The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
clevelandfed.org/research/workpaper/index.cfm
Working Paper 9408
THE IMPACT OF AFDC ON BIRTH
DECISIONS AND PROGRAM PARTICIPATION
by Elizabeth T. Powers
Elizabeth T. Powers is an economist at the Federal Reserve
Bank of Cleveland. The author thanks Alan Auerbach,
David Neumark, and Stephen Zeldes for helpful comments
and suggestions, and Jeff Gray for providing a state-matched
subset of observations for the National Longitudinal Survey
of Women. Kristin Roberts provided research assistance.
Working papers of the Federal Reserve Bank of Cleveland
are preliminary materials circulated to stimulate discussion
and critical comment. The views stated herein are those of
the author and not necessarily those of the Federal Reserve
Bank of Cleveland or of the Board of Governors of the
Federal Reserve System.
June 1994
clevelandfed.org/research/workpaper/index.cfm
Abstract
Recently, New Jersey and Wisconsin eliminated the practice of increasing the AFDC
benefits of families that bear additional children while on the program. Policymakers seem
to accept the notion that added benefits encourage participants to bear more children,
despite little direct formal evidence. This paper uses data fiom the National Longitudinal
Survey of Women to examine the impact of both the level of AFDC benefits and the per
child increment on births, as well as the effect of benefit policy and childbearing on AFDC
participation. Single-equation probit estimates suggest that women on AFDC are no more
likely than nonparticipants to give birth over the five years following the observation, but
that those births which do occur are positively associated with incremental AFDC benefits.
When birth and welfare participation decisions are estimated sequentially in a nested logit
framework, AFDC benefits are found to be a significant factor in the post-birth
participation decision, and empirical support emerges for the hypothesis that AFDC
benefits also encourage additional births. The estimated parameters are used to simulate
the impact on participation and births of eliminating incremental benefits for both new
program entrants and continuing participants. Even though the specification supports the
"AFDCbenefits cause births" hypothesis, eliminating the new-birth increment would
reduce total program costs by less than 3 percent, since both the per dollar effect of
benefits on births and the per child increments themselves are small.
clevelandfed.org/research/workpaper/index.cfm
I. Introduction
In the past two years, an environment of sluggish economic growth and tight state budgets has
led to renewed interest in welfare reform, overturning the apparent state and federal consensus on the
go-slow, human-capital-oriented approach to reducing welfare dependency as represented in the Family
Support Act of 1988. Many reformers have centered their proposals around the allegedly detrimental
effects of the Aid to Families with Dependent Children (AFDC) program on family structure. In 1992,
legislatures in New Jersey and Wisconsin voted to deny incremental benefits to AFDC recipients who
bear additional children.' The Wisconsin law also allows AFDC recipients who marry to retain some
Proposals introduced but not passed
benefits for a fixed period, even if their husbands are emp~oyed.~
into law include paying a one-time $500 bonus to AFDC recipients in Kansas who agree to be
implanted with the contraceptive Norplant, and paid childbirth expenses for unwed mothers in
Wyomjng who agree to put their child up for adoption. Ten more states have introduced legislation
linking welfare and Norplant in the past year.
What is the impetus for this seemingly sudden clamor for welfare reform'? Fiscal pressures,
particularly on state budgets, have undoubtedly played an important role. Nationwide, the AFDC
caseload experienced largely unanticipated growth of 27 percent between 1989 and 1992,
encompassing 4.8 million families by late 1992. Worsening economic prospects for low-skilled
workers -- not growth in the population of female-headed households -- are primarily responsible for
'New Jersey does loosen the earned income restrictions on new mothers to offset the lost benefits.
However, most AFDC recipients are not in the labor force and would doubtless face difficulties reentering it soon after giving birth. New Jersey is being sued over "the $64 question" ($64 being the
previously automatic per child monthly benefit adjustment) as of this writing (Wall Street Journal
[1994]). Georgia and Arkansas have recently joined New Jersey in this "experiment," and there are
motions on both the Republican and Democratic sides of the U.S. Congress to impose similar policies
nationally.
'while men with children and some two-parent families are potentially eligible for AFDC benefits,
more than 90'~ercentof recipient households are headed by women.
clevelandfed.org/research/workpaper/index.cfm
this sudden upturn. Changes in federal support for the unemployment insurance (UI) program in the
early 1980s prompted many states to tighten their eligibility standards, which also aggravated the
movement into elfa are.^
In 1991, 31 states responded to the dramatic growth in welfare expenses by freezing nominal
AFDC benefits, while nine others actually cut them. Growth in Medicaid program obligations also
seems to have squeezed AFDC spending. Not only are health care costs difficult to control, but most
state obligations to Medicaid are federally mandated but federally unfunded. Not surprisingly, other
welfare programs' funding has fallen since Medicaid's introduction in 1965. Figure 1 shows that, on
average, states have consistently spent about $0.90 to $1.00 per $100 of personal income on all forms
of welfare over the past 15 years. The shaded regions illustrate how AFDC (and other welfare)
spending has shrunk as Medicaid costs have c ~ i m b e d .On
~ average, states devoted 12.3 percent of total
expenditures to welfare by 1988, with two-thirds of that going to Medicaid.
It is little wonder, then, that following the institution of a lenient federal approval policy in
1992, many states eagerly came forward with new welfare experiments. Federal encouragement of
state-level experimentation has continued even as the current administration plots its own
comprehensive reform. Capping benefits regardless of the number of children may be a palatable way
to limit program costs, as long as new entrants are aware of the consequences of additional births.
However, the cost savings from such a policy depend on the share of program costs accounted for by
per child increments and the propensity of welfare mothers (and other women who may potentially
enter the program) to bear additional children. In this paper, I ask not only if there is empirical
support for the common belief that AFDC policy encourages fertility, but ,dso whether significant
3 ~ 1981,
n
the federal government instituted a 10 percent charge on states' borrowing from the U.S.
Treasury to cover their UI trust funds. Many states responded by increasing their base-period earnings
requirements, reducing the availability of UI to part-time and intermittently employed workers.
