Chapter34
WELFARE PROGRAMS AND LABOR SUPPLY*
ROBERT A. MOFFITT
Johns Hopkins University; and NBER
Contents
Abstract 2394
Keywords 2394
Introduction 2395
1. US policy and institutional background 2395
2. Theoretical models and issues 2402
2.1. Introduction 2402
2.2. Basic static model 2403
2.3. Dynamics 2408
2.4. Work requirements and tagging models 2411
2.5. Wage and earnings subsidies 2413
2.6. In-kind transfers 2417
3. Empirical evidence: a short review 2417
3.1. Introduction 2417
3.2. Pre-1995 estimates 2419
3.3. Post-1995 structural estimates 2420
3.4. Post-1995 reduced-form policy impact estimates 2422
4. Summary 2426
References 2427
' The author would like to thank the participants of a Handbook conference in Berkeley, December 1-2,
2000, for comments and Kara Levine for SIPP tabulations.
Handbookof PublicEconomics, Volume 4, Editedby A.J Auerbach and M. Feldstein
© 2002 Elseuier Science B. V All rights reserved
2394 R.A. Moffitt
Abstract
The labor supply and other work incentive effects ofwelfare programs have long been a
central concern in economic research. Work has also been an increasing focus of policy
reforms in the USA, culminating with a number of major policy changes in the 1990s
whose intent was to increase employment and earnings levels of welfare recipients
and other disadvantaged individuals. This chapter reviews the economic research on
this topic, covering both the theoretical models that have been developed as well as the
empirical findings from econometric studies of the effects of existing welfare programs
on labor supply.
Keywords
welfare programs, labor supply, work incentives
JEL classification: 13, J2
Ch. 34: Welfare Programsand Labor Supply
Introduction
The work incentives of programs which provide means-tested transfer benefits to the
low-income population has been a subject of increasing concern to voters and policymakers
in the USA. Interest in work incentives first arose in the 1960s, when caseloads
in the Aid to Families with Dependent Children (AFDC) program rose dramatically,
and Congress lowered the tax rate on earnings in the program to encourage work.
Increased interest in encouraging work among welfare recipients gradually grew in
the 1970s and 1980s, with a shift in focus toward work requirements rather than
lowered tax rates. In the 1990s, major new policy developments occurred whose
focus was on increasing work, including 1996 legislation introducing major new work
requirements into the AFDC program as well as the expansion of the Earned Income
Tax Credit (EITC), an earnings subsidy program. Increased interest in encouraging
work simultaneously has occurred in disability programs and the Food Stamp program.
Similar reforms have developed in Europe, particularly in the United Kingdom with
the recent introduction of the Working Families Tax Credit and with the welfare reform
there termed the "New Deal".
Research on work and labor supply issues related to welfare reform has also long
been the subject of attention by economists. In the 1960s, both James Tobin and Milton
Friedman noted that the 100% marginal tax rates which were, at that time, imposed by
some welfare programs discouraged work and that a negative income tax (NIT) with
tax rates less than 100% was one solution [Friedman (1962), Tobin (1965)]. The NIT
is now a staple of undergraduate textbooks. The economics profession conducted an
enormous amount of research on the NIT in the 1970s, with much of the focus on
the development of new econometric techniques for the analysis of work incentives,
as well as a focus on the analysis of several random-assignment experimental tests of
a negative income tax. After something of a research lull in the late 1970s and 1980s,
there has been a resurgence of research interest in the 1990s in these issues which is
still underway. Econometric methods have shifted toward more reduced form methods
which make identification more transparent but whose results are less convenient for
generalizability than the older structural methods. Much of the analysis has likewise
shifted from the estimation of underlying models of behavior to the impact estimation
of specific programmatic reforms.
1. US policy and institutional background
Table 1shows some of the major welfare programs in the USA and their characteristics
relevant to labor supply - who is eligible, the form of assistance, average transfer
amounts, and marginal tax rates. The most well-known cash program, the Temporary
Assistance for Needy Families (TANF) program, provides benefits primarily to single
mothers and their children; the Supplemental Security Income (SSI) program serves
low income aged, blind, and disabled individuals and families; the Earned Income
2395
Tax Credit (EITC) serves only those with positive earnings, albeit all individuals
and families in that category; child-care programs likewise serve only those who are
working and with children; and the Medicaid program serves the aged and disabled
as well as single mothers and their children, the latter overlapping heavily with the
TANF population. Only the Food Stamp program and housing programs are universal
in eligibility (aside from low income and assets), and even the latter is unique because
it is rationed in quantity and hence not universally eligible in that sense. That leaves
the Food Stamp program as the only truly universal means-tested transfer programs in
the USA.
This categorization of the population raises two issues. One is whether individuals
can alter their eligibility status and change their labor supply in the process by making
themselves eligible for a higher level of benefits. In some cases this might be desirable,
as in the case of the EITC or child care where individuals might increase their labor
supply to make themselves eligible for subsidies, while in other cases it might not
be, as in the case of TANF or Medicaid where individuals (primarily women) might
take actions to become, or remain, single mothers in order to retain eligibility for
benefits, in most cases reducing their labor supply in the process . A second issue
is whether, even if eligibility status is unalterable or at least alterable but exogenous
to labor supply, the set of eligibility categories, benefit levels, and marginal tax rates
assigned to the different groups in the population makes sense from a normative point
of view. Answering this question requires a framework to assess normative issues and
will not be discussed in this review, although it is a key design issue for the transfer
system as a whole.
The form of assistance varies as well, from cash to in-kind transfers for food,
medical care, housing, and child care. This is a particular feature of the US system not
present in some European countries, where there are no specific food programs and
where health care is more typically covered by a universal program. The labor supply
effects of in-kind transfers differ from those of cash transfers because they subsidize
a good which may be a complement or substitute for leisure, and because the in-kind
transfers typically provide minimum or fixed quantities of the good in question and
hence have the potential to constrain the consumption choices of the family. What
evidence there is on the cash-equivalent value of in-kind transfers suggest that Food
Stamps are very nearly equivalent to cash while Medicaid, housing, and child care are
not [Smeeding (1982)].
Expenditures vary tremendously across programs, with Medicaid being the dominant
program in this respect. However, the bulk of Medicaid expenditures go for the aged
and disabled; single mothers and their children, the group of most research focus in
the literature, constitute only a little over 10% of total Medicaid expenditures. Among
I In all cases, of course, because benefits are only eligible for those with incomes below specified
amounts, there is a labor supply disincentive. This is the standard labor supply effect to be discussed in
Section 2 below.
2396 R.A. Moffitt
Ch. 34: Welfare Programsand Labor Supply
Table I
Characteristics of major means-tested transfer programs in the USA
a
Program Main eligible Form of Annual Average monthly Marginal tax
populationb assistance expenditures" expenditure for rate on
(FY2000) family of 3 earnings
(FY2000)
Temporary
Assistance for
Needy Families
Supplemental
Security
Income
Earned Income
Tax Credit
Mostly singlemother
families
Aged, blind, and
disabled
individuals and
families
Individuals with
positive earnings
Food Stamps All individuals
and families
Cash 14490
Cash 35066
Cash 30000
d
Food coupons
Medicaid Families with Health care
dependent services
children,
disabled, elderly
Subsidized All individuals
Housing g and families
Child Care
Child Care Working parents
Block Grant of children
under 13
Housing units
Child care
assistance
20341
207 195
22498
6934
600 Ranges across
states from
0 to 100%
1326
135
e
50%
Ranges from
-40% to 21%
279 30%
2238'
422 h
861
0% or >100%
Ranges from
20% to 30%
Sliding fee
scale set by
states (can be
zero)
Working parents
of children
under 13
Nonrefundable
credit in
federal
income tax
2200 75 Credit is
20% to 30%
of eligible
expenses
a Sources: Blau (2003), Burke (2001), Rowe and Roberts (2002), U.S. House of Representatives (2000).
b In addition to low income and assets.
c Combined federal and state and local; in millions.
d Includes tax reduction as well as refundable portion.
e Per filing unit, tax year 2000.
f FY1999.
g Combined Section 8 and public housing.
h Family or dwelling unit.
iFY1998, for 2 children in child care.
Dependent Care
Tax Credit
2397
the other programs, most are in the same general range except for child care, which
is considerably smaller. The expenditures shown in Table I have changed dramatically
over the last three decades, as TANF expenditures (equal to Aid to Families with
Dependent Children, AFDC, prior to 1996) have declined from the largest in the
late 1960s to one of the smallest today; as Food Stamps and Medicaid have grown,
particularly the latter; and as the EITC and SSI expenditures have mushroomed [for
more discussion of the history, see Moffitt (2002)].
The last two columns in the table, showing average expenditures per family of 3
(a rough measure of benefits) and the marginal tax rate, give some indication of the
magnitudes of income and substitution effects on labor supply, respectively. Medicaid
has the largest average benefits but, again, are much smaller for a family of three
consisting of a single mother and two children, which are approximately $427 2
This makes SSI the largest average benefit program, perhaps not surprisingly since
it serves a population which is capable of much less labor supply than the eligible
populations for the other programs. Transfers for in-kind programs are generally
smaller than those for TANF, necessarily because they are intended to subsidize only
part of consumption. An exception is the child care block grant program, although
benefits in this program are poorly measured. The EITC, despite its relative high
expenditure level, has, on average, very low benefits because its recipient base is
so large. As for absolute amounts, the TANF, Food Stamps, and housing programs
(ignoring child care) are sizable relative to full-time minimum wage work of $893
per month.
The marginal tax rates (MTRs) in the programs vary greatly. Those in the TANF program
are now set by the states, and range from 0 to 100% (a zero percent MTR is
possible only ifthe state imposes a maximum income level, which by itself imposes a
cliff, or notch, in the benefit schedule). SSI imposes a 50% rate, Food Stamps imposes a
30% MTR, and public housing and Section 8 programs typically impose double MTRs,
one on income net of deductions and one on gross income, ranging from 10 to 30%.
The Medicaid program is the most extreme form of a cliff, or notch, program, which
is available in its entirety (i.e., full subsidized care) until eligibility is ended - such as
income exceeding a fixed amount3
. The EITC is an earnings subsidy program which
has negative MTRs as long as -40% in its initial subsidy range and positive MTRs
up to 21% in its phaseout range.
Virtually all ofthese rates are the same as they were when the program was created,
with occasional important exceptions. The MTR in the TANF-AFDC program was
100% from its inception in 1935 to 1969, when it was lowered to 67%; the MTR was
2 This estimate equals the average expenditure per adult plus two times the average expenditure per
child. It is larger than the amount for many single-mother Medicaid families off TANF, for whom only
the children are eligible.
3 Medicaid has many different subprograms and a few currently require copayments, but they are the
exception.
