), 202–226.
Eissa, Nada, and Hilary W. Hoynes, “Taxes and the Labor Market Participation
of Married Couples: The Earned Income Tax Credit,” Journal of Public Eco-
nomics, 88 (2004), 1931–1958.
Eissa, Nada, and Jeffrey Liebman, “Labor Supply Responses to the Earned Income
Tax Credit,” Quarterly Journal of Economics, 111 (1996), 605–637.
Fein, David, and Jill Hamadyk, “Bridging the Opportunity Divide for Low-Income
Youth: Implementation and Early Impacts of the Year Up Program,” OPRE
Report 2018-65, 2018.
Finkelstein, Amy, Nathaniel Hendren, and Erzo F. P. Luttmer, “The Value of Med-
icaid: Interpreting Results from the Oregon Health Insurance Experiment,”
Journal of Political Economy, 127 (2019).
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1313
Finkelstein, Amy, Nathaniel Hendren, and Mark Shepard, “Subsidizing Health
Insurance for Low-Income Adults: Evidence from Massachusetts,” American
Economic Review, 109 (2019), 1530–1567.
Finkelstein, Amy, and Robin McKnight, “What Did Medicare Do? The Initial Im-
pact of Medicare on Mortality and Out of Pocket Medical Spending,” Journal
of Public Economics, 92 (2008), 1644–1668.
Finkelstein, Amy, and Matthew J. Notowidigdo, “Take-Up and Targeting: Exper-
imental Evidence from SNAP,” Quarterly Journal of Economics, 134 (2019),
1505–1556.
Finkelstein, Amy, Sarah Taubman, Bill Wright, Mira Bernstein, Jonathan Gruber,
Joseph P. Newhouse, Heidi Allen, and Katherine Baicker, and Oregon Health
Study Group, “The Oregon Health Insurance Experiment: Evidence from the
First Year,” Quarterly Journal of Economics, 127 (2012), 1057–1106.
Freedman, Stephen, Daniel Friedlander, Winston Lin, and Amanda Schweder,
“The GAIN Evaluation: Five-Year Impacts on Employment, Earnings
and
AFDC
Receipt,”
Manpower
Demonstration
Research
Corporation,
1996.
Freedman, Stephen, Jean Tansey Knab, Lisa A. Gennetian, and David Navarro,
“The Los Angeles Jobs-First GAIN Evaluation: Final Report on a Work First
Program in a Major Urban Center,” Manpower Demonstration Research Cor-
poration, 2000.
French, Eric, and Jae Song, “The Effect of Disability Insurance Receipt on Labor
Supply,” American Economic Journal: Economic Policy, 6 (2014), 291–337.
Ganong, Peter, and Pascal J. Noel, “Consumer Spending during Unemployment:
Positive and Normative Implications,” American Economic Review, 109 (2019),
2383–2424.
Garc´
ıa, Jorge Luis, James J. Heckman, Andres Hojman, Yu Kyung Koh, Joshua
Shea, and Anna Ziff, “Documentation of full ABC/CARE Treatment Effects,”
Unpublished Manuscript, University of Chicago, 2016.
Garc´
ıa, Jorge Luis, James J. Heckman, Duncan Ermini Leaf, and Mar´
ıa Jos´
e
Prados, “Quantifying the Life-Cycle Benefits of a Prototypical Early Childhood
Program,” NBER Working Paper no. 23479, 2017.
Gelber, Alexander, Timothy J. Moore, and Alexander Strand, “The Effect of Dis-
ability Insurance Payments on Beneficiaries’ Earnings,” American Economic
Journal: Economic Policy, 9 (2017), 229–261.
Goering, John, Joan Kraft, Judith Feins, Debra McInnis, Mary Joel Holin, and
Huda Elhassan, “Moving to Opportunity for Fair Housing Demonstration Pro-
gram: Current Status and Initial Findings,” U.S. Department of Housing and
Urban Development, 1999.
