Does the length of maternity leave affect maternal health?

Markowitz, Sara

1. Introduction

In the United States, 51% of mothers of infants currently work outside the home (Bureau of Labor Statistics 2003). Among mothers who return to work during the first year after childbirth, almost all return to work by the third month (Klerman and Leibowitz 1994; Cantor et al. 2001). The large number of infants with employed mothers has led to an increased interest in the effects of maternal employment during infancy on child health and development. Recent studies suggest that some forms of maternal employment during the child's first year are detrimental to children's cognitive development and lead to more behavioral problems (Blau and Grossberg 1992; BrooksGunn, Han, and Waldfogel, 2002; Waldfogel, Han, and Brooks-Gunn 2002; Baum 2003). These studies imply that longer maternal leaves will benefit children.

Previous research, however, does not consider that in addition to improving children's health and development, longer maternity leaves also may affect the health and well-being of mothers. A few correlational studies in the public health literature show that women who are employed postpartum or who return to work soon after childbirth experience more mental and physical health symptoms than other women (Gjerdingen et al. 1993, 1995; Hyde et al. 1995), perhaps because of increased stress and obligations. While the detrimental effects of physical and mental health problems to the mother are obvious, these conditions also may affect the child and other family members through emotional and financial distress. We know very little about this aspect of maternal employment despite the large number of women in the United States who currently balance a job outside the home with the care of a young infant (Hyde 1995).

From a policy perspective, it is useful to consider the effect of maternity leave length on both mothers and children. Concerns about the health of infants and postpartum women were motivating forces behind the Family and Medical Leave Act (FMLA) of 1993. The case for longer leaves is bolstered if longer leaves benefit mothers as well as children. In the case that longer leaves have neutral or detrimental effects on maternal health, this information still is needed to inform the debate over family leave policy. To date, however, despite a number of recent studies on maternal employment and child health, there is little empirical evidence regarding whether longer maternity leave affects maternal health (Hyde 1995). This evidence is still needed today despite the passage of the FMLA because states currently are passing or are considering legislation that would provide paid family leave. This policy change would likely increase the length of maternity leave but at a cost to states, employees, and businesses. Without information about the health impact of longer maternal leave after childbirth, it is difficult to weigh the costs and benefits of these proposed state-level policy changes.

This paper investigates how the length of maternity leave affects maternal health in a sample of mothers who returned to work after childbirth. Data come from the National Maternal and Infant Health Survey (NMIHS) of 1988. This survey is particularly useful because it was conducted before the FMLA was enacted in 1993, allowing us to use empirical methods that take advantage of pre-FMLA variation in maternal leave policies across states. Maternal health is represented by three measures. As discussed further here, the first two examine depressive symptoms using the Center for Epidemiological Studies Depression (CES-D) Scale, a widely used screening tool for depression. The third measure of maternal health represents overall health and is a dummy variable indicating whether the mother had at least three outpatient visits for any health problems (mental or physical) during the first six months after childbirth. We estimate baseline models using ordinary least squares (OLS) methods and then address the potential endogeneity of the return-to-work decision using instrumental variables (IV) methods.

The results indicate that among employed mothers of infants, delaying the return to work decreases the number of depressive symptoms. Holding other factors constant, a one-week increase in the length of maternal leave from work would reduce a scale of depressive symptoms on average by 6-7%; however, it is not clear whether this reduction has clinical significance because we find only weak evidence that the length of maternity leave is significantly associated with a reduction in the probability of meeting a threshold of depressive symptoms that is indicative of clinical depression. We also find a negative but statistically insignificant association between the length of maternal leave and having had at least three postpartum outpatient visits for mental or physical health problems. These findings contribute to the growing literature on maternal leave policy, which focuses primarily on the benefits of leave for child health and development, by evaluating the influence of longer maternal leave on the health of mothers.

2. Returning to Work and Maternal Health

To the best of our knowledge, no previous study in the economics literature has explored the effect of the length of maternity leave on maternal well-being. In the economics literature, most of the research on maternal leave has focused on the impact of leave and leave policies on labor market outcomes, such as employment, wages and job continuity (Waldfogel 1998; Klerman and Leibowitz 1999), and child health and development (Winegarden and Bracy 1995; Ruhm 2000; Baum 2003). These latter studies suggest that longer maternity leave has positive effects on children's physical health (proxied by mortality) and cognitive development.

Winegarden and Bracy (1995) and Ruhm (2000) use time-series data from European countries to study the effect of paid maternal leave on child health. Both Winegarden and Bracy and Ruhm find that longer paid leave is associated with reductions in infant mortality; Ruhm additionally finds that longer maternal leave is associated with lower rates of young child mortality. Baum (2003), using data from the National Longitudinal Survey of Youth, demonstrates that returning to work within the first three months of life is associated with lower cognitive test scores during childhood. These studies suggest that longer maternal leave after childbirth may benefit child health and development.

A few studies from other disciplines have explored the impact of returning to work on the mother's health. In regard to physical health, employed postpartum women have higher rates of respiratory infections, breast symptoms, and gynecologic problems compared to postpartum women who are not employed (Gjerdingen et al. 1993, 1995). This research on physical health is based on a sample of 436 first-time mothers in Minnesota. In regard to mental health, there is some mixed evidence that among employed mothers, returning to work earlier increases depressive symptoms. Hyde et al. (1995), for example, uses a sample of 570 mostly white mothers in Wisconsin to explore the postpartum employment experience. They find that among mothers who are back at work four months postpartum, a short length of maternal leave increases the probability of depression but only among mothers who also have marital concerns and mothers who feel their jobs are unrewarding. Gjerdingen and Chaloner (1994), based on a sample of 436 married, employed, first-time mothers in Minnesota, find that returning to work within 24 weeks alter childbirth, as well as longer work hours, is associated with poor mental health. These studies are based on small, nonrepresentative samples. Moreover, it is not clear whether the association between shorter maternity leave and increased depressive symptoms is causal.