2398 R.A. Moffitt
Ch. 34: Welfare ProgramsandLabor Supply
raised back to 100% in 1981, where it remained until the 1996, when states were given
freedom to set it at their discretion 4
. The EITC rates were increased in 1993 as well.
Employment rates in transfer programs generally vary with MTRs in the direction
one should expect. In TANF, the average employment rate is currently approximately
30%, and is higher in those states with lower MTRs. Prior to 1996, when MTRs were
closer to 100%, employment rates among AFDC recipients were approximately 9%.
As will be discussed below, this relationship does not prove that there is an effect of
MTRs on labor supply because a lower MTR has a higher "breakeven" level of hoursworked
(i.e., the level at which eligibility ends) and thus more workers are eligible
in that case. But it does show that the MTRs are actually enforced and do affect the
numbers of recipients combining work and welfare. The employment rate in SSI is
only 6.7%, no doubt reflecting the high rates of disability in the program. Food Stamp
employment rates are 24%.
The actual MTRs in most programs differ in significant ways from the "nominal"
MTRs shown in Table 1, complicating the picture. Virtually all have exemption
amounts of earnings, below which benefits are not reduced and hence the MTR is
zero. The TANF program, Food Stamps, Medicaid, and child-care programs also all
have maximum income limits that create notches, as benefits and eligibility go to zero
for earnings that cause income to exceed those levels. In the TANF program, states
are allowed to reduce calculated benefits by a fixed percentage, thereby reducing the
MTR by that percent, as well as putting maximums on the benefit, which effectively
creates a region of zero MTR. Most programs also allow deductions for work-related
expenses and others for consumption items like housing or rental costs, which vary to
some degree with income and hence lower the effective MTR 5. A further wrinkle in
some programs, such as TANF and SSI, is that a different MTR is applied at initial
application for the program as is applied thereafter. In most cases a 100% MTR is
applied at eligibility - that is, earnings are counted directly against the benefit - and
then the lower nominal MTR, shown in Table 1,is applied after eligibility is established
and recipiency has begun. These types of rules are designed to discourage entry into
the program by workers. Their theoretical effects will be discussed below.
The MTRs may be considerably higher for families participating in more than
one program, for the individual MTRs may add up to a considerably higher
cumulative MTR. Table 2 shows multiple benefit receipt among nonelderly singlemother
families in 1997, excluding the EITC and child-care programs. Around 16% of
all single mothers received AFDC, Medicaid and Food Stamps and possibly one other
program. About 32% of all single mothers who receive benefits from any program at
4 Under so-called waiver programs, states began to be allowed to set their MTR in the early 1990s. See
Moffitt (2003) for a detailed discussion.
5 Work-related expense deductions are somewhat ambiguous because, if they do reflect true costs of
working, they do not lower the MTR ifthe latter were calculated on net income in the first place. But
the deductions vary considerable from state to state, from program to program, and over time, and are
unlikely to represent true costs of working.
2399
R.A. Moffitr
Table 2
Multiple benefit receipt by nonelderly single-mother households, 1997 a
Source of benefit Percent distribution
No Program 48.0
AFDC, Food Stamps, Medicaid, and another program 10.4
AFDC, Food Stamps, and Medicaid only 6.0
AFDC, Medicaid, and another program 1.1
AFDC and Medicaid only 0.7
Medicaid only 6.2
Food Stamps only 1.8
Other cash transfers only 4.9
Other 18.0
Total 100.0
a Source: Tabulations from the Survey of Income and Program Participation by Kara Levine, University
of Wisconsin.
all are in this multiple receipt category; thus about a third of all welfare recipients
in this group have significant multiple benefit receipt. This is thus the major multiple
recipient category among the programs listed in Table 1.
The cumulative MTR for families receiving these three benefits is not as high
as might be thought, however. The MTR for Medicaid is zero until the eligibility
point for AFDC is reached (Medicaid eligibility is automatic with AFDC receipt).
Also, the Food Stamp program includes AFDC benefits as income. As a result, the
cumulative MTR for TANF and Food Stamps combined is [t + 0.3(1 - t)] where t is
the TANF MTR. Prior to 1996, when t= 1.0, the cumulative MTR was thus simply
1.0, not 1.3.
Cumulative MTRs after 1996 are shown in Table 3 for 12 states. The income
calculations include the EITC and positive tax payments as well as TANF and Food
Stamps, but not Medicaid. In the absence of the EITC, moving from 0 hours to
part-time hours at the minimum wage results in MTRs ranging from 34% to 71%.
While these rates are less than 100%, they are not low by most standards (e.g.,
the positive income tax faced by most US families). Moving from part-time work
to full-time work results sometimes in higher MTRs and sometimes in lower ones,
depending on the state. Higher MTRs typically occur when the individual hits the
TANF income eligibility cliff and loses all benefits, while lower MTRs typically occur
when the individual hits that limit in the neighborhood ofpart-time hours, implying that
earnings obtained by moving to full-time work are taxed at lower, nonwelfare rates.
Moving from minimum wage to $9/hour at full-time generally results in somewhat
lower MTRs because usually the individual is off welfare by that point and hence
only the nonwelfare rates almost always apply. The nonconvex budget set created by
2400
Ch. 34: Welfare Programsand Labor Supply
Table 3
Cumulative marginal tax rates for recipients of TANF and Food Stamps in 12 states, 19 9 7 a,b
State Rate (%)
From no work to From part-time work From minimum wage
part-time work at to full-time work to $9 hourly wage
minimum wage at minimum wage at full-time work
w/o EITC with EITC w/o EITC with EITC w/o EITC with EITC
Alabama 46 6 33 9 24 58
California 50 9 67 33 67 89
Colorado 57 17 71 39 29 59
Florida 46 6 59 28 35 63
Massachusetts 57 13 64 28 64 87
Michigan 63 23 84 47 35 63
Minnesota 55 8 65 27 69 89
Mississippi 34 -6 32 7 24 55
New Jersey 64 23 62 30 41 67
New York 65 16 67 27 55 84
Texas 50 10 24 0 25 57
Washington 71 30 67 33 50 76
a Source: Coe, Acs, Lerman and Watson (1998, Tables 4, 5).
b Income includes earnings, TANF and Food Stamp benefits, federal and state EITC amounts, less
employee payroll and federal and state income taxes. Minimum Wage is $5.15 per hour. Family size of
three assumed.
this pattern of rising then falling marginal rates tends to drive labor supply decisions
to either extreme (i.e., either down to part-time work and up to an earnings level
considerably above eligibility), a result to be shown formally below6
.
The EITC has drastic effects on these rates7
. Moving from no work to part-time
work or from part-time work to full-time work, rates are lowered by as much as 40%,
which is the largest subsidy rate in the EITC. The resulting MTRs are quite low,
no more than 30% in the first case and 47% in the second. However, when moving
from the minimum wage to $9/hour at full-time work, the EITC actually increases the
cumulative MTR because the individual is in the phaseout range of the EITC by that
point. Cumulative MTRs are never lower than 55% and are often in the 80% range
when moving to this higher earnings level. Thus, the EITC effectively convexities the
budget set, which should be expected to drive labor supply decisions toward the middle
6 Adding Medicaid into the calculation would increase MTRs greatly in the second and third categories,
where TANF eligibility is lost.
7 After 1996 states are allowed to count the EITC in TANF income calculations but they generally do
not do so.
2401
point. Thus, there is an incentive to increase earnings from no work to full-time work
but also to decrease earnings from high-wage work to lower-wage work, with expected
ambiguous effects on labor supply (see below).
Finally, it should be emphasized that the benefit formula is not the only source of
labor supply incentives in many US transfer programs, for work requirements have
an additional, perhaps stronger, effect. In the TANF program, recipients are required
to work some minimum number of hours (usually 20 per week or more) after some
minimum stay on the welfare rolls, and sanctions (i.e., benefit reductions) are applied
vigorously if recipients fail to meet the requirements. The Food Stamp program also
has work requirements for all recipients and particularly strong requirements for nondisabled
single individuals. Models of work requirements will be considered below.
2. Theoretical models and issues
2.1. Introduction
The basic static model of labor supply familiar to undergraduate textbooks has been
the workhorse of the literature on work incentives of welfare programs for over thirty
years. Despite the clear importance of dynamics, human capital, job search, and other
considerations in the study of the effects of welfare programs, the endurance of the
static model is testimony to its usefulness for the analysis of a wide range of types of
welfare program alternatives and the analysis of the comparative statics of an equally
wide range of effects of simple changes in program parameters. Its ease of graphical
analysis increases its usefulness as an analytic tool.
Relative to the same model used for the work incentives ofincome and payroll taxes,
the welfare application has some unique features, however. One is that the meanstesting
inherent in a welfare program necessarily creates a nonconvexity in the budget
set somewhere over the range of earnings, at the very least at the point where income
rises to the point ofineligibility. Changes in welfare reform parameters inevitably either
change this eligibility point or change the incentives for individuals to locate above
or below it, and this sets off work incentives which greatly complicate the analysis
relative to that ofincome and payroll taxes. A related unique feature of welfare program
analysis is that participation in welfare itself is a choice variable, partly because of the
decision to locate above or below the eligibility point but also, it turns out, because
even some of those whose income is below that point choose not to go onto welfare.
This also adds some complexity to the model and to the analysis of labor supply
effects.
These unique features have important ramifications for normative questions concerning
optimal design of welfare programs. It turns out that almost all changes in welfare
program parameters have labor supply effects that differ for different individuals
in the distribution, and the desirability of implementing those changes requires a
social welfare function or some other public choice mechanism which explicitly or
2402 R.A. Moffitt
Ch. 34. Welfare Programsand LaborSupply
implicitly assigns weights to distributional objectives. Almost no program reforms
have unambiguously desirable labor supply effects on their own. Even a program as
well-known and popular among economists as the negative income tax shares this
feature, and its normative advantages in terms of labor supply are quite questionable
without such an analysis. While the exposition presented below will stick fairly closely
to positive analysis, normative issues will necessarily be referred to in appropriate
places.
A final feature of the welfare model also relating to optimal design arises from
the above-noted feature that welfare program participation is a choice and that some
choose, voluntarily, not to participate. This necessarily self-categorizes the eligible
population into those on welfare and those not, which implies that modifications of
the welfare system in the form of adding new programs - e.g., for food, housing,
medical assistance, or job training - must confront how those programs should be
made available to those on the initial welfare program (e.g., cash) and those not. Both
equity considerations play a role in that decision as well as efficiency ones, for if the
different programs are at all tied together, incentives to join the various programs will
be altered. A universalist policy which keeps all programs separate and makes all of
them available to all of the low-income population is one extreme, but one that has
been rarely chosen by US policy-makers.