Gold, Rachel, and Asta Kenney, “Paying for Maternity Care,” Family Planning
Perspectives, 17 (1985), 103–111.
Goldrick-Rab, Sara, Robert Kelchen, Douglas N. Harris, and James Benson, “Re-
ducing Income Inequality in Educational Attainment: Experimental Evidence
on the Impact of Financial Aid on College Completion,” American Journal of
Sociology, 121 (2016), 1762–1817.
Goodman, Joshua, “Who Merits Financial Aid?: Massachusetts’ Adams Scholar-
ship,” Journal of Public Economics, 92 (2008), 2121–2131.
Goodman-Bacon,
Andrew,
“The
Long-Run
Effects
of
Childhood
Insurance
Coverage: Medicaid Implementation, Adult Health, and Labor Market
Outcomes,” NBER Working Paper no 22899, 2017.
Greenberg, David H., Victoria Deitch, and Gayle Hamilton, “A Synthesis of Ran-
dom Assignment Benefit-Cost Studies of Welfare-to-Work Programs,” Journal
of Benefit-Cost Analysis, 1 (2010), 1–30.
Grogger, Jeff, and Lynn A. Karoly, Welfare Reform (Cambridge, MA: Harvard
University Press, 2005).
Grogger, Jeffrey, “The Effects of Time Limits, the EITC, and Other Policy Changes
on Welfare Use, Work, and Income among Female-Headed Families,” Review
of Economics and Statistics, 85 (2003), 394–408.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1314
THE QUARTERLY JOURNAL OF ECONOMICS
Gruber, Jonathan, “The Consumption Smoothing Benefits of Unemployment In-
surance,” American Economic Review, 87 (1997), 192–205.
Hamilton, Gayle, Stephen Freedman, Lisa Gennetian, Charles Michalopoulos, Jo-
hanna Walter, Diana Adams-Ciardullo, Anna Gassman-Pines, and Sharon
McGroder et al., “How Effective Are Different Welfare-to-Work Approaches?
Five-Year Adult and Child Impacts for Eleven Programs. National Evaluation
of Welfare-to-Work Strategies,” Manpower Demonstration Research Corpora-
tion, 2001.
Heckman, James J., “Skill Formation and the Economics of Investing in Disad-
vantaged Children,” Science, 312 (2006), 1900–1902.
Heckman, James J., Seong Hyeok Moon, Rodrigo Pinto, Peter A. Savelyev, and
Adam Yavitz, “The Rate of Return to the High Scope Perry Preschool Program,”
Journal of Public Economics, 94 (2010), 114–128.
Heckman, James J., Rodrigo Pinto, Azeem M. Shaikh, and Adam Yavitz, “Inference
with Imperfect Randomization: The Case of the Perry Preschool Program,”
NBER Working Paper no. 16935, 2011.
Heim, Bradley T., “The Effect of Recent Tax Changes on Taxable Income: Evidence
from a New Panel of Tax Returns,” Journal of Policy Analysis and Manage-
ment, 28 (2009), 147–163.
Helburn, Suzanne W., “Cost, Quality and Child Outcomes in Child Care Cen-
ters. Technical Report, Public Report, and Executive Summary,” Manpower
Demonstration Research Corporation, 1995.
Hendra, Richard, David H. Greenberg, Gayle Hamilton, Ari Oppenheim, Alexan-
dra Pennington, Kelsey Schaberg, and Betsy L. Tessler, “Encouraging
Evidence on a Sector-Focused Advancement Strategy: Two-Year Impacts from
the Work Advance Demonstration,” Manpower Demonstration Research Cor-
poration, 2016.
Hendren, Nathaniel, “The Policy Elasticity,” Tax Policy and the Economy, 30 (2016),
51–89.
———, “Efficient Welfare Weights,” NBER Working Paper no. 20351, 2017a.
