Within-mother estimates of the effects of WIC on birth outcomes in New York City.
Currie, Janet ; Rajani, Ishita
I. INTRODUCTION
There is growing acceptance of the idea that prenatal conditions
can have health effects not only at the time of the birth, but also
lasting into adulthood (Almond and Currie 2011). This observation leads
people to question whether there are "shovel ready" policies
that can be implemented to improve children's life chances? The
U.S. Supplemental Nutrition Program for Women, Infants, and Children
(called WIC) would appear to be ideally suited to play such a role. The
program provides nutritious food, counseling, and assistance with access
to medical care to pregnant women, as well as to their infants and young
children. Bitler, Currie, and Scholz (2003) show that the program serves
a large fraction of pregnant women in the United States. In fiscal year
2011, 9 million participants were served each month including 2.1
million pregnant or lactating women which can be compared with the
roughly 4 million births in that year (USDA 2012).
Yet, despite decades of research suggesting that "WIC
works" in the sense that women who use WIC during pregnancy have
healthier babies than other similar women, the suspicion persists that
the observed positive association between WIC participation and outcomes
might not be causal. (1) There are two limitations to most WIC research.
(2) First, as women choose whether or not to participate in WIC and when
to begin participating, it is possible that women who seek out WIC
benefits are different from other observationally similar women in some
unobservable way that is conducive to better birth outcomes. Despite
this concern however, there has been little direct examination of the
determinants of selection into the program.
Second, the longer a woman is pregnant, the longer she has to
enroll in WIC. Hence, a positive correlation between WIC and length of
gestation could reflect this "longer window of opportunity"
effect. Moreover, as the fetus gains most of its weight in the third
trimester, if women with longer gestations are more likely to sign up
for WIC, then we would also expect their babies to have higher birth
weights.
This article addresses these limitations of the previous
literature. Starting with a sample of all New York City births from 1994
to 2004, we focus on women with more than one singleton birth over the
sample period and estimate models with mother fixed effects. In these
models, the effect of WIC is identified by comparing births to the same
mother, with and without WIC. This specification controls for the fact
that women who use WIC may be different in some unobservable fixed
respect than women who do not. It is still possible however, that
time-varying characteristics of mothers are associated with both WIC
participation and positive birth outcomes. Hence, we also use our data
to directly investigate this type of selection and ask when mothers are
most likely to use WIC in terms of these time-varying characteristics.
In order to deal with the possibility of a mechanical correlation
between gestation length and probability of WIC enrollment, we focus
much of our attention on full-term births as well as asking whether
infants were "small for dates" (i.e., below the 10th
percentile of national weight standards given gestational age). We also
ask whether the effects of WIC are bigger for first born children. Women
pregnant with their first born child may lack information about healthy
practices and may not know that they are at risk for poor birth
outcomes, so it is possible that WIC would have particularly large
effects in this group.
We are able to document important facts about selection into the
WIC program--mothers who change WIC status between births are more
likely to use WIC when they are younger, unemployed, and/or unmarried.
Women are also much more likely to be referred to WIC if they have had a
previous poor pregnancy outcome. However, mothers are also more likely
to be diagnosed with chronic hypertension or chronic diabetes when they
are on WIC suggesting that WIC may improve access to medical care and/or
screening for these conditions.
We find that WIC is associated with reduction in low birth weight,
even among full-term infants, and with reductions in the probability
that a child is "small for dates." These improvements are
associated with a reduction in the probability that the mother gained
too little weight during pregnancy. Effects on low birth weight are
concentrated in first born infants, although effects on being
"small for dates" and weight gain are present for all WIC
mothers. These findings suggest that WIC counseling may be particularly
useful for first time mothers. We also find that women on WIC and their
infants receive more intensive medical services, a finding that may
reflect improved access to medical care.
II. BACKGROUND
This section does not attempt to review all of the voluminous
literature on WIC (see Currie, 2003 for a review of the older
literature). Instead, we first provide an overview of the program, and
then focus on studies that have attempted to address some or all of the
empirical problems outlined above, and discuss how our study builds on
and adds to these efforts.
WIC is a supplemental feeding program for pregnant and lactating
women, infants, and children up to age 5. In addition to belonging to
one of these groups, a WIC participant must either have income less than
185% of the federal poverty line, or must have Medicaid coverage
(available to pregnant women with incomes less than 200% of the federal
poverty line. In New York, the threshold was about $38,000 for a family
of three in 2012). Women must also be evaluated and deemed to be
"at nutritional risk," though in practice this requirement
does not appear to be very binding (Bitler, Currie, and Scholz 2003).
WIC clients receive coupons or a debit card which can be used to
purchase only specifically selected nutritious food items. (3) In 2012,
New York WIC provided food benefits worth an average of $55 per
participant per month. (4) To maintain their benefits, clients are
required to attend nutritional counseling and breastfeeding education
sessions at regular intervals. One of the open questions about WIC is
why it works (if indeed it does) given that the benefits are small
relative to people's monthly food budgets. Possible answers are
that the healthy WIC foods available or the nutritional counseling
"nudge" people's diets in a healthier direction, or that
the availability of WIC benefits helps to facilitate access to medical
care.
As discussed above, most studies of WIC have difficulty dealing
with possible nonrandom selection of women into WIC benefits. If women
who enroll in WIC are, for example, more concerned about their
child's health, or more motivated to change their behaviors than
other women, then the measured positive effects of WIC could reflect
these unobserved characteristics of mothers who enroll in the program.
In view of this concern, it is surprising that few WIC studies have
estimated models with maternal fixed effects because these models offer
a way to control for all observed and unobserved fixed characteristics
of mothers. (5) The study closest in spirit to ours is Kowaleski-Jones
and Duncan (2002) who use data from the Children of the National
Longitudinal Survey of Youth and estimate mother fixed effects models.
Unfortunately, they had only 104 sibling pairs, and only 71 with
discordant WIC participation. Nevertheless, they found a statistically
significant effect on birth weight (of 7 ounces) and a positive effect
on temperament scores. They did not find a statistically significant
effect on measures of motor and social skills.