McGovern et al. (1997) address some of these problems by accounting for the possibility that the timing of the return-to-work decision is endogenous. They find that maternity leave length has a positive effect on a mother's well-being, measured at about seven months postpartum using a generic measure of mental health, vitality, and role function. As identifying instruments, these researchers use a set of variables that measure the infant's health endowment (birth weight and gestation, congenital anomalies), the infant's race, health insurance, maternal leave policies, child care arrangements, and job characteristics. These variables are shown in the analysis to be reasonably adequate predictors of maternal leave length. However, it seems unlikely that they can be validly left out of the maternal health equation. For example, there is evidence from other studies that infant health and child care arrangements affect maternal stress and depression (Gjerdingen et al. 1995; Mandl et al. 1999: McLennan, Kotelchuck, and Cho 2001). No results from overidentification tests are shown to justify these exclusions.

The present study addresses the endogeneity problem with a different set of instruments. We use state-level labor market conditions and state-level maternal leave policies as identifying instruments rather than the potentially endogenous individual characteristics used by McGovern et al. (1997). State-level variables are more likely than individual-level variables to be exogenous to the model. We test the set of identifying instruments to gauge whether they can be validly left out of the maternal health equation and to determine whether they are reasonably strong predictors of the length of maternal leave from work. All models are estimated using several sets of independent variables to see whether the estimates are sensitive to the variables included in the model, some of which may be endogenous.

We use data from the NMIHS, which includes a national, racially diverse sample of mothers. The McGovern et al. sample is limited to the Twin Cities region of Minnesota, and 91% of the sample respondents are white. The NMIHS is a national survey that oversampled blacks and low-birth-weight infants, and our sample respondents come from all 50 states. The analysis sample used in the paper is not necessarily representative of employed mothers in 1988. Consequently, the results should be generalized with caution.

3. Modeling the Return-to-Work and Maternal Health Relationship

This paper is based on the hypothesis that among women who were employed while pregnant and who return to work during the first six months of the child's life, longer leave from work will influence maternal health. For some mothers, time at work may be more complementary to health than time at home; for other mothers, the opposite may be true. Intuitively, the direction of the impact is indeterminate, and it remains an empirical question.

The study focuses on estimating the following equation:

H = [b.sub.0] + [b.sub.1]E + [b.sub.2]X + u + e. (1)

This equation is specific to the mother/child dyad. The dependent variable H is a measure of maternal health, which in our case is represented by two measures of depressive symptoms and a measure indicating whether the mother had at least three outpatient visits for mental or physical health problems during the first six months after childbirth.

The main independent variable of interest is E, the length of time after the birth of the child when the mother returns to work. We hypothesize that returning to work alters maternal health, generating measurable differences in health status among women with varying durations of time away from the labor force. The coefficient on E shows the direction and magnitude of this effect.

The vector X includes observed maternal factors that may affect maternal health, such as the mother's age, marital status, number of children, education, occupation, and income, and observed child-specific factors that may influence maternal health, such as the child's health endowment. Specific details about the variables included are discussed here. In addition to these measured variables, there may exist unobserved, individual-level factors that are associated with both health status and employment decisions. These unobserved factors are represented by u in Equation 1, and e is a random disturbance term.

Initially, a standard OLS model is used to estimate Equation 1. Estimating Equation 1 by OLS, however, can lead to biased and inconsistent estimates if a problem of reverse causality exists (e.g., postpartum health affects the timing of returning to work) or if unobserved, mother-specific factors exist that influence both maternal health and return-to-work decisions (e.g., u is correlated with E and H). Examples of unobserved, mother-specific factors might include the degree of stress in home or work environment, social support, or marital discord. The direction of any bias, however, is unknown--mothers experiencing depressive symptoms and other health problems may return to work later because of their health (see Lennon, Blome, and English 2001), but this may not necessarily be the case.

We attempt to account for this problem using IV methods, which purge the potentially endogenous return-to-work variable of its correlation with the error term. We use the Durbin-Wu-Hausman test to test whether the endogeneity of maternity leave length with respect to maternal health affects the consistency of the estimates. The results from this test are useful in deciding which estimates, OLS or IV, are preferred. Additionally, the validity of the overidentifying restrictions is tested, and the predictive power of the identifying instrumental variables is assessed. In all models, t-statistics are computed from Huber-White standard errors with adjustment for clustering on state of residence. This adjustment helps account for any correlation in the error term among residents of the same state.

The OLS and IV models initially are estimated with a set of basic covariates that are exogenous from the mother's perspective (such as age and race). We then estimate the same models with two larger sets of covariates that may include potentially endogenous variables (such as smoking and occupation) that are treated as exogenous. Excluding these variables could potentially lead to a spurious association between maternity leave length and maternal health in the OLS models. Comparing results across these models allows us to gauge whether the estimates are sensitive to the inclusion of potentially endogenous covariates.

Even though black respondents and mothers of low-birth-weight babies are oversampled, we do not use sample weights in the regression analyses. The NMIHS sample weights are designed so that researchers can make inferences about the population of women with live births and fetal or infant deaths. We have limited the sample to a very specific group of respondents--women who worked at any point during pregnancy and who had returned to work by the time the infant was six months old. This group accounts for only 7% of the original sample of 26,355 women. Employing the sample weights will not make the results generalizable to the population of employed women. Also, Maddala (1983) shows that the estimation of weighted regressions is not required in the case of exogenous stratification (oversampling based on exogenous regressors such as race), and DuMouchel and Duncan (1983) show that weighted regressions are not appropriate if averages of strata-specific regression coefficients are desired.

4. The NMIHS

This study uses data from the 1988 NMIHS. The objective of the NMIHS was to investigate the determinants of negative pregnancy outcomes (for more information, see U.S. Department of Health and Human Services [USDHHS] 1992). The survey respondents were a national sample of women between 15 and 49 years old who had a pregnancy in 1988. The NMIHS oversampled very low-birth-weight, low-birth-weight, and black intents. Initially, 26,355 women were sampled based on birth certificates, death certificates, and reports of fetal death from 1988. The sample includes 13,417 women who had live births, 4772 women who had fetal deaths, and 8166 women who had infant deaths. This paper uses data only from NMIHS respondents who had live births in 1988 (USDHHS 1992).

Of the 13,417 mothers who had live births, 9953 completed the survey, a response rate of 74%. On average, mothers of live births completed the NMIHS survey 17 months after the child's birth (USDHHS 1992). The NMIHS dealt with item nonresponse by imputing many variables using the hot-deck imputation procedure (for more details about this procedure, see USDHHS 1992). In most cases, this imputation affected less than 1% of respondents, although for a few key variables used in this study, such as income, imputation was performed in more than 10% of cases (USDHHS 1992).