———, “Knowledge of Future Job Loss and Implications for Unemployment In-
surance,” American Economic Review, 107 (2017b), 1778–1823.
———, “Measuring Ex-Ante Welfare in Insurance Markets,” NBER Working Paper
no. 24470, 2017c.
Hendren, Nathaniel, and Ben Sprung-Keyser, “Replication Data for: ‘A Unified
Welfare Analysis of Government Policies’,” (2020), Harvard Dataverse, doi:
10.7910/DVN/ZHOSGC.
Hoekstra, Mark, “The Effect of Attending the Flagship State University on Earn-
ings: A Discontinuity-Based Approach,” The Review of Economics and Statis-
tics, 91 (2009), 717–724.
Hollister, Robinson G., Peter Kemper, and Rebecca A. Maynard, “The National
Supported Work Demonstration,” 1984.
Hotz, V. Joseph, and John K. Scholz, “The Earned Income Tax Credit,” in Means-
Tested Transfer Programs in the United States, Robert A. Moffitt, ed. (Chicago:
University of Chicago Press, 2003), 141–198.
Hoxby, Caroline M., and George B. Bulman, “The Effects of the Tax Deduction
for Postsecondary Tuition: Implications for Structuring Tax-Based Aid,” Eco-
nomics of Education Review, 51 (2016), 23–60.
Hoynes, Hilary W., Marianne Page, and Ann Huff Stevens, “Can Targeted Trans-
fers Improve Birth Outcomes?: Evidence from the Introduction of the WIC
Program,” Journal of Public Economics, 95 (2011), 813–827.
Hoynes, Hilary W., and Ankur J. Patel, “Effective Policy for Reducing Poverty and
Inequality? The Earned Income Tax Credit and the Distribution of Income,”
Journal of Human Resources, 53 (2018), 859–890.
Hoynes, Hilary W., and Diana W. Schanzenbach, “Consumption Responses to In-
Kind Transfers: Evidence from the Introduction of the Food Stamp Program,”
American Economic Journal: Economic Policy, 1 (2009), 109–139.
———, “Work Incentives and the Food Stamp Program,” Journal of Public Eco-
nomics, 96 (2012), 151–162.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1315
———, “Safety Net Investments in Children,” Brookings Papers on Economic
Activity (2018), 89–133.
Hoynes, Hilary W., Diane Whitmore Schanzenbach, and Douglas Almond, “Long-
Run Impacts of Childhood Access to the Safety Net,” American Economic
Review, 106 (2016), 903–934.
Hylland, Aanund, and Richard Zeckhauser, “Distributional Objectives Should Af-
fect Taxes but not Program Choice or Design,” in Measurement in Public
Choice, Steinar Strøm, ed. (London: Palgrave Macmillan, 1981), 123–143.
Hyman, Joshua, “Does Money Matter in the Long Run? Effects of School Spending
on Educational Attainment,” American Economic Journal: Economic Policy, 9
(2017), 256–280.
Jackson, C. Kirabo, Claudia Persico, and Rucker C. Johnson, “The Effects of
School Spending on Educational and Economic Outcomes: Evidence from
School Finance Reforms,” Quarterly Journal of Economics, 131 (2016),
157–218.
Jacob, Brian A., and Jens Ludwig, “The Effects of Housing Assistance on Labor
Supply: Evidence from a Voucher Lottery,” American Economic Review, 102
(2012), 272–304.
Jacob, Brian A., Jens Ludwig, and Max Kapustin, “The Impact of Housing As-
sistance on Child Outcomes: Evidence from a Randomized Housing Lottery,”
Quarterly Journal of Economics, 130 (2014), 465–506.
Johnson, Rucker C., and C. Kirabo Jackson, “Reducing Inequality through Dy-
namic Complementarity: Evidence from Head Start and Public School Spend-
ing,” American Economic Journal: Economic Policy, 11 (2019), 310–349.