Rossin-Slater (2013) estimates maternal fixed effects models in a
study of the effects of the closure of WIC clinics in Texas. She finds
that having a WIC clinic in one's zip code increases maternal
weight gain and birth weight, but is also associated with more diagnoses
of diabetes and hypertension during pregnancy. She speculates that this
latter result might reflect better access to medical care, which in turn
leads to higher diagnosis rates.
Joyce, Gibson, and Colrnan (2005) also examine birth records for
New York City. In order to deal with selection, they focus on relatively
narrowly defined groups, for example, Hispanic mothers with a first
pregnancy, and use matching models with propensity scores. In order to
deal with the mechanical correlation between gestation and the window
available for WIC enrollment (i.e., the fact that longer pregnancies
both allow more time to enroll and also generally result in healthier
infants), they focus on birth weight conditional on gestational age.
Joyce, Racine, and Yunzal-Butler (2008) examine data from nine states
and ask whether earlier initiation
of WIC is associated with better maternal behaviors and better
birth outcomes. Both papers argue that a positive association between
WIC and gestation is suspect for the reasons given above. Both of these
papers do however find positive effects of WIC on infant outcomes,
though they argue that the effects are smaller than those found in the
previous literature.
Bitler and Currie (2005) and Figlio, Hamersma, and Roth (2009) also
attempt to deal with selection by narrowing the comparison group. Bitler
and Currie focus on a large national sample of mothers whose births were
paid for by Medicaid (so that all of the mothers were adjunctively
eligible for WIC). Figlio et al. use administrative data from Florida
and focus on families in a narrow income band around the WIC eligibility
threshold. They also use a tightening of WIC income documentation
requirements to try to identify the effect of WIC receipt. Using this
design, they find a strong effect of WIC on the incidence of low birth
weight (birth weight less than 2,500 g) even though, as they point out,
the marginal mother eligible for WIC may benefit less than inframarginal
recipients.
Hoynes, Page, and Stevens (2011) focus on the roll out of the WIC
program in the late 1970s in a difference-in-differences framework and
show that in counties that adopted WIC, the incidence of lower birth
weight fell. However, arguably the factors affecting the nutritional
status of pregnant women could be quite different today than they were
in the 1970s, given sharp declines in the relative price of food over
time, as well as large increases in obesity.
This brief review indicates that our study is one of very few to
examine the effects of WIC in models that control for mother fixed
effects. It is also the first large-scale analysis of selection into
WIC, and the first to examine effects of WIC by birth order. The
exploration of within-mother selection into the WIC program and of the
effects of WIC on firstborns is of independent interest and sheds light
on the way that "WIC works."
III. METHODS
In order to investigate the way that timevarying maternal
characteristics are associated with participation in the WIC program, we
estimate models of the form:
(1) [Z.sub.it] = [a.sub.i] + [a.sub.1] WIC + [a.sub.2] [X.sub.it] +
[a.sub.3]Month + [a.sub.4]Year + [a.sub.5]Neighborhood + [e.sub.it],
where [Z.sub.it] is one of a vector of maternal characteristics
including prepregnancy weight, whether she has chronic diabetes, whether
she has chronic hypertension, age, whether she smoked at all during the
pregnancy, whether she was employed, whether she was married, and
whether the mother had a previous poor pregnancy outcome (the later
outcome can only be examined in mothers who had a previous birth that is
included in our data set). The vector [a.sub.i] represents a fixed
effect for each mother. The vector [X.sub.it] includes time-varying
characteristics of the mother and child which might be associated with
selection into the program including child gender, maternal education
(<12, 12, 13-15, 16+, missing), parity (1, 2, 3, 4+, missing), and
maternal age (<20, 20-24, 25-29, 30-34, 35+, missing). (Note that
these age dummies are omitted when the dependent variable in question is
maternal age.) The coefficient of interest is a,, which measures the
correlation between changes in the time-varying dependent variables and
changes in WIC status between pregnancies.
We also include dummy variables for each month and year of birth in
order to control for time trends in these variables, and for maternal
neighborhoods within New York City, because neighborhood is an important
indicator of economic and social status. (6) Finally, [e.sub.it]
represents a random error term.
Our object in estimating Equation (1) is to determine whether
time-varying characteristics of mothers are systematically related to
WIC enrollment. If we find that a mother is more likely to be on WIC
when she has a dangerous chronic condition that complicates her
pregnancy, then we should take this information into account when
interpreting the within-mother estimates of the effects of WIC
participation on birth outcomes. For example, chronic diabetes is
associated both with a higher risk of a small and preterm baby, and with
a higher risk of a very large baby with a complicated delivery. Children
of diabetic mothers also have a higher incidence of congenital anomalies
and so may be at higher risk of negative developmental outcomes (Nold
and Georgieff 2004). If mothers diagnosed with diabetes were more likely
to be referred to WIC, then this type of selection could bias the
estimated effect of WIC on birth outcomes.
Turning to outcomes of pregnancy that we might expect to be
affected by the WIC program, we estimate models of the form:
(2)
[Y.sub.it] = [a.sub.i] + [a.sub.2] [WIC.sub.it] +
[a.sub.3][Z'.sub.it] + [a.sub.4][X.sub.it] + [a.sub.5] Month +
[a.sub.6] Year + [a.sub.7] Neighborhood + [e.sub.it],
where [Y.sub.it] is a vector of outcomes including whether the baby
is low birth weight, whether the baby is small for dates, whether the
baby had a low APGAR (Appearance, Pulse, Grimace, Activity, Respiration)
score, whether the baby was preterm (gestation less than 38 weeks),
whether the baby was admitted to the neonatal intensive care unit
(NICU), and whether there were complications of labor and delivery; WIC
is an indicator for whether the mother used WIC during the pregnancy,
and Z' includes the following time-varying characteristics of the
mother: prepregnancy weight, whether she has chronic diabetes, whether
she has chronic hypertension, whether she smoked at all during the
pregnancy, whether she was employed, whether she was married. The other
variables are defined as in Equation (1) except that [X.sub.it] includes
one additional variable, which is whether the information about the
father is missing from the birth certificate. This variable can be
viewed as an indicator of father involvement.