The NMIHS is particularly suitable for this study because it contains detailed information on prenatal and childbirth characteristics (which may be important predictors of length of maternity leave), depression, and the dates when the mother left and returned to employment. Another advantage of using these data is the timing of the survey. Because all infants were born in 1988, the mothers returned to work before the passage of the FMLA of 1993, which mandates 12 weeks of leave for eligible mothers. This feature of the survey is important because we use variation in state-level maternity leave policies to instrument for the length of maternity leave.

Analysis Sample

We limit the sample to eligible respondents of at least 18 years of age who had worked at any point during pregnancy and who had returned to work by the time the infant was six months old. We exclude mothers with infants older than 24 months at the time of the survey, mothers who were no longer employed at the time of the interview, and mothers who were pregnant with another child at the time of the survey. These exclusions reduce the sample size to 1762 mothers.

The reason for limiting the sample to mothers who returned to work within six months is because most mothers who return to work during the first year do so within three months of childbirth (Klerman and Leibowitz 1994). Mothers who return to work within six months are more likely than mothers who return to work later to be returning to their prechildbirth employer; maternity leave statutes pertain only to mothers who return to the same employer after childbirth. By limiting the sample to those returning within six months, we exclude about 296 mothers who met all other sample inclusion criteria (aside from being currently employed). (1) We limit the sample to adults because the focus of the study is employment. Mothers who are currently pregnant with another child are excluded because the new pregnancy may affect their depressive symptoms and health services utilization.

There were 411 mothers who were excluded from the analysis sample because even though they were employed before the child's birth and returned to work within six months, they were no longer employed at the time of the survey. It is possible that these mothers may have experienced health problems after returning to work that caused them to eventually stop working. To gauge whether this issue affected the results shown in the paper, we estimated the OLS and IV models including these mothers. The estimates are very similar to those shown in the paper, although the Hausman test statistic indicates that OLS estimates are preferred (results available on request).

Dependent Variables

CES-D. We focus on depression because this condition is a leading cause of lost years of healthy life as measured by disability-adjusted life years (Murray and Lopez 1996). (2) Furthermore, depression is particularly common among women with young infants, 10-20% of whom develop postpartum depression within six months of delivery (Miller 2002). Postpartum depression is defined as major depression that has its onset during the postpartum period, within four weeks after delivery (American Psychiatric Association 2000). Such depression is not to be confused with postpartum blues, in which new mothers experience short-term increases in emotional reactivity for up to several weeks following the birth of a child (Miller 2002). It is important to note that in this study, we measure mothers' depressive symptoms when their infants are between 6 and 24 months old (as described later in this section), which is somewhat outside the period during which postpartum depression per se is diagnosed.

The NMIHS survey includes the CES-D to measure depressive symptoms. The CES-D is one of the most widely used psychiatric scales. The scale captures symptoms of depression and includes 20 items that focus on mood, somatic problems, interactions with others, and motor functioning, such as "I felt lonely," "my sleep was restless," and "I could not get going." (3) The CES-D has been widely used in studies of postpartum depression (Campbell and Cohn 1991; Civic and Holt 2000; Weinberg et al. 2001; Beeghly et al. 2002, 2003) and has demonstrated excellent psychometric properties in studies involving diverse populations, including postpartum women (Radloff 1977; Husaini et al. 1980).

The respondent is asked to respond to each item in the CES-D scale according to a four-point Likert scale, with higher values corresponding to higher frequency of the item in the past week. For example, for the item "I felt lonely," mothers responded either "less than 1 day" (zero points), "1-2 days" (one point), 3-4 days (two points), or 5-7 days (three points). The final CES-D score is computed by adding the points assigned to each item. The maximum score is 60 (20 items x maximum of three points per item), and a score of 16 or higher is generally considered a likely case of clinically defined depression. Scores on the CES-D between 2 and 12 are considered to be in the normative range (Beeghly et al. 2002). The CES-D scale does not correspond to a diagnosis of major depression according to the Diagnostic and Statistical Manual of Mental Disorders (4th ed.; American Psychiatric Association 2000). It is used primarily as a screening tool for depression, not as a diagnostic tool (Eaton et al. 2003).

Because the CES-D is skewed to the right in these data, we use the natural log of the total CES-D score as a dependent variable in this analysis. All models also were estimated with the CES-D score in its natural units (not logged) to gauge whether the estimates are sensitive to this issue. Results are discussed here. We also consider a dummy variable indicating whether the respondents' CES-D score is 16 or higher. This dummy variable is not equivalent to a psychiatric diagnosis of depression, but it does capture respondents who are experiencing many symptoms of depression or several symptoms with high frequency in the past week (Eaton et al. 2003).

Ideally, we would have liked to measure depression at the same point in time for all mothers (i.e., when all infants were one year old). Unfortunately, this approach is not possible because although all the infants were born in 1988, the mothers did not complete the depression screener when all their infants were a particular age. Because it is possible that the timing of return to work impacts the mother's depressive symptoms differently depending on the current age of the child, we limit the sample to mothers whose children are 24 months old or younger, and we control for the child's age in months. Since the youngest infant in the sample was six months old at the time of the survey, the sample is effectively limited to mothers of infants who are between 6 and 24 months old.

Postpartum Utilization of Outpatient Health Services. We capture another dimension of maternal health using a measure of the mother's postpartum health services utilization. The NMIHS respondents were asked to report the number of outpatient visits they made to a clinic or physician concerning their own physical or mental health during the first six months after childbirth. The American College of Obstetricians and Gynecologists recommends that healthy postpartum women have one outpatient visit for physical health four to six weeks after childbirth (American Academy of Pediatrics and the American College of Obstetricians and Gynecologists 1997). Since the NMIHS oversampled low-birth-weight infants, who may be more likely than normal-weight infants to have had complicated deliveries, outpatient utilization may be higher than normal for the analysis sample even if the mothers are not experiencing postpartum health problems.

As seen in Figure 1, 57% of the respondents have zero or one visit within six months of childbirth, 25% have two visits, and 18% have three or more visits. The median number of outpatient visits in the data is 1, the mean is 1.8, and the standard deviation is 2.5. Because one or two postpartum visits for physical health would be expected for healthy women, we measure maternal health using a dummy variable set equal to one if the mother had at least three outpatient visits during the first six months after childbirth. This variable is a crude indicator of poor postpartum health.