Johnston, Andrew C., and Alexandre Mas, “Potential Unemployment Insurance
Duration and Labor Supply: The Individual and Market-Level Response to a
Benefit Cut,” Journal of Political Economy, 126 (2018), 2480–2522.
Jones, Damon, and Ioana Marinescu, “The Labor Market Impacts of Universal
and Permanent Cash Transfers: Evidence from the Alaska Permanent Fund,”
NBER Working Paper no. 24312, 2018.
Kane, Thomas J., “College Entry by Blacks since 1970: The Role of College Costs,
Family Background, and the Returns to Education,” Journal of Political Econ-
omy, 102 (1994), 878–911.
Katz, Lawrence F., and Bruce D. Meyer, “The Impact of the Potential Duration
of Unemployment Benefits on the Duration of Unemployment,” Journal of
Public Economics, 41 (1990), 45–72.
Kawano, Laura, Caroline Weber, and Andrew Whitten, “Estimating the Elas-
ticity of Broad Income for High-Income Taxpayers,” 2016, https://papers.
ssrn.com/sol3/papers.cfm?abstract_id=2852048 (accessed July 7, 2019).
Kemple, James J., Daniel Friedlander, and Veronica Fellerath, “Florida’s Project
Independence. Benefits, Costs, and Two-Year Impacts of Florida’s JOBS Pro-
gram,” ERIC, 1995.
Kleven, Henrik, “The EITC and the Extensive Margin: A Reappraisal,” NBER
Working Paper no. 26405, 2019.
Kleven, Henrik J., and Claus Thustrup Kreiner, “The Marginal Cost of Public
Funds: Hours of Work versus Labor Force Participation,” Journal of Public
Economics, 90 (2006), 1955–1973.
Kline, Patrick, and Christopher R. Walters, “Evaluating Public Programs with
Close Substitutes: The Case of Head Start,” Quarterly Journal of Economics,
131 (2016), 1795–1848.
Kroft, Kory, and Matthew J. Notowidigdo, “Should Unemployment Insurance Vary
with the Unemployment Rate? Theory and Evidence,” Review of Economic
Studies, 83 (2016), 1092–1124.
LaLumia, Sara, “Tax Preferences for Higher Education and Adult College Enroll-
ment,” National Tax Journal, 65 (2012), 59–92.
Landais, Camille, “Assessing the Welfare Effects of Unemployment Benefits Using
the Regression Kink Design,” American Economic Journal: Economic Policy,
7 (2015), 243–278.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1316
THE QUARTERLY JOURNAL OF ECONOMICS
Lo Sasso, Anthony T., and Dorian G. Seamster, “How Federal and State Policies
Affected Hospital Uncompensated Care Provision in the 1990s,” Medical Care
Research and Review, 64 (2007), 731–744.
Long, Bridget, “The Impact of Federal Tax Credits for Higher Education Expenses,”
in College Choices: The Economics of Where to Go, When to Go, and How to Pay
for It, Caroline Hoxby, ed. (Chicago: The University of Chicago Press, 2004),
101–168.
Ludwig, Jens, and Douglas L. Miller, “Does Head Start Improve Children’s Life
Chances? Evidence from a Regression Discontinuity Design,” Quarterly Jour-
nal of Economics, 122 (2007), 159–208.
Maestas, Nicole, Kathleen J. Mullen, and Alexander Strand. “Does Disability In-
surance Receipt Discourage Work? Using Examiner Assignment to Estimate
Causal Effects of SSDI Receipt,” American Economic Review, 103 (2013),
1797–1829.
Manoli, Day, and Nicholas Turner, “Cash-on-hand and College Enrollment: Evi-
dence from Population Tax Data and the Earned Income Tax Credit,” American
Economic Journal: Economic Policy, 10 (2018), 242–271.
Marx, Benjamin M., and Lesley J. Turner, “Borrowing Trouble? Human Capital
Investment with Opt-In Costs and Implications for the Effectiveness of Grant
Aid,” American Economic Journal: Applied Economics, 10 (2018), 163–201.