Finally, we also estimate models that allow the effects of WIC to
differ for first born children. There are two reasons why this exercise
is of interest. One is that first born children may be more affected by
WIC because their mothers do not know what to expect and so the guidance
offered by WIC might be of particular value. Second, as we show below,
mothers who have had a previous poor pregnancy outcome (such as an
infant death) are more likely to use WIC in the current pregnancy.
Failure to control for this negative selection into WIC could result in
downward bias in the estimated effect on infant health; therefore, it is
of interest to compare the effects of WIC on firstborns with the effects
of WIC for subsequent pregnancies. These models take the form:
(3) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]
where Firstborn is an indicator equal to one if the child is the
mother's first born, and all other variables are defined as
described above. In this regression, [a.sub.2] is the estimated effect
of WIC on all children, and a significant value for [a.sub.3] indicates
that the effect is different for firstborns.
We conduct our analyses for all mothers, and for subsamples of
mothers defined by Medicaid coverage, education less than 12 years, and
by race. We examine these subsamples largely for the sake of
comparability with previous research. Some previous analyses have found
larger effects of WIC among African American mothers. As discussed
above, several previous analyses have confined their attention to
mothers with Medicaid coverage. Finally, one might think that the
counseling aspect of WIC could be particularly valuable to less educated
mothers, so we also examine this group separately.
IV. DATA
The data for this study come from the New York City Department of
Health and Mental Hygiene's Vital Statistics Natality system for
collecting and recording information from the certificate of live birth.
Data for these certificates come from two worksheets. One is completed
by the mother and asks information about her circumstances and behaviors
(such as marital status, smoking during pregnancy, and prepregnancy
weight). The other worksheet is completed by the medical facility where
the birth takes place using medical records. This worksheet includes
information about prenatal care visits, risk factors for the pregnancy,
complications of labor and delivery, and newborn health. We start with
all live singleton births in New York City between 1994 and 2004, which
yields approximately 1.2 million records.
Table 1 provides an overview of our data. In the full sample, 42%
of the births were to mothers who used WIC at some point during
pregnancy. The second column of Table 1 shows means for those children
with a sibling in the sample, roughly 45% of the sample births. The
remaining two columns focus on this subsample of sibling births, and
divide them into births in which the mother used WIC and births in which
she did not. Focusing on these last two columns, we can see that the
incidence of negative outcomes is generally higher for the WIC infants:
They are more likely to be low birth weight, small for dates, to have
low APGAR scores (a measure of the health of the infant at birth that is
taken out of 10), and to be premature. (7) Strikingly, the WIC infants
are almost twice as likely as the non-WIC infants to end up in the NICU.
Of course much of the gap between the WIC infants and the non-WIC
infants is likely to be due to differing maternal characteristics. Table
1 indicates that mothers of the WIC infants were much less educated (a
mean difference of more than 2 years), younger, half as likely to be
married, twice as likely to be missing the father's information on
the birth certificate (an indicator of paternal involvement), and more
likely to smoke. They also have higher incidence of chronic diabetes and
chronic hypertension, are less likely to receive prenatal care in the
first trimester of pregnancy, and are almost entirely dependent on
Medicaid coverage. The WIC mothers are a little heavier on average
before the pregnancy but less likely to have gained an adequate amount
of weight during pregnancy. (8) They are also more likely to have
complications of labor and delivery. Finally WIC mothers who have
already had one child are more likely to have had a poor outcome in a
previous pregnancy. We define a poor outcome to include: A previous
infant death (very rare), a prior child admitted to the NICU, a previous
infant who was premature, or a previous low birth weight infant. These
stark differences between the circumstances of WIC infants and non-WIC
infants illustrate the importance of adequately controlling for the
characteristics of WIC mothers when evaluating the effects of the
program on outcomes.
V. RESULTS
Table 2 examines the relationship between maternal characteristics
and participation in the WIC program using mother fixed effects models.
Therefore, these estimates rely on the sample of mothers who used WIC
for one pregnancy but did not use it for another to tell us when they
were relatively more likely to use WIC. The table shows estimates for
all births, full-term births only, and for different groups of mothers
defined by insurance status, race, and education.
The first three columns of Table 2 show that mothers tend to use
WIC when they are not employed, when they are unmarried, and when they
are younger. These differences exist for all the groups we examine,
except that there is no difference in age among less educated mothers.
All but African American mothers have lower prepregnancy weights when
they go on WIC, whereas African American mothers, who are the highest
mean weight group, are more likely to use WIC when they are relatively
heavy. Together these results suggest a scenario in which mothers use
WIC (and smoke) when they are relatively young, unemployed, and
unmarried. As they grow older they may be more likely to quit smoking,
become employed, and/or marry and they become less likely to use WIC.
It is surprising then that we also find that women are more likely
to have chronic diabetes and chronic hypertension when they are on WIC
because these are conditions that increase in prevalence with age. An
examination of the subgroups shows that this relationship is strongest
among African American women. We think that, consistent with
Rossin-Slater's interpretation of her results from Texas, women on
WIC may be more likely to be properly diagnosed and treated if they have
chronic diabetes or hypertension, so that these conditions are more
likely to be subsequently recorded on the birth certificate. Consistent
with this interpretation, Table 2 also shows that women on WIC are more
likely to get prenatal care in the first trimester. (9)
The last column of Table 2 shows that in the subset of women who
have already had a birth that we can observe, women who had a previous
poor pregnancy outcome are much more likely to be on WIC. This finding
could reflect women being more motivated to try to improve their
pregnancy outcomes, or a pattern of referring women with poor prior
outcomes to WIC.