2.2. Basic static model
The canonical static model of labor supply considers an individual with well-behaved
preference function over hours of leisure (L) and consumption (C) which we denote
as U(L,C) and budget constraint N + W(T - L) =PC, where N is exogenous unearned
income, W is the hourly wage rate, P is the price of consumption goods, and T is the
total time available in the interval. With normalization by P we may relabel C as
income, Y With hours of work defined as H= T - L, we can equivalently consider a
preference function U(H,Y) maximized w.r.t. N + WH = Y A generic welfare program
provides benefit B = G - t(WH +N), where G is the guarantee amount given to those
with zero income and t is the marginal tax rate. With benefits added into the budget
constraint, we have W(1 - t)H +G- tN =Y
Figure 1 shows the familiar budget constraint so created as segment CD, where the
intercept with vertical distance AC represents G and where the slope of segment CD is
-W(1 -t). The nonwelfare constraint is AE with slope -W and intercept N (assumed
to be zero for illustration). The arrows labeled 1 and 2 denote the two types of labor
supply response to the creation of the welfare program, and both result in reductions
in labor supply unambiguously.
The nonconvexity of the budget set clearly is a defining characteristics ofFigure 1.It
could be argued that individuals can convexify that constraint by moving on and offthe
program over time, thereby achieving an average hours of work and average income
that lies above the constraint CDE. Put differently, they could "build" a constraint
to bridge the nonconvexity. Ignoring preference discounting, higher utility could be
2403
R.A. Moffitt
Income
E 2
C
A
< Hours worked Fig. 1. Generic welfare program.
gained in that way than by choosing a fixed hours along CDE. One theory of welfare
turnover is that it is induced precisely by this feature of the constraint, explaining
why many individuals alternate between not working at all and being on welfare with
working and not being on welfare, treating the two as mutually exclusive alternatives.
The existence of nonparticipating eligibles as a complicating factor may be
mentioned at this point. For virtually all welfare programs, some individuals are
observed to locate on segment AD in Figure 1, in a location which would appear
to be inferior to being on welfare. One rationalization for this behavior is to assume
that being on welfare carries some disutility, possibly arising from the stigma of being
on welfare [Moffitt (1983)]. This can be modeled by inserting a welfare participation
dummy, P (equal to I if on welfare and 0 if not) into the preference function as in
U(H, P), where P has negative marginal utility. With heterogeneity of preferences
w.r.t. P across the population of a sufficiently wide range, some individuals will choose
not to participate despite the income and leisure gains to doing so. An alternative
rationalization is to assume that there are fixed costs of going onto welfare in the
form of time, money, or "hassle" costs of complying with the myriad rules of welfare
and requirements to visit the welfare office periodically. With heterogeneity of such
costs across the population, and with a sufficiently wide range that for some individuals
the costs will exceed the utility gains of welfare participation, some individuals will
again choose not to go onto welfare. These two models are observationally equivalent
without some further structure imposed on them or some actual data on stigma or
participation costs 8.
The labor supply implications of such nonparticipating eligibles are, at this stage,
straightforward, for participation in welfare is more likely, the greater the utility gains
from income and leisure, holding the stigma or fixed costs constant. Whether labor
supply falls or rises upon joining the welfare program depends on the substitutability
I One approach in the literature to the stigma model is to consider stigma to be a function of how
many other families are on welfare. This sets up a social interactions model with interdependent choices
in the population. See Besley and Coate (1992a), Lindbeck, Nyberg and Weibull (1999) and Nechyba
(1999).
2404
Ch. 34. Welfare Programsand Labor Supply
Income
< Hours worked Fig. 2. Effect of an increase in G.
or complementarity of P and L in the preference function, but if P is separable then
labor supply will fall, as illustrated in Figure 1 by the arrow 1.
The most important two comparative statics of the model are those involving the
effects of a change in G and a change in . Figure 2 illustrates the effect of an increase
in G. All three types of labor-supply effects, shown by the three arrows, are negative.
Note that the change in labor supply is not a measure of traditional pure income effects
because those represented by arrows 2 and 3 experience substitution as well as income
effects, though these operate in the same direction as the income effects. Not shown in
the figure are the effects on nonparticipating eligibles, for some individuals will move
from segment AD to C'D' because the utility gains will now outweigh the costs of
participating.
A reduction in t is illustrated in Figure 3, where the pre-change constraint is shown
to have t= 1.0 (segment CD). Again, there are three types of response illustrated in the
figure, but in this case the second two are opposite in sign to the first, resulting in an
ambiguous net change in hours worked. The average change depends on the sizes of
the different responses as well as the relative numbers of individuals at different initial
points along the constraint. Those represented by arrow 2 are made newly eligible by
the reduction in t and hence reduce their labor supply whereas those represented by
arrow 3 now find the higher benefits available over the positive hours range to be
sufficiently great as to reduce labor supply to join the program 9
. Because the region
around points D and D' often falls in the part-time hours range, it can be said that this
reform essentially subsidizes part-time work.
Among other things, this result shows why the employment rates or hours of work
of those on welfare may be correlated with the level of t even if there is no labor
supply response. For example, if arrows 1 and 3 are zero and arrow 2 is vertical (i.e.,
labor supply does not change when going onto the program for these individuals),
9 Ashenfelter (1983) termed the arrow-2 response a "mechanical" change and the arrow-3 response a
"behavioral" change. This interpretation is useful but is complicated by the presence of nonparticipating
eligibles, some of whom in the range DD' will choose not to go onto the program despite the potential
income and leisure gains.
2405
R.A. Moffittr
Income
< Hours worked Fig. 3. Effect of a decrease in t.
then employment rates and mean hours of work of those on welfare will rise after the
reduction in t, but this is just a compositional effect resulting from the addition of a
group of high-hours workers to the welfare rolls.
The implication of this analysis is that the negative income tax, and general
reductions in t, do not necessarily increase average labor supply [Levy (1979), Moffitt
(1992)]. A corollary of this result is that it is possible that 100% tax rates maximize
average labor supply. However, the distributional impacts of the tax rate reduction must
be considered. The reform does achieve its objective of encouraging the lowest-laborsupply
individuals to work more, but at the cost of some labor supply reduction among
higher-labor-supply individuals, many of whom are in the group termed the "working
poor". Incomes are increased among the latter group (at least for those represented by
arrow 2) as well, representing a shift in the income distributional impact of the welfare
program away from the lowest-income individuals and toward somewhat higher-income
individuals, at least in relative terms ". Much of the policy debate over the desirability
ofa negative income tax and similar reforms revolves around the relative distributional
weights assigned to the very poorest (sometimes called the "target efficiency" of the
program) versus the "working poor", who receive no assistance in a t = 1.0 program.
In the US TANF program, states have tried at least two measures to avoid the
labor-supply-reducing effects of t reductions. One is to disallow the lower t when
considering initial eligibility, and to allow the lower t only for those individuals who
have established eligibility under the stricter standard. In a world without transactions
costs, these barriers would have no effect and individuals who wished to locate in the
region DD' in Figure 3 would simply reduce their labor supply to point C to become
eligible and then increase their labor supply to the preferred point". With costs to
entry and application, this process will be slowed but not eliminated. A second strategy
10An expenditure-constant change in G and t, for example, would rotate the segment CD' in a direction
increasing the slope, showing more clearly the existence of such a distributional shift.
1l Random shocks to labor supply would achieve the same result, e.g., if every individual has a
probability p of being laid off or losing her job each period, after which labor supply can be freely
chosen until the next layoff occurs, individuals will eventually end up at the global utility maximum
regardless of the stricter eligibility rules.
2406
Ch. 34: Welfare Programsand Labor Supply
Income
- Hours worked Fig. 4. Welfare program with a notch.
by states is simply to impose an upper limit on income independent of the benefit
formula, as illustrated in Figure 4. In that figure, the notch occurs at point D and the
segment DD' is eliminated, discouraging workers above that point from coming onto
the welfare program. However, some will still reduce their labor supply to point D in
any case, and labor-supply gains that might have been experienced by those initially
on welfare who would have otherwise increased work into the DD' region are lost, so
there is a cost to this approach. However, a number of states have taken this approach
to its extreme by setting t = 0 at the same time as imposing such an eligibility limit.
Figure 4 also illustrates notch constraints in general, as occur in the Medicaid and
other programs. The dashed line from D to D' represents the effect of removing the
notch and allowing a gradual phaseout of benefits. This change has ambiguous effects
on labor supply for the same reason that a general reduction in t does. While it is true
that those initially at D are encouraged to work more (arrow 1), those represented by
arrows 2 and 3 experience labor-supply reductions. Thus, the labor-supply effect of
eliminating welfare-program notches is ambiguous in sign.
A different method of providing work incentives is by increasing the rewards of
being off welfare rather than on welfare. Whereas the reduction in t attempts to
provide incentives by encouraging individuals to combine welfare and work, increasing
subsidies for those who are off welfare and work could seemingly provide greater
labor-supply increases. Figure 5 illustrates one real-world case of such a reform, in
this case one in which Medicaid is offered to low-income families off welfare up to
some maximum income point (the dashed line C'D'). An increase in child-support
income off welfare has similar effects, but without the notch. Arrow 1 represents the
labor-supply increases generated by this reform as some individuals move off welfare
and receive Medicaid while off. However, labor-supply reductions also occur, as shown
by arrows 2, 3 and 4 (arrow 4 arises as some of those who did not want to participate
in the initial welfare program find Medicaid acceptable). Thus, the labor supply effects
of this reform are again ambiguous in sign.
Of course, providing a new form of welfare "off welfare" is just a matter of
semantics, for in essence this reform just adds another layer onto the initial welfare
system. Ignoring the possible responses ofnonparticipating eligibles or different stigma
2407
R.A. Moffitt
Income
< Hours worked Fig. 5. Medicaid expansion.
effects of the programs, the effect of this reform is no different than simply extending
Medicaid eligibility for those initially on welfare up to point D', which is very similar
to a reduction in t. Alternatively, simply combining the two programs and having
one program with the constraint CDD' would be equivalent. The popularity of some
of these reforms to the general public appears instead to be based on the idea that
Medicaid, child support, and other forms of welfare represented by the constraint C'D'
are preferable to cash welfare possibly represented by CD. But that requires bringing
voter preferences into the model, which has not been done here. In the recipientoriented
model used here, there is no important difference between "on welfare" and
"off welfare" reforms if they are all essentially welfare, and it is merely a matter of
government form whether assistance is provided in one program or multiple programs
so long as the eligibility groups are identical. The fundamental design issue of how to
construct the overall constraint, where assistance should be given across the income
and labor-supply distribution, and where labor-supply incentives and disincentives are
preferred to occur, remains and requires addressing the same distributional questions
already discussed.