Masse, Leonard N., “A Benefit Cost Analysis of the Carolina Abecedar-
ian
Preschool
Program,”
2003,
https://www.minneapolisfed.org/∼/media/
files/publications/studies/earlychild/2003conf/barnettdoc.doc?la=en.
Masse, Leonard N., and W. Steven Barnett, “A Benefit Cost Analysis of the
Abecedarian Early Childhood Intervention,” National Institute for Early Ed-
ucation Research working paper, 2002.
Maxfield, Michelle, “The Effects of the Earned Income Tax Credit on Child Achieve-
ment and Long-Term Educational Attainment,” Michigan State University
Working Paper, 2013.
Mayshar, Joram, “On Measures of Excess Burden and Their Applications,” Journal
of Public Economics, 43 (1990), 263–289.
Meyer, Bruce D., and Wallace K. C. Mok, “Quasi-Experimental Evidence on the
Effects of Unemployment Insurance from New York State,” NBER Working
Paper no. 12865, 2007.
Meyer, Bruce D., and Dan T. Rosenbaum, “Welfare, the Earned Income Tax Credit,
and the Labor Supply of Single Mothers,” Quarterly Journal of Economics, 116
(2001), 1063–1114.
Michelmore, Katherine, “The Effect of Income on Educational Attainment:
Evidence
from
State
Earned
Income
Tax
Credit
Expansions,”
2013,
https://papers.ssrn.com/sol3/papers.cfm?abstract id=2356444.
Miller, Cynthia, Lawrence F. Katz, Gilda Azurdia, Adam Isen, and Caroline B.
Schultz, “Expanding the Earned Income Tax Credit for Workers without De-
pendent Children: Interim Findings from the Paycheck Plus Demonstration
in New York City,” MDRC, 2017.
Miller, Cynthia, Virginia Knox, Lisa A. Gennetian, Martey Dodoo, Jo Anna Hunter,
and Cindy Redcross, “Reforming Welfare and Rewarding Work: Final Report
on the Minnesota Family Investment Program. Vol. 1: Effects on Adults and
Volume 2: Effects on Children,” Manpower Demonstration Research Corpora-
tion, 2000.
Miller, Sarah, and Laura R. Wherry, “The Long-Term Effects of Early Life Medi-
caid Coverage,” Journal of Human Resources, 54 (2019), 785–824.
Mills, Gregory, Daniel Gubits, Larry Orr, David Long, Judie Feins, Bulbul Kaul,
Michelle Wood, and Amy Jones et al., “Effects of Housing Vouchers on Welfare
Families,” U.S. Department of Housing and Urban Development, Office of
Policy Development and Research, 2006.
Mirrlees, James A., “An Exploration into the Theory of Optimal Income Taxation,”
Review of Economic Studies, 38 (1971), 175–208.
———, “Optimal Tax Theory: A Synthesis,” Journal of Public Economics, 6 (1976),
327–358.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


UNIFIED WELFARE ANALYSIS OF GOVERNMENT POLICIES 1317
Moffitt, Robert A., “Welfare Programs and Labor Supply,” Handbook of Public
Economics, 4 (2002), 2393–2430.
———, Means-Tested Transfer Programs in the United States (Chicago: University
of Chicago Press, 2003).
Okun, Arthur M., Equality and Efficiency (Washington, DC: Brookings Institution
Press, 1975).
Pavetti, LaDonna, “Who is Affected by Time Limits?” in Welfare Reform: An Anal-
ysis of the Issues, Isabel V. Sawhill, ed. (Washington, DC: The Urban Institute,
1995), 31–34.
Price, David J., and Jae Song, “The Long-Term Effects of Cash Assistance,” Work-
ing Paper, 2018.
Rea, David, and Tony Burton, “New Evidence on the Heckman Curve,” Journal of
Economic Surveys, 34 (2020), 241–262.
