Effects of child health on sources of public support.

1. Introduction

A growing body of research reveals that not only does low income lead to poor child health, but poor child health can have deleterious effects on family resources that may contribute to health and economic disadvantages over the life course. Corman and Kaestner (1992), Mauldon (1992), Joesch and Smith (1997), and Fertig (2004) found that married couples are more likely to divorce when their child has a serious health problem, and Reichman, Corman, and Noonan (2004) found that one-year-old children with serious health problems are less likely than their healthy peers to live with their fathers. Having a child in poor health reduces mothers' labor force participation (Norberg 1998, Powers 2003, Corman, Noonan, and Reichman 2005), and it also appears to have increased reliance on cash assistance through the former Aid to Families with Dependent Children (AFDC) program (Wolfe and Hill 1995). It is not known whether mothers with unhealthy children are more likely than those with healthy children to rely on welfare in the current public assistance environment. The answer to this question has implications for mothers' ability to make ends meet and to invest in their children's health.

The Personal Responsibility and Work Opportunity Reconciliation Act (PRWORA) of 1996 dramatically restricted eligibility for cash assistance in the United States by establishing time limits on the receipt of welfare, expanding work requirements for those receiving welfare, and allowing states to impose stricter sanctions for noncompliance with work requirements or other guidelines. Reflecting this new paradigm, it replaced the former Aid to Families with Dependent Children (AFDC) program with Temporary Assistance to Needy Families (TANF). Other features of the 1996 legislation included the reimposition of a more stringent definition of child disability for Supplemental Security Income (SSI) benefits (after the criteria had been eased in 1990) and the de-linking of eligibility for Medicaid from that for cash assistance so that nonpregnant mothers can be eligible for Medicaid even if they do not meet their state's new cash assistance requirements.

In this paper, we estimate the effect of having a young child in poor health on a mother's reliance on TANF and other sources of public support within this new regime. The effects on maternal labor supply found in other studies suggest that having a child in poor health will increase reliance on welfare. However, this may not be the case under TANF because mothers with children in poor health may (1) have difficulty complying with TANF work requirements and therefore choose not to participate (or see their benefits eliminated for this reason); (2) be more likely to delay participating in order to "bank" their lifetime benefit allotment (in states that have such limits); (3) have less incentive to participate in TANF because nonpregnant mothers' eligibility for Medicaid is no longer tied to welfare participation; or (4) be more likely than before PRWORA to substitute SSI for TANF because the former, despite being more difficult to obtain, has higher financial benefits, fewer restrictions, and more permanency than the latter. In addition, having a child in poor health may increase reliance on other sources of public support, such as food stamps or the Supplemental Nutrition Program for Women, Infants, and Children (WIC), which have fewer restrictions, such as work requirements or time limits.

We estimate the effects of having a child in poor health on the mother's receipt of TANF and cash assistance through SSI and in-kind subsidies in the form of food, health insurance, and shelter (WIC, food stamps, Medicaid, and housing). We control for a rich set of covariates, consider alternative definitions of poor child health, and test for the potential endogeneity of child health. The results contribute to the literature on determinants of welfare participation and further our understanding of how vulnerable families make ends meet under the current welfare regime. They also have implications for our understanding of the processes underlying children's health and income trajectories in low-income families.

2. Background

In this paper, we estimate the effects of poor child health on participation in TANF and other programs that provide public support. An extensive literature examines the determinants of participation in both TANF and its predecessor, AFDC. Many other researchers have evaluated the effects of policies, including the 1996 PRWORA legislation, on welfare caseloads. (1) Both of these bodies of research include studies relevant to our research question, but few have examined the causal effect of poor child health on welfare participation. We summarize the relevant research, first focusing on pre-welfare reform literature and then reviewing relevant post-welfare reform studies.

Pre-Welfare Reform Studies

A seminal study by Blank (1989) examined the relationship between medical need and AFDC participation. In the AFDC era, health insurance through the Medicaid program was strongly linked to AFDC participation. Blank posited that a mother who had medical problems (or had a family member with medical problems) would have an incentive to participate in AFDC for two reasons: a diminished capacity to work (directly, as a result of her own disability, or indirectly, due to the burden of caring for a disabled family member) and to obtain public health insurance through Medicaid. Using the National Medical Care Utilization and Expenditure Survey, Blank compared individuals' likelihood of AFDC participation in states with and without programs that had Medically Needy provisions for Medicaid; such provisions expanded coverage to certain financially or categorically ineligible individuals. She found that poor health of a mother or one of her family members increased the likelihood of AFDC participation among single mothers, but participation was unrelated to whether or not the state had a Medically Needy provision. Thus, poor health increased AFDC participation because it limited the mother's ability to work, not because AFDC provided access to public health insurance.

Wolfe and Hill (1995) separately analyzed the effects of child disability (a serious activity limitation) and mother's poor or fair health on labor force participation. Using the 1984 panel of the Survey of Income and Program Participation (SIPP), they simulated mothers' eligibility for a number of hypothetical public health insurance plans and found that those that would cover children for health insurance regardless of AFDC participation would have a large positive impact (as much as 22 percentage points) on labor force participation of single mothers with disabled children. They inferred that welfare participation would decrease substantially with increased health insurance coverage for working mothers. Their findings indicate much stronger health policy effects than those found by Blank; the divergent results may reflect the different measures of health/disability that were used. Blank used an average measure of health for the family, whereas Wolfe and Hill used specific and distinct measures of child disability and maternal health status. The effects for children may be smaller today because eligibility for Medicaid has been de-linked from welfare participation for children since the 1980s.

Two recent studies assessed the effects of poor child health on exits from AFDC. Using data from the 1990 SIPP, Acs and Loprest (1999) found that child disability (defined as a limitation that prevented the child from performing usual activities) increased the probability that a mother of a child under six years of age was able to leave welfare by about 14 percentage points within a short period (one to two and one-half years), but it did not increase the probability that she left welfare to work. This result suggests that mothers may have left welfare to obtain more generous benefits from other sources, such as SSI, for which individuals can be eligible if they are both poor and disabled. Meyers, Brady, and Seto (2000) estimated the likelihood of transitioning from welfare using panel data on participants in the California Work Pays Demonstration Project from 1992 to 1996 (just before implementation of PRWORA). Using a hazard model and holding constant the mother's age, ethnicity, number of children, presence of a partner in the household, county of residence, and own disability status, they found that having a severely disabled child reduced the probability of discontinuing cash assistance at all (AFDC or SSI) but increased the likelihood of leaving AFDC for SSI.

Two studies examined the effects of AFDC or SSI generosity on child SSI participation. Black, McKinnish, and Sanders (1998) found that states with lower levels of AFDC generosity had higher rates of child SSI participation, holding constant other factors. Garrett and Glied (2000), who also used state-level data, obtained results similar to those of Black, McKinnish, and Sanders and found that the Zebley decision of 1990, which liberalized the definition of child disability until PRWORA made it more stringent in 1996, resulted in a significant increase in child SSI participation. In a related study that used data from the National Health Interview Survey (from 1987 to 1994, covering pre- and post-Zebley decision years), Kubik (1999) found that greater numbers of children were diagnosed with chronic impairments after the Zebley decision. That study also found that children in states with high SSI benefits (compared with AFDC benefits) were more likely to be diagnosed with a disability than those in states with low net SSI benefits. These results indicate that policies that make welfare less generous or less accessible, or that make SSI more generous or more accessible, tend to shift participation from welfare to SSI. In a study that did not focus on children, Schmidt and Sevak (2004), using data from the March Current Population Survey from 1988-97, found that female-headed households in states aggressively pursuing welfare reform were 21.6 percent more likely to receive SSI than those in less-aggressive states. An implication of this set of studies is that welfare reform may have increased SSI participation and decreased TANF participation among mothers of young children with severe health problems.