2.3. Dynamics
We shall consider three separate topics in this section: (1) welfare-program effects
on labor supply in a life-cycle context, (2) effects of welfare programs on human
capital investment, (3) effects of welfare-based human capital investment programs,
and (4) time limits on welfare receipt. There has been relatively little research on these
issues so this section will be necessarily briefer than the last.
Life-cycle labor supply models are generally considered to be separable in order
to apply two-stage budgeting techniques familiar from the literature on consumption
[Blundell and MaCurdy (1999)]. If an individual is on welfare in all periods, the
analysis of labor-supply effects of welfare programs is simple because those programs
just reduce the net wage and increase nonlabor income. This leads to reductions in
labor supply in all periods and to intertemporal wealth effects which also reduce labor
supply. If the individual is on welfare in some periods and off in others, this also can
be accounted for within the framework by the use of virtual income and virtual net
2408
Ch. 34: Welfare Programs andLabor Supply
wage formulations (Blundell and MaCurdy). Intertemporal asset allocations can be, as
before, separated from the within-period decisions, and the latter are essentially the
same as those in the static model with an adjustment in full income to take account
of the asset flows. Thus, the net result of these models in the life-cycle labor-supply
literature is that there is little if any alteration in the basic labor-supply effects found
in the static model.
Two different types of effects are possible, however. One is the possibility of
"building" a convexified constraint over the nonconvex budget set by moving on and
off the program in different periods, as mentioned previously. This case has not been
analyzed in the literature. Another is the effect ofthe income floor in transfer programs,
which reduces precautionary saving and hence labor supply, as noted by Hubbard,
Skinner and Zeldes (1995).
Effects of welfare programs on human capital investment have also been little
studied 2 .
Kesselman (1976) analyzed the effects of a welfare program on human
capital, assuming that the individual was on welfare both before and after the human
capital investment decision. Under this assumption, the effect of welfare programs is
to reduce the incentive for human capital investment because both the opportunity
cost of time changes from W to W(1 - t) and the return to investment changes from
W to W(l - t); thus the rate of return is simply reduced to (1 - t) of what it is in the
absence of the program. This result is identical to that of the effect of a proportional
income tax on human capital investment in an equally simple model. Income effects
play no role, as is typical in human capital investments which assume perfect capital
markets.
However, an investment in human capital which has a sufficiently high return as
to move the individual above the income eligibility point and hence off welfare has
a higher return than this, because the return, at the margin, remains W rather than
W(1 -t). Thus, human capital investment is encouraged relative to what it would be
in the absence of the program. This is entirely the result of the nonconvexity of the
budget set, which leads to a type of increasing returns to investment.
The effects of human capital investment programs, or job training programs, for
welfare recipients has been discussed by Moffitt (2003). The relevant case is that in
which a recipient is required to undergo training as a condition of welfare receipt, and
thus human capital investment becomes a type of work requirement. In this case the
net present value of the investment opportunity becomes part of the welfare package.
In a two-period model, that net present value is
NPV = W(1 - t)I
+ I+r {P2 [(W - W)(1 - t)H2] + (1 -P2) [(W2 - W) H2 - (G - tWIH2)},
(1)
12 See Miller and Sanders (1997) for an exception.
2409
where W1 is the wage if the recipient were not to undergo the training program, W2
is the (higher) wage in period two if she does, I is the amount of investment time
required in period one (assumed to come out of work time rather than leisure), H2 is
hours worked in period 2, and P2 is a welfare participation dummy in period 2 if the
recipient undergoes the program 13. The investment cost is represented by the first term
and the return by the term in curly brackets, which depends on whether the individual
is still on welfare after the wage increase. If not, the return is equal to the earnings gain
minus the welfare benefit lost. If this net present value is positive, the welfare program
is enhanced in value because it offers a training program with a positive return which
is not available elsewhere, leading to an increase in welfare participation; but if it is
negative, the value of the welfare program is enhanced and welfare participation will
decline.
If the program is voluntary for welfare recipients, then the value of welfare cannot
decline and welfare participation will ambiguously increase. Both of these cases
illustrate the issue of tied transfers discussed previously, for a universal human capital
program available to all low-income individuals would not have an effect on the welfare
caseload as it does in these cases, when the program is only offered, or mandated, to
welfare recipients.
Time limits on welfare receipt have dynamic effects that are more interesting than
their long-run static effects. The long-run static effects of time limits are simply to
eliminate welfare completely, and this should be expected to increase labor supply for
the same reasons that welfare decreases labor supply in the first place. Supporters of
time limits sometimes argue that mandatory human capital investment programs of
the type just discussed would, if required during all available years ofwelfare receipt,
increase potential wages enough that a former welfare recipient could be better off
after the end of the time limit by working than he or she would have been on welfare.
The empirical evidence on the rates of return to these programs does not support this
conjecture, at least on average. Instead, time limits are best interpreted as simply a
desire to reduce redistribution 14
There are two types of dynamics effects that could arise from the presence of time
limits. One would occur if individuals on welfare anticipate the approach of the limit
and begin to leave welfare and accept job offers at an increasing rate as the limit nears.
This behavior has been noted for unemployment insurance recipients approaching the
point at which their benefits will be exhausted [Moffitt (1985)]. In the UI case, this
behavior is generally explained by the randomness of wage offers and the desire to
accept an attractive offer when it arrives even if it does so somewhat in advance of
13The change inH2 resulting from the increase inthe wage isignored for simplicity.
14 A more interesting question iswhy time limits would be preferred by the voters to simple reductions
inbenefit levels. This preference isprobably based on the assumption that individuals experience a finite
number of discrete negative shocks over their lifetimes and that full support should be given during
those periods but not during others. A lower but permanent benefit level would simply lower the amount
of support given during the periods during which negative shocks occur.
2410 R.A. Moffitt
Ch. 34: Welfare Programs and LaborSupply
the benefit exhaustion date. The same may apply for welfare recipients approaching
a time limit. The result is that labor supply will begin to increase in advance of the
actual time limit, and that labor supply effects would be observed to occur even for
those who do not actually exhaust their benefits 15
A second response can occur if recipients "bank" their benefits by going off the rolls
during good (labor market) times and saving their benefits for bad times (downturn
in labor market, unexpected negative income shock, etc.). This result again requires
the presence of uncertainty in future wage offers. A simple artificial model proves
the possibility of this response. Suppose that the lifetime consists of two periods and
that welfare can only be received in one of them. The per-period utility function is
Y - aH, where Y is income and H is a dummy equal to 1 if the individual works and
Oif not. Let B be the welfare benefit and suppose that an individual has a per-period
probability p of a wage draw W >0 and probability 1-p of a wage draw W = 0. To
make the comparison stark, let the positive wage draw equal B. Then, in the absence
of any time limit, the individual will be on welfare both periods regardless ofthe wage
draw because even if a positive wage is drawn, the disutility from working will make
welfare participation the preferred choice. But with a time limit, an individual who has
drawn a positive wage of B in the first period and is choosing whether to work and
stay off welfare vs. going onto welfare and using up her entitlement compares utility
for these two alternatives, which are, respectively
B - a +/3[pB +(l -p)B] if work in period 1, (2)
B +-/[p(B - a) +(1 -p) 0O] if go onto welfare in period 1, (3)
where fB < 1 is the discount rate. The value of working in period 1 is therefore
-a( - p) + B(1 -p). (4)
If p <1 then this quantity can be positive if the second term is larger than the first.
Thus the person will choose to work today and go onto welfare tomorrow even though
the myopic decision today would be to go onto welfare.
2.4. Work requirements and tagging models
In several US welfare programs, work requirements have become an alternative means
by which labor supply of welfare recipients can be increased. Figure 6 shows the effect
of such a program which requires Hmin hours ofwork on the part of the recipient. This
eliminates the portion of the budget BC. Arrows 1 and 2 show possible responses,
both of which lead to increases in hours of work. A work requirement, therefore,
15 See Gottschalk (1988) for a model of the effect of welfare programs on job search and Krueger and
Meyers (Chapter 33, this volume) for a review of the unemployment insurance literature.
2411
R.A. Moffitt
Income
< Hours worked Hmn Fig. 6. Work requirement.
unambiguously increases labor supply. This should not be surprising because it is
equivalent to eliminating welfare altogether over a certain hours range, and welfare
is known to decrease labor supply unambiguously.
This favorable result is often contrasted by policy-makers with a reduction in t,
which yields ambiguous effects on labor supply. However, the comparison is not
appropriate because the conventional welfare program attempts to provide support to
those who have sufficiently low wages (perhaps because of health reasons or very
young children) that their hours of work are low or zero whereas a work requirement
system provides no such support. Work requirement programs instead must necessarily
be accompanied by a categorization of the welfare population into those who can
work, who are given the budget constraint shown in Figure 6 with a minimum hours
requirement, and those who cannot work, who are give support even at zero hours.
Work requirements also stand in contrast to the idea of a negative income tax, which
attempts to provide income support to those with low wages at the same time as
providing some work incentives (through a reduced t). The favorable labor supply
effects of work requirements are entirely dependent on the ability to adequately
categorize the welfare population into the two groups.
There is a literature on tagging and related issues which considers the properties
and desirability of these types of categorical systems. Akerlof (1978) showed that if
individuals can be "tagged" as nonemployable, they can be given a greater G and
lower t than they could under a noncategorical negative income tax. He, and others
in this literature, directly dealt with the incentive problem to change categories by
requiring that an incentive compatibility constraint be set that would discourage such
behavior. Parsons (1996) extended the Akerlof model to consider what he termed
"two-sided" error, meaning that not only are some of the untagged individuals truly
nonemployable, but some of the tagged individuals are in fact employable. This
leads to an optimal structure in which a positive G is given to both tagged and
untagged individuals, but a higher G and lower t is given to the tagged group. More
direct consideration of workfare as an alternative is considered in other papers. For
example, Besley and Coate (1992b, 1995) showed that, under a different optimization
2412
Ch. 34: Welfare ProgramsandLabor Supply
criterion, workfare can be used as a screening device to ensure that higher-wage
individuals do not take advantage of the program. In this rather different justification
for work requirements, all recipients must undergo the cost of complying with work
requirements, but benefits can be higher because high-wage individuals have been
screened out 16
.
The literature surrounding the negative income tax in the late 1960s and early 1970s
emphasized the difficulties in tagging. It was generally argued that the administrative
difficulty in assigning recipients to categories is too great for work requirements to
avoid large error. It was also argued that work requirements would inevitably end up
giving individual caseworkers in local welfare offices great discretion in deciding who
is able to work and who is not, and that this would create an unacceptable level of
inequitable treatment across individuals who are observationally identical [Barth and
Greenberg (1971), Browning (1975), Lurie (1975)].