[FIGURE 1 OMITTED]

Clearly, using a measure of health services utilization to proxy maternal health has limitations. Health care utilization is influenced by many factors other than health, and although we can control for many of these factors (i.e., insurance status, health behaviors), some remain unobserved. The use of IV methods addresses the possibility that unobserved factors that are associated with health services utilization are also correlated with the timing of return to work. The NMIHS respondents were not asked about the exact timing of outpatient visits, the reasons for their outpatient visits, or physical health symptoms they experienced during the first six months after childbirth. Moreover, they provided this information on health care utilization retrospectively. (4)

Despite these limitations, considering health services utilization as an outcome in addition to depressive symptoms enhances this analysis for several reasons. First, the timing of returning to work may impact physical as well as mental health, and the utilization measure may capture physical health problems. Second, respondents were asked about health care utilization that took place within the first six months after childbirth. Since approximately 50% of the sample returned to work within eight weeks and over 75% returned within 12 weeks, this outcome captures much of the short-term health impact of returning to work. In contrast, depressive symptoms were measured more than a year (on average) after the mother has returned to work. Focusing on depressive symptoms alone, therefore, would limit the analysis to studying the effect of the timing of returning to work on the long-term depressive symptoms of mothers. Considering both outcomes allows one to study both the short- and the long-term effects of the timing returning to work on maternal health.

Independent Variables

The main independent variable of interest in this study is the number of weeks after giving birth when the mother returns to work. This variable was constructed by NMIHS based on the mother's reported date of return to work and the child's date of birth, which is confidential and not provided to researchers. We do not have information regarding whether the mother returned to the same employer; however, previous research by Klerman and Leibowitz (1999) suggests that during the time period when NMIHS mothers gave birth, most mothers who worked full time during pregnancy continued to work for the same employer after childbirth. We also do not have information regarding whether the mother took paid or unpaid leave from work. During the time period when these data were collected, paid maternity leave was very rare (for more details, see Family and Medical Leave Commission 1995), but it is possible that some mothers used accrued vacation and sick time for maternity leave.

In the analysis sample, the mean child age when the mother returned to work was nine weeks. To proxy the intensity of work, we also include as a covariate whether the mother worked part time (defined as less than 35 hours) at the time of the interview. Although we treat this variable as exogenous, it is possibly endogenous to the model. For this reason, we examine the sensitivity of the estimates to this variable by estimating models with and without part-time work (as well as other job characteristics) as covariates.

In addition to the length of leave from work, maternal depressive symptoms and outpatient services utilization are likely to be influenced by numerous other personal and family-level factors. Previous research suggests that important predictors of postpartum depression include poor prenatal mental and physical health, low social support, concerns about child care arrangements, young maternal age, and low income. (Gjerdingen and Froberg 1991; Gjerdingen et al. 1993; Gjerdingen and Chaloner 1994; Gjerdingen et al. 1995; McGovern et al. 1997; Deal and Holt 1998; Chaudron et al. 2001). To proxy these factors, we include the following variables in all the models: (i) mother's age in years, (ii) mother's education (dummy indicators with high school graduate as the baseline, dropout, some college completed, four-year college degree), (iii) gross household income 12 months prior to the child's birth (measured in approximate quartiles, with the lowest quartile as the baseline), (5) (iv) race/ethnicity (dummy indicators with white as the baseline, black, Hispanic, Asian), (v) a dummy variable indicating whether the mother is married, (vi) the number of other children in the household, (vii) a dummy variable indicating a multiple birth, and (viii) the age of the child in months at the time of the interview.

Ideally, we also would want to control for how long the mother has been back at work at the time of the interview (which ranges from 3 to 24 months) since this factor may affect depressive symptoms. However, it is not possible to control for both the age of the child and how long the mother has been back at work since the age of the child is the sum of how long the mother has been back at work and the length of maternity leave. Thus, the estimated coefficient on maternal leave in our models captures both the effect of the length of leave and the effect of how long the mother has been back at work. Clearly, mothers who take relatively long maternity leaves will have returned to work more recently than mothers who took relatively short maternity leaves. If long maternal leave has health benefits, these benefits may be dampened by the fact that these mothers have returned to work relatively recently and may be experiencing stress related to the transition.

Alternatively, if we had included how long the mother had been back at work as a covariate and excluded the child's age, the estimated coefficient on maternal leave would have captured both the effect of the child's age and the effect of the length of maternity leave. We estimated models this way to examine the sensitivity of the estimates to this change. Qualitatively, the results are very similar. The main change is that the OLS results for the logged CES-D score become statistically significant, and the magnitudes of the IV results are somewhat smaller. These results are available on request.

Previous research suggests that other factors, such as socioeconomic stresses, insurance status, preexisting depression and health problems, and poor infant health, may affect maternal depression as well as health services use (Mandl et al. 1999; McLennan, Kotelchuck, and Cho 2001). For this reason, in some models, we include the following measures of socioeconomic stress: (ix) whether the mother receives welfare and (x) whether the mother has any kind of health insurance. Although we have no direct measures of the mother's physical and mental health before the child was born, we have proxies for prenatal and preconception health behaviors that may be correlated with her health status at the time. These proxies are (xi) whether the mother smoked during pregnancy, (xii) whether the mother initiated prenatal care during the first trimester, (xiii) whether the mother exercised regularly before pregnancy, and (xiv) whether the mother was a daily smoker before pregnancy. Finally, to proxy the mother's prenatal health and child's initial health endowment, we include (xv) whether the mother was advised by a doctor to stay in bed for at least a week during the pregnancy, (xvi) whether the child was born prematurely (before 37 weeks' gestation), and (xvii) whether the child was low birth weight (less than or equal to 2500 grams). It is arguable as to whether these variables are endogenous to the return-to-work decision. By both excluding and including this set of variables, we are able to gauge the sensitivity of the return to work coefficient to these factors (items 9-17) in the OLS regressions.