Post-Welfare Reform Studies

Two recent studies examined the associations of having a child with chronic illness with employment, TANF receipt, and medical coverage in the post-welfare reform era. Smith et al. (2002) studied 500 low-income parents of children with chronic illnesses (asthma, diabetes, sickle cell anemia, epilepsy, hemophilia, cerebral palsy, or cystic fibrosis) in Texas in 2001 and found that mothers have extreme difficulty combining work and caring for their unhealthy children, and they are more likely than mothers with healthy children to rely on TANF. Wise et al. (2002) used the 1998 National Health Interview Survey to investigate the association of chronic illnesses in children (asthma, mental retardation, cerebral palsy, autism, attention-deficit disorder, muscular dystrophy, cystic fibrosis, sickle-cell anemia, diabetes, arthritis, and congenital heart disease) with labor force participation and program receipt of poor families. Like Smith et al., they found that mothers of children with chronic conditions are more likely to receive TANF and less likely to work than those with healthy children. They also found that although cash assistance and medical assistance are supposed to be de-linked, TANF-enrolled children are more likely to have health insurance than poor children who are not receiving TANF. In a study that did not consider the role of child health or disability, Pati, Romero, and Chavkin (2002) found that declining enrollments in Medicaid and food stamp programs coincided with decreased enrollment in TANF in poor Manhattan neighborhoods after PRWORA. Kaestner and Kaushal (2003) found that welfare reform was responsible for a 3-4% increase in the proportion of low-income women and children without health insurance in the United States.

Brandon (2000) and Brandon and Hogan (2004) investigated the effects of child disability on exits from TANF using data from the SIPP. Both studies used proportional hazard models to estimate the probability of leaving welfare over a four-year period beginning in 1996. Both found that child disability, based on measures that included limitations in communication, mobility, self-care, or cognition, was negatively associated with exiting from welfare.

In this paper, we estimate the effects of having a young child with a severe health problem on whether mothers receive TANF, SSI, WIC, Medicaid, food stamps, and housing in the post-welfare reform era. We contribute to previous research by (1) using data from a population-based birth cohort, which allows us to assess the effects of poor child health among a representative sample of children, all of whom are the same age; (2) using rich data that allow us to accurately characterize poor child health and incorporate information on both parents' health and human capital; (3) considering the extent to which poor child health affects TANF participation through entry into and remaining on welfare; and (4) testing for the endogeneity of child health.

Although needs of children vary at different ages, most data sets do not include enough children of the same age to have representative samples or to allow for multivariate analyses on a single cohort. Most studies control for age, assuming by default that the effects of poor child heath do not vary by age. Using family fixed effects models, Norberg (1998) found that certain high-risk health problems at birth reduce the mother's labor force participation, and the effects are stronger in ages 3, 4, and 5 than at ages 1 and 2. This result suggests that the effects on reliance on public support may also vary by child's age.

Most studies include conditions, such as asthma, that are not likely to be exogenous and include few covariates. Most do not include paternal characteristics at all, or do so only for fathers who are present in the household or married to the mother. We characterize poor child health from both survey reports and medical records, and include measures of both parents' human capital, with particularly detailed information on the health status of the mother. Whereas many previous studies have addressed participation in or exits from welfare, none has considered the take-up of welfare in response to the birth of an unhealthy child. In terms of (4), we assess the robustness of our results, explore competing explanations, and test for endogeneity.

3. Analytical Framework

We consider the following model to estimate the effect of poor child health on whether mothers rely on a specific type of public support:

Reliance on public support = f 'mother's and father's wages, parents' relationship characteristics, quantity and characteristics of their children, labor market opportunities, availability of public support, [mu]) (1)

Reliance on public support is a function of the parents' earning capacity (wages), the number and characteristics of their children (together and with other partners), their labor market opportunities, and the availability of public support. The public support function may also contain another set of factors, [micro], that are unobserved. To estimate this model, we need good measures of or proxies for parents' wages, the number and characteristics of their children, their local labor markets, and the availability of public assistance. For wages, we use a set of characteristics that include age, race/ethnicity, nativity, education, work history, and health status. We include measures of the parents' relationship status, which may play a role in decisions about program participation. We focus on the effects of one characteristic of the focal child--health--and also consider the child's sex. We include whether the mother has other children and whether the father has children with another partner. For local labor markets, we include city-level unemployment rates, average full-time wages of females, and cost of living. For availability of public support, we include state-fixed effects in certain models.

4. Data

The Fragile Families and Child Wellbeing Study (FFCWB) follows a cohort of newborns and their parents in 20 large U.S. cities in 15 states. In-hospital baseline interviews were conducted from 1998 to 2000 with 4898 sets of parents shortly after their children's births. (2) These births were sampled randomly from 75 hospitals in the 20 cities. Mothers were eligible for the study if the father of the baby was living, they were able to complete the interview in English or Spanish, and they were not planning to place the child for adoption. Baseline response rates were 86% among eligible mothers and 78% among the partners of those mothers; 89% of the mothers who completed baseline interviews were re-interviewed when their children were between 12 and 18 months old. By design, about three quarters of the births were to unmarried parents. The data, when weighted, are representative of births in large U.S. cities. (3)

The Fragile Families data are well suited for analyzing the effects of child health on reliance on public assistance because they were collected as part of a longitudinal birth cohort study and include (1) data on the child's health from birth; (2) information about receipt of different types of public assistance; (3) measures of human capital (e.g., health status, education) of both fathers and mothers; (4) data on the fathers regardless of whether they ever lived with their children; and (5) detailed information on the parents' relationship status, living arrangements, and other children (together and with other partners) at the time of the birth. We also were able to include local labor market characteristics in our models. In addition, some of our analyses incorporate hospital medical records data that were collected from a subsample of mother respondents and their newborns. (4) Because the FFCWB study sampled births exclusively in large cities and oversampled nonmarital births within those cities, the mothers in our sample are more likely than the general population of mothers of newborns in the United States to receive public assistance.

5. Descriptive Analysis

The purpose of this paper is to estimate the effects of poor child health on mothers' reliance on public assistance. Unless indicated otherwise, all covariates are measured at baseline. In general, we use mother reports for information about the mother and father reports for information about the father. However, in cases where father data are missing, we use mother reports about the father if these are available. We restrict our sample to nonmultiple births for which there were no missing data on any of the main analysis variables. In Table 1, the characteristics of the primary analysis sample are presented, both overall and stratified by parity (first births vs. higher-order births). Selected characteristics of the subsample of births for which medical record data are available, which we use in supplemental analyses, are presented in Table 2.

In most previous studies, child health was ascertained through survey questions to mothers. We construct measures of poor child health from both the FFCWB surveys and medical records and assess the robustness of our results to the choice of measure. For our first measure, that used by Reichman, Corman and Noonan (2004), we consider a child to have poor health if at least one of the following criteria is met (all are from mothers' reports): the child weighed less than 4 pounds at birth (2.4%); the mother reported at follow-up that the child had a physical disability, such as Down Syndrome or cerebral palsy (2.0%); or, if the child was at least 12 months old at follow-up, he or she had neither walked nor crawled (0.8%) (figures not shown in table). We use a stringent definition of low birth weight rather than the typical 5.5-pound (2500 gram) cutoff in order to better identify cases of serious and chronic health problems because most moderately low-birth weight (between 3.3 and 5.5 pounds, or 1500 and 2500 grams) children do not experience severe health problems (Hack, Klein, and Taylor 1995). (5) Almost 5% of the children in our sample met at least one of these three criteria and are therefore characterized as having poor health. (6) We refer to this measure, our primary measure of poor child health, as reported poor child health.

Our second measure of poor child health is whether the child weighed less than 5.5 pounds (2500 grams) at birth--the standard cutoff for low birth weight. This measure is readily obtained from maternal reports or medical records (7) but is not very specific because, as indicated, few moderately low-birth weight children (the majority of low-birth weight children) have severe health problems. The advantage of this measure is that it conforms to the standard definition of low birth weight and is comparable across studies. About nine percent of the infants in our urban, mostly nonmarital sample were low birth weight. We refer to this measure of poor child health as low birth weight.

Our third measure of poor child health, coded from the medical records, is whether the infant had a severe abnormal condition at birth or was very low birth weight. The coding of the abnormal conditions was conducted by an outside pediatric consultant who was directed to code a case as having poor child health if the child had a condition that is severe, long-term, and likely not related to maternal prenatal behavior. Although this measure of poor child health is our most precise, it is available for only 1430 cases in the sample at this time. About 4% of the infants in the sample of 1430 were born with a severe abnormal condition or very low birth weight (figure not shown in any table). (8) We refer to this measure of poor child health as abnormal newborn condition.