Changes in these time-varying maternal characteristics are likely
to bias estimated WIC effects if they are not controlled. The fact that
women use WIC when they are relatively disadvantaged (young, unmarried,
and unemployed) would be expected to bias estimated effects of WIC
downward in the absence of controls for these variables. Similarly, if
women who know that they are at risk of future poor outcomes are more
likely to use WIC, then estimates that do not take this selection into
account would understate the benefits of WIC. If women on WIC are really
more likely to have chronic diabetes and/or hypertension, then this
would also tend to be reflected in poorer birth outcomes. However, as
discussed above, it is possible that this pattern reflects differential
reporting of chronic conditions for women on WIC rather than an actual
increased prevalence of these conditions.
Table 3 shows estimated effects of WIC on infant outcomes. The
estimates in this table and in Table 4 are based on Equation (2) and
include observable time-varying characteristics associated with
selection into the WIC program: maternal age, employment, marital
status, smoking during pregnancy, prepregnancy weight, (10) chronic
diabetes, chronic hypertension, and whether women received prenatal care
in the first trimester. We have also estimated these models excluding
the prenatal care variable, and find that this exclusion has little
effect on the estimates.
As we only observe previous poor pregnancy outcomes for a subset of
women in our data set, we have not controlled for that variable here.
Below, we show estimates which allow the effect of WIC to vary for first
born children and other children. These estimates will shed light on the
potential importance of any information that mothers may have gained
from previous pregnancies, because no mother has such information for
first births.
Column 1 in the first panel suggests that consistent with the vast
majority of WIC studies, WIC participation is estimated to have a
negative effect on the incidence of birth weight less than 2,500 g: The
coefficient in the first row and column of .0041 can be compared with
the fraction low birth weight which is also given in the table and thus
implies a 6.12% reduction in the overall probability of low birth
weight. However, it is possible that some of the estimated
"effect" is driven by the mechanical correlation between
gestation and the probability of enrollment in WIC (due to having a
longer time period in which to enroll) that was discussed earlier.
Column 2 suggests that WIC has a large effect on prematurity, reducing
it by 10.05% which again, may reflect this mechanical correlation.
Hence, column 3 shows estimates of whether the baby was "small for
dates." Conditioning on gestation length in this way suggests that
WIC reduces the overall probability of a small for dates baby by 2.92%.
This smaller effect suggests that some of the estimated WIC effect in
low birth weight and prematurity is indeed due to the mechanical
relationships between gestation length, a longer window in which to
enroll in WIC, and positive birth outcomes. However, the significant
effect on whether the baby is small for dates indicates that WIC has a
significant effect even conditional on gestation. The fourth column
provides some evidence regarding potential mechanisms, suggesting that
the probability of low weight gain during pregnancy (defined as weight
gain less than 6 kg) is significantly reduced by participation in WIC,
falling by 9.18%.
Columns 5 and 6 examine complications of labor and delivery, and
the probability that the infant was admitted to the NICU, respectively.
Surprisingly, the estimated effect of WIC is significantly positive in
both cases, indicating that receiving WIC prenatally increases the
probability of these outcomes. Because not even the harshest critics of
WIC have suggested that it actually harms women and children, it appears
likely that these estimates reflect improved access to medical care
during and after the birth among WIC recipients rather than a causal
effect of prenatal nutritional assistance or counseling.
The second panel of Table 3 shows estimates for mothers who were
covered by Medicaid in both pregnancies. The estimated effects of WIC
are almost identical to those in the full sample of mothers for all the
outcomes examined. Hence, it appears that although Medicaid participants
are adjunctively eligible for WIC, WIC participation is not simply a
proxy for Medicaid coverage. Hence any improvements in access to medical
care that are associated with WIC are not coming solely through the
provision of health insurance, but through some other channel. The
estimates for mothers with less than 12 years of education are very
similar to those for Medicaid covered mothers.
Turning to the estimates by race, the effect on "small for
dates," which is the most reliable measure of infant health in this
table, is about one-third larger in terms of the point estimate for
African Americans, but is not precisely estimated given the smaller
subsample. The effect of WIC on the probability that the mother gained
too little weight is also negative and about a third larger than for the
sample as a whole. This estimate is statistically significant. Thus, the
table provides some support for the hypothesis that the effects of WIC
are larger for African American women than for other groups.
Aside from the regressions using "small for dates" as the
dependent variable, the estimates in Table 3 are potentially colored by
the mechanical correlation between gestation and WIC enrollment that has
been discussed. Hence, Table 4 repeats the same analyses on the subset
of infants who are full term, that is, born after 37 weeks gestation.
Perhaps surprisingly, these estimates are similar to those in Table 3,
and in some cases are larger. For example, in this sample WIC is
estimated to reduce the probability that an infant is small for dates by
4.88%. Turning to the subsamples, the effect is significant for Medicaid
mothers and less educated mothers. When we look by race, the impact of
WIC on low birth weight is over twice as high for African Americans as
for other mothers. The effects are also larger for African Americans in
the models of low weight gain, small for dates, and NICU use.
Table 5 shows estimates from model (3), which allows the effect of
WIC to be different for first born children than for children of higher
birth order. Columns 2 through 5 suggest that the estimates are, for the
most part, quite similar to Table 4. There is no significant difference
in the effect of WIC on being small for dates between first born and
other children: Both groups of children are 4.81% less likely to be
small for dates compared with siblings who did not receive WIC in utero.
The effects of WIC on low weight gain are somewhat larger for
firstborns, but statistically significant for all children.
Moreover, there appear to be significant effects of WIC on
complications of labor and delivery and the probability of NICU use for
most children, including first born children. The sole exception is
among African American mothers. In this group, WIC has no significant
effect on complications of labor and delivery, and has a much weaker
effect on NICU use among first borns than among other children. Recall,
that we presented some evidence earlier suggesting that women with
previous poor pregnancy outcomes are referred to WIC. Because mothers of
first born children have not had previous poor pregnancy outcomes, the
effect of WIC on the use of these medical services cannot be driven by
such referrals in this group. Stronger effects of WIC on the utilization
of medical care among later born African American children are
consistent with such a pattern of referrals, however.
One result that is quite different for firstborns than for other
children appears in column 1. Only first born children have
significantly lower probabilities of being low birth weight when they
received WIC. Moreover, WIC is estimated to reduce the probability of
low birth weight in this group by a third.