Because previous work shows that employment factors, work intensity, and child care arrangements are associated with maternal postpartum depression and health, we also include in some models (xviii) the mother's occupational class (manager, service, or technical, with other occupation as the baseline), (xix) whether the mother currently works part time, and (xx) child care arrangements (day care center as the baseline, nonrelative babysitter, relative babysitter, and other type of child care). As with the socioeconomic and health variables, all these independent variables are potentially endogenous. Consequently, the OLS and IV models are also estimated with and without this richer set of variables.

Identifying Instrumental Variables

The NMIHS respondents gave birth in 1988, when the United States was one of just two industrialized countries that did not have a national maternal leave policy (Hyde 1995). The FMLA of 1993 guarantees 12 weeks of unpaid leave for eligible mothers and the right to return to their jobs. However, before this national legislation was passed, many states had laws that provided some of the leave provisions (or more generous ones) that currently are covered by the FMLA (Department of Labor 1990: RAND Labor and Population Program Research Brief 1995). As of 1990, 30 states had some kind of maternity or parental leave law, ranging from laws that allow only for leave for the mother during recovery from childbirth to laws that allow for up to one year of leave for either parent to care for an infant (Department of Labor 1990). Of the 30 states with maternity/paternal laws of some kind, 12 had laws that applied to state employees only. Most state laws regarding leave exempted small businesses, but the definition of a small business varied by state (Family and Medical Leave Commission 1995). Several states in 1990 also had temporary disability laws that provided some salary support during leave from work (Department of Labor 1990). The temporary disability laws covered all employers with at least one employee (Family and Medical Leave Commission 1995).

In this study, we use the cross-sectional variation in these state-level policies to instrument for the length of the mother's leave from work. We use three dummy indicators to represent these state policies: (i) whether the state had any kind of unpaid, job-protected maternity leave law in 1988 that applied to private-sector workers, not just state employees (states with salary replacement laws are excluded here): (ii) an interaction term between this maternity leave law and the number of weeks of unpaid leave provided by the law: and (iii) whether the state had a temporary disability law in 1988 that would provide some level of salary replacement for non-work-related disabilities, including pregnancy related conditions and childbirth. These data come from the Department of Labor (1990) and Waldfogel (1999). We expect that mothers who lived in states with maternity leave laws and disability laws will take longer leaves from work compared to mothers who lived in states without these laws.

Following Baum's (2003) work on maternal employment and child development, we use additional instruments that are intended to proxy local labor market conditions. Mothers living in more economically depressed labor markets are expected to return to work earlier than other mothers because of concerns about retaining their jobs. Also, women with higher potential earnings in the market, as proxied by local per capita income, are expected to return to work sooner than other women. However, state-level labor market variables are not expected to directly impact maternal health after controlling for a range of individual-level socioeconomic factors.

To proxy local labor market conditions, Baum (2003) uses measures such as the local unemployment rate, the percentage of the local labor market that is female, local per capita income, and the percentage of the local population that has a high school and college degree. We have access to state but not local identifiers for NMIHS respondents. Therefore, we proxy local labor market conditions by using state-level measures of unemployment, the percentage of women in the labor force, the percentage of the population with a college degree, and average real per capita income.

5. Results

Table 1 displays means and standard deviations for all variables used in the analyses. The average CES-D score in the sample is 9.5, and 20% of the respondents have a CES-D score of at least 16, which is considered to be an elevated rate of depressive symptoms that may be indicative of clinical depression. This rate of depression is much higher than rates reported in community samples of women of childbearing age (O'Hara et al. 1990; Campbell et al. 1992). However, it is consistent with other research on disadvantaged mothers (Siefert et al. 2000: Lennon, Blome, and English 2001; Reading and Reynolds 2001: Beeghly et al. 2003), some of which is based on NMIHS (McLennan, Kotelchuck, and Cho 2001). McLennan. Kotelchuck, and Cho (2001), for example, use a sample of 7537 mothers from NMIHS and report that 24% had a CES-D score of at least 16. About 18% of mothers in the sample report having made at least three visits to an outpatient provider during the first six months after childbirth.

On average, the sample mothers returned to work nine weeks after childbirth, and more than 75% had returned to work by the time their infants were 12 weeks old (Figure 2). This finding is consistent with the work of Klerman and Leibowitz (1994) and Cantor et al. (2001), who find that most mothers who return to work during the first year do so within three months of childbirth. Almost all mothers in the sample have at least a high school degree (97%), and 44% have completed some college or a college degree. The sample includes a large proportion of black mothers (39%) and low-birth-weight infants (23%) because the NMIHS oversampled these groups. However, the sample is only 5% Hispanic and 3% Asian.

[FIGURE 2 OMITTED]

Table 2 shows results from all models that are estimated with the log of the CES-D score as the dependent variable. Columns 1-3 display OLS estimates with increasingly richer specifications. Column 1 presents a model with only basic, sociodemographic variables included on the right-hand side. Column 2 shows a model that also includes potentially endogenous socioeconomic and infant health endowment variables. Finally, column 3 displays a model that additionally includes employment characteristics and child care arrangements as covariates. Columns 4-6 show IV models that correspond to each of the OLS specifications presented in columns 1-3.

All the models indicate that returning to work later is associated with fewer depressive symptoms (Table 2, columns 1-6). In the OLS models, returning to work a week later is associated with a 1% decline in the mother's CES-D score, but this association is not statistically significant. Moreover, because the CES-D is a scale and not a symptom count, the clinical significance of this improvement is hard to assess. The reduction in depressive symptoms could correspond to no longer validating a particular depressive symptom in the past week or experiencing a depressive symptom less frequently in the past week. The OLS models show no evidence that the timing of returning to work is correlated with other, observed characteristics that also affect depressive symptoms. The magnitude of the estimated effect remains virtually the same regardless of the model specification. When the models are reestimated with the CES-D measure in its natural units (unlogged), returning to work later has a larger but still statistically insignificant negative effect on depression. These results are available on request.

The OLS models do not account for the possibility of reverse causality--mothers may return to work later or earlier as a response to their depressive symptoms. Also, the OLS estimates may be confounded by unmeasured characteristics that are correlated with both the timing of returning to work and depression. The IV methods account for these problems by purging the potentially endogenous return-to-work variable of its correlation with the error term.