Mothers were asked at the one-year follow-up interview whether they had received any food stamps or assistance from TANF in the past 12 months. They were also asked whether they had received any benefits from WIC since the child was born or if anyone in the family received SSI. Positive responses to these questions were used to code the receipt of TANF, food stamps, WIC, and SSI, respectively. Mothers were asked whether they or their child(ren) who live with them currently receive health insurance through Medicaid or another "public, federal, or state assistance program that pays for medical care." If the mother responded affirmatively to that question, she was coded as having received Medicaid. Finally, mothers were asked whether the federal, state, or local government was currently helping them pay their rent and whether they live in a housing project. We coded them as receiving housing support if they responded affirmatively to either of these two questions or if they specified "housing assistance" in response to the question about other kinds of local state or federal agencies (other than welfare or TANF) that have helped them since their child was born.

Almost one quarter (24%) of mothers received some cash assistance through TANF. Over one third of mothers (36%) received food stamps and over half (56%) received Medicaid. A very large proportion (72%) of the mothers in the sample participated in the WIC program. In this sample, housing was relied on less commonly as a source of assistance. Twenty percent of mothers reported at one year that they were receiving housing assistance or living in a housing project. Only 3% of families received SSI. Among those who reported poor child health, 21% reported receiving SSI (for any family member), compared to 2% among those without reported poor child health (figures not shown in table). All forms of public assistance were more prevalent among mothers with higher-order births than among those with first births, with the largest differences for TANF, food stamps, and housing.

We include several additional measures related to the focal child or the parents' other children that may affect the mother's reliance on public assistance, and possibly child health: whether the focal child is male, (9) whether the mother has other children (with the newborn's father or another partner), and whether the father had at least one child with another partner. According to the mothers' reports, about one third of fathers had at least one child with another partner at the time of the mother's follow-up interview. (10) In analyses using the medical records subsample, we include whether the mother had a previous preterm or low-birth weight baby (see Table 2).

We take advantage of the longitudinal nature of the data by estimating models that control for whether the birth was financed by Medicaid or another government health insurance program (63%) and whether the mother reported at baseline that she had received any benefits from public assistance, TANF, or food stamps (hereafter, public assistance at baseline) in the year before the birth (35%). The former is an indicator for being poor or near-poor. The latter is the closest measure available in the data for whether the mother received TANF at baseline, which is of interest to us because we are interested not only in TANF participation at one year, but also entry and exit from the program. Thus, if the mother reported not receiving public assistance at baseline, we know that she did not receive TANF. However, if she reported receiving public assistance at baseline, we cannot be sure whether she received TANF unless it was a first birth (in this case, she would not have been eligible for TANF at baseline). The least clear situation is when it is a higher-order birth and the mother reported receiving public assistance at baseline. In that case, we cannot determine whether the mother received TANF. We control for public assistance at baseline in all models of follow-up TANF participation and estimate separate models for first- and higher-order births to investigate entry into and staying on TANF.

We include sociodemographic characteristics that may be associated with both reliance on public assistance and child health, as well as three direct measures of the mother's human capital--her education, whether she worked within the two-year period preceding the child's birth, and her health status. We include many of the same characteristics for fathers. Due to the oversampiing of nonmarital births exclusively in large cities, the sample consists predominantly of minority and disadvantaged parents. Indeed, only 22% of the mothers are white and non-Hispanic. One third of the mothers did not complete high school, and more than half did not live with both parents when they were 15 years old. (11) In our primary analyses, we use the mother's self-reported health status (excellent or very good versus good, fair, or poor) from the baseline interview. In analyses using the medical records subsample, we also include whether the mother has each of the following preexisting health conditions: anemia, lung disease, morbid obesity (BMI [greater than or equal to] 39), underweight (BMI < 18.5), and any other serious condition (see Table 2).

We go beyond whether the father was present in the mother's household to characterize the parents' relationship at the time of the birth; we consider whether the parents were married, cohabiting, romantically involved or friends, or rarely or never talked. About three quarters (74%) of the parents were unmarried at baseline; half (51%) of the unmarried parents lived together. Overall, 95% of the parents were in some type of relationship.

In all models, we include the number of months between the mother's baseline and follow-up interviews. We characterize local markets through city-level unemployment rates, mean full-time wages of females, and the cost of living. Finally, in our two-stage models (discussed later), we use two identifiers for poor child health: the number of level III neonatal intensive care unit (NICU) beds per mile in the city of the birth in fiscal year 1998 and per capita expenditures on tobacco prevention in the state in 2002.

6. Modeling Strategy

We are interested in estimating the effect of a child's poor health on the mother's reliance on public assistance. We can express Equation 1 from our analytical framework as follows:

Reliance on public assistance = f (child health, other child characteristics, number of children, parent relationship status, mother and father characteristics, city labor market characteristics, [mu]) (2)

We use probit models to estimate Equation 2, which is straightforward as long as the measure of child health is truly random (exogenous). It is possible, however, that there are non-random components of child health that are correlated with unobserved determinants of the receipt of public assistance ([mu]). If so, child health would be endogenous and its estimated effect on receipt of public assistance would be biased. We address the potential endogeneity problem in several ways. First, we include a rich set of covariates (including baseline Medicaid and public assistance at baseline) that may be related to both child health and reliance on public assistance at follow-up. Second, we attempt to define poor child health as an exogenous variable and assess robustness to three different definitions. Third, we address the possible endogenity of child health in a two-equation system.

7. Results

The results from single-equation probit models of the mother's TANF participation in the 12 months preceding her follow-up interview are presented in Table 3. The models include all of the measures in Table 1 other than the non-TANF sources of support and the measures we use as identifiers in supplementary two-stage models. The results are based on reported poor child health (constructed from survey measures of birth weight, disability, and development) as the measure of poor child health. Table 4 shows corresponding estimates using the low birth weight and abnormal newborn conditions measures of poor child health. (12) Estimates for abnormal newborn conditions are based on a much smaller sample because that measure has been coded for only a subset of the medical records cases. In both tables, results for all births are in the first column of figures, results for first births are in the second column, and results for higher-order births are in the third column. Probit coefficients, marginal effects, and standard errors (corrected for city clustering observations using Huber-White method) are shown.

Mothers of children in poor health were between 2 and 6 percentage points more likely to receive TANF than those with healthy children, and there is a striking difference by parity (Tables 3 and 4). As expected, we found stronger effects using the more precise measures of poor child health than by using low birth weight. Poor child health is not a significant predictor of TANF participation at one year for the subsample of first births, but it increases the likelihood of TANF receipt for the higher-order births by 5 to 9 percentage points. Thus, poor child health is not related to entry into TANF among first-time mothers (who were not previously eligible to receive TANF). Because our measure of baseline public assistance is broader than just TANF, the effects for higher-order births could reflect entry, staying, or some combination of the two. (13)

In supplementary analyses of mothers with higher-order births who did not receive baseline public assistance (results not shown), we found significant but smaller marginal effects of poor child health than in the full sample of higher-order births. This indicates that poor child health increases TANF participation not only by making mothers less likely to leave welfare, but also by causing them to enter the welfare system.

The coefficients for the other covariates generally have the expected signs. Having lower educational attainment, being born in the United States, and not having worked in two years before baseline all are associated with an increased likelihood that a mother received TANF at follow-up. Interestingly, the weaker the tie between the mother and father at baseline, the more likely it is that the mother received TANF. Mothers who had no relationship with the baby's father at the time of the birth were about 17 percentage points more likely to receive TANF one year later than those who cohabited with the baby's father at the time of the birth and about 28 percentage points more likely than those who were married to the father at baseline. Father's education is associated with the mother's receipt of TANF; in general, the more educated the father, the less likely that the mother relied on TANF. Both Medicaid and public assistance at baseline are significant predictors of TANF at follow-up.

Auxiliary Analyses

We conducted a number of additional analyses to explore competing explanations, further assess the robustness of our results, and examine the effects of poor child health on other sources of support that may substitute for or complement TANF.