"The findings of Moffitt (1990).support the contention that total state welfare spending is
remarkably constant over long periods and that states are quick to reduce the AFDC component of.
welfare spending in response to new federal programs or mandates.
clevelandfed.org/research/workpaper/index.cfm
budget savings arise from denying per child benefit increments.
Ourline
Section I1 describes an economic framework for thinking about the effects of welfare on the
childbirth decisions of female heads of household. This necessitates a discussion of AFDC's potential
impact on marriage and labor force participation. In search of a basis for public opinion on welfare, I
first use data from the National Longitudinal Survey of Women (NLSW) to compare the childbearing
behavior of AFDC recipients and nonrecipients.' Although many stereotypes about welfare mothers
are confirmed on a prima facie basis, contrary to public opinion, I find that participants are no more
likely than nonparticipants to bear children. This surprising result holds up even after controlling for
many other characteristics in a regression framework. However, for AFDC mothers as a group, probit
models do indicate a statistically significant and positive relationship between incremental benefits and
births, suggesting that even fewer children might be bom to them if incremental benefits were reduced
or eliminated.
A single-equation approach, however, ignores the fact that increased numbers of children and
the presence of very young children normally enhance the likelihood of AFDC participation,
independent of the benefit policy. To isolate the effect of incremental benefits on births and
participation, the subsequent birth and participation decisions of a sample of female heads of
household from the NLSW are modeled sequentially and are estimated using a nested logit model.
First, I calculate the optimal AFDC participation rule as a function of family size, total AFDC
benefits, and other variables. I then estimate the optimal birth choice under the assumption that post
'I use the words "recipient" and "participant" interchangeably to refer to a person reporting AFDC
income in a given year. It is technically possible to participate in AFDC without receiving cash
payments simply to qualify for Medicaid or other services associated with the program, but these cases
are not discemable in the data set.
clevelandfed.org/research/workpaper/index.cfm
birth, the optimal participation rule will be followed. The effects of denying incremental benefits to
AFDC recipients with new births on their future childbearing behavior and continued participation are
simulated in. a natural way in this framework, as is the effect on total participation if benefits are
frozen for each family according to 1978 family size. In the concluding section, I summarize the
findings and place them in the context of previous work on fertility and AFDC policy.
11. Economic Models of the Family
Because female-headed households dominate the poierty population, an understanding of
marriage, birth, and work decisions is crucial for poverty policy. In this section, I discuss how AFDC
benefits can affect birth, maniage, and time allocation choice^.^
Children are assumed'to yield direct utility to parents, and children's consumption may also be
an argument in parents' utility functions. While current earnings are obviously affected, the primary
cost of children is thought to be forgone human capital development (and hence a lower and perhaps
flatter future wage profile) by the mother, who presumably devotes more time than the father to
childrearing, even if married.7 Work on the number and spacing of births focuses on the joint
determination of fertility and the path of labor market returns, usually holding marital status constant.
One of the first dynamic empirical treatments of this issue was by Moffitt.(l984), who looked at the
fertility and labor supply decisions of mamed women over lengthy periods and found support for this
-
hypothesized relationship between lifetime fertility patterns and wage profiles in the NLSW data.
The free or highly subsidized child care provided to working AFDC recipients has an ambiguous
effect on births. On one hand, it relieves the mother of the worklchild care trade-off, but at the same
time, it enables human capital investment, which may lead to reduced future births. The model
presented here could potentially be extended to incorporate this feature of policy as well.
7 ~ the
n Becker (1973, 1974) mamage model, the returns to specialization of the woman in home
work can be shown to be decreasing in .the ratio of the wife's to the husband's wages. Most women
presumably e m a lower wage than their husbands do.
clevelandfed.org/research/workpaper/index.cfm
In the absence of welfare, the ability of a woman to raise more children than her own income
can support comes from the option of marrying a man with higher earnings. Economic models of
maniage are commonly organized around the principle of comparing utility inside and outside of
maniage, following Becker (1973, 1974). Because the utility from a woman's own low prospective
labor market earnings is outweighed by the returns to specializing in home work and sharing her
husband's income, the standard model predicts stable family relationships for low-wage women, cet.
par., in the absence of welfare (e.g., Johnson and Skinner [1988]). Van der Klaauw (1993) estimates a
dynamic model of marriage and finds empirical support for these predictions. However, welfare may
provide a i acceptable alternative to maniage for low-wage women who do not want to sacrifice the
enjoyment of children. Not surprisingly, studies have shown that when their own and potential
husband's labor market prospects are poor, very young women tend to have children out of wedlock,
subsequently supporting the new family with AFDC benefits. Welfare is also predicted to raise the
probability of divorce for low-wage women by decreasing the returns to specialization in maniage and
raising the level of consumption (both own and children's) attainable alone. Finally, married women
who would otherwise choose to remain childless may "insure" against the income risk of divorce by
having a child, thus guaranteeing contingent AFDC eligibility. Single women who would -otherwise
remain childless may also insure against income risk in this way.*
Welfare policy has implications for the timing of births as well. Consider, for instance, the
effect of AFDC on an always-single woman. The option of welfare participation tends to flatten the
age-income profile by smoothing downside income fluctuations. Hence, if wage profiles take on a
traditiohal hump shape with respect to age, and if borrowing against future labor earnings is not
permitted, women can also afford to bear and raise children earlier in life in the presence of welfare.
&This paper examines only the second or subsequent birth decisions of unmanied women. Thus, I
do not address the insurance effect of AFDC on first births or on births to married women.
clevelandfed.org/research/workpaper/index.cfm
The AFDC program may encourage both married and unmarried women to space births farther apart in
order to extend wage- or divorce-contingent program eligibility.
There is one possible source of savings in program costs from encouraging (or failing to
discourage) births that the economic model described so far does not consider. Pregnancy potentially
provides an impetus to marriage for some men and women. In this regard, it could plausibly play a
positive role, since half of all program exits are accounted for by mamage (Hutchens [1981]).