VI. DISCUSSION AND CONCLUSIONS
This article presents the first large-sample mother fixed effects
estimates of the effects of WIC on infant outcomes and use of health
care. Because fixed effects estimates only control for fixed
characteristics of mothers, we also pay careful attention to
time-varying factors associated with within-mother changes in prenatal
WIC participation. The portrait that emerges is one in which young,
unmarried, unemployed women use WIC initially, but are less likely to
use it for subsequent pregnancies in which they are older and married or
employed. These results also show that investigation of the determinants
of WIC participation is useful when interpreting its estimated effects.
The evidence suggests that women receive more intensive medical
care when they are on WIC, at least in New York City. Women are more
likely to be diagnosed with chronic diabetes and hypertension when they
are on WIC despite being younger; they are more likely to be diagnosed
with complications of labor and delivery; and their infants are more
likely than non-WIC siblings to spend any time in the NICU. These
results contrast with earlier estimates (e.g., Bitler and Currie 2005)
which suggested that WIC resulted
in immediate cost savings due to reductions in use of medical care
among women and newborns. Moreover, even firstborns are more likely to
be diagnosed with complications of labor and delivery and to use the
NICU if their mothers were on WIC during pregnancy, which suggests that
the overall effect cannot be an artifact of women with poor previous
pregnancy outcomes being referred to WIC, though it is possible that the
pattern among African Americans reflects such a referral pattern.
We also find, consistent with most previous WIC studies, that WIC
participation has positive effects on infant health: WIC infants are
5.6% less likely to be low birth weight and 4.9% less likely to be small
for dates (even when we only look at full-term infants). However,
consistent with the critiques of Joyce, Gibson, and Colman (2005),
Joyce, Racine, and Yunzal-Butler (2008), and others, these estimates are
quite a bit smaller than many that are in the literature. (11) An
exception is among firstborns, where WIC reduces the probability of low
birth weight by a third. Thus, the results suggest that WIC promotes
weight gain and reduces the probability that a child is "small for
dates" among most women, but has especially strong effects on women
delivering for the first time.
One possible caveat to this study is that the estimated effects may
actually be too small. Measurement error that was independent of WIC
status, which is self-reported by the mother, could bias the estimated
coefficients toward zero. Also, it is not clear that one should expect
the estimated effects of WIC to be the same in every time and place as
they must depend on the average health and nutritional status of women
in the program, the composition of program benefits, and the medical and
nutritional benefits available outside the WIC program.
In summary, we find that infants whose mothers used WIC during
pregnancy are healthier than siblings born when the mothers were not
using WIC. Some of the effect appears to come through a reduced
probability of low weight gain during pregnancy, which is reflected in
heavier babies and a lower incidence of "small for dates"
babies. WIC mothers in our sample were also heavier users of medical
care, and were more likely to be diagnosed with chronic conditions
(probably reflecting this more intensive use of care). We believe that
mother fixed effects techniques applied to richer administrative data
(such as future mergers of educational records with birth records) could
yield even greater insights into the long-term benefits of WIC.
ABBREVIATIONS
APGAR: Appearance, Pulse, Grimace, Activity, Respiration
NICU: Neonatal Intensive Care Unit
WIC: Women, Infants, and Children
doi: 10.1111/ecin. 12219
REFERENCES
Almond, D., and J. Currie "Human Capital Development before
Age Five," in Handbook of Labor Economics, Vol. 4, edited by O.
Ashenfelter and D. Card. Amsterdam: North Holland, 2011, 1315-486.
Besharov, D. J., and P. Germanis. Rethinking WIC: An Evaluation of
the Women, Infants, and Children Program. Washington, DC: American
Enterprise Institute, 2001.
Bitler, M. P., and J. Currie. "Does WIC Work? The Effects of
WIC on Pregnancy and Birth Outcomes." Journal of Policy Analysis
and Management, 24(1), 2005, 73-91.
Bitler, M. P, J. Currie, and J. K. Scholz. "WIC Eligibility
and Participation." Journal of Human Resources, 38, 2003, 1139-79.
Currie, J. "U.S. Food and Nutrition Programs," in
MeansTested Transfer Programs in the United States, edited by Robert
Moffitt. Chicago: University of Chicago Press for NBER, 2003, 199-290.
Figlio, D., S. Hamersma, and J. Roth. "Does Prenatal WIC
Participation Improve Birth Outcomes? New Evidence from Florida."
Journal of Public Economics, 93(1-2), 2009, 235-45.
Hicks, L. E., and R. A. Langham. "Cognitive Measure Stability
in Siblings Following Early Nutritional Supplementation." Public
Health Reports, 100(6), 1985,656-62.
Hoynes, H., M. Page, and A. H. Stevens. "Can Targeted
Transfers Improve Birth Outcomes? Evidence from the Introduction of the
WIC Program." Journal of Public Economics, 95, 2011, 813-27.
Joyce, T.. D. Gibson, and S. Colman. "The Changing Association
between Prenatal Participation in WIC and Birth Outcomes in New York
City." Journal of Policy Analysis and Management, 24(4), 2005,
661-85.
Joyce, T., A. D. Racine, and C. Yunzal-Butler. "Reassessing
the WIC Effect: Evidence from the Pregnancy Nutrition Surveillance
System." Journal of Policy Analysis and Management, 25(2), 2008,
277-303.
Kowaleski-Jones, L., and G. J. Duncan. "Effects of
Participation in the WIC Program on Birthweight: Evidence from the
National Longitudinal Survey of Youth." American Journal of Public
Health, 92(5), 2002, 799-804.
Ludwig, J., and M. Miller. "Interpreting the WIC Debate."
Journal of Policy Analysis and Management, 24(4), 2005,691-701.
Nold, J. L., and M. K. Georgieff. "Infants of Diabetic
Mothers." Pediatric Clinics of North America, 51, 2004, 619-37.
Rossin-Slater, M. "WIC in Your Neighborhood: New Evidence on
the Impacts of Geographic Access to Clinics." Journal of Public
Economics, 102, 2013, 51-69.