Including Additional Covariates

The FFCWB surveys do not include detailed information on the mother's pre-birth health status or the health of her other children. To allow us to rule out the possibility that our estimated effects of poor child health reflect poor health endowments of the mother or her other children, we ran probit models for the medical records subsample that included information from detailed health histories in addition to the other covariates (including maternal self-reported health status). (14) The maternal health information from the medical records is likely more exogenous than the baseline survey measure of overall health status, which was self-reported shortly after delivery. The results presented in Table 5 are based on reported poor child health, low birth weight, and abnormal newborn condition, respectively. The sample sizes for the abnormal newborn condition models are smaller than those in Table 4 because of missing data on additional covariates. The estimated effects of poor child health on TANF receipt in these models are similar to those in Tables 3 and 4, as are the patterns across measures of poor child health and parity. These similarities indicate that our estimated effects of the focal child's health are not driven by mother's pre-birth health status or the health of her other children. (15)

Modeling Poor Child Health

Although the potential endogeneity of child health has likely been adequately addressed through our carefully defined measures of child health and rich set of covariates, as a further test we model poor child health as follows and estimate it jointly with TANF participation (Equation 2):

Poor Child Health = g (parents's characteristics, prenatal & perinatal health inputs, e). (3)

Because the outcome measures in both Equations 2 and 3 are dichotomous, we use a bivariate probit specification that allows us to test for the endogeneity of child health. We allow for correlation between the error terms in the child health and TANF equations, which we assume are normally distributed. (16)

To identify the bivariate probit model, we include certain variables (identifiers) in the child health equation but not in the TANF equation. For identifiers to be valid, they need to satisfy two conditions: They must be significant predictors of poor child health (prediction), and they must be uncorrelated with the mother's reliance on TANF after controlling for poor child health and the other covariates (exclusion). If these two conditions are met, and [rho] (the correlation of the error terms in the two equations) is not significantly different from zero, it follows that poor child health can be considered exogenous and that a standard probit is the more appropriate model. (17) The following two identifiers satisfied both conditions: The number of level III NICU beds per mile in fiscal year 1998 in the city where the mother gave birth and the per capita state expenditures for tobacco prevention in 2002. (18) The former reflects availability of postnatal resources that improve child health and the latter is an exogenous proxy for prenatal smoking, which past research has shown leads to reduced birth weight and other adverse infant health outcomes (Chomitz, Cheung, and Lieberman 1995).

Results of the bivariate probit estimation using reported poor child health for the full sample (corresponding to the first single equation probit model in Table 3) appear in the Appendix. The first column shows estimates of TANF participation (Equation 2), and the second column shows estimates of poor child health (Equation 3), which include all variables in the TANF equation (except the local labor market characteristics) plus the two identifiers.

Both identifiers are negative and significant predictors of poor child health and are excludable from the TANF equation (latter test results not shown). Based on a likelihood ratio test, the value of [rho] is not significantly different from zero (p-value = 0.63), indicating that poor child health is not endogenous in this two-equation system and that the corresponding single-equation probit is the preferred model.

Other Sources of Public Support

Table 6 shows probit estimates of the effects of poor child health on receipt of SSI, food stamps, Medicaid, WIC, and housing support using each of the three definitions of poor child health. Each cell indicates the marginal effect of poor child health from a probit model using the corresponding samples from Tables 3 and 4. To the extent possible, we controlled for the baseline level of the corresponding source of support in addition to Medicaid and the other covariates in Tables 3 and 4.

The purpose of the SSI program is to provide cash assistance to poor families with a disabled family member. Therefore, it is not surprising that we find positive and significant effects of poor child health on SSI receipt. By using our least-specific definition of poor child health, which is low birth weight, we find that poor child health increases the likelihood of SSI receipt by 3 percentage points, or 100%. Using our more-specific definitions, poor child health increases the likelihood of SSI receipt (of any family member) by 11 to 17 percentage points, or triple to quintuple the rate of families with healthy children. (19) Given the strong effects on SSI and findings from past research suggesting that mothers with disabled family members substitute SSI for welfare, it is noteworthy that we still find positive effects of poor child health on TANF. These results suggest either that mothers are not fully substituting SSI for TANF or that some families are eligible for both TANF and SSI because they have another child who is not disabled. The latter interpretation is consistent with the significant effect of poor child health on TANF receipt for higher-order births.

Poor child health is not significantly associated with participation in either WIC or food stamps. Although not always statistically significant, families with children in poor health are more likely to report receiving Medicaid at follow-up. Therefore, we cannot rule out the possibility that mothers with unhealthy children are more likely to participate in TANF to obtain benefits through Medicaid as appeared to be the case in the mid-1980s (Wolfe and Hill 1995), despite the legislative de-linking of the two programs in the mid-to-late 1980s and again in 1996. Poor child health is also positively related to receipt of housing assistance at follow-up.

As a test for endogeneity, we ran a model for receipt of housing assistance in the 12-month period preceding the baseline interview as a function of reported poor child health and the other covariates using the full sample (results not shown). We found that poor child health is not associated with receipt of housing assistance before the child was born (p-value = .54). This finding strengthens our interpretation that poor child health has a causal effect on mothers' receipt of housing at one year and refutes an alternative interpretation that poor child health is associated with factors that affect reliance on public housing.

Overall, the results indicate that having a young child in poor health not only increases the likelihood a mother will receive TANF (and, of course, SSI) one year later, but it also increases the likelihood that she will rely on Medicaid and housing assistance. In contrast, poor child health does not increase the likelihood of receiving assistance in the form of food stamps or WIC benefits. Subsample analyses for SSI, food stamps, WIC, Medicaid, and housing (not shown) reveal no clear patterns by parity.

8. Conclusion

We estimated the effects of having a child in poor health on whether mothers with one-year-old children rely on TANF and other sources of public support in the post-PRWORA program environment. We found that mothers with infants in poor health are about 2 to 8 percentage points (8-33%) more likely to receive TANF than mothers with healthy children. The effects are confined to mothers with older children (there were no effects for first-time mothers). The effects for mothers with older children primarily reflect staying on welfare, but we also found evidence of entry effects for this group. These results suggest that the decision to remain on (or enter) welfare may be the result of cumulative or multiple barriers to work. In other words, the birth of a seriously unhealthy child on top of the responsibilities of caring for other children may represent a critical "tipping point." However, it may simply reflect eligibility for TANF (a mother with only one child who is both disabled and receiving SSI would not be eligible for TANF). As expected, we found strong effects of poor child health on receipt of SSI and also found that mothers of children in poor health are more likely than those with healthy children to receive Medicaid and housing assistance. All of these effects pertain to a very short period of time (12 to 18 months after the child's birth). Since some serious health problems can be diagnosed or occur at older ages and there is evidence that the labor supply effects of poor child health increase with children's age (Norberg 1998), we may be underestimating the effects of poor child health on public assistance participation.

Our results for TANF suggest that, overall, the labor supply effects on welfare participation of having a seriously unhealthy child outweigh the potential competing effects outlined earlier (compliance or deterrence issues surrounding work requirements or time limits under TANF, the delinking of Medicaid from welfare, and the relatively more generous and permanent benefits available under SSI than TANF). More generally, our results have both positive and negative implications for families with children in poor health. The positive implication is that families who are particularly needy because they have unhealthy children are getting extra financial help in the form of cash assistance, health insurance, and housing. The negative implication is that they are more likely than those with healthy children to rely on TANF, which is a much more precarious source of support than its predecessor, AFDC, because welfare is now time limited in most states. In addition, TANF recipients are subject to work requirements and other restrictions with which mothers who have children in poor health may have difficulty complying. Even in states that grant exemptions from such requirements, mothers with children in poor health may not qualify. It is even possible that mothers do not apply for exemptions because they are unaware that exemptions exist or because they choose to avoid a potentially onerous application process. Housing assistance also is increasingly being provided with strings attached, such as community service requirements, and offers fewer preferences for people with extenuating circumstances. (20) Ultimately, families with unhealthy children may have insufficient resources to invest in their children's health, which may have negative effects on the children's health and economic trajectories.

Our finding of a positive effect of poor child health on reliance on TANF, together with Norberg's (1998) finding that the negative effects on mothers' labor supply increase from birth to five years of age, suggest that it is vitally important that exemptions be not only available but implemented for mothers of children with serious health problems in the current welfare environment of time-limited benefits and work requirements. Since many (but not all) serious child health problems are diagnosed at birth, the hospital of delivery would be a convenient checkpoint for educating mothers with unhealthy children about exemptions from TANF time limits and work requirements and about other public benefits to which they may be entitled.