111. Characteristics of Female Heads of Household
The weight of the empirical research on AFDC and family structure-provides only mixed
support for the notion that the program significantly affects childbirth decision^.^ However, public
opinion strongly favors the theory that AFDC policy has important and detrimental effects on the
family. In a New York TimeslCBS news poll conducted in May 1992, most respondents agreed that
the welfare system encourages people to have larger families. That attitude is obviously shared and
reinforced by many elected officials.
What is the basis for this opinion? Prima facie evidence from the NLSW confirms many of
the hypothesized effects of welfare on fertility, and this is one way the public might form its ideas
about the behavior of welfare mother^.'^ The NLSW is a panel data set that follows a group of
women between the ages of 14 and 24 in 1968. Information on AFDC
is collected in the
1978 and 1983 surveys. Table 1 presents some comparisons of the family characteristics of femaleheaded households both on and off welfare in 1978. The findings reveal that welfare mothers do have
See Moffitt (1992) and An, Haveman, and Wolfe (1993) for discussions of the literature.
I
hold family structure constant in the comparison. However, it is also possible that the general
public perceives female-headed households negatively, whether the family is on welfare or not: This
is a potential source.of additional stigma for families receiving aid.
clevelandfed.org/research/workpaper/index.cfm
significantly more children (34 percent on average) than nonrecipients and that they begin bearing
them at significantly younger ages (AFDC recipients had their first child at around age IS, while
-
. .
nonrecipients on average had their first child about a year later). The shares of family sizes between
the two groups are essentially reversed. While the same proportion of women in each group have two
children, 47 percent of AFDC recipients have three or more children, while 44 percent of nonrecipients
have only'one child. AFDC recipients are more likely to have never been married and are also less
likely to many in the five years following the observation. Studies using other samples suggest that
recipients tend to be younger women with younger children (e.g., Blank [1989]). This is not reflected
in the NLSW data (the fraction of women with children under age six is not significantly different
across the two groups) due to the age restriction on the sample.
The age spread between the youngest and oldest child is significantly higher for AFDC
recipients. This is consistent with the notion that women who might use welfare may space their
births farther apart to lengthen contingent eligibility. However, these figures need to be broken down
by family size, since the spread is definitionally increasing in the number of births. When this is
done, the only significant difference between the participating and nonparticipating groups is for
female heads with three children; in this case, the age spread for nonparticipating families is nearly
one year longer.
The final line of table 1 compares the guarantee across participating and nonparticipating
groups. The maximum benefit or guarantee is the2paymentto a zero-earning family of a given size. It
is the highest possible payment to the family, from which are deducted variables such as labor and
property income, child support, and alimony to anive at the final benefit payment. The mean values
are for a family size of three (one parent and two children). AFDC participants tend to live in states
with significantly higher guarantees. One explanation for this is that more generous benefits induce
clevelandfed.org/research/workpaper/index.cfm
greater participation." Holding family size constant, recipients live in states where benefits are on
average 12 percent higher. The returns to additional children do not vary according to participation
status and are as follows: 24 percent for the second child, 19 percent for the third, 15 percent for the
fourth, and 13 percent for the fifth.
Subsequent Births to 1978 Female Heads
.
.
For now, let us ignore any "insurance effect" of welfare benefits, which may manifest itself in
the nonparticipating portion of female-headed households. In this case, the effects of policy will be
evident if there are significant differences between the childbearing characteristics of participants and
nonparticipants. If welfare encourages careless contraception or the active creation of additional
children, one might expect to observe marked differences in the fertility patterns of AFDC participants
versus nonparticipants. Because the NLSW does not contain enough detailed participation data to hold
recipiency status constant over an extended period, I compare the subsequent childbirth experiences of
participants and nonparticipants as of 1978.
The first line of table 2 presents the fraction of each group bearing at least one child between
1978 and 1983 by 1978 participation status --I5 percent of participants and 18 percent of
nonparticipants. This difference is statistically insignificant, which might surprise those predisposed to
think that life on AFDC encourages childbearing, either through direct monetary rewards or by
indulging a careless lifestyle. However, it is still possible that of those women who do have children,
AFDC participants have more. Though the findings in the next line refute this, it could be that the
comp'bison is not yet specific enough. We know that AFDC mothers have more children to begin
" ~ alternative
n
hypothesis is that high-benefit states are "welfare magnets." Gramlich and Laren
(1984) find some support for significant but very small population movements in response to welfare
policy. To what extent this can explain the large differences in mean benefits noted here remains an
open issue.
clevelandfed.org/research/workpaper/index.cfm
with. Since the propensity to bear additional children might decline with family size, it may be
inappropriate to compare the subsequent birth patterns of women with different family sizes in 1978.
Lines 4-7 of table 2 reveal that even after adjusting for initial family size, the AFDC group is no more
likely to bear an additional child than the rest of the population of female heads. In fact, although the
differences are not statistically significant, the fraction of AFDC mothers giving birth is actually
smaller than that of nonparticipants at every given initial family size. This is an intriguing result in
light of the rhetoric surrounding welfare mothers and pregnancy. Finally, nonrecipients tend to be
older. Further age restrictions make the difference between the two groups more pronounced, although
it remains insignificant. It may be that older women have deliberately delayed conception and thus are
more likely to give birth over the next five years. Because there are many other characteristics for
which one should control, I shift to a regression framework below to investigate this phenomenon
further.
To do this, I estimate a probit model with a binary dependent variable that equals one if a
birth occurs within the next five years. The coefficient of interest is on a binary variable for AFDC
participation in 1978. It is significant and positive if participants tend to have more children, all else
equal. I maintain the number of children in 1978 and the age of the mother as explanatory variables
and also add income, education, race, and demographic characteristics thought to affect fertility. Table
3 summarizes the findings.
Twelve variables are included in the final specification. Eight are significantly different from
zero at the 10 percent level or less. The last column of the table presents the results of converting the
coefficients to percentage-point changes. As expected, the mother's age significantly reduces the
probability of an additional birth (by 2.3 percentage points per year), while the younger the mother is
at her first birth, the less likely she is to continue to bear children over the period of interest.