Reynolds, Arthur J., Judy A. Temple, Dylan L. Robertson, and Emily A. Mann,
“Age 21 Cost-Benefit Analysis of the Title I Chicago Child-Parent Centers,”
Educational Evaluation and Policy Analysis, 24 (2002), 267–303.
Reynolds, Arthur J., Judy A. Temple, Suh-Ruu Ou, Irma A. Arteaga, and Barry
A.B. White, “School-Based Early Childhood Education and Age-28 Well-Being:
Effects by Timing, Dosage, and Subgroups,” Science, 333 (2011), 360–364.
Riccio, James A., “Jobs–Plus: A Promising Strategy for Increasing Employment
and Self–Sufficiency among Public Housing Residents,” Technical Report, pre-
sented before the Subcommittee on Federalism and the Census, House Com-
mittee on Government Reform, 2006.
Saez, Emmanuel, “Using Elasticities to Derive Optimal Income Tax Rates,” Review
of Economic Studies, 68 (2001), 205–229.
———, “The Effect of Marginal Tax Rates on Income: A Panel Study of ‘Bracket
Creep’,” Journal of Public Economics, 87 (2003), 1231–1258.
Saez, Emmanuel, Joel Slemrod, and Seth H. Giertz, “The Elasticity of Taxable
Income with Respect to Marginal Tax Rates: A Critical Review,” Journal of
Economic Literature, 50 (2012), 3–50.
Sanbonmatsu, Lisa, Lawrence F. Katz, Jens Ludwig, Lisa A. Gennetian, Greg J.
Duncan, Ronald C. Kessler, Emma K. Adam, and Thomas McDade et al., “Mov-
ing to Opportunity for Fair Housing Demonstration Program: Final Impacts
Evaluation,” U.S. Department of Housing and Urban Development, 2011.
Schaberg, Kelsey, “Can Sector Strategies Promote Longer-Term Effects? Three-
Year Impacts from the WorkAdvance Demonstration,” Manpower Demonstra-
tion Research Corporation, 2017.
Schmieder, Johannes F., and Till Von Wachter, “The Effects of Unemployment In-
surance Benefits: New Evidence and Interpretation,” Annual Review of Eco-
nomics, 106 (2016), 547–581.
Schochet, Peter Z., “National Job Corps Study: 20-Year Follow-Up Study Using
Tax Data,” Mathematica Policy Research Report, 2018.
Schochet, Peter Z., John Burghardt, and Sheena McConnell, “Does Job Corps
Work? Impact Findings from the National Job Corps Study,” American Eco-
nomic Review, 98 (2008), 1864–1886.
Schochet, Peter Z., John A. Burghardt, and Sheena M. McConnell et al., “National
Job Corps Study and Longer-Term Follow-Up Study: Impact and Benefit-Cost
Findings Using Survey and Summary Earnings Records Data,” U.S. Depart-
ment of Labor, Employment and Training Administration, 2006.
Scholz, John Karl, “The Earned Income Tax Credit: Participation, Compliance,
and Antipoverty Effectiveness,” Institute for Research on Poverty Discussion
Papers 1020-93, 1993.
Scrivener, Susan, Richard Hendra, Cindy Redcross, Dan Bloom, Charles
Michalopoulos, and Johanna Walter, “WRP: Final Report on Vermont’s Wel-
fare Restructuring Project, Manpower Demonstration Research Corporation,
2002.
Seftor, Neil S., and Sarah E. Turner, “Back to School: Federal Student Aid Policy
and Adult College Enrollment,” Journal of Human Resources, (2002), 336–352.
Slemrod, Joel, and Shlomo Yitzhaki, “The Social Cost of Taxation and the Marginal
Cost of Funds,” International Monetary Fund Staff Papers, 43 (1996), 172–198.