We conclude with a caveat. It is possible that the short-term effects we found for TANF will not persist over a longer time period because families with infants in poor health who rely on TANF may ultimately move from TANF to SSI. In other words, they may use TANF as a stopgap measure while applying for SSI, which is more difficult to obtain but more generous and permanent than TANF. To the extent that this may be the case, the families' cash assistance status would be less precarious than previously suggested. Although, as indicated earlier, there was a substantial shift of children from welfare to SSI in the early to mid-1990s, the extent of such changes after the PRWORA legislation is not yet known. It is an important question to answer, however, as Duggan and Keamey (2004) recently found that the increased SSI participation among female-headed families before PRWORA resulted in both increased maternal labor supply and decreased child poverty.

Appendix

Bivariate Probit Estimates for TANF Participation at Follow-Up and
Poor Child Health

                                           TANF Participation

                                                  Marginal   Standard
                                    Coefficient    Effect      Error

Child measures
  Reported poor child health         0.52            0.15      0.64
  Child is male                     -0.00           -0.00      0.06
  Mother has other children          0.08            0.02      0.08
  Father has children with
    other mothers                    0.15 **         0.04      0.08

Mother's characteristics
  Medicaid birth                     0.22 ***        0.05      0.08
  Public assistance at baseline      0.70 ***        0.18      0.08
  Age                               -0.08 **        -0.02      0.03
  Age squared                        0.00 **         0.00      0.00
  High school graduate              -0.16 *         -0.04      0.08
  Some college                      -0.34 ***       -0.07      0.11
  College                           -0.68 ***       -0.12      0.15
  Hispanic                           0.09            0.02      0.11
  African-American                   0.31 **         0.07      0.14
  Other nonwhite/non-Hispanic        0.00            0.00      0.19
  Immigrant                         -0.44 ***       -0.09      0.12
  Lived with both parents
    at age 15                       -0.10 **        -0.02      0.04
  Worked within 2 years
    before birth                    -0.25 ***       -0.06      0.06
  Health very good or excellent     -0.09           -0.02      0.06
  Attends religious services
    several times per month         -0.06           -0.01      0.05
  Number of months since
    baseline interview
    (child's age)                   -0.02            0.00      0.01

Father's characteristics
  Age                                0.04 *          0.01      0.02
  Age squared                       -0.00 **        -0.00      0.00
  High school graduate              -0.15 ***       -0.03      0.05
  Some college                      -0.17 **        -0.04      0.09
  College                           -0.18           -0.04      0.18
  Hispanic                          -0.03           -0.01      0.08
  African-American                   0.24 **         0.06      0.10
  Other nonwhite/non-Hispanic        0.26 *          0.07      0.16
  Health very good or excellent     -0.00           -0.00     -0.05
  Health status missing              0.21 **         0.05     -0.10

Parents' relationship at baseline
  Months mother knew father         -0.00           -0.00      0.00
  Cohabiting                         0.45 ***        0.11      0.09
  Romantic or friends                0.70 ***        0.19      0.11
  Rarely/never talk                  0.88 ***        0.28      0.13

City characteristics
  Unemployment rate                  0.02            0.00      0.03
  Average full-time
    female earnings                  0.00            0.00      0.00
  Cost of living                     0.00            0.00      0.00

Identifiers
  Level III NICU beds per
    mile in city of birth in
    fiscal year 1998                    N/A         N/A        N/A
  Per capita expenditure
    on tobacco prevention
    in state in 2002                    N/A         N/A        N/A

Rho                                       -0.14 (p-value = 0.63)
Number of observations                              3717
Log likelihood                                     -2222.87

                                       Reported Poor
                                        Child Health

                                                  Standard
                                    Coefficient    Error

Child measures
  Reported poor child health            N/A         N/A
  Child is male                        0.11 *       0.06
  Mother has other children           -0.15         0.11
  Father has children with
    other mothers                     -0.09         0.08

Mother's characteristics
  Medicaid birth                       0.16         0.10
  Public assistance at baseline        0.02         0.09
  Age                                 -0.01         0.06
  Age squared                          0.00         0.00
  High school graduate                -0.02         0.08
  Some college                        -0.04         0.08
  College                              0.07         0.12
  Hispanic                            -0.36 ***     0.10
  African-American                     0.03         0.10
  Other nonwhite/non-Hispanic          0.05         0.23
  Immigrant                           -0.06         0.14
  Lived with both parents
    at age 15                         -0.02         0.09
  Worked within 2 years
    before birth                      -0.07         0.12
  Health very good or excellent       -0.20 ***     0.08
  Attends religious services
    several times per month            0.15 **      0.07
  Number of months since
    baseline interview
    (child's age)                       N/A         N/A

Father's characteristics
  Age                                  0.06 *       0.04
  Age squared                         -0.00         0.00
  High school graduate                -0.04         0.08
  Some college                        -0.27 ***     0.09
  College                             -0.40 **      0.20
  Hispanic                             0.14         0.12
  African-American                    -0.03         0.13
  Other nonwhite/non-Hispanic         -0.17         0.25
  Health very good or excellent        0.08         0.07
  Health status missing               -0.13         0.11

Parents' relationship at baseline
  Months mother knew father            0.00         0.00
  Cohabiting                           0.15         0.11
  Romantic or friends                  0.31 **      0.14
  Rarely/never talk                    0.12         0.22

City characteristics
  Unemployment rate                     N/A         N/A
  Average full-time
    female earnings                     N/A         N/A
  Cost of living                        N/A         N/A

Identifiers
  Level III NICU beds per
    mile in city of birth in
    fiscal year 1998                  -0.06 *       0.04
  Per capita expenditure
    on tobacco prevention
    in state in 2002                  -0.04 *       0.02

Rho                                 -0.14 (p-value = 0.63)
Number of observations                      3717
Log likelihood                             -2222.87

* Significant at 10% level.

** Significant at 5% level.

*** Significant at 1% level.

This research was supported by Grants #R01-HD-45630 and #R01-HD-35301 from the National Institute of Child Health and Human Development and Grant #KN03-NPC02 from the National Poverty Center at the University of Michigan. We are grateful for valuable input from Janet Currie, William Greene, Michael Grossman, Jennifer Marogi, Magdalena Ostatkiewicz, Ofira Schwartz-Soicher, Erdal Tekin, Kerwin Charles, two anonymous reviewers, and other conference participants at the National Poverty Center at the University of Michigan and the health economics group at the Institute for Health, Health Care Policy, and Aging Research at Rutgers University.

Received January 2005; accepted November 2005.

References

Acs, Geza, and Pamela Loprest. 1999. The effect of disabilities on exits from AFDC. Journal of Policy Analysis and Management 18(1):28-49.

American Hospital Association. 1998. AHA Annual Survey Database FY 1998. Chicago: American Hospital Association.

Black, Dan, Terra McKinnish, and Seth Sanders. 1998. Are AFDC and SSI substitutes? Working Paper, University of Kentucky. Accessed 1 August 2004. Available http://gatton.uky.edu/Academic/Economics/WorkingPapers/BlackssiAFDC.pdf.

Blank, Rebecca M. 1989. The effect of medical need and Medicaid on AFDC participation. The Journal of Human Resources 24(1):54-87.

Blank, Rebecca M. 2002. Evaluating welfare reform in the United States. Journal of Economic Literature XL: 1105-66.

Brandon, Peter D. 2000. Child care utilization among working mothers raising children with disabilities. Journal of Family and Economic Issues 21 (4):343-64.

Brandon, Peter D., and Dennis P. Hogan. 2004. Impediments to mothers leaving welfare: The role of maternal and child disability. Population Research and Policy Review 23:419-36.

Campaign for Tobacco Free Kids (2005). Special Reports, State Tobacco Settlement (Appendix A). Washington DC. Accessed 5 May 2005. Available http://www.tobaccofreekids.org/reports/settlements/2005/history.pdf.

Chomitz, Virginia, Lillian W. Cheung, and Elice Lieberman. 1995. The role of lifestyle in preventing low birth weight. The Future of Children 5:121-38.

Corman, Hope, and Robert Kaestner. 1992. The effects of child health on marital status and family structure. Demography 29(3):389-408.

Corman, Hope, Kelly Noonan, and Nancy E. Reichman. (2005). Mothers' labor supply in fragile families: The role of child health. Eastern Economic Journal 31:601-6.