2.5. Wage and earningssubsidies
Wage and earnings subsidy programs have been discussed for many years as a means
to increase labor supply incentives. In a wage subsidy, a government subsidy increases
the hourly wage of an individual by [a - /3W], where W is the individual's initial wage
and /3< 1.The resulting remuneration per hour for an individual is a +(1 - /) W. The
subsidy declines as W rises and reaches zero at W =a//3. In an earnings subsidy, an
individual receives a subsidy S=sWH for earnings up to some cutoff level C, where
s < 1, and after that cutoff level the subsidy is recalculated as S=sC-r(WH-C),
which declines at rate r for every dollar of earnings. The subsidy reaches zero at
earnings equal to [C(s-r)/r].The net wage up to C is W(1 +s) while the net wage
above C is W(1 -r) 17.
Figure 7 shows how both types of subsidy affect the budget constraint. The wage
rate subsidy creates the budget constraint AC while the earnings subsidy creates the
constraint ABD. The labor supply responses are indicated by the arrows. The response
for the wage subsidy is represented by arrows 1 and 2. While those entering work from
nonwork clearly increase labor supply, the income and substitution effects of those
initially at positive hours go in opposite directions, leading to a change ambiguous
in sign. However, most of the groups made eligible for these subsidies have forwardbending
labor supply curves, so the impact of the subsidy is generally expected to
be positive. The earnings subsidy has the same effect in low ranges but a negative
effect in the higher ranges of earnings and hours worked, where the earnings subsidy
16 For other papers in this literature, see Beaudry and Blackorby (1998), Chone and Laroque (2001),
Cuff (2000) and Immonen, Kanbur, Keen and Tuomala (1998).
17 In some earnings subsidies, like the US EITC, there is a flat range of earnings in the middle over
which the subsidy stays at its maximum, before starting to decline. There is a large literature on wage
and earnings subsidies dating from the 1960s and 1970s. For some of the early discussions, see Barth
and Greenberg (1971), Garfinkel (1973), Kesselman (1969, 1973) and Zeckhauser (1971).
2413
R.A. Moffitt
Income
< Hours worked Fig. 7. Earnings and wage subsidies.
is reduced. Arrows 3 and 4 both represent reductions in labor supply. The net effect of
the subsidy is ambiguous in sign. As for the welfare programs discussed earlier, the
distributional impact of the program is important because the positive and negative
labor supply effects occur in different ranges.
A simple comparison of the wage and earnings subsidies in the figure indicates
that wage subsidies, because they do not have the taxback region BD, are more
likely to increase labor supply than earnings subsidies. However, the wage-rate subsidy
must nevertheless be phased out as the wage rises, and this creates a disincentive for
human capital investment, job search, and other activities seeking higher wages. As
it has turned out in the USA, wage-rate subsidies have foundered in any case on the
administrative difficulties of measuring hourly wages and basing a transfer on them.
Many individuals are not paid by the hour, for example, and even those that are often
have an agreement with the employer for a stipulated number of hours ofwork, making
the contract closer to an earnings agreement than an hourly wage agreement. Further,
a wage-rate subsidy creates opportunities for fraudulent collusion on the part of the
employee and employer to set the pay in terms of earnings rather than wages and
then to overstate hours of work, pushing the reported wage rate down and the subsidy
up. Preventing manipulations of the reported hourly wage would be quite difficult. For
this reason, earnings subsidies, which are administratively simpler to conduct, are more
common, even though they, too, have some incentives for overreporting of earnings
[Hotz and Scholz (2003)].
The goal of wage and earnings subsidies is explicitly to increase earnings and labor
supply. A secondary goal - though a primary one to some voters and policy-makers - is
to increase government support for the "working poor", roughly meaning those families
and individuals with significant hours of work but still low earnings. An earnings
subsidy can be designed so that the maximum subsidy at point B in Figure 7 is in the
middle of the range of earnings that define that group. This is a purely distributional
goal that, in principle, could be at odds with the goal of increasing labor supply,
particularly if the labor supply disincentives in the taxback region are significant. As
has been demonstrated in several programs already in this review, subsidizing one part
2414
Ch. 34: Welfare Programsand Labor Supply
Income
Fie. 8. Earnings subsidy nlus welfare ro<
Hours worked gram.
of the hours or earnings distribution tends to draw individuals to that region from both
above and below.
Earnings subsidies have also been proposed for those demographic groups who
are eligible for an existing income transfer program, in the hopes that the earnings
subsidy can counter the labor supply disincentives created by the tax rate of the
welfare program. The combination of the two is illustrated in Figure 8, which is
drawn on the assumption that the maximum earnings subsidy occurs at an hours
level in excess of the hours level which ends the income transfer . Compared to
the income-transfer-only, the addition of the earnings subsidy has the same mix
of positive and negative labor supply effects as the pure earnings subsidy, leading
to ambiguous effects on labor supply. However, labor supply incentives for initial
welfare recipients are all positive (assuming forward-bending labor supply curves)
and it is possible as well that those recipients may "leave welfare" in some cases 19
As illustrated in Table 3 discussed earlier, earnings subsidies have greatly reduced
the cumulative marginal tax rate on earnings for welfare recipients, although they
have also raised that rate at higher level of earnings. This effect is clear from
Figure 8.
The shift in Figure 8 is the same as that created by a reduction in t in a
welfare program at point F and below and differs only above E The labor supply
disincentives shown by arrow 4 are also the same as those created by the reduction in t.
However, the labor supply disincentives represented by that arrow occur at a higher
level of earnings than would occur from a reduction in t equivalent in magnitude
18 In the USA, most welfare programs do not include the EITC as part of income. If they did, the effect
of the earnings subsidy in pivoting the segment CD in Figure 8 around to CF would be much smaller.
Indeed, if t= 1.0 in the welfare program, there would be no effect at all of the earnings subsidy on the
slope of this segment if the earnings subsidy is included as income.
'9 As noted previously in the discussion of the Medicaid expansions and child support programs, many
policy makers and analysts regard earnings subsidies as "not welfare" because they are not stigmatizing
to the recipient and are more highly valued by the voter. This discussion ignores those effects. If stigma
is less for the earnings subsidy than for income transfers, it is also possible that an individual on welfare
may move off the welfare rolls to segment AD in Figure 8 and receive only the earnings subsidy,
becoming an eligible nonparticipant for welfare.
2415
R.A. Moffitt
Income
- 5. . L.rninl -UhUrA , Itlh - Wr-t ICJnAl,i
rig. 7. L lllllllgb 3UVlUy WILII d1 wl iW tn. lU .CIIIJ-
Hours worked ment plus welfare program.
to that created by segment CF (i.e., if CF were extended upward, as in the case
of a t reduction). In this sense, the earnings subsidy simply pushes the labor
supply disincentives further up the earnings and hours distribution than they would
be under a reduction in t. Once again, the desirability of this shift depends on
the distributional weights assigned to the groups in the population. On the other
hand, if point G falls closer to point F than is shown in the figure, then there is
little difference in an earnings subsidy addition and a reduction in t in the first
place.
The comparison of the program shown in Figure 8 with a reduction in t in a
welfare program is also complicated by the fact that public expenditures are almost
surely higher for the Figure 8 earnings subsidy addition than by the reduction in t,
for subsidies are paid out at much higher earnings levels which are typically in
much thicker regions of the income distribution. This makes it appear as though
all individuals can be made better off and no one worse off. A fairer comparison
would be an expenditure-constant comparison of an earnings subsidy addition versus
a reduction in t, for in that case the level of G would have to be lower at the time the
earnings subsidy is introduced to reduce expenditures. This demonstrates more clearly
that the earnings subsidy addition is a regressive program relative to a program which
reduces t, for it shifts support, in relative terms, away from the worst-off families and
individuals and toward those who are better off. This is the underlying reason for its
favorable effects on labor supply.
Finally, Figure 9 shows the effect of offering the earnings subsidy only to welfare
recipients who work a minimum number of hours, a suggestion that has been made by
some welfare policy analysts. For the same reasons as work requirements in general,
this will result in an increase in labor supply relative to an earnings subsidy program
without any such minimum hours rule, as shown by the arrows in the figure. However,
some individuals who would have increased their labor supply by a smaller number
of hours will fail to do so under this program, so some labor supply gains will be
lost. Also, once again, the distribution of government support shifts in this case, in
relative terms, away from the worst off and toward the better off, in return for the
more favorable labor supply effects.
2416
Ch. 34: Welfare Programsand LaborSupply
2.6. In-kind transfers
There is a small literature on the effects of in-kind transfers on labor supply relative to
the effects of cash transfers [Gavhari (1994), Leonesio (1988), Murray (1980)]. At first
glance, there does not seem to be any issue, given that in-kind transfers in developed
countries like the USA do not subsidize the market price of the commodity directly,
as in developing countries, but rather simply provide families with a fixed amount
of the good. That fixed amount initially seems close to having the same effect as an
increase in income. For example, suppose the utility function is U(L, S, C) where L
is hours of leisure, S is the quantity of the subsidized good, and C is the quantity
of other consumption goods. Let the amount of the gift of the subsidized good be
pS*=pG- t(WH +N), where p is the market price of S, G is the maximum amount
of the good given, t is the MTR in the program, and WIH and N are as defined in the
static labor-supply model (H +L = T). Then the resulting budget constraint is
W(l - t)H +N(l - t) +pG =pS +qC, (5)
where q is the price of C. This constraint looks very much like the constraint for a
cash transfer, for the relative prices of S, C and H (or L) are the same as they are in
a cash transfer.
However, as the above-cited studies demonstrate, the more interesting case is when
the optimal S resulting from the maximum problem presented by the budget constraint
in Equation (5) is less than the S* amount granted by the government. In this case
the individual is constrained to consume more of the good than he would have if the
transfer had been provided in the form of a lump sum cash grant. This distortion of
the consumption of S has effects on the consumption of L and C. The basic result
from the studies is that ifS and L are Hicks-Allen substitutes, then the effects of such
an "over-provided" in-kind transfer is to increase hours of work rather than decrease
them, because L is reduced by the excess consumption ofS. IfS and L are complements
however, the opposite is likely to occur.
There is no strong empirical evidence on this issue to date. The Food Stamp program
in the USA is generally considered to be nearly equivalent to cash because the amounts
of food coupons provided to families are generally less than they appear to spend,
i.e., they are inframarginal [Currie (2003)]. On the other hand, Medicaid and housing
subsidies have cash-equivalent values less than one because individuals would in
almost all cases consume less than the quantities provided by the government if the
transfer were in cash. But what evidence there is suggests that these programs have
work disincentives, but there has been no strong comparison of those effects relative
to those of cash transfers.
3. Empirical evidence: a short review
3.1. Introduction
The volume of research on the effects of welfare programs on labor supply has been
cyclical over the last three decades. While there was a large amount of research in the
2417
late 1960s and 1970s, there was relatively little in the 1980s and early 1990s. There
has been a resurgence of interest in these issues in the 1990s, however, accompanied
by an increase in the volume of new work.