Significant variables having large effects on new-birth probability include race (whites are 9 percent
clevelandfed.org/research/workpaper/index.cfm
less likely to bear-additional children) and future marital status (those who many are 14 percent more
likely to bear additional children).'' The initial number of children also significantly influences the
probability of future births. Women with four or more children in the home in 1978 are the least
likely to add to their families, while those with two children are more likely to bear additional children
than women with only one child. The coefficients for the presence of young children, income, and
education are not significantly different from zero. Surprisingly, participation significantly reduces the
probability of a subsequent birth in this specification. All else equal,'participants are 6 percent less
likely to give birth over the next five years than n~nparticipants.'~
IV. An Ad Hoc Test for the Influence of Policy
while participants.appear to be no more likely than nonparticipants to bear children over the
subsequent five years, it is plausible that they would experience even fewer births if incremental
benefits were unavailable. As a preliminary test of this hypothesis, I estimate a probit model identical
to.the one above, except that 1) the sample is split by 1978 participation status, and 2) the maximum
benefit'to a zero-earning family (of appropriate size) and the increment to benefits that it receives if an
additional birth occurs enter as explanatory variables. Policy variables are predicted to have little or
no effect on the fertility behavior of 1978 nonparticipant^.'^ If children are a "normal good," one
would expect additional benefits to have a positive effect on births. Consequently, additional births
should be more likely in states that offer a higher per child increment.
'Vheory suggests that future marital status may be endogenous with births. This issue is ignored here.
1 3 ~ c(1993),
s
working with a group of 14 to 23 year-olds in the National Longitudinal Survey of
Youth--Young Women (NLSY), also finds that AFDC recipiency around the time of a first birth has
little effect on the likelihood of a second birth, suggesting that this result may be robust even for
teenage mothers.
'"This crude assumption is relaxed below.
10
clevelandfed.org/research/workpaper/index.cfm
Table 4 presents the findings for a sample of 134 participants in 1978.15 Age, race, mother's
age at first birth, and education are insignificant. Ultimate marital status and the initial number of
children continue to be significant, as are maximum benefits and the increment to new births. The
split between the effects of base and incremental benefits may be capturing nonlinearities in the birth
response, which would account for the negative coefficient on base benefits. The predictive power of
the model (presented in the frequency table and summarized by a pseudo-R') is far superior to what is
essentially the same model excluding the policy variables presented in table 3. The probit coefficients
suggest that an additional $10 in base benefits is associated with a 1-percentage-point lower probability
of subsequent births, while a $10 increase in the per child increment raises the probability of an
additional birth by 6 percentage points. This suggests that New Jersey could reduce births to AFDC
mothers by 25 percent if incremental' benefits were cut an average of $41.78 ($64 in nominal 1994
dollars).
However, there is reason to believe that this estimate is overstated. Family size, the presence
of children less than six years of age, and the AFDC guarantee have been shown to have a positive
influence on welfare participation across many studies and data sets. Consequently, new births may be
associated with higher incremental benefits, not because benefits cause new births, but because both
the addition to the family and the higher (final) benefits in a state make AFDC participation more
attrnctive for given levels of initial benefits and other factors. Even the extreme case of random births
may generate a psitive relationship between incremental benefits and births. All else equal, one
would still expect women facing higher birth increments to be more likely program participants,
simply because higher birth increments mean higher final family benefits, .yhich have a demonstrated
positive effect on participation.
"similar specifications applied to the 1978 nonparticipating group yield insignificant coefficients
on the policy variables.
clevelandfed.org/research/workpaper/index.cfm
The nested logit approach presented in the next section addresses these concerns in two ways.
First, the participation decision is incorporated into the estimation subsequent to the birth decision, so
that family size and children's age variables are included as explanatory variables in the participation
stage. Second, the birth increment may affect the participation decision only after a birth has
occurred, while having little impact on the birth decision itself. The sequential structure of the
estimation leaves open the possibility that this indirect effect of the guarantee increment on births is
minimal relative to other factors that both directly and indirectly affect births. Finally, the nested logit
specification allows for the possibility that benefits to 1983 participants influence the birth decision
even if they did not participate in 1978, and that benefits may influence the birth decisions of those
who did not participate ex ante or ex post.
V. A Sequential Choice Model
Taking initial status as a single female with one or more children as given in 1978, I estimate
the probability of an additional birth and, contingent upon whether the birth occurs, the probability of
participating in the AFDC program in 1983. Since policymakers' primary interest is in program
participants, it will be interesting to contrast estimates for initial participants and nonparticipants. If
those who view welfare as a "way of life" are most responsive to program rules, substantial reductions
in births and program costs could occur if additional benefits are denied to current (but not new)
participants.
The model estimates can be used to examine the effect of the per child increment on both the
decision to have an additional child and the decision to participate in AFDC contingent upon an
additional birth. Thus, I can address the primary question that seems to be on lawmakers' minds: Do
incremental benefits promote participant fertility'? I can also assess the impact of the incremental
benefit on total program participation (i.e., both continued and new).
clevelandfed.org/research/workpaper/index.cfm
Model
The nested logit model is governed by equations for the binary variables "have a child" and
"participate in. AFDC." Let "N" and "n"denote these choices. There are i = 1.2 choices for N (i.e.,
give or do not give birth to an additional child -- or children -- over the period 1978-83) and j = 1,2
choices for n (participate in 1983 or not). The sequential choices form the decision tree illustrated in
figure 2. The indirect utility from the final outcome (ij) of the decision process is specified as
where Xij contains variables specific to the (birth-contingent) participation decision, and Yi contains
variables that determine childbirth but not subsequent decisions. The random utility model makes
explicit the inability of agents to optimize perfectly, both because in a realistic setting their actions
cannot yield precisely the theoretically possible utility value, and because changed circumstances may
lead to changes in preferences in the future that are unpredictable a priori. However, the maintained
assumption is that consumers' underlying behavior is constrained optimization of perceived expected
utility. That is, the observed choice (iJ) corresponds to
The parameter estimates are obtained by maximum likelihood using the joint extreme
-
value distribution for the cij, which yields a probability for outcome (ij) of
and a conditional probability for participation choice j given birth outcome i of
clevelandfed.org/research/workpaper/index.cfm
For derivations of these probability expressions, see Maddala (1983), section 3.6.