Dahl, Gordon, and Enrico Moretti. 2004. The demand for sons: Evidence from divorce, fertility, and shotgun marriage. NBER Working Paper No. 10281.

Duggan, Mark G., and Melissa S. Kearney. 2004. The rise in SSI participation among children: Assessing the impact on poverty and labor supply. Unpublished paper, University of Maryland and Wellesley College.

Fertig, Angela. 2004. Healthy Baby, Healthy Marriage? The effect of children's health on divorce. NBER Working Paper No. 2003-15.

Garrett, Bowen, and Sherry Glied. 2000. Does state AFDC generosity affect child SSI participation? Journal of Policy Analysis and Management 19(2):275-95.

Hack, Maureen, Nancy Klein, and H. Gerry Taylor. 1995. Long-term developmental outcomes of low birth weight infants. The Future of Children 5:176-96.

Hunt, Louise, Mary Schulhof, and Steven Holmquist. 1998. Summary of the Quality Housing and Work Responsibility Act of 1998 (Title V of P.L. 105-276), Office of Policy, Program and Legislative Initiatives, Office of Public and Indian Housing. Accessed 20 November 2004. Available http://www.hud.gov/offices/pih/phr/about/titlev.pdf.

Joesch, Jutta M., and Ken R. Smith. 1997. Children's health and their mothers' risk of divorce or separation. Social Biology 44(3-4): 159-69.

Kaestner, Robert, and Neeraj Kaushal. 2003. The effect of welfare reform on health insurance coverage of low-income women and children. Journal of Health Economics 22(6):959-81.

Kubik, Jeffrey D. 1999. Incentives for the identification and treatment of children with disabilities: The Supplemental Security Income program. Journal of Public Economics 73:187-215.

Mauldon, Jane. 1992. Children's risks of experiencing divorce and remarriage: Do disabled children destabilize marriages? Population Studies 46(2):349-62.

Meyers, Marcia K., H. E. Brady, and E. Y. Seto. 2000. Expensive children in poor families: The intersection of childhood disabilities and welfare. San Francisco, CA: Public Policy Institute of California.

Moffitt, Robert A. 2002. The temporary assistance for needy families program. NBER Working Paper No. 8749.

Norberg, Karen. 1998. The effects of daycare reconsidered. NBER Working Paper No. W6769.

Pati, Susmita, Diana Romero, and Wendy Chavkin. 2002. Changes in use of health insurance and food assistance programs in medically underserved communities in the era of welfare reform: An urban study. American Journal of Public Health 92(9):1441-5.

Powers, Evan T. 2003. Children's health and maternal work activity: Static and dynamic estimates under alternative disability definitions. Journal of Human Resources 38(3):522-56.

Reichman, Nancy E., Hope Corman, and Kelly Noonan. 2003. Effects of child health on parents' relationship status. NBER Working Paper No. 9610.

Reichman, Nancy E., Hope Corman, and Kelly Noonan. 2004. Effects of child health on parents' relationship status. Demography 41(3):569-84.

Reichman, Nancy E., Julien O. Teitler, Irvin Garfinkel, and Sara McLanahan. 2001. Fragile families: Sample and design. Children and Youth Services Review 23(4/5):303-26.

Schmidt, Lucie, and Purvi Sevak. 2004. AFDC, SSI, and welfare reform aggressiveness: Caseload reductions versus caseload shifting. Journal of Human Resources 39(3):792-812.

Smith, Lauren A., Dina Romero, P. R. Wood, N. A. Wampler, Wendy Chavkin, and P. H. Wise. 2002. Employment barriers among welfare recipients and applicants with chronically ill children. American Journal of Public Health 92(9):1453-7.

Solon, Gary, Mary Corcoran, Roger H. Gordon, and D. Laren. 1988. Sibling and intergenerational correlations in welfare program participation. The Journal of Human Resources 23(3):388-96.

U.S. Census (2000). QuickFacts. Accessed 5 May 2005. Available http://quickfacts.census.gov/qfd/.

Wise, P. H., N. A. Wampler, Wendy Chavkin, and Diana Romero. 2002. Chronic illness among poor children enrolled in the temporary assistance for needy families program. American Journal of Public Health 92(9):1458-61.

Wolfe, Barbara L., and Steven C. Hill. 1995. The effect of health on the work effort of single mothers. The Journal of Human Resources 30(1):42-62.

(1) See Blank (2002) and Moffitt (2002) for reviews of the major changes in U.S. welfare policy during the 1990s and related research.

(2) Because the FFCWB study is a birth cohort study, the births are not necessarily first births.

(3) Background on the research design of the FFCWB study is available in Reichman et al. (2001).

(4) The medical records data collection is ongoing as part of an "add-on" study to the FFCWB survey.

(5) Using very low-birth weight (<3.3 pounds or 1500 grams) as a cutoff in constructing this measure would have resulted in too few cases of poor child health for meaningful analysis.

(6) The individual measures add up to greater than 5% because some children fit more than one criterion.

(7) The correlation of birth weights from the two sources in our data is 0.98.

(8) The sample is not large enough to separately analyze specific conditions, such as Down syndrome.

(9) Recent studies (for example, Dahl and Moretti [2004]) indicate that fathers of sons tend to have stronger commitment to their families than fathers of daughters.

(10) Data limitations make it impossible to ascertain whether the father had any children with another partner at the time of the baseline interview.

(11) See Solon et al. (1988) for a discussion about the intergenerational aspects of welfare dependence.

(12) Results for covariates are not presented in Table 4, but are available from the authors upon request. The covariate estimates in models using low birth weight and abnormal newborn conditions are very similar to those presented in Table 3 that used reported pool child health.

(13) Baseline public assistance is likely endogenous and the coefficient of that variable may therefore be biased. The coefficients of poor child health are very similar in models with and without the baseline public assistance variable, and the difference between the two coefficients is not statistically significant.

(14) The additional variables are listed in Table lb and described earlier in the text.

(15) We also estimated full sets of models (corresponding to Tables 2-4) that included state fixed effects and obtained similar results to those presented that do not include state fixed effects.

(16) For a more detailed description of this estimation strategy, see Reichman, Corman, and Noonan (2003).

(17) Passing these two tests indicates that the equation has been overidentified.

(18) Data on the number of NICU In beds were obtained from the American Hospital Association (1998). Data on square mileage and population were obtained from the U.S. Census (2000). Data on state-level tobacco prevention expenditures were obtained from Tobacco Free Kids (2005).

(19) It seems unlikely that the focal child's severe poor health would affect the likelihood that another member of the family receives SSI. However, we cannot rule out that possibility.

(20) See Hunt, Schulhof, and Holmquist (1998), which describes recent mandatory community service activity requirements and the repeal of mandatory Federal preferences for public housing and public housing agency-administered section 8 assistance programs.

Nancy E. Reichman, * Hope Corman, ([dagger]) and Kelly Noonan ([double dagger])

* Robert Wood Johnson Medical School, University of Medicine and Dentistry of New Jersey, 97 Paterson St., Room 435, New Brunswick, NJ 08903 USA; E-mail [e-mail]; corresponding author.

([dagger]) Rider University and National Bureau of Economic Research, 2083 Lawrenceville Road, Lawrenceville, NJ 08648 USA; E-mail [e-mail].

([double dagger]) Rider University and National Bureau of Economic Research, 2083 Lawrenceville Road, Lawrenceville, NJ0 8648 USA; E-mail [e-mail].