USDA. "Frequently Asked Questions About WIC." 2012.
Accessed March 31, 2015. http://www.fns.usda.gov/
wic/frequently-asked-questions-about-wic.
(1.) See Currie (2003) for a review of the older WIC literature and
a discussion of its limitations. Also see Besharov and Germanis (2001)
and Joyce, Gibson, and Colman (2005) for further discussion of the
limitations of the WIC literature.
(2.) See Ludwig and Miller (2005).
(3.) As an example, the New York regulations state the following
about infant cereal: "This is the only brand of cereal allowed for
infants up to 12 months old: Gerber Cereal for Baby; Plain varieties
only: barley, oatmeal, rice, whole wheat, or mixed; 8 and 16-ounce
containers are allowed; NOT ALLOWED: Organic, extra ingredients such as
DHA, fruit, formula, or added protein."
(4.) http://www.fns.usda.gov/pd/25wifyavgfdS.htm.
(5.) A third study, by Hicks and Langham (1985) focuses on the
beginning of the WIC program in Louisiana. They look at 19 sibling pairs
in which, due to the timing of the phase in, one child received WIC in
utero, and the other did not receive it until they were already a year
old. They find positive effects of prenatal WIC participation on IQ and
cognitive test scores.
(6.) Neighborhoods are defined by the Department of City Planning
and are at the level of areas such as "Brighton Beach,"
"Clinton Hill," or "Hunts Point." See
http://www.nyc.gov/html/dcp/html/bytes/dwn_npa.shtml.
(7.) Birth weights <700 g and >7.000 g have been recoded as
missing. The dummy for gestation <37 weeks has been recoded as
missing for gestation < 20 weeks and > 45 weeks.
(8.) Prepregnancy weight has missing values in the data and in
addition values <91 pounds have been coded as missing when this is
used as an outcome.
(9.) An alternative possibility is that women on WIC are being
unnecessarily diagnosed with these conditions and treated, but we think
this is unlikely given the clear diagnostic criterion and lack of debate
about appropriate treatment for high blood pressure and diabetes.
(10.) When used as a control, indicators for the following
categories of prepregnancy weight have been included: missing,
>90-< 110 pounds, 110-<120 pounds, 120-<135 pounds,
135-<159 pounds, 159-<185 pounds, and > = 185 pounds.
(11.) For example, Bitler and Currie (2005) find a 29% reduction in
the incidence of low birth weight. It may be that the effects of WIC are
quite different in different populations, depending in part on their
underlying nutritional status and access to medical care.
JANET CURRIE and ISHITA RAJANI *
* The authors would like to thank Katherine McVeigh, Melissa
Pfeiffer, Maushumi Mavinkurve, Jisen Ho, Meredith Slopen, and Slavenka
Sedlar for their roles in making the data available. Thu Vu provided
excellent programming assistance.
Currie: Professor, Department of Economics, Princeton University,
Princeton, NJ 08540, NBER, Cambridge, MA 02138. Phone 609 258 7393, Fax
609 258 5974, E-mail
[email protected]
Rajani: Graduate Student, Department of Economics, Princeton
University, Princeton, NJ 08540, Phone 609 258 5566, Fax 609 258 5874,
E-mail
[email protected]
TABLE 1
Variable Means
Moms
Sample Full Sample w >1 Child
On WIC during pregnancy 0.420 0.423
Number of observations [1,232,007] [559,935]
Outcomes
Low birth weight 0.068 0.061
Infant small for dates 0.099 0.093
Gestation < 37 weeks 0.080 0.074
Weight gain < 6 kg 0.068 0.072
Complications of labor 0.348 0.322
Infant to NICU 0.081 0.072
Mother and child characteristics
Child female 0.488 0.490
Child's parity 1.994 2.275
Mother education 12.475 12.344
Mother age 27.923 27.492
Marital status 0.511 0.546
Father's information missing 0.212 0.190
Employed 0.342 0.320
Smoked during pregnancy 0.039 0.039
Prepregnancv weight (kg) 64.943 65.179
Chronic diabetes 0.003 0.003
Chronic hypertension 0.009 0.007
African American 0.278 0.259
Medicaid coverage 0.604 0.669
First trimester prenatal care 0.645 0.651
Previous poor outcome (a) 0.056 0.109
Moms w >1 Moms w > 1
Sample Child, on WIC Child not on WIC
On WIC during pregnancy 1.000 0.000
Number of observations [236,680] [323,255]
Outcomes
Low birth weight 0.068 0.055
Infant small for dates 0.102 0.086
Gestation < 37 weeks 0.079 0.070
Weight gain < 6 kg 0.085 0.062
Complications of labor 0.345 0.306
Infant to NICU 0.094 0.056
Mother and child characteristics
Child female 0.491 0.490
Child's parity 2.357 2.216
Mother education 11.093 13.259
Mother age 25.802 28.729
Marital status 0.347 0.691
Father's information missing 0.266 0.135
Employed 0.153 0.444
Smoked during pregnancy 0.052 0.030
Prepregnancv weight (kg) 65.888 64.655
Chronic diabetes 0.004 0.002
Chronic hypertension 0.009 0.007
African American 0.323 0.211
Medicaid coverage 0.968 0.451
First trimester prenatal care 0.567 0.714
Previous poor outcome (a) 0.137 0.091
(a) Applies only to 746,491 births for which we observe
a previous birth in our sample.