Variables
'
The binary choice variables are whether a birth occurs over the 1978-83 period and whether
the agent is a 1983 AFDC participant. Figure 2 shows the sequence of decisions and the number of
observations on each decision. Previous estimates suggest that biological variables have an important
influence on fertility. State policy toward abortion and the availability and generosity of family
planning or prenatal services have also been cited as important determinants (e.g., Moore and Caldwell
[I9771 and Lundberg and Plotnick [1990]). The mother's current age, her age at first birth, the current
number of children, and the number of small children in the home are variables that reflect preferences
about family structure and that indicate the mother's biological ability to bear additional children.
State policies that affect births include the availability and cost to the mother of abortion and other
family planning services, the availability and generosity of'both Medicaid and private health insurance,
,
and the generosity of the Women, Infants, and Children (WIC) program. WIC supplements food
stamps and has been available to pregnant women in all states but Utah since 1976. Income and total
net wealth in 1978 are included in the birth decision because they indicate resources potentially
available for children's consumption (part of the effect of 1978 AFDC participation on births should
be transmitted through extremely low resources). Finally, race, religion, prior marital status, and
education reflect heterogeneity in childbearing behavior as well as awareness about contraception.
clevelandfed.org/research/workpaper/index.cfm
I follow the previous literatureI6 in assuming that the primary observable determinant of
welfare participation is the trade-off between income from welfare and income from the labor market:
The important variables are the benefit guarantee in the state, the wage that the feinale head can obtain
in the marketplace, and other income (such as alimony) that is heavily "taxed" by the AFDC program.
The AFDC guarantee varies by family size and takes on different values according to which branch of
the decision tree is chosen in the birth stage. While the hours worked choice is not explicitly
modeled, the presence of preschool children, who pose the most significant child care problem,
depends on the 1978-83 birth decision and is reflected in the relative importance of the number of
children across the participation branches. Variables thought to be influential at both levels of choice
are education and age (reflecting fertility and work experience),.number of children, race, and
Medicaid coverage. Variables thought to affect participation directly are child care policy, prior
experience with welfare, and the local unemployment rate.
Estimated Wages
Wages are an important indicator of the trade-off between welfare and work. NLSW
respondents were asked about wages in their current or previous job during the 1983 interview.
However, about one-third did not report wages because they had never worked or because they did not
respond to the question. In those instances where 1978 wages are reported but 1983 wages are not,
inflated wages from the earlier year are used. For the remaining observations that do not report wages
in either period, wage rates ace imputed from a standard human-capital wage equation.
'?or. example, see Moffitt (1983) and Blank (1985).
15
clevelandfed.org/research/workpaper/index.cfm
Empirical Findings
Estimates of the effects of personal characteristics and policy variables on births and
participation are obtained from the nested logit model. Mechanically, this amounts to estimating a
discietechoice model for 1983 participation (with the coefficients of the conditional choices
constrained to be equal), constructing an "inclusive value" from the fitted results, and estimating a
discrete choice model for 1978-83 births with the inclusive value as an additional explanatory variable.
Standard errors are corrected for the fact that the inclusive value is estimated." All dollar figures are
deflated using the Consumer Price Index.
The top panel of table 5 presents the coefficient estimates of the welfare participation problem
given the birth choice. Only four of the 10 included variables are significant at the 5 percent level or
more, but each has the anticipated effect on participation. Larger benefits and more children lead to a
greater likelihood of participation, while higher wages make work more attractive and reduce the
chances of participation. The local unemployment rate is significant at the 10 percent level and
increases the value of participation by reducing the return to being in the labor market. Coefficients
on age, non-labor income, education, race, and a constant are not significant in the discrete choice
model for participation.
Future marital status is an important determinant of participation status, since married women
are effectively removed from the prospective AFDC population. Ideally, one would like to incorporate
-
maniage as an endogenous choice, but the s m p l e is too small for this to be feasible. Instead, it enters
as a highly significant explanatory variable in the participation model. To investigate the possible bias
introduced in the coefficients by the inclusion of 1983 marital status, I reran the participation phase of
the model without this variable (the results are not reported). The only coefficient that changed
he multinomial logit model is obtained by restricting the coefficient on the inclusive value to
one.
clevelandfed.org/research/workpaper/index.cfm
significantly was for race. Nonwhite female heads are much more likely participants than whites if
future marital status is excluded, indicating that it is mostly whites exiting the eligible population
through marriage. All of the second-stage estimation is based on the specification with maniage.
The second part of table 5 reports the findings of the birth choice estimation. Policy variables
in the birth stage include the value of WIC benefits and family planning services, Medicaid
expenditures, and an index of state abortion policy. Because women who are in the AFDC program
.
-
may have more ready access to or may be more comfortable using other public programs, WIC
benefits and family planning services are also interacted with 1978 AFDC participation, although WIC
and family planning are not strictly governed by income and asset tests, as the larger programs are.
One would expect higher WIC benefits to increase births, while the availability of family planning
should increase the value of the "no birth" choice by subsidizing the effort and expense of
contraception. State-averaged Medicaid benefits for one woman and one child are interacted with
AFDC participation, since this is the primary method of access to Medicaid. Medicaid provides
abortion services and covers pre- and postnatal care, so the overall effect on births is ambigu~us.'~
However, I find that none of these policy variables has a significant influence on births.
In contrast to the findings of the simple birth probits in tables 3 and 4, the mother's age at
first birth, initial number of children, and prior marital status are insignificant in the nested logit
specification. Income and education are strongly significant, while they were not in the earlier singleequation specifications, suggesting that the nested logit model gives more credit to economic, rather
than biological, circumstances at the time of the birth decision. In fact, total resource variables (1978
income and total net wealth) are the most influential of all, implying that current AFDC participants
are less likely to have additional children because they are at the bonom of the income and wealth
181attempt to separate these effects by interacting the Medicaid variable with both the
restrictiveness and continuity variables for abortion policy. None of the interactions is significantly
different from zero. The findings are not reported.
clevelandfed.org/research/workpaper/index.cfm
distributions.