Table 1. Sample Characteristics

                                              Full Sample
                                              (N = 3717)

Sources of public support (reported at follow-up interview)
  TANF                                           0.24
  SSI                                            0.03
  Food stamps                                    0.36
  WIC                                            0.72
  Medicaid                                       0.56
  Housing                                        0.20

Child measures
  Reported poor child health                     0.05
  Low birth weight                               0.09
  Child is male                                  0.53
  Mother has other children                      0.61
  Father has children with other mothers         0.34

Mother characteristics
  Medicaid birth                                 0.63
  Public assistance at baseline                  0.35
  Age (years)                                25.15 (6.04)
  Less than high school (a)                      0.33
  High school graduate                           0.30
  Some college                                   0.25
  College graduate                               0.12
  White (a)                                      0.22
  African-American                               0.47
  Hispanic                                       0.27
  Other race/ethnicity                           0.04
  Immigrant                                      0.16
  Lived with both parents at age 15              0.43
  Worked within 2 years before birth             0.81
  Health very good or excellent                  0.67
  Attends religious services
    several times per month                      0.39
  Months since baseline interview
    (child's age)                             14.9 (3.38)

Father characteristics
  Age (years)                                 27.7 (7.16)
  Less than high school (a)                      0.32
  High school graduate                           0.34
  Some college                                   0.23
  College graduate                               0.11
  White (a)                                      0.20
  African-American                               0.49
  Hispanic                                       0.27
  Other race/ethnicity                           0.04
  Health very good or excellent                  0.63
  Health status missing                          0.12

Parents' relationship at baseline
  Months mother knew father                  59.07 (56.77)
  Married (a)                                    0.26
  Cohabiting                                     0.38
  Romantic or friends                            0.31
  Rarely/never talk                              0.05

City characteristics
  Unemployment rate                           5.38 (1.93)
  Average full-time female
    earnings (dollars)                    28,332.7 (3,958.68)
  Cost of living index                      115.59 (24.04)

Identifiers
  Level III NICU beds per
    mile in city of birth in
    fiscal year 1998                          1.22 (.98)
  Per capita expenditure on
    tobacco prevention in
    state in 2002                             2.62 (1.64)

                                              First Births
                                               (N = 1440)

Sources of public support (reported at follow-up interview)
  TANF                                            0.19
  SSI                                             0.03
  Food stamps                                     0.26
  WIC                                             0.71
  Medicaid                                        0.53
  Housing                                         0.14

Child measures
  Reported poor child health                      0.05
  Low birth weight                                0.10
  Child is male                                   0.54
  Mother has other children                          0
  Father has children with other mothers          0.28

Mother characteristics
  Medicaid birth                                  0.58
  Public assistance at baseline                   0.23
  Age (years)                                 22.98 (5.51)
  Less than high school (a)                       0.29
  High school graduate                            0.27
  Some college                                    0.29
  College graduate                                0.15
  White (a)                                       0.27
  African-American                                0.41
  Hispanic                                        0.28
  Other race/ethnicity                            0.05
  Immigrant                                       0.17
  Lived with both parents at age 15               0.46
  Worked within 2 years before birth              0.86
  Health very good or excellent                   0.71
  Attends religious services
    several times per month                       0.37
  Months since baseline interview
    (child's age)                             14.56 (3.21)

Father characteristics
  Age (years)                                 25.7 (6.50)
  Less than high school (a)                       0.30
  High school graduate                            0.32
  Some college                                    0.25
  College graduate                                0.13
  White (a)                                       0.25
  African-American                                0.42
  Hispanic                                        0.28
  Other race/ethnicity                            0.05
  Health very good or excellent                   0.65
  Health status missing                           0.12

Parents' relationship at baseline
  Months mother knew father                  45.28 (47.74)
  Married (a)                                     0.23
  Cohabiting                                      0.36
  Romantic or friends                             0.35
  Rarely/never talk                               0.06

City characteristics
  Unemployment rate                           5.26 (1.92)
  Average full-time female
    earnings (dollars)                    28,439.54 (3,992.83)
  Cost of living index                       115.60 (24.07)

Identifiers
  Level III NICU beds per
    mile in city of birth in
    fiscal year 1998                           1.28 (.99)
  Per capita expenditure on
    tobacco prevention in
    state in 2002                             2.64 (1.66)

                                                Higher
                                              Order Births
                                               (N = 2277)

Sources of public support (reported at follow-up interview)
  TANF                                            0.27
  SSI                                             0.04
  Food stamps                                     0.43
  WIC                                             0.73
  Medicaid                                        0.58
  Housing                                         0.23

Child measures
  Reported poor child health                      0.05
  Low birth weight                                0.09
  Child is male                                   0.53
  Mother has other children                          1
  Father has children with other mothers          0.37

Mother characteristics
  Medicaid birth                                  0.65
  Public assistance at baseline                   0.43
  Age (years)                                 26.52 (5.96)
  Less than high school (a)                       0.34
  High school graduate                            0.33
  Some college                                    0.23
  College graduate                                0.10
  White (a)                                       0.21
  African-American                                0.50
  Hispanic                                        0.26
  Other race/ethnicity                            0.03
  Immigrant                                       0.14
  Lived with both parents at age 15               0.41
  Worked within 2 years before birth              0.77
  Health very good or excellent                   0.64
  Attends religious services
    several times per month                       0.40
  Months since baseline interview
    (child's age)                             15.06 (3.48)

Father characteristics
  Age (years)                                 29.0 (7.27)
  Less than high school (a)                       0.35
  High school graduate                            0.35
  Some college                                    0.21
  College graduate                                0.09
  White (a)                                       0.16
  African-American                                0.53
  Hispanic                                        0.27
  Other race/ethnicity                            0.04
  Health very good or excellent                   0.62
  Health status missing                           0.12

Parents' relationship at baseline
  Months mother knew father                   67.8 (60.20)
  Married (a)                                     0.28
  Cohabiting                                      0.39
  Romantic or friends                             0.29
  Rarely/never talk                               0.04

City characteristics
  Unemployment rate                           5.46 (1.94)
  Average full-time female
    earnings (dollars)                    28,265.13 (3,936.31)
  Cost of living index                       115.58 (24.03)

Identifiers
  Level III NICU beds per
    mile in city of birth in
    fiscal year 1998                           1.18 (.97)
  Per capita expenditure on
    tobacco prevention in
    state in 2002                             2.61 (1.63)

Standard deviations are in parentheses.

(a) Indicates omitted category in multivariate analyses.

Table 2. Selected Characteristics of Medical Records Subsample

                         Full Medical                     Higher Order
                        Records Sample    First Births       Births
                          (N = 1845)       (N = 692)       (N = 1156)

Sources of public support (reported at follow-up interview)
  TANF                       0.24             0.20            0.27
  SSI                        0.02             0.02            0.02
  Food stamps                0.36             0.27            0.42
  WIC                        0.75             0.73            0.76
  Medicaid                   0.56             0.53            0.58
  Housing                    0.20             0.14            0.23

Mother's pre-existing conditions
  Anemia                     0.23             0.20            0.25
  Lung disease               0.13             0.11            0.14
  Underweight (BMI <
    18.5)                    0.04             0.06            0.03
  Morbidly obese (BMI
    [greater than or
    equal to] 39)            0.02             0.01            0.02
  Other health
    condition                0.17             0.15            0.18
  Previous low birth
    weight or preterm
    infant                   0.08        Not applicable       0.12

Table 3. Probit Estimates of TANF Participation

                                                 Full Sample

                                                    Marginal   Standard
                                      Coefficient    Effect     Error

Child measures
  Reported poor child health           0.23 **       0.06        0.10
  Child is male                        0.00          0.00        0.06
  Mother has other children            0.08          0.02        0.08
  Father has children with other
    mothers                            0.15 *        0.04        0.08

Mother's characteristics
  Medicaid birth                       0.22 ***      0.05        0.08
  Public assistance at baseline        0.71 ***      0.18        0.08
  Age                                 -0.08 **      -0.02        0.03
  Age squared                          0.00 **       0.00        0.00
  High school graduate                -0.16 *       -0.04        0.08
  Some college                        -0.34 ***     -0.07        0.11
  College graduate                    -0.68 ***     -0.12        0.15
  Hispanic                             0.08          0.02        0.11
  Other nonwhite/non-Hispanic          0.00          0.00        0.19
  Immigrant                           -0.44 ***     -0.09        0.12
  Lived with both parents at age 15   -0.10 **      -0.02        0.04
  Worked within 2 years before
    birth                             -0.25 ***     -0.06        0.06
  Health very good or excellent       -0.10 *       -0.02        0.05
  Attends religious services
    several times per month           -0.05         -0.01        0.05
  Months since baseline interview
    (child's age)                     -0.02         -0.00        0.01

Father's characteristics
  Age                                  0.04 *        0.01        0.02
  Age squared                          0.00 **       0.00        0.00
  High school graduate                -0.15 ***     -0.03        0.05
  Some college                        -0.18 **      -0.04        0.08
  College graduate                    -0.19         -0.04        0.19
  African-American                     0.24 **       0.06        0.10
  Other nonwhite/non-Hispanic          0.25          0.07        0.16
  Health very good or excellent        0.00          0.00        0.05
  Health status missing                0.20 **       0.05        0.10