The IV results demonstrate in every case that returning to work later is associated with a reduction in depressive symptoms that is statistically significant at the 0.10 level (Table 2, columns 4-6). The magnitudes of the IV estimates, however, are six to seven times larger than the OLS estimates, which would suggest that mothers experiencing more depressive symptoms return to work later (and/or that mothers with unobserved characteristics that are positively associated with depression return to work later). Returning to work one week later is associated with a 6-7% reduction in depressive symptoms. Like the OLS estimates, the IV estimates are not sensitive to the covariates included in the models. When the IV models are reestimated with the unlogged version of the CES-D score, returning to work later is still associated with a negative albeit marginally statistically significant decline in CES-D score (results available on request).

The identifying instrumental variables perform reasonably well in these models. First-stage estimates are presented in the Appendix. The first-stage results show that respondents living in states with a maternity leave policy or a temporary disability law take about a week longer of maternity leave. Respondents living in states with higher per capita incomes also take longer maternity leaves. The F-test on the identifying instruments ranges from approximately 14 to 16 and are statistically significant at the 0.001 level. The overidentification test suggests that the instruments can be validly excluded from the depression equation. The Durbin-Wu-Hausman test is used to test for the consistency of the OLS estimate. The null hypothesis is rejected in every case at the 5% level. Thus, there is evidence that the IV estimates are the preferred estimates.

Table 3 shows results from models where the dependent variable is a dummy variable indicating whether the mother had a CES-D score of at least 16. This threshold is commonly used as a cutoff for a likely clinical case of depression (Eaton et al. 2003). In all the OLS and IV models, returning to work later is associated with a small reduction in the probability of being a depressive case. (6) The estimated effects are not statistically significant in any of the models, but this is not entirely surprising given that we lose information when dichotomizing the outcome. It is notable that many of the previously statistically significant coefficients also lose their statistical significance in these models. Therefore, we believe that these findings are still suggestive of an impact of returning to work later on the probability of being a likely case of clinical depression.

We consider health care utilization in Table 4. The dependent variable in these models is a dummy variable indicating whether the respondent visited an outpatient physician or clinic at least three times during the six months after childbirth. This measure is intended to proxy the mother's physical and mental status during the time period when she first returns to work. In contrast, maternal depressive symptoms, which were the focus of Tables 2 and 3, were measured on average about a year after the mother returned to work.

The OLS results in Table 4 (columns 1-3) indicate that returning to work later is associated with a very small, statistically significant increase in the probability of having had at least three outpatient visits. It is likely that these results are confounded by effect of health on the timing of returning to work--mothers in poor health may postpone their return to employment. The IV results, which address this potential problem, suggest the opposite. Returning to work later is associated with a reduction in the probability of having at least three outpatient visits in every model. The size of the effect indicates about a one- to two-percentage-point reduction in the probability, which is about a 6-11% reduction in the probability of having at least three outpatient visits when measured at the sample mean of 0.18. The estimated coefficients on maternity leave length are not statistically significant in these models, but, as in the case of clinical depression, most coefficients lose statistical significance in these models. The F-tests on the identifying instruments are statistically significant at the 0.001 level, and the overidentification test suggests that the instruments can be validly excluded from the second-stage equation. Based on the Durbin-Wu-Hausman test results, the consistency of the OLS estimates is rejected at about the 5% level, making the IV estimates preferred in this case.

6. Discussion and Conclusions

In 2002, California became the first state to provide up to six weeks of paid family leave to care for a newborn or a seriously ill family member. In 2001 and 2002, paid leave bills were introduced in at least 28 states. States are considering a variety of different options to finance paid family leave, including using general funds from state budgets, giving tax credits to employers who provide paid leave, extending existing temporary' disability systems, and expanding unemployment insurance programs to families with newborn children (National Partnership for Women and Families 2004). All these policy initiatives are intended to help families actually take advantage of the FMLA. which currently guarantees 12 weeks of unpaid leave.

To understand the net impact of these policies, states need information on the benefits of maternity leave to families. Previous economic research on maternal employment has focused on understanding how the length of maternal leave after childbirth influences children's health and development. This study extends this literature by examining the effect of maternal leave length on the health of the mother. We focus on depression because of its very high prevalence among women of childbearing age, because of its potential negative effects on children, and because this disorder tends to be recurrent. We also consider outpatient health services utilization in the first six months after childbirth as a measure of the mother's overall health.

The results suggest that longer leave from work is associated with a reduction in the number or frequency of depressive symptoms. Specifically, increasing maternal leave by one week is associated with a 6-7% decline in depressive symptoms. This result, which comes from IV models, means that mothers are experiencing fewer symptoms of depression, are experiencing depressive symptoms with less frequency, or both. It is difficult to determine what these reductions in the CES-D scale mean in terms of maternal functioning and well-being. Indeed, there is only suggestive evidence that returning to work later lowers the probability, of having a CES-D score of 16 or higher. There is also only suggestive evidence that returning to work later alters the probability of having at least three outpatient visits in the first six months alter childbirth. In sum, the findings indicate that longer maternal leave may have some lasting benefits for maternal health.

This paper offers the first evidence that policies that lengthen maternity leaves, such as state initiatives to offer paid leave, may have the added benefit of reducing symptoms of depression among employed mothers. We caution, however, that most of the variation in maternity leave in our sample is small, making it difficult to assess the effects of substantial changes in maternity leave policy, such as paid family leave. Furthermore, data limitations may have affected our ability to find statistically significant effects of maternity leave length on maternal health. As we note in the paper, we measure maternal depressive symptoms relatively late in terms of when mothers have returned to work. The results for the measures of depression may be diluted by the fact that mothers who return to work late have been back at work for a shorter period of time. Also, our measures of maternal health have limitations; the use of the CES-D, a screening tool for depression, captures many women experiencing transient stresses, and our measure of outpatient health services use is quite crude. To better understand how maternity leave length affects maternal health, better data are needed on mothers' mental and physical health in the months immediately after they return to work. These data would allow future researchers to focus on refining our understanding of the ways in which returning to work affects maternal health.