Consistent with earlier findings, the mother's current age significantly reduces the probability
of another birth. Nonwhites are more likely to give birth over the next five years. For 1978, AFDC
participation was entered directly in some specifications of the birth choice (not reported), but as
before, did not positively influence births. Finally, the inclusive value camed over from the
participation stage significantly and positively affects the birth decision and is significantly different
from one, supporting the nested over the multinomial logit specification. In combination with the
findings from the first stage, this implies that increased AFDC benefits result in a significantly higher
unconditional probability of a birth. I now proceed to investigate the magnitude of this effect.
Simulations
To be clear about the simulation exercise below, it is worth spelling out the role of AFDC
benefits in the model explicitly. Letting Uij denote the utility from the outcome of birth choice i and
participation choice j, we have
where No is the initial number of children, K is the change in the number of children between 1978
and 1983, G(N) is the AFDC guarantee for a family of one adult plus N children, E is autonomous
income, and X,j and Yi are as defined above. Thus, the first term, for example, specifies utility from
the decisions to have a child and participate in welfare as a function of the welfare guarantee for a
family of size "No + K" and the other choice-specific variables.
The specific policy changes in New Jersey and Wisconsin disallow incremental benefits for
births occurring while the mother is on AFDC. Presumably, those entering the program still face
clevelandfed.org/research/workpaper/index.cfm
guarantees that vary by family size. If G(No) denotes the maximum benefit to a family of size No, this
policy change can be simulated by setting G(No+K) = G(No) for all K in the lower branches of the
decision tree for current (1978) participants, but not for 1978 nonparticipants. From the new implied
probabilities ofparticipation, the number of 1978 participants changing their birth choice can be
inferred.
The proposed policy change affects participation rates in two ways. First, the probability of
participation conditional upon giving birth is reduced because welfare benefits are now lower,
dropping 1.8 percentage points on average and 2.5 percentage points for previous participants (see
table 6). Second, the probability of giving birth is indirectly reduced by the adverse policy change,
falling 1.5 percentage points on average and 2.3 percentage points for previous participants. However,
participation increases through another channel: While participation probabilities conditional upon no
birth are unaffected by the new policy, unconditional b!rth probabilities must rise.19 Therefore, the
joint probability of observing participation without a birth rises above that of the base case, up 0.3
percentage point for the entire sample and 0.7 percentage point for the previously participating
subgroup. The net effect of the policy change is that the 1983 participation rate drops by only a very
small amount: 0.4 percentage point for the entire sample (from 2 1.9 percent to 2 1.5 percent) and 0.7
for the subsample of 1978 participants (from 34.7 percent to 34.0 percent).
The above estimates can be combined with benefit information to provide some idea of the
total cost savings of denying incremental benefits to participants alone. Given actual 1983
participation data, the average monthly benefit cost is $322 per participant. For the subgroup who
participate in both 1978 and 1983, comprising 71 percent of the 1983 participant group, the average
cost is slightly higher ($340), accounting for 75 percent of total costs. If the policy change does
-'g~ntuitively,
many of those who are discouraged by the policy change from having a child will
nevertheless participate in AFDC.
clevelandfed.org/research/workpaper/index.cfm
nothing to discourage births among this group, the average cost of a continuing participant would be
reduced by $9.65 per month, or 2.8 percent."
For the group of 1978 participants, the total expected
participation rate drops from 34.7 to 34.0 percent, implying at most a 3.5 percent (= 2.8
+ 0.7)
cost
reduction for the previously participating group ($1 1.90 per person), or a 2.6 percent (= 0.75 x 3.5) .
reduction in total costs. Hence, although the empirical findings support a significant effect of policy
on births, it does not seem possible to generate large cost savings from the proposed policy change,
simply because the relative size of incremental benefits and the propensity of AFDC participants to
give birth are both quite small.
VI. Conclusion
Summary of Findings
This paper finds support for the notion that birth decisions respond to welfare program
incentives, but the magnitude of the response is modest. At least among mature (i.e., 24 to 34 year-.
olds in 1978) mothers, the potential cost savings of denying birth increments are small, both because
relatively few welfare mothers give birth (at least in this sample) and because although benefits
significantly and positively affect participation, birth increments are not of sufficient magnitude to
discourage participation by much. In fact, denying incremental benefits to AFDC recipients who bear
more children while on the program would save just $1 1.90 per month per continuing participant
under the 1983 benefit schedule, or 2.62 percent of their average payment. If the maximum benefit
were frozen for all female heads at the 1978 family size, total participation would be reduced by less
than half of 1 percent. In the remainder of this section,. I compare my findings with the related
literature.
'%is is less than the average per child increment because the cost savings are obtained only for
those actually giving birth, and very few 1978 participants give birth by 1983.
clevelandfed.org/research/workpaper/index.cfm
Related Literature
Acs (1993), using a sample of women between the ages of 14 and 16 in 1979 from the NLSY,
concludes that there is little support for the hypothesis that incremental benefits encourage births
among female heads of household. Acs examines first and second births occurring by age 23 using
the single-equation hazard approach of Plotnick (1990). The welfare participation decision is not
explicitly modeled, and the particular empirical specifications are similar to the model reported in table
4." Acs' policy variables are the maximum benefit for a family of two and the "AFDC gap" between
family sizes of two and three. His findings on the effect of policy variables are overall quite similar
to mine, but he dismisses significant results for some groups as an artifact of the omission of the
separate influence of children on participation. This paper goes further to demonstrate that benefit
policy has an independent and significant effect on births, but that findings from the single-equation
approach are grossly overstated due to a type of simultaneous-equations bias.