Methodological controversies have been a major focus in this literature and continue
to be at the present time. One controversy has been over the relative merits of structural
versus reduced-form estimation of the effects of policy variation, and another has
been over the best sources of policy variation to use when estimating either structural
or reduced-form models. This distinction is not always fully understood. The first
issue concerns whether some particular source of policy variation - such as crossstate
variation in welfare rules, for example - should be used to estimate black-box
impacts of policy difference without using economic models in any formal sense
and without estimating any underlying behavioral elasticities, such as income and
substitution effects. The primary argument for structural estimation in this sense has
always been that it affords a superior method of conducting out-of-sample forecasting
of untested and unimplemented policies than can the estimates from reduced-form
models20
. While the evidence on this issue is not conclusive at this time, it is a fact
that estimation of reduced-form impacts ofpolicy variation has become more common
than structural estimation in the literature. The second issue concerns whether the best
source of variation to use is pure cross-sectional variation in policies across areas
(when it is available), variation across areas in how policies change over time (e.g.,
the state-fixed effects model), or variation between somehow-defined "eligible" and
"ineligible" populations, either in pure cross section or over time (the difference-indifference
method). The latter two sources of variation have been more heavily used
than the first. When these different sources of variation are each used to estimate justidentified
models, it is generally not possible to test them against one another, which is
one reason for the continued disagreement on which source is preferable. But whatever
source of variation is used, either structural or reduced form estimation can be applied
to it21
.
The issue ofthe source ofpolicy variation is the more important one in the analysis
of the effect of welfare programs on labor supply. Some programs in the USA
(TANF, Medicaid, child-care subsidies, housing programs) have considerable crosssectional
variation as well as variation over time that differs across areas, furnishing
the opportunity for the use of more than one type for identification. Yet other programs
(the EITC, Food Stamps, SSI) have essentially no cross-sectional variation because
they are national in scope and the same rules apply to everyone. These latter programs
must use some other source of variation for estimation and usually it is a source which
will rest on more tenuous assumptions.
20 See Heckman (2000) for a discussion.
21 Using non-policy sources of variation to identify policy impacts is in a different class. The main
argument for using such non-policy variation (e.g., variation in wages alone to identify the effects of
W(l -t) when there is no variation in ) is that it is the only alternative if there is no direct policy
variation cross-sectionally or over time.
2418 R.A. Moffitt
Ch. 34: Welfare Programsand Labor Supply
The review of the empirical literature below is divided into sections based partly
on these issues2 2
. The pre-1995 literature will first be reviewed, for that literature
tended to estimate structural models and use certain types of sources of variation for
identification. The post-1995 literature will then be reviewed, first covering structural
estimation and then estimation of reduced-form policy impacts. For the most part, the
labor supply literature in this area is much thinner than might be supposed and hence
there are not many studies to review.
3.2. Pre-1995 estimates
There have been several major reviews of the pre-1995 literature on the effects of
welfare programs on labor supply [Danziger, Haveman and Plotnick (1981), Moffitt
(1992), Hoynes (1997)] which exhaustively cover the studies. That literature will
therefore not be rereviewed here in detail but rather merely summarized. Most of
the studies estimated the effects of the AFDC program and consisted of structural
or quasi-structural models, using pure cross-state variation in AFDC benefits along
with other sources of identification to obtain parameter estimates 23
. They found almost
universally that the AFDC program reduced labor supply by from 10 to 50% of nonAFDC
levels. The substitution and income elasticities estimated from the literature fell
into the general range of those elasticities obtained from the literature on substitution
and income effects estimated from wage and nonlabor income variation, respectively,
suggesting that the simple static theory - which presumes these elasticities to be the
same - is roughly verified.
This literature also addressed the effects of a reduction in t and the ambiguity of
labor supply response noted in the previous section. In general, it was found that
for single mothers, the primary eligibility group for benefits, average labor supply
was quite inelastic with respect to changes in t holding G fixed, suggesting that
the labor-supply-inducing and labor-supply-decreasing effects roughly cancelled each
other out (see Moffitt (1992), for a discussion). While these effects were obtained
by simulation, using structural estimates of substitution and income elasticities to
forecast the positive and negative labor supply responses to a reduction in t over
the income and hours distribution, they are consistent with reduced form estimates
from the negative income tax experiments 24
. The NIT experiments tested multiple
G and t combinations and hence it was possible to determine the average effect of a
change in t, holding G fixed, by comparing treatment-control outcomes across different
experimental groups. The results showed a very mixed pattern of results, with labor
22 For reasons of length, and not because of any lack of inherent interest, the literature from countries
other than the USA will not be reviewed.
23 There was one state-fixed effects estimate from this literature [Moffitt (1986)].
24 The results of the NIT experiments will also not be reviewed here. See Moffitt and Kehrer (1981)
and Burtless (1987) for reviews.
2419
supply levels sometimes higher and sometimes lower with no consistent pattern and
with few significant differences [SRI International (1983, Table 3.9)].
Fortin, Truchon and Beausejour (1993) addressed the social welfare issue referred to
previously by simulating the effects of different combinations of G and t on a variety
of social welfare indices, most of them based on Atkinson-style inequality aversion
scales. Using a calibrated model with assumed elasticities, and a data set of Canadian
families, the results showed a large number of "perverse" effects of a reduction in t on
labor supply, consistent with prior evidence. However, the programs that most often
maximized social welfare were those with relatively low levels of both G and t. At
these levels, the amount of redistribution is not very high and, further, they occur in
a range where reductions in t tend to increase average labor supply. The paper thus
presents a useful exercise on how to address this important question, even though
the results may be specific to the data set used and to the income distribution in the
Canadian data set used for the analysis.
A scattering of estimates were obtained for other programs. Estimates of the effect
of the Food Stamp program showed very little labor supply response, possibly because
the benefit amounts in the program are quite small relative to those of cash programs.
There were a fairly large number of estimates of the effect of Medicaid on labor supply,
with the majority indicating some significant work disincentives of the program, albeit
concentrated among those in poor health [see Gruber (2003), for a review of these
studies]. One study attempted to estimate the labor supply effects of housing programs
using a very indirect method ofidentification, rather than variation in housing programs
per se, and found the program to reduce hours of work by about 4% [Murray (1980)].
There were also a number of structural estimates of the effect of the price of child
care on labor supply, both before and after 1995 and using similar methologies. These
studies almost uniformly show negative price effects on employment, implying that
child care subsidies will indeed increase employment [see Blau (2003), for a review
of this literature].
3.3. Post-1995 structuralestimates
Because of the decline in structural estimation in the labor supply literature, there
have been few structural estimates of labor supply responsiveness. Table 4 shows
four of the more well-known studies. Hoynes (1996) studied the AFDC-UP program
and found it to have significant negative effects on the labor supply of husbands and
wives, but that marginal reductions in t had little effect, consistent with prior work.
Hagstrom (1996) estimated the effect of the Food Stamp program on labor supply and
found it to have very small effects, even smaller than those found in the pre-1995
literature. This reinforces the sense that the Food Stamp program has little effect on
work disincentives. Keane and Moffitt (1998) focused on the labor supply effects of
participating in multiple programs, including not only AFDC but also Food Stamps,
subsidized housing, and the Medicaid program. They showed that cumulative tax rates
were generally greater than 100% in this case. Nevertheless, while their estimated
2420 R.A. Moffitt
Ch. 34: Welfare Programsand Labor Supply
. 0 O
C
N -*X 0 = as
U t
6 A,,
a0 0 C;M. M._
0.
M
C
>a;:
0I
g~~~a~
U
C). C 0 CO
0 0
ZS > c
e1U = W
CO
0
- - x
U a Ci CO
~44 M
I 0U 6 y0f
0C cO
D tI =A C X
Ca,
73
a aa. 3o
CO o 0.X t:C)C a
m o a a
e ae9 O
C3 d vl 3: a 5 C
) C) C
IR
'Oa.-
U
E.~ o
gUC~, 0 Dtc
Y ic 54B 8~~~~~~~c
eg O X .g O
D iv a Da aD E
V1X~C -rV C
· a i, (n a. . U I CA .0.
COCOa ~ lCO O
B·~o o E O40.uL;
C)
C CAa.
0
P
C)
C)z
0:.
n
3U
c
o
UE E C, E O
0 0 - 0 -
02 a0Z Z: -,a
0 00 O 0 CO 0 0zCO
E. 7 > =
c)0 C) Iw r C) C.
D C flO-5 a^
0 a O 0 0 a C On cd X a C
'0
g g
0
o
0v
0-
::.
to
v va
E0 0
0
o cCO O
C),'U a.
4 z: a EO < v:
I
cO
0
C)
a '.
x
U
o
- I
0 m .
0 a,
Uu
-" U
0 aU c.
2421
aE
DO
i
r1
a
0
C
M
o00
d
M
0
a.
C)
C)
C)
0
C)
c
f
E
~0
0
F
C)
a
C)
C)
C)
M)
3
-O
2
rE
-C
C)
C)
3
CO
a
a.
0
.
0
9to
0.
a
a
C
0
5I
0c
substitution and income elasticities were sizable, the net effect on labor supply of
reducing the marginal tax rates to a level below 100% was negligible. Meyer and
Rosenbaum (2001) focused on an attempt to explain the increase in employment rates
among single mothers from 1984-1996. They found that AFDC benefits and tax rates
(the latter affecting potential benefits if working) had expected effects on employment
probabilities, but that the time series increase in single-mother employment was less
affected by changes in those parameters and other welfare variables than a change in
the generosity of the EITC over the period25
.
Some of the simulations of alternative programs conducted by Keane and Moffitt
bore on one ofthe issues raised in the theoretical section above, which is whether wage
and earnings subsidies are likely to draw welfare recipients off welfare and into the
(subsidized) private labor market. Keane and Moffitt found AFDC and Food Stamps
to have heavy stigma attached to them but they assumed that there would be no stigma
associated with a wage or earnings subsidy program. As a consequence, they found
the latter types of programs to have much higher forecasts of work incentives than
reductions in the welfare tax rate t. Whether there would be any stigma attached to
such programs is not clear, however [see also Keane (1995)].