Downloaded from https://academic.oup.com/qje/article/135/3/1209/5781614 by guest on 29 September 2024


1318
THE QUARTERLY JOURNAL OF ECONOMICS
———, “Integrating Expenditure and Tax Decisions: The Marginal Cost of Funds
and the Marginal Benefit of Projects,” National Tax Journal, 54 (2001), 189–
202.
Solon, Gary, “Work Incentive Effects of Taxing Unemployment Benefits,” Econo-
metrica, 53 (1985), 295–306.
Stiglitz, Joseph E., and Parthaa Dasgupta, “Differential Taxation, Public Goods,
and Economic Efficiency,” Review of Economic Studies, 38 (1971), 151–174.
Tax Foundation, “U.S. Federal Individual Income Tax Rates History, 1862–
2013,”
2013,
https://taxfoundation.org/us-federal-individual-income-tax-
rates-history-1913-2013-nominal-and-inflation-adjusted-brackets/.
Tax Policy Center, “Earned Income Tax Credit Parameters, 1975–2016,” 2016,
https://www.taxpolicycenter.org/sites/default/files/legacy/taxfacts/content/pdf/
historical eitc parameters.pdf.
Turner, Nicholas, “The Effect of Tax-Based Federal Student Aid on College En-
rollment,” National Tax Journal, 64 (2011), 839–861.
U.S. Census Bureau, “Current Population Reports: Consumer Income,” 1996,
https://www2.census.gov/prod2/popscan/p60-049.pdf.
U.S. Department of Education Office of Postsecondary Education, “2009–
2010 Federal Pell Grant Program End-of-Year Report,” 2010, https://
www2.ed.gov/finaid/prof/resources/data/pell-2009-10/pell-eoy-09-10.pdf.
U.S. Department of Health & Human Services, “The Final Report of the
Seattle-Denver Income Maintenance Experiment,” 1983, https://aspe.hhs.gov/
report/overview-final-report-seattle-denver-income-maintenance-experiment.
U.S. Social Security Administration, “SSI Annual Statistical Report, 2013,” SSA
Publication No. 13-11827, 2014.
——–, “Annual Statistical Supplement to the Social Security Bulletin, 2017,” Pub-
lication No. 13-11700, 2018.
Von Wachter, Till, Jae Song, and Joyce Manchester, “Trends in Employment and
Earnings of Allowed and Rejected Applicants to the Social Security Disability
Insurance Program,” American Economic Review, 101 (2011), 3308–3329.
Weimer, David (ed.), Cost-Benefit Analysis and Public Policy, vol. 1 (New York:
John Wiley & Sons, 2009).
Weimer, David L., and Aidan R. Vining (eds.), Investing in the Disadvantaged
(Washington, DC: Georgetown University Press, 2009).
Werning, Ivan, “Optimal Fiscal Policy with Redistribution,” Quarterly Journal of
Economics, 122 (2007), 925–967.
Wherry, Laura R., and Bruce D. Meyer, “Saving Teens: Using a Policy Discon-
tinuity to Estimate the Effects of Medicaid Eligibility,” Journal of Human
Resources, 51 (2016), 556–588.
Wherry, Laura R., Sarah Miller, Robert Kaestner, and Bruce D. Meyer, “Child-
hood Medicaid Coverage and Later-Life Health Care Utilization,” Review of
Economics and Statistics, 100 (2018), 287–302.
Whitmore, Diane, “What Are Food Stamps Worth?,” Princeton University Indus-
trial Relations Section Working Paper no. 468, 2002.
Wood, Michelle, Jennifer Turnham, and Gregory Mills, “Housing Affordability and
Family Well-Being: Results from the Housing Voucher Evaluation,” Housing
Policy Debate, 19 (2008), 367–412.
WSIPP, “Benefit-Cost Technical Documentation,” Washington State Institute for
Public Policy Technical Report, 2019.
Zimmerman, Seth D., “The Returns to College Admission for Academically
Marginal Students,” Journal of Labor Economics, 32 (2014), 711–754.