In contrast with the results of Lundberg and Plotnick (1990), I find little evidence that policies
such as WIC benefits, family planning, and state abortion laws influence fertility. However,
Lundberg and Plotnick's data set (the NLSY) allows them to implement a more specific test: They
have sufficient data to model conception and birth decisions separately. It is possible that realized
births are not sufficiently informative to test the effects of these policies. Lundberg and Plotnick
(1990) also examine younger women, who presumably have more to learn about family planning. It
may be that the mature women in the NLSW sample have little knowledge to gain from statesponsored programs, and hence these programs are of little relevance for their birth decisions.
In recent work, Murray (1994) revisits the basic time series evidence on welfare policy and
illegitimate births. He suggests that while the number of births per (black) woman has been falling
''~echnically, the primary difference is that Acs estimates a logit specification with corrections for
censored observations.
clevelandfed.org/research/workpaper/index.cfm
over the last 20 years, commensurate with the population as a whole, the share 'of illegitimate births in
all births has been following increases in welfare generosity with a two-year lag. The results of my analysis are typical of the type that frustrate Murray about cross-sectional studies: Policy effects are
found to be significant but minuscule. While the finding that single welfare mothers are no more
likely to bear additional children than their nonparticipating counterparts seems to ,support earlier
evidence that the illegitimacy rate has not been driven by welfare policy, plausible competing
hypotheses are that the sample is from a period when the "culture of poverty" had seeped into the nonwelfare-participating groups, or that female-headed households are culturally quite similar, regardless
of whether they participate in welfare.
clevelandfed.org/research/workpaper/index.cfm
Appendix A: Policy Variables
Policy variables are culled from various sources, including Bush (1983), Gold (1982), Sollom
(1994), Torres, Forrest, and Eisman (1981), Torres and Forrest (1983). and several government
agencies.
ABLAW: An overall score of the restrictiveness of state law with regard to abortions. A point is
added if 1) abortions became legal only after 1969, 2) parental consent or notification is required, or
3) second-trimester abortions must be performed in a hospital. A higher score reflects a more
restrictive policy, which may discourage women from seeking abortion services.
HYDE80: The Hyde amendment, passed in 1977, virtually eliminated the federal role in providing
subsidized abortion services for women on Medicaid. During 1980, the amendment was temporarily
suspended by court order. The dummy variable HYDE80 is zero if states continued to provide
funding for Medicaid abortions when the amendment was in force during 1980-81. This variable
should capture both the acceptability of abortion in the state (i.e., a willingness to continue to provide
the same level of service offered by the federal government before 1977) and any strong
discontinuities in abortion funding over the periods before, during, and after the amendment's
suspension.
FP79: Title IX provides federal funds for family planning. While low-income women are its primary
target, anyone can receive services. The expected value of family planning services is defined 9 the
percentage of "at-risk" low-income women served by family planning services in a state times an
estimate of per patient expenditure. An at-risk woman is sexually active. Total expenditures are
clevelandfed.org/research/workpaper/index.cfm
divided by the sum of at-risk low-income women plus teens served. The denominator excludes
higher-income women served, which may lead to a. slight overstatement of per patient expenditure.
(Note that since funds come from federal sources, state-level variation arises from different
probabilities of participation, which may in part reflect the state's ability to distribute aid efficiently.)
WIC78: The Women, Infants, and Children's program provides food to pregnant and nursing mothers.
Monthly data on participation and food expenditures by state are averaged over the year. Average
monthly food expenditures are divided by the average monthly number of participants to arrive at a
per-recipient food expenditure amount. Data are from the U.S. Department of Agriculture (1979).
MEDIC78 and MEDIC83: variables on statewide Medicaid expenditures per AFDC adult and child
are from the Joint Tax Committee "Green Book" (various editions). These are combined to yield
expected values of Medicaid for AFDC families of various sizes.
clevelandfed.org/research/workpaper/index.cfm
References
Acs, G., "The Impact of AFDC on Young Women's Childbearing Decisions," Working Paper, The
Urban Institute, March 1993.
An, C.,Haveman, R., and Wolfe, B., "Teen Out-of-Wedlock Births and W.elfare Receipt: The Role of
Childhood Events and Economic Circumstances," Review of Economics and Statistics 75(2),
1993, 195-208.
Becker, G.S., "A Theory of Maniage: Part I," Journal of Political Economy 81(4), -1973, 813-846.
, "A Theory of Marriage: Part 11," Journal of Political Economy 82(2, Part 11), 1974,
5 11-526.
Blank, R.M., "The Impact of State Economic Differentials on Household Welfare and Labor Force
Behavior," Journal of Public Economics 28, 1985, 25-58.
, "The Effect of Medical Need and Medicaid on AFDC Participation," Journal of Human
Resources 24, 1989, 25-58.
Bush, D., "Fertility-Related State Laws Enacted in 1982," Family Planning Perspectives 15(3),
1983, 111- 116.
Gold, R., "Publicly Funded Abortions in FY 1980 and FY 1981," Family Planning
Perspectives 14(4), 1982, 204-207.
Gramlich, E.M., and Laren, D.S., "Migration and Income Redistribution Responsibilities,"
Journal of Human Resources 19, 1984, 489-5 11.
Hutchens, R., "Entry and Exit Transitions in a Government Transfer Program: The Case of AFDC,"
Journal of Human Resources 16, 1981, 217-237.
'
Johnson, W.R., and Skinner, J., "Labor Supply and Marital Separation," American Economic Review
76(3), 1986, 455-469.
Lundberg, S., and Plotnick, R., "Effects of State Welfare, Abortion, and Family Planning Policies on
Premarital Childbearing among White Adolescents," Family Planning Perspectives 22(6),
1990, 246-25 1.
Maddala, G.S., Limited-Dependent and Qualitative Variables in Economics, Econometric Society
Monograph No. 3. Cambridge, England: Cambridge University Press, 1983.
Moffitt, R., "An Economic Model of Welfare Stigma," American Economic Review 73(5), 1983, 10231035.
, "Profiles of Fertility, Labour Supply, and Wages of Mamed Women: A Complete
Life-Cycle Model," Review of Economic Studies 51, 1984, 263-278.
clevelandfed.org/research/workpaper/index.cfm
, "Has State Redistributive Policy Grown More Conservative?" National Tax Journal 43(2),
1990, 123-142.