Appendix
First-Stage Results Estimate (t-Statistic)

 (1) (2)
 Basic Full Set of
 Covariates Covariates

State unemployment rate -0.117 (-1.36) -0.132 (-1.57)
State female labor force
 participation -0.079 (-1.58) -0.084 (-1.62)
State income 0.0002 (2.37) 0.0002 (2.31)
State college degree -0.073 (-0.82) -0.074 (-0.83)
State leave law 0.958 (1.79) 0.949 (1.79)
State temporary disability law 1.331 (2.04) 1.239 (1.86)
Interaction term between state
 leave law and number of weeks
 of leave permitted -0.024 (-0.86) -0.023 (-0.81)
Mother's age 0.082 (2.78) 0.074 (2.58)
High school dropout -0.047 (-0.07) -0.003 (0.00)
Some college 0.321 (1.23) 0.267 (1.03)
College graduate 0.429 (1.31) 0.324 (1.03)
Income in low-middle
 approximate quartile -0.864 (-2.38) -1.011 (-2.71)
Income in high-middle
 approximate quartile -0.081 (-0.23) -0.285 (-0.76)
Income in highest approximate
 quartile 0.355 (0.66) 0.125 (0.23)
Hispanic 0.467 (0.94) 0.545 (1.12)
Black 1.086 (4.63) 1.030 (4.35)
Asian -0.155 (-0.21) -0.215 (-0.29)
Married 0.823 (2.54) 0.621 (2.11)
Number of children -0.361 (-3.46) -0.323 (-3.21)
Multiple birth -0.034 (-0.07) -0.279 (-0.58)
Age of child in months 0.027 (0.76) 0.026 (0.73)
Welfare recipient -0.667 (-0.86)
Prescribed bed rest 0.269 (0.79)
Premature infant -0.475 (-1.83)
Low birth weight 0.501 (1.26)
Prenatal care in first trimester 0.622 (1.66)
Insured 0.585 (1.72)
Smoked daily during pregnancy -0.284 (-0.61)
Exercised before pregnancy 0.010 (0.03)
Smoked daily during pregnancy -0.240 (-0.56)
Works part time
Manager
Technical
Service
Relative babysitter
Nonrelated babysitter
Other child care
F-test on instruments
 (test statistic
 and p-value) 15.040 (0.000) 15.640 (0.000)

 (3)
 Model (2) plus
 Occupation and
 Child Care
 Variables

State unemployment rate -0.115 (-1.36)
State female labor force
 participation -0.072 (-1.40)
State income 0.0002 (2.11)
State college degree -0.079 (-0.89)
State leave law 0.845 (1.55)
State temporary disability law 1.247 (1.85)
Interaction term between state
 leave law and number of weeks
 of leave permitted -0.021 (-0.74)
Mother's age 0.081 (2.78)
High school dropout -0.102 (-0.15)
Some college 0.245 (0.91)
College graduate 0.401 (1.14)
Income in low-middle
 approximate quartile -0.938 (-2.53)
Income in high-middle
 approximate quartile -0.258 (-0.68)
Income in highest approximate
 quartile 0.206 (0.38)
Hispanic 0.495 (0.98)
Black 1.070 (4.51)
Asian -0.219 (-0.31)
Married 0.629 (2.16)
Number of children -0.351 (-3.53)
Multiple birth -0.241 (-0.49)
Age of child in months 0.035 (0.96)
Welfare recipient -0.829 (-1.05)
Prescribed bed rest 0.250 (0.73)
Premature infant -0.448 (-1.67)
Low birth weight 0.437 (1.08)
Prenatal care in first trimester 0.581 (1.54)
Insured 0.587 (1.73)
Smoked daily during pregnancy -0.257 (-0.55)
Exercised before pregnancy 0.013 (0.04)
Smoked daily during pregnancy -0.246 (-0.61)
Works part time 0.498 (1.45)
Manager 0.0005 (0.00)
Technical 0.031 (0.10)
Service 0.085 (0.17)
Relative babysitter 0.910 (2.69)
Nonrelated babysitter -0.372 (-0.70)
Other child care 0.675 (2.07)
F-test on instruments
 (test statistic
 and p-value) 13.920 (0.000)

Table 1. Sample Means and Standard Deviations (N = 1762)

 Mean
 (Standard
Variable Definition Deviation)

Maternal health
 Center for
 Epidemiological
 Studies Depression
 (CES-D) Score on CES-D screener 9.47 (9.37)

Scale score
 Depressive case Dummy variable = 1 if 0.198
 respondent reports a
 score of at least 16 on
 the CES-D, 0 otherwise

 At least three Dummy variable = 1 if 0.184
 outpatient visits in respondent reports having
 first six months after visited a clinic or
 childbirth physician for mental or
 physical health problems
 at least three times in
 the first six months
 after childbirth,
 0 otherwise

Length of maternal leave
 Number of weeks since The number of weeks after 9.18 (4.99)
 birth when mother the birth when the
 returned to work respondent returned to
 work

Other independent variables
 Mother's age Mother's age in years at 27.81 (5.03)
 time of birth

 High school dropout Dummy variable = 1 if 0.034
 respondent is a high
 school dropout,
 0 otherwise

 Some college Dummy variable = 1 if 0.242
 respondent completed some
 college but did not
 graduate, 0 otherwise

 College graduate Dummy variable = 1 if 0.192
 respondent is a college
 graduate, 0 otherwise

 Income in low-middle Dummy variable = 1 if 0.253
 approximate quartile respondent's gross
 household income in the
 12 months preceding
 childbirth is between
 $17,001 and $27,500,
 which corresponds
 approximately to the
 second-to-lowest quartile
 in the analysis sample
 distribution, 0 otherwise

 Income in high-middle Dummy variable = 1 if 0.325
 approximate quartile respondent's gross
 household income in the
 12 months preceding
 childbirth is between
 $27,501 and $45,000,
 which corresponds
 approximately to the
 second-to-highest
 quartile in the analysis
 sample distribution, 0
 otherwise

 Income in highest Dummy variable = 1 if 0.168
 approximate respondent's gross
 quartile household income in the
 12 months preceding
 childbirth is greater
 than $45,000, which
 corresponds approximately
 to the highest quartile
 in the analysis sample
 distribution, 0 otherwise

 Hispanic Dummy variable = 1 if 0.053
 respondent is Hispanic, 0
 otherwise

 Black Dummy variable = 1 if 0.389
 respondent is black and
 not Hispanic, 0 otherwise

 Asian Dummy variable = 1 if 0.025
 respondent is Asian
 (Chinese, Japanese,
 Hawaiian, Filipino, or
 other Asian) and not
 Hispanic, 0 otherwise