3.4. Post-1995 reduced-formpolicy impact estimates
The literature on estimating the reduced-form impact ofpolicy changes on employment
and earnings is also exceeding modest. Yelowitz (1995) used the expansion of
Medicaid in different states and in different age groups of children to estimate their
impact on employment, and found those expansions to have a positive effect on
employment rates. A number of policy-impact studies ofthe effect ofthe EITC on labor
supply have also been conducted [e.g., Eissa and Liebman (1996), Eissa and Hoynes
(1998)]. The results show that the EITC increases employment probabilities for single
mothers but has no net effect on their hours of work, if working. This is consistent
with the notion that labor supply is encouraged among initial nonworkers but that the
taxback region of the program reduces labor supply enough to cancel out the effects
among workers. As noted above, Meyer and Rosenbaum (2001) also estimated the
effects of the EITC on single mothers and also found it to have significantly positive
effects on employment probabilities. The results from the Eissa-Hoynes study shed
light on the effects ofthe EITC on married men and married women. The findings were
that the EITC increased the employment probabilities of married men but decreased
those of married women, and that it decreased the hours of work of workers of both
men and women. The results for married women may be because many of them are
located in the taxback region because their husbands have significant earnings in and
25 Because Meyer and Rosenbaum examined employment rather than hours of work, the "perverse"
effect of a change in t could not occur. They briefly examined effects on hours of work as a sensitivity
test but they noted that the model-independent variables were not set up for that dependent variable.
2422 R..4. Moffitt
Ch. 34: Welfare Programsand LaborSupply
of themselves and the wives may be regarded as the marginal worker, even at the first
dollar. The hours results for men may be a result of the same taxback region26
.
These constitute the main body of this type of work and is, as already noted, quite
small in volume. There have been no studies of this type for the SSI program at all,
nor any new studies of the Food Stamp or housing programs.
Many more studies have been conducted of the effects of the 1996 welfare
legislation, the Personal Responsibility and Work Opportunity Reconciliation Act
(PRWORA), which transformed the AFDC program into the Temporary Assistance for
Needy Families (TANF) program by introducing time limits, work requirements, and
devolved responsibility and block grant funding to the states [see Moffitt (2003), for
a detailed review of its provisions]. There were also a number of studies of so-called
"waiver" programs just prior to 1996 under which states undertook to test programs
that were in most cases similar to the later national PRWORA legislation. Evaluating
these waiver programs was made possible by the fact that different states tested
different types of programs and did so at different calendar times, furnishing variation
in policy which could be used to estimate impacts on labor supply and other outcomes.
However, evaluating the 1996 legislation is difficult because it was implemented
nationally and all states had to come into compliance with its main provisions. Thus
there was no cross-state variation in the overall nature of the program.
As a consequence, difference-in-difference methods have generally been used for the
evaluation of the effects of TANF. As Ellwood (2000) and Schoeni and Blank (2000)
note, use of these methods is particularly problematic when other reforms, such as the
EITC, were occurring roughly simultaneously, and when business cycle and economywide
trends were occurring which could affect different groups differently. A further
difficulty in evaluation has arisen because the separate impacts of time limits, work
requirements, and other provisions has been difficult to determine even in the pre-1996
waiver period. This is partly because a state's choice of those components is correlated
with other characteristics ofthe state and partly because their implementation was often
significantly different than what would be expected from their formal definition. As a
consequence, while there are a number of estimates of the overall impact of welfare
reform in this period, taken as a whole, there are almost no credible studies of the
impact of different individual components of reform taken individually27
.
Table 5 shows the main studies estimating the overall impact of the pre-1996 waiver
programs and of TANF. With a few exceptions, the studies show waivers to have had
26 See Hotz and Scholz (2003) for a more detailed review of these EITC studies and a number of others
that used indirect estimation methods to simulate the effect ofthe EITC on labor supply. It should be
noted that the Eissa-Hoynes study, while using direct policy variation for reduced form estimation, also
translated their impacts into estimated income and substitution elasticities.
27 An exception is Grogger (2000, 2001), who has attempted to estimate the independent effects of time
limits by using age variation in children combined with assumptions that that variation does not interact
identically with other welfare reform features. The validity of the assumptions needed for these methods
to be unbiased is unknown.
2423
R.A. Moffitt
a0
o o
G a ~0
0N
0 0 o - 0
,, G 0 a r.. .D., ,Coe , S!
0 Ga N b E o
o 0 - Ga 0 X e
tJ G
E
> 4 0
0 00 a E G
Galfl
Ga 0G o55
1 ~~~~~S
Ga~ a a G00GGa a~ G
e, ~ Oa, a
M~)~r ) 00.
0 0
o o
0 E c
Ga Ga ° .o S
po pfE x
°BB 0i E 0EE
O e a -S
-0
ta
" 0.
S
Ga
0. 0
-P
EGa
" E
P u.rl
-,
0
_ a
Ea
".E
E
.t
C CD
Gae
a~~~j r~r
o
a~~~~~~~~G
a~ ~ ~ a
0
0 C:.
0 4
O C
Ga3
0o ls ce
o gE ^ W cGaG
3
B .sE
O Ga
a
a -070
. 0 G
Go
= o
o
.E
a'
.
a
Ga
- E
GOp
< EY
._.
00
E
Ga
o
50
GaO
G .0
5
m
O~a
0-e
0 .C
E
30
o
Ga
Ga Ga
- p
0
0
0
c
Ga
Ga
. El
z
0
= p
-0F:w
.6
F:
0
2424
c
a
o
0.
-
0.
0
g0
E
· o
C.
0o
-0
a
a
o
Ga
.-
Ca
q
.0
LnGa
Ga
Gaou
Ga2
5E
LB0
o-
Ga
CJ
0
o
.
0
Fo
0
m
Ga
40
u
Ga
o
U:
-
m
Ga
d
.c
a
0
E0
CL~
Eas
o
.
0
1~
Ga
Ck
i,
Ch. 34: Welfare Programsand Labor Supply
' I s a=
6
oF -a
00a e
a a. Zo a a
,, Ar Cn
C Z ._u Uo >
OC CZo
6 b c O c 65 to
E E 00
z Zr 0 °Z a Z7
0 Z Z
a
0
_ Oc
O NC,
aO _
C*
= ada mCC
_ 0
J
z -
U
E) Gn
2425
o E
0
t'C
00
o
o e
a
0-
aU
r-U
C 3O
Pe
:0
0
a
0
0
H11
0
v0
0
vo
t
E
m
0a
a
a0
0
a.
P
Z
0
rd
o
a
0A
0
v
M
0
Vn
EQ
o
C<
a.
w:0
ZC4s
r
Q
U
-0
a
o E.
fl 0
UC A
o_
X .2z
.E a
Q US
g
CL B
0 a 0>
0i .E
ca
m Y,
- oo
- a
.eE
6~0 1f
a.~.
M 0
aama
0U
0o
a 0
<a.
a
a
C.
-aa
co
o
0
U
mn
a
a
P.
:
S
o
(N E
M e0)
g ag
M _
positive effects on most measures of labor supply and negative effects on measures of
AFDC participation, as expected. These studies all control for the state ofthe economy,
usually by controlling for the unemployment rate, so the estimated effects of welfare
reform are all intended to be net of the strong economy.
Two entries in Table 5 are for experiments which made use of traditional randomassignment
methods rather than cross-state variation in the presence of reform. These
studies generally also find positive effects on employment and earnings and negative
effects on welfare participation 28
. However, random-assignment methods are not wellsuited
for major structural reforms like the pre-1996 welfare waivers - or for TANF
itself - because such structural reforms tend to cause changes in local labor markets
and local communities that feed back onto the control group, and because structural
reforms tend to have significant effects on entry into welfare. Experiments produce
biased estimates of total reform effects under these circumstances 29
.
The estimates of the effects of TANF are generally positive on employment and
earnings but not always. Further, in some cases the effects ofTANF cannot be separated
from the effects of other policy changes occurring at the same time, as emphasized
by Ellwood (2000); he concludes that these difficulties are sufficiently severe that
the separate contributions of welfare reform, the EITC, and the economy cannot be
identified. McKernan, Lerman, Pindus and Valente (2000) and Schoeni and Blank
(2000) are the other two studies using difference-in-difference methods, one of which
finds TANF to have increased employment while the other finds it not to have done
so but to have affected family earnings, income, and AFDC participation. The two
studies used different control groups so this may be the source of the difference. What
evidence there is, therefore, indicates some TANF effects in the expected direction but
the small number of studies and problems in statistical inference make the conclusions
rather uncertain.
4. Summary
The labor supply and other work incentive effects of welfare programs have long been
a central concern in economic research. Work has also been an increasing focus of
policy reforms in the USA, culminating with a number of major policy changes in
the 1990s whose intent was to increase employment and earnings levels of welfare
recipients and other disadvantaged individuals. This review of the theoretical and
28 There have been many more random-assignment studies in this period but those listed in Table 5 are
those which had all of the main features of PRWORA, namely, time limits, work requirements, sanctions,
and enhanced earnings disregards, and which made these reforms within the AFDC system rather than
outside of it.
29 Another difficulty in the use of experiments for evaluating structural welfare reform is that the control
group is often contaminated by the general atmosphere of reform which changes the expectations of the
eligible population as a whole.
2426 R.A. Moffitt
Ch. 34: Welfare Programsand Labor Supply
empirical literature on this issue reveals that there is much that remains to be done
to gain a full understanding of these programs. Many issues relating to the optimal
levels of welfare program parameters and the social desirability of labor supply effects
in different parts of the income distribution remain to be studied, a key issue. New
policy initiatives in the area of work requirements, time limits, and other topics have
been understudied, as have dynamic models of labor supply response. The proper
integration of the complex multi-program environment in the USA is also a needed
area of research.
Even more on the empirical side, more research is needed in a number of areas.
While traditional studies of the effect of AFDC guarantee and tax rates are reasonably
plentiful, structural or quasi-structural models of the effects of welfare reforms in the
1990s are rare and yet are needed to understand the mechanisms by which effects of
that reform have taken place as well as to forecast the effects of new policies. Many
programs other than TANF are quite understudied, especially the SSI program but also
including the Food Stamp and Medicaid programs. Even the EITC, which has seen a
considerable amount of research attention, has at present been the subject of only a
modest number of studies. Thus there are many areas of new research to be conducted
on the effects of welfare programs on labor supply.
References
Akerlof, G. (1978), "The economics of 'tagging' as applied to the optimal income tax, welfare programs,
and manpower planning", American Economic Review 68:8-19.
Ashenfelter, 0. (1983), "Determining participation in income-tested social programs", Journal of the
American Statistical Association 78:517 525.
Barth, M., and D. Greenberg (1971), "Incentive effects of some pure and mixed transfer systems",
Journal of Human Resources 6:149-170.
Beaudry, P., and C. Blackorby (1998), "Taxes and employment subsidies in optimal redistribution
programs", Working Paper 6355 (National Bureau of Economic Research, Cambridge, MA).
Besley, T.J., and S. Coate (1992a), "Understanding welfare stigma: taxpayer resentment and statistical
discrimination", Journal of Public Economics 48:165-163.