, "Incentive Effects of .the U.S. Welfare System: A Review," Journal of Economic
Literature 30, 1992, 1-6 1 .
Moore, K., and Caldwell, S., "The Effect of Government Policies on Out-of-Wedlock Sex and
Pregnancy," Family Planning Perspectives 9(4), 1977, 164-169.
Murray, C.;"Does Welfare Bring More Babies'?" The American Enterprise 5(1), January/
February, 1994, 52-59.
Plotnick, R., "Welfare and Out-of-Wedlock Childbearing: Evidence from the 1980s."Journal
of Marriage and the Family 52, 1990, 735-746.
Sollom, T., unpublished data, The Alan Guttrnacher Institute, 1994.
Torres, A, and Forrest, J.D., "Family Planning Clinic Services in the United States, 1981," Family
Planning Perspectives 15(6), 1983, 272-278.
Torres, A., Forrest, J.D., and Eisman, S., "Family Planning Services in the United States, 1978-1979,"
Family Planning Perspectives 13(3), 198 1, 132- 14 1 .
U.S. Congress, House Committee on Ways and Means, Overview of Entitlement Programs:
Background Material and Data on Programs within the Jurisdiction of the Committee
on Ways and Means ("Green Book"). Washington, D.C.: U.S. Government Printing Office,
1983, 1985, 1993.
U.S. Department of Agriculture, Food and Nutrition Service, Special Supplemental Food
Program for Women, Infants, and Children. Washington, D.C.: FNS Management Information
Division, 1979.
Van der Klaauw, W., "Female Labor Supply and Marital Status Decisions: A Life Cycle Model,"
Working Paper No. RR#93-23, New York University, May 1993.
Wall Street Journal, "The $64 Question," Editorial, March 28, 1994.
clevelandfed.org/research/workpaper/index.cfm
clevelandfed.org/research/workpaper/index.cfm
clevelandfed.org/research/workpaper/index.cfm
Table 1: Characteristics o f Female-Headed Households by AFDC Participation Status, 1978
b
a
c
Participant
Nonparticipant
t-statistic
2.62
1.95
Fraction with one child
0.24
0.44
Fraction with two children
0.30
0.32
0.52
Fraction with h e children
0.24
0.13
3.57
Fraction with four+ children
0.23
0.11
4.15
28.46
28.74 (N=435)
1.14
0.40
0.09
9.94
267.62
238.78
2.53
4
Number of children
6.74
)
5.08
4
4
4
Mother's age
hlother's age at first birth
Fractionwith children under six
Age of youngest child
Spread between oldest and youngest child (years)
SpreaM children
Spreadn children
Spreaa4 children
Spread4 or more children
Pcmenl of mothera who are while
Fraction never married
Fraction marrying, 1978-83
AFDC participant. 1983
Maximum benefit. two children
a N=217 except where otherwise noted
b N=442 except where otherwise noted
o 1-test for equality of mean, of samples are drawn from two populations with the same variance,
Exocedr t for difference at 5 percent d ~ d e n c level.
e
Souroc: Author's comp~itationsfrom the NLSW.
t . m ,= 1.96.
4
clevelandfed.org/research/workpaper/index.cfm
clevelandfed.org/research/workpaper/index.cfm
Table 3: Probit Estimates of Births Using AFDC Participation as an Explanatory Variablea
Variable
Coefficient
Probability Change
Mean (x)
Std. Dev. (x)
8.25E-02
(2.13)
1.77E-02
18.823
2.737
Children 1.t. 6'
Married by 1983C
Mother's age at f ~ sbirth
t
1978 income
Education:
Not high school graduateC
High school graduateC
Children:
One child, 1978'
Two children, 1978c
Four children, 1978C
AFM= participant, 1978'
a
Sample of 640 women who are heads of household in 1978.
b t 025 a, - 1.96. t 05,m = 1.645.
Bin& variable equal to one if statement is true for household.
Source: Author's computations fiom the NLSW.
*
clevelandfed.org/research/workpaper/index.cfm
Table 4: Probit Estimates of Births Using AFDC Benefits as an Explanatory Variablea
Variable
Coefficient
Probability Change
Mean (x)
Std. Dev. (x)
3.74E-02
(2.57)
6.09E-03
56.71
22.08
WhiteC
Children 1.t. 6'
Married by 1983'
Mother's age at first birth
1978 income
Education:
Not high school graduateC
High school graduateC
Children:
One child, 197SC
Two children, 1 97SC
Four children, 197SC
AFDC benefit, 1978
Incremental benefit, 1978
a Sample of1.96,
134 state-matched, AFDC-participating female household heads in 1978.
1.645.
b tQ2sa =
tos,a
=
B I ~ &variable equal to one if statement Is true for household.
source:-~uthor'scomputations fiom the NLSW.
clevelandfed.org/research/workpaper/index.cfm
Table 5: Nested Logit Estimates
Coefficient
Participation
Constant a
Age a
AFDC benefit a
Future marriage
Nonwage income
Education (years)
Unemployment
White
Number of children a
(continued)
clevelandfed.org/research/workpaper/index.cfm
Table .5: Continued
.
Births
Constant
Prenatal C.d
Age
Education (years)d
Income
Net wealth
Age @ first birth
Number of children
Number of preschool children
Never married
MedicaidZparticipant
White
Abortion law index
Inclusive value C*d
a Variable affects participation.
b Variable affects nonparticipation.
c Variable affects birth.
d Variable affects no birth.
e Family planning and WIC are combined into a single variable.
Source: Author's computations&om the NLSW.
clevelandfed.org/research/workpaper/index.cfm
Table 6: Policy Simulations
Sample
Subsample of 1978 AFDC Participants
Status Quo
New Policy: No Incremental Benefits for New Births
0.170
=
0.208 * 0.819
a B=bi.
b P=participant.
c NB=no birth.
Source: Author's computations fiom the NLSW.
0.272
=
0.332
*
0.822
@FedFRASER