TABLE 2
Mother Fixed Effects Estimates Describing Selection into the
WIC Program (Coefficient, (SE), Percentage Effect) (All
Coefficients and Standard Errors Multiplied by 100)
Dependent
Variable Married Employed Age
All births -2.454 -6.094 -0.535
N = 1,226,026 (0.127) (0.165) (0.109)
-4.80% -17.78% -5.53%
Gestation > 37, -2.551 -6.277 -0.455
<42 weeks (0.137) (0.179) (0.117)
At = 1,119,928 -4.90% -18.20% -4.80%
Medicaid coverage -2.536 -6.370 -0.660
N = 739,990 (0.141) (0.164) (0.128)
-7.05% -37.96% -4.80%
Mother's -1.566 -3.453 -0.077
education <=12 (0.154) (0.183) (0.151)
N = 732,852 -3.83% -17.90% -0.52%
African American -3.262 -7.712 -1.641
N = 340,650 (0.240) (0.304) (0.216)
-10.86% -21.07% -12.51%
Number 1,225,999 1,166,803 1,225,837
nonmissing
Dependent Prepregnancy Chronic
Variable Weight Smoked Diabetes
All births -8.518 0.582 0.071
N = 1,226,026 (3.732) (0.078) (0.023)
-0.13% 15.08% 21.80%
Gestation > 37, -9.954 0.638 0.050
<42 weeks (4.007) (0.080) (0.022)
At = 1,119,928 -0.13% 18.03% 17.90%
Medicaid coverage -8.845 0.584 0.060
N = 739,990 (4.180) (0.090) (0.025)
-0.14% 12.33% 17.12%
Mother's -13.287 0.629 0.042
education <=12 (4.829) (0.109) (0.028)
N = 732,852 -0.20% 12.59% 12.35%
African American 31.649 0.941 0.067
N = 340,650 (7.906) (0.181) (0.047)
0.45% 15.16% 9.53%
Number 1,092,221 1,217,216 1,226,026
nonmissing
Prenatal Poor
Dependent Chronic Care First Previous
Variable Hypertension Trimester Outcome (a)
All births 0.120 0.620 2.536
N = 1,226,026 (0.037) (0.235) (0.223)
13.51% 0.96% 13.64%
Gestation > 37, 0.053 0.488 2.976
<42 weeks (0.036) (0.253) (2.243)
At = 1,119,928 7.23% 0.75% 17.66%
Medicaid coverage 0.106 0.852 1.884
N = 739,990 (0.040) (0.262) (0.265)
11.86% 1.52% 8.75%
Mother's 0.050 1.305 2.684
education <=12 (0.044) (0.303) (0.285)
N = 732,852 5.83% 2.25% 12.63%
African American 0.160 0.491 2.149
N = 340,650 (0.091) (0.450) (0.457)
9.72% 0.84% 8.14%
Number 1,226.026 1.096,100 163,639
nonmissing
Notes: Each entry shows the coefficient on WIC from a
separate regression of Equation (1). Standard errors in
parentheses. All models controlled for the child's gender,
maternal education (<12, 12, 13-15, 16 + years), child
parity (1,2,3,4+), maternal age (<20, 20-24, 25-29, 30-34,
35+), month of birth, year of birth, 193 neighborhoods, and
indicators for missing values of these variables.
(a) This variable is defined only for women with a previous
birth that is included in our data set.
TABLE 3
Mother Fixed Effects Estimates of the Effects
of WIC Participation on Outcomes
Low Birth Gestation Small
Weight <37 Weeks for Dates
All births, N = 1,226,026
Coeff x 100 -0.414 -0.800 -0.288
(SE) x 100 (0.106) (0.118) (0.131)
Mean dep. var. 0.067 0.080 0.099
Implied % change -6.12% -10.05% -2.92%
Medicaid covered mothers only, N = 739,990
Coeff x 100 -0.426 -0.862 -0.265
(SE) x 100 (0.116) (0.129) (0.143)
Mean dep. var. 0.073 0.084 0.106
Implied % change -5.81% -10.21% -2.50%
Mothers with < = 12 years of education only, N = 732,852
Coeff x 100 -0.451 -0.966 -0.187
(SE) x 100 (0.135) (0.150) (0.168)
Mean dep. var. 0.073 0.084 0.106
Implied % change -6.15% -11.44% --
African American mothers only, N = 340.650
Coeff x 100 -1.142 -1.437 -0.398
(SE) x 100 (0.228) (0.249) (0.263)
Mean dep. var. 0.099 0.110 0.124
Implied % change -11.50% -13.03% --
# Nonmissing 1,223,284 1,216.781 1,212,655
Low Weight Complications Any
Gain of Labor NICU
All births, N = 1,226,026
Coeff x 100 -0.625 2.115 1.143
(SE) x 100 (0.131) (0.217) (0.125)
Mean dep. var. 0.068 0.348 0.081
Implied % change -9.18% 6.07% 14.17%
Medicaid covered mothers only, N = 739,990
Coeff x 100 -0.676 2.285 1.222
(SE) x 100 (0.150) (0.227) (0.139)
Mean dep. var. 0.080 0.348 0.091
Implied % change -8.47% 6.56% 13.38%
Mothers with < = 12 years of education only, N = 732,852
Coeff x 100 -0.767 2.224 1.334
(SE) x 100 (0.176) (0.264) (0.159)
Mean dep. var. 0.080 0.341 0.089
Implied % change -9.54% 6.53% 14.95%
African American mothers only, N = 340.650
Coeff x 100 -0.875 0.386 0.856
(SE) x 100 (0.268) (0.401) (0.272)
Mean dep. var. 0.088 0.345 0.116
Implied % change -9.90% -- 7.35%
# Nonmissing 1,077,828 1,226,026 1,112,908
Notes: Each coefficient is from a separate regression of
Equation (2). Standard errors in parentheses. In addition to
the variables listed in the notes to Table 2, these
regressions controlled for chronic diabetes, chronic
hypertension, prepregnancy weight, maternal smoking,
maternal employment, marital status, whether the mother
obtained prenatal care in the first trimester, and whether
the father was listed on the birth certificate. Implied %
change in the dependent variable is shown if the coefficient
estimate is significantly different than zero.