Besley, T.J., and S. Coate (1992b), "Workfare versus welfare: incentive arguments for work requirements
in poverty-alleviation programs", American Economic Review 82:249-261.
Besley, T.J., and S.Coate (1995), "The design of income maintenance programmes", Review ofEconomic
Studies 62:187-221.
Blau, D.M. (2003), "Child care subsidy programs", in: R. Moffitt, ed., Means-Tested Transfer Programs
in the U.S. (University of Chicago Press, Chicago) forthcoming.
Bloom, D., and C. Michalopoulos (2001), How Welfare and Work Policies Affect Employment and
Income: A Synthesis of Research (Manpower Demonstration Research Corporation, New York).
Blundell, R., and T. MaCurdy (1999), "Labor supply: a review of alternative approaches", in:
O. Ashenfelter and D. Card, eds., Handbook of Labor Economics, Vol. 3A (Elsevier, Amsterdam)
pp. 1559 1695.
Browning, E.K. (1975), Redistribution and the Welfare System (American Enterprise Institute,
Washington, DC).
Burke, V (2001), Cash and Non-Cash Benefits for Persons with Limited Income: Eligibility
Rules, Recipient and Expenditure Data, FY1998-FY2000 (Congressional Research Service,
Washington, DC).
2427
R.A. Moffitt
Burtless, G. (1987), "The work response to a guaranteed income: a survey of the experimental evidence",
in: A. Munnell, ed., Lessons from the Income Maintenance Experiments (Federal Reserve Bank and
Brookings, Boston) pp. 22 52.
Chone, P., and G. Laroque (2001), "Optimal incentives for labor force participation", Working Paper
2001-25 (Institut National de la Statistique et des Etudes Economique (INSEE), Paris, France).
Coe, N., G. Acs, R. Lerman and K. Watson (1998), Does Work Pay? A Summary of the Work Incentives
Under TANF (Urban Institute, Washington, DC).
Cuff, K. (2000), "Optimality of workfare with heterogeneous preferences", Canadian Journal of
Economics 33:149-174.
Currie, J. (2003), "Food and nutrition programs", in: R. Moffitt, ed., Means-Tested Transfers in the U.S.
(University of Chicago Press, Chicago) forthcoming.
Danziger, S., R. Haveman and R. Plotnick (1981), "How income transfers affect work, savings, and the
income distribution: a critical review", Journal of Economic Literature 19:975-1028.
Eissa, N., and H. Hoynes (1998), "The earned income tax credit and the labor supply of married couples",
Working Paper 6856 (National Bureau of Economic Research, Cambridge, MA).
Eissa, N., and J. Liebman (1996), "Labor supply response to the earned income tax credit", Quarterly
Journal of Economics 11:605-637.
Ellwood, D. (2000), "The impact of the earned income tax credit and social policy reforms on work,
marriage, and living arrangements", National Tax Journal 53(Pt. 2):1063-1105.
Fein, D., D. Long, J. Behrens and W Lee (2001), The ABC Evaluation: Turning the Corner: Delaware's
A Better Chance Welfare Reform Program at Four Years (Abt Associates, Cambridge, MA).
Fortin, B., M. Truchon and L. Beausejour (1993), "On reforming the welfare system: workfare meets
the negative income tax", Journal of Public Economics 51:119-151.
Friedman, M. (1962), Capitalism and Freedom (University of Chicago Press, Chicago).
Garfinkel, I. (1973), "A skeptical note on 'the optimality' ofwage subsidy programs", American Economic
Review 63:447-453.
Gavhari, E (1994), "In-kind transfers, cash grants and labor supply", Journal of Public Economics
55:495-504.
Gottschalk, P. (1988), "The impact of taxes and transfers on job search", Journal of Labor Economics
6:362-375.
Grogger, J.T. (2000), "Time limits and welfare use", Working Paper 7709 (National Bureau of Economic
Research, Cambridge, MA).
Grogger, J.T. (2001), "The effects of time limits and other policy changes on welfare use, work, and
income among female-headed families", Working Paper 8153 (National Bureau of Economic Research.
Cambridge, MA).
Gruber, J. (2003), "Medicaid", in: R. Moffitt, ed., Means-Tested Transfer Programs in the U.S. (University
of Chicago Press, Chicago) forthcoming.
Hagstrom, P. (1996), 'The food stamp participation and labor supply of married couples: an empirical
analysis of joint decisions", Journal of Human Resources 31:331-358.
Heckman, J.J. (2000), "Causal parameters and policy analysis in economics: a twentieth century
retrospective", Quarterly Journal of Economics 115:45-97.
Hotz, VJ., and J.K. Scholz (2003), "The earned income tax credit", in: R. Moffitt, ed., Means-Tested
Transfers in the U.S. (University of Chicago Press, Chicago) forthcoming.
Hoynes, H. (1996), "Welfare transfers in two-parent families: labor supply and welfare participation
under AFDC-UP", Econometrica 64:295-332.
Hoynes, H. (1997), "Work, welfare, and family structure: what have we learned?" in: A. Auerbach, ed.,
Fiscal Policy: Lessons from Economic Research (MIT Press, Cambridge, MA).
Hubbard, R.G., J. Skinner and S.P Zeldes (1995), "Precautionary saving and social insurance", Journal
of Political Economy 103:360-399.
Immonen, R., R. Kanbur, M. Keen and M. Tuomala (1998), "Tagging and taxing: the optimal use of
categorical and income information in designing tax/transfer schemes", Economica 65:179-192.
2428
Ch. 34: Welfare Programsand Labor Supply
Keane, M. (1995), "A new idea for welfare reform", Federal Reserve Bank of Minneapolis Quarterly
Review 19:2-28.
Keane, M., and R. Moffitt (1998), "A structural model of multiple welfare program participation and
labor supply", International Economic Review 39:553-589.
Kesselman, J.R. (1969), "Labor-supply effects of income, income-work, and wage subsidies", Journal
of Human Resources 4:275-292.
Kesselman, J.R. (1973), "Incentive effects of transfer systems once again", Journal of Human Resources
8:119 129.
Kesselman, J.R. (1976), "Tax effects on job search, training, and work effort", Journal of Public Economics
6:255-272.
Leonesio, M. (1988), "In-kind transfers and work incentives", Journal of Labor Economics 4:515-529.
Levy, E (1979), "The labor supply effects of female heads, or AFDC work incentives don't work too
well", Journal of Human Resources 14:76-97.
Lindbeck, A., S. Nyberg and J. Weibull (1999), "Social norms, the welfare state, and voting", Quarterly
Journal of Economics 114:1 35.
Lurie, I., ed. (1975), Integrating Income Maintenance Programs (Academic Press, NY).
McKernan, S.-M., R. Lerman, N. Pindus and J. Valente (2000), "The relationship between metropolitan
and non-metropolitan locations, changing welfare policies, and the employment of single mothers",
Mimeo (Urban Institute, Washington, DC).
Meyer, B.D., and D.T. Rosenbaum (2001), "Welfare, the earned income tax credit, and the labor supply
of single mothers", Quarterly Journal of Economics 116:1063-1114.
Miller, R., and S. Sanders (1997), "Human capital development and welfare participation", CarnegieRochester
Conference Series on Public Policy 46:1-43.
Moffitt, R. (1983), "An economic model ofwelfare stigma", American Economic Review 73:1023-1035.
Moffitt, R. (1985), "Unemployment insurance and the distribution of unemployment spells", Journal of
Econometrics 28:85-101.
Moffitt, R. (1986), "Work incentives in transfer programs (revisited): a study of the AFDC program",
in: R. Ehrenberg, ed., Research in Labor Economics, Vol. 8 (JAI Press, Greenwich, CT).
Moffitt, R. (1992), "Incentive effects of the U.S. welfare system: a review", Journal of Economic
Literature 30:1-61.
Moffitt, R. (1999), "The effect of pre-PRWORA waivers on AFDC caseloads and female earnings,
income, and labor force behavior", in: S. Danziger, ed., Economic Conditions and Welfare Reform
(Upjohn Institute, Kalamazoo, MI) pp. 91-118.
Moffitt, R. (2002), "Economic effects of means-tested transfers in the U.S.", Tax Policy and the Economy
16:1 35.
Moffitt, R. (2003), "The temporary assistance for needy families program", in: R. Moffitt, ed., MeansTested
Transfer Programs in the U.S., (University of Chicago Press, Chicago) forthcoming.
Moffitt, R., and K. Kehrer (1981), "The effect of tax and transfer programs on labor supply: the evidence
from the income maintenance experiments", in: R. Ehrenberg, ed., Research in Labor Economics,
Vol. IV (JAI Press, Greenwich, CT).
Mueser, P., J. Hotchkiss, C. King, P Rokicki and D. Stevens (2000), "The welfare caseload, economic
growth and welfare-to-work policies: an analysis of five urban areas", Mimeo (University of Missouri,
Columbia, MO).
Murray, M. (1980), "A reinterpretation of the traditional income-leisure model, with application to
in-kind subsidy programs", Journal of Public Economics 14:69-81.
Nechyba, T.J. (1999), "Social approval, values, and AFDC: a re-examination of the illegitimacy debate",
Working Paper 7240 (National Bureau of Economic Research, Cambridge, MA).
O'Neill, J., and M.A. Hill (2001), Gaining Ground? Measuring the Impact of Welfare Reform on Welfare
and Work (Manhattan Institute, NY).
Parsons, D.O. (1996), "Imperfect 'tagging' in social insurance programs", Journal of Public Economics
62:183-208.
2429
2430 R.A. Moffitt
Rowe, G., and T. Roberts (2002), Welfare Rules Databook Tables: State Policies as of July 2000 (Urban
Institute, Washington, DC).
Schoeni, R., and R.M. Blank (2000), "What has welfare reform accomplished? Impacts on welfare
participation, employment, income, poverty, and family structure", Working Paper 7627 (National
Bureau of Economic Research, Cambridge, MA).
Smeeding, T. (1982), "Alternative methods for valuing selected in-kind transfer benefits and measuring
their effects on poverty", Technical Paper 50 (Bureau of the Census, Washington, DC).
SRI International, Inc. (1983), Final Report of the Seattle/Denver Income Maintenance Experiment,
Vol. I (SRI International, Inc., Menlo Park, CA).
Tobin, J. (1965), "On improving the status of the negro", Daedalus 94:878-897.
U.S. House of Representatives (2000), Background Material and Data on Programs Within the Jurisdiction
of the Committee on Ways and Means (Government Printing Office, Washington, DC).
Yelowitz, A. (1995), "The Medicaid notch, labor supply, and welfare participation: evidence from
eligibility expansions", Quarterly Journal of Economics 105:909-940.
Zeckhauser, R.J. (1971), "Optimal mechanisms for income transfer", American Economic Review 61:
324-334.