Parents' relationship at baseline
  Months mother knew father            0.00          0.00        0.00
  Cohabiting                           0.45 ***      0.11        0.09
  Romantic or friends                  0.71 ***      0.19        0.11
  Rarely/never talk                    0.88 ***      0.28        0.13

City characteristics
  Unemployment rate                    0.02          0.00        0.03
  Average full-time female earnings    0.00          0.00        0.00
  Cost of living                       0.00          0.00        0.00

Number of observations                 3717
Log likelihood                        -1517.63

                                                  First Birth

                                                    Marginal   Standard
                                      Coefficient    Effect     Error

Child measures
  Reported poor child health           0.10          0.02        0.17
  Child is male                       -0.01         -0.00        0.10
  Mother has other children             N/A           N/A         N/A
  Father has children with other
    mothers                            0.03          0.01        0.11

Mother's characteristics
  Medicaid birth                       0.30 **       0.05        0.13
  Public assistance at baseline        0.58 ***      0.13        0.09
  Age                                 -0.04         -0.01        0.11
  Age squared                          0.00          0.00        0.00
  High school graduate                -0.03         -0.01        0.12
  Some college                        -0.25         -0.04        0.16
  College graduate                    -0.79 **      -0.10        0.33
  Hispanic                             0.33 *        0.07        0.20
  Other nonwhite/non-Hispanic          0.34          0.07        0.36
  Immigrant                           -0.68 ***     -0.09        0.24
  Lived with both parents at age 15   -0.16 **      -0.03        0.08
  Worked within 2 years before
    birth                             -0.08         -0.01        0.10
  Health very good or excellent       -0.07         -0.01        0.09
  Attends religious services
    several times per month           -0.05         -0.01        0.11
  Months since baseline interview
    (child's age)                     -0.02          0.00        0.02

Father's characteristics
  Age                                  0.03          0.01        0.05
  Age squared                         -0.00         -0.00        0.00
  High school graduate                -0.11         -0.02        0.12
  Some college                        -0.22         -0.04        0.15
  College graduate                    -0.43         -0.06        0.28
  African-American                     0.30 *        0.06        0.18
  Other nonwhite/non-Hispanic          0.13          0.03        0.20
  Health very good or excellent        0.03          0.01        0.11
  Health status missing                0.43 ***      0.10        0.13

Parents' relationship at baseline
  Months mother knew father            0.00          0.00        0.00
  Cohabiting                           0.39 *        0.08        0.23
  Romantic or friends                  0.62 **       0.13        0.26
  Rarely/never talk                    0.78 ***      0.21        0.28

City characteristics
  Unemployment rate                    0.09 **       0.02        0.04
  Average full-time female earnings    0.00 ***      0.00        0.00
  Cost of living                      -0.01         -0.00        0.00

Number of observations                 1440
Log likelihood                        -531.83

                                            Higher Order Births

                                                    Marginal   Standard
                                      Coefficient    Effect     Error

Child measures
  Reported poor child health           0.29 **       0.08       0.14
  Child is male                        0.01          0.00       0.06
  Mother has other children             N/A           N/A        N/A
  Father has children with other
    mothers                            0.22 **       0.06       0.10

Mother's characteristics
  Medicaid birth                       0.15 *        0.04       0.09
  Public assistance at baseline        0.79 ***      0.22       0.09
  Age                                 -0.12 ***     -0.03       0.03
  Age squared                          0.00 ***      0.00       0.00
  High school graduate                -0.22 **      -0.05       0.08
  Some college                        -0.38 ***     -0.09       0.12
  College graduate                    -0.59 ***     -0.12       0.18
  Hispanic                            -0.10         -0.03       0.23
  Other nonwhite/non-Hispanic         -0.23         -0.06       0.26
  Immigrant                           -0.35 ***     -0.08       0.11
  Lived with both parents at age 15   -0.06         -0.02       0.07
  Worked within 2 years before
    birth                             -0.35 ***     -0.10       0.08
  Health very good or excellent       -0.12 **      -0.03       0.05
  Attends religious services
    several times per month           -0.06         -0.02       0.08
  Months since baseline interview
    (child's age)                     -0.01         -0.00       0.01

Father's characteristics
  Age                                  0.04 *        0.01       0.03
  Age squared                         -0.00 **      -0.00       0.00
  High school graduate                -0.18 ***     -0.05       0.06
  Some college                        -0.18 *       -0.05       0.10
  College graduate                    -0.06         -0.01       0.17
  African-American                     0.23          0.06       0.17
  Other nonwhite/non-Hispanic          0.36          0.11       0.26
  Health very good or excellent       -0.01         -0.00       0.07
  Health status missing                0.07          0.02       0.14

Parents' relationship at baseline
  Months mother knew father           -0.00 *        0.00       0.00
  Cohabiting                           0.49 ***      0.13       0.08
  Romantic or friends                  0.75 ***      0.22       0.11
  Rarely/never talk                    0.99 ***      0.34       0.19

City characteristics
  Unemployment rate                   -0.02         -0.01       0.03
  Average full-time female earnings   -0.00         -0.00      -0.00
  Cost of living                       0.01 **       0.00       0.00

Number of observations                 2277
Log likelihood                        -963.07

* Significant at 10% level.

** Significant at 5% level.

*** Significant at 1% level.

Table 4. Probit Estimates of TANF Participation Using Alternative
Measures of Poor Child Health

                                   Full Sample

                                     Marginal   Standard
                       Coefficient    Effect     Error

Low birth weight (a)      0.10         0.02       0.08
Abnormal
  newborn
  condition (b)           0.24         0.06       0.18

                                     First Birth

                                     Marginal   Standard
                       Coefficient    Effect     Error

Low birth weight (a)      -0.07       -0.01       0.14
Abnormal
  newborn
  condition (b)            0.35        0.06       0.42

                             Higher Order Births

                                     Marginal   Standard
                       Coefficient    Effect     Error

Low birth weight (a)   0.17 **         0.05       0.08
Abnormal
  newborn
  condition (b)        0.32            0.09       0.22

(a) Includes full sample (N = 3717): 1440 first births and 2277
higher order births.

(b) Includes medical record subsample (N= 1430): 544 first births and
886 higher order births.

* Significant at 10% level.

** Significant at 5% level.

*** Significant at 1% level.

Table 5. Probit Estimates of TANF Participation Using Alternative
Measures of Poor Child Health and Additional Covariates (Medical
Records Subsample)

                                  Full Sample

                                     Marginal   Standard
                       Coefficient    Effect     Error
Reported poor
  child health (a)     0.36 **         0.10       0.14
Low birth weight (a)   0.12            0.03       0.14
Abnormal
  newborn
  condition (b)        0.29            0.08       0.21

                                   First Birth

                                     Marginal   Standard
                       Coefficient    Effect     Error
Reported poor
  child health (a)        0.32         0.07       0.24
Low birth weight (a)      0.16         0.03       0.24
Abnormal
  newborn
  condition (b)           0.26         0.04       0.41

                             Higher Order Births

                                     Marginal   Standard
                       Coefficient   Effect      Error
Reported poor
  child health (a)       0.42 **       0.13       0.19
Low birth weight (a)      0.11         0.03       0.11
Abnormal
  newborn
  condition (b)           0.36         0.10       0.23

(a) N = 1845: 692 first births and 1156 higher order births.

(b) N = 1386: 532 first births and 857 higher order births.

* Significant at 10% level.

** Significant at 5% level.

*** Significant at 1% level.

Table 6. Marginal Effects of Poor Child Health on Receipt of SSI,
Food Stamps, WIC, Medicaid, and Housing

                               SSI      Food Stamps   WIC

Reported poor child health   0.11 ***    0.01         0.00
Low birth weight             0.03 ***    0.00         0.01
Abnormal newborn condition   0.17 ***   -0.04         0.02

                             Medicaid   Housing    N

Reported poor child health   0.08       0.04 **   3717
Low birth weight             0.05 *     0.00      3717
Abnormal newborn condition   0.20 **    0.10      1430

* Significant at 10% level.

** Significant at 5% level.

*** Significant at 1% level.