TABLE 4
Mother Fixed Effects Estimates of the Effects of WIC
Participation on Outcomes (Gestation > 37 Weeks Only)
Low Birth Small for Low Weight
Weight Dates Gain
All births, AT = 1,119,928
Coeffx 100 -0.160 -0.482 -0.511
(SE) x 100 (0.077) (0.138) (0.136)
0.028 0.099 0.063
Implied % change -5.63% -4.88% -8.17%
Medicaid covered mothers only. A = 672,213
Coeffx 100 -0.155 -0.447 -0.590
(SE)x 100 (0.085) (0.150) (0.156)
0.031 0.107 0.073
Implied % change -4.97% -4.18% -8.03%
Mothers with < =12 years of education only, N = 664,951
Coeffx 100 -0.136 -0.412 -0.682
(SE)x 100 (0.100) (0.177) (0.183)
0.031 0.107 0.074
Implied % change -- -3.84% -9.21%
African American mothers only, N = 300,423
Coeff x 100 -0.376 -0.630 -0.798
(SE)x 100 (0.169) (0.281) (0.285)
0.040 0.126 0.080
Implied % change -9.37% -5.02% -9.95%
Number nonmissing 1,119,723 1,119,723 990,317
Complications Any
of Labor NICU
All births, AT = 1,119,928
Coeffx 100 2.147 1.268
(SE) x 100 (0.233) (0.110)
0.336 0.051
Implied % change 6.39% 24.80%
Medicaid covered mothers only. A = 672,213
Coeffx 100 2.324 1.327
(SE)x 100 (0.243) (0.123)
0.336 0.060
Implied % change 6.92% 22.10%
Mothers with < =12 years of education only, N = 664,951
Coeffx 100 2.352 1.384
(SE)x 100 (0.283) (0.141)
0.328 0.058
Implied % change 7.18% 24.04%
African American mothers only, N = 300,423
Coeff x 100 0.528 1.425
(SE)x 100 (0.440) (0.246)
0.368 0.071
Implied % change -- 20.19%
Number nonmissing 1,119,928 1,016,222
Notes: Each coefficient is from a separate regression of
Equation (2). Standard errors in parentheses. In addition to
the variables listed in the notes to Table 2, these
regressions controlled for chronic diabetes, chronic
hypertension, prepregnancy weight, maternal smoking,
maternal employment, marital status, whether the mother
obtained prenatal care in the first trimester, and whether
the father was listed on the birth certificate. Implied %
change in the dependent variable is shown if the coefficient
estimate is significantly different than zero.
TABLE 5
Mother Fixed Effects Estimates of the Effects of WIC
Participation on Outcomes (Gestation > 37 Weeks Only)
Low Birth Small for Low Weight
Weight Dates Gain
All births, N =1,119,928
WIC coeff x 100 0.077 -0.475 -0.382
(SE)x 100 (0.078) (0.147) (0.146)
WIC x first birth x 100 -1.103 -0.171 -0.498
(SE)x 100 (0.111) (0.209) (0.204)
Mean of dep. var. 0.028 0.099 0.063
Implied % change -- -4.81% -6.11%
% Change for first birth -36.06% -- -14.08%
Medicaid covered mothers only. N = 672,213
WIC coeff x 100 0.022 -0.573 -0.536
(SE) x 100 (0.089) (0.165) (0.173)
WIC x first birth x 100 -0.908 0.325 -0.210
(SE)x 100 (0.152) (0.281) (0.294)
Mean of dep. var. 0.031 0.107 0.073
Implied % change -- -5.36% -7.30%
% Change for first birth -29.03% -- --
Mothers with <= 12 years of education only, N = 664,951
WIC coeff x 100 0.063 -0.454 -0.611
(SE) x 100 (0.102) (0.189) (0.198)
WIC x first birth x 100 -1.100 0.037 -0.288
(SE) x 100 (0.163) (0.301) (0.310)
Mean of dep. var. 0.031 0.107 0.074
Implied % change -- -4.23% -8.26%
% Change for first birth -33.14% -- --
African American mothers only, N = 300,423
WIC coeff x 100 -0.129 -0.679 -0.597
(SE)x 100 (0.176) (0.305) (0.312)
WIC x first birth x 100 -0.971 0.217 -0.814
(SE) x 100 (0.290) (0.504) (0.506)
Mean of dep. var. 0.040 0.126 0.080
Implied % change -- -5.41% -7.44%
% Change for first birth -27.39% -- --
Number nonmissing 1,119,723 1,119,723 990,317
Complications Any
of Labor NICU
All births, N =1,119,928
WIC coeff x 100 2.354 1.296
(SE)x 100 (0.250) (0.104)
WIC x first birth x 100 -0.886 -0.431
(SE)x 100 (0.357) (0.148)
Mean of dep. var. 0.336 0.051
Implied % change 7.00% 25.34%
% Change for first birth 4.31% 16.91%
Medicaid covered mothers only. N = 672,213
WIC coeff x 100 1.954 1.273
(SE) x 100 (0.270) (0.121)
WIC x first birth x 100 1.330 -0.187
(SE)x 100 (0.460) (0.206)
Mean of dep. var. 0.336 0.060
Implied % change 5.82% 21.20%
% Change for first birth 9.79% --
Mothers with <= 12 years of education only, N = 664,951
WIC coeff x 100 2.171 1.244
(SE) x 100 (0.306) (0.137)
WIC x first birth x 100 0.611 0.075
(SE) x 100 (0.486) (0.217)
Mean of dep. var. 0.328 0.058
Implied % change 6.63% 21.61%
% Change for first birth -- --
African American mothers only, N = 300,423
WIC coeff x 100 0.866 1.716
(SE)x 100 (0.482) (0.240)
WIC x first birth x 100 -1.352 -1.087
(SE) x 100 (0.796) (0.395)
Mean of dep. var. 0.368 0.071
Implied % change -- 24.31%
% Change for first birth -- 8.91%
Number nonmissing 1,119,928 1.016,222
Notes: Each set of coefficients is from a separate
regression of Equation (3). Standard errors in parentheses.
In addition to the variables listed in the notes to Table 2,
these models controlled for chronic diabetes, chronic
hypertension, prepregnancy weight, maternal smoking,
maternal employment, marital status, whether the mother
obtained prenatal care in the first trimester, and whether
the father is listed on the birth certificate. Implied %
change in the dependent variable is shown if the coefficient
estimate is significantly different than zero. % Change for
first births shown if significantly different than all
births.