 White/other Dummy variable = 1 if 0.533
 respondent is white/not
 Hispanic or of other
 race/not Hispanic (such
 as other nonwhite, Native
 American)

 Married Dummy variable = 1 if 0.775
 respondent is married at
 the time of the birth, 0
 otherwise

 Number of children Number of children in 1.50
 household

 Multiple birth Dummy variable = 1 if 0.038
 respondent's child was
 part of a multiple birth,
 0 otherwise

 Age of child in Age of child in months at 15.26
 months time of interview

 Welfare recipient Dummy variable = 1 if 0.026
 respondent receives AFDC
 at time of survey,
 0 otherwise

 Prescribed bed rest Dummy variable = I if 0.236
 respondent reports that
 her physician advised her
 to stay in bed for at
 least one week during her
 pregnancy, 0 otherwise

 Premature infant Dummy variable = 1 if 0.210
 respondent's child was
 born earlier than 37
 weeks' gestation, 0
 otherwise

 Low birth weight Dummy variable = 1 if 0.228
 respondent's child was
 low birth weight, 0
 otherwise

 Prenatal care in Dummy variable = 1 if 0.895
 first trimester respondent initiated
 prenatal care during
 first trimester, 0
 otherwise

 Insured Dummy variable = 1 if 0.850
 respondent has health
 insurance, 0 otherwise

 Smoked daily during Dummy variable = 1 if 0.178
 pregnancy respondent Smoked daily
 during pregnancy 0
 otherwise

 Exercised before Dummy variable = 1 if 0.457
 pregnancy respondent exercised or
 played sports at least
 three times per week
 before finding out she
 was pregnant, 0 otherwise

 Smoked daily before Dummy variable = 1 if 0.260
 pregnancy respondent smoked daily
 during the three months
 before finding out she
 was pregnant, 0 otherwise

 Works part time Dummy variable = 1 if 0.231
 respondent worked less
 than 35 hours per week at
 the time of the
 interview, 0 otherwise

 Manager Dummy variable = 1 if 0.261
 respondent has a
 managerial occupation, 0
 otherwise

 Technical Dummy variable = 1 if 0.460
 respondent has a
 technical occupation, 0
 otherwise

 Service Dummy variable = 1 if 0.148
 respondent has a service
 occupation, 0 otherwise

 Relative babysitter Dummy variable = 1 if 0.487
 respondent has a relative
 who watches child on
 workdays, 0 otherwise

 Nonrelated babysitter Dummy variable = 1 if 0.310
 respondent has a baby-
 sitter (not a relative)
 who watches child on
 workdays, 0 otherwise

 Other child care Dummy variable = 1 if 0.071
 respondent uses other
 child care arrangements

 State unemployment rate State unemployment rate 5.62
 in 1988

 State female labor force State female labor force 0.568
 participation participation in 1988

 State college degree Percent of state 0.189
 population with college
 degree or higher

 State income Average real per capita 16,924 (2475)
 income in state in 1988

 State leave law Dummy variable = 1 if 0.098
 state had passed by 1988
 any type of maternity
 leave law that applies to
 private-sector employees
 (not just state
 employees) and does not
 have a temporary
 disability law, 0
 otherwise

 State temporary Dummy variable = 1 if 0.145
 disability law state had passed by 1988
 a temporary disability
 law (in addition to
 having a state leave
 law), 0 otherwise

 Interaction term between Interaction between the 2.37 (5.29)
 state leave law and state leave law dummy
 number of weeks of leave variable and the maximum
 permitted number of weeks of leave
 mandated by the law
 (e.g., 12 weeks, six
 weeks)

Table 2. Depression Score and Length of Maternal Leave Estimate
(t-Statistic) (a)

Dependent Variable: Log CES-D Score

 OLS

 (3)
 Model
 (2) plus
 (2) Occupation
 (1) Full and Child
 Basic Set of Care
 Covariates Covariates Variables

Number of weeks -0.010 -0.010 -0.010
 since birth when (-1.16) (-1.10) (-1.14)
 mother returned
 to work
Mother's age -0.009 -0.009 -0.009
 (-1.74) (-1.89) (-1.87)
High school dropout 0.196 0.144 0.122
 (1.88) (1.31) (1.13)
Some college -0.098 -0.091 -0.082
 (-1.69) (-1.59) (-1.32)
College graduate -0.140 -0.117 -0.096
 (-2.01) (-1.67) (-1.22)
Income in low-middle -0.237 -0.238 -0.231
 approximate (-2.96) (-3.07) (-2.93)
 quartile
Income in -0.202 -0.200 -0.188
 high-middle (-2.49) (-2.31) (-2.19)
 approximate
 quartile
Income in highest -0.416 -0.413 -0.392
 approximate (-4.48) (-4.24) (-3.79)
 quartile
Hispanic 0.299 0.292 0.295
 (2.89) (2.65) (2.61)
Black 0.289 0.287 0.287
 (4.96) (4.69) (4.24)
Asian 0.191 0.197 0.201
 (1.99) (1.81) (1.82)
Married -0.137 -0.108 -0.109
 (-2.14) (-1.73) (-1.73)
Number of children -0.034 -0.028 -0.030
 (-1.54) (-1.21) (-1.40)
Multiple birth 0.020 -0.033 -0.026
 (0.180) (-2.900) (-0.230)
Age of child -0.018 -0.017 -0.017
 in months (-3.46) (-3.20) (-3.08)
Welfare recipient 0.360 0.348
 (2.29) (2.27)
Prescribed bed rest 0.148 0.147
 (2.54) (2.52)
Premature infant -0.010 -0.010
 (-0.110) (-0.110)
Low birth weight -0.010 -0.001
 (-0.120) (-0.120)
Prenatal care in -0.106 -0.102
 first trimester (-1.32) (-1.23)
Insured 0.053 0.061
 (0.900) (0.990)
Smoked daily -0.098 -0.103
 during pregnancy (-1.24) (-1.28)
Exercised before -0.008 -0.007
 pregnancy (-0.190) (-0.170)
Smoked daily 0.177 0.177
 before pregnancy (2.33) (2.34)
Works part time 0.044
 (0.650)
Manager -0.048
 -(-0.500)
Technical -0.029
 (-0.400)
Service 0.036
 (0.490)
Relative babysitter -0.017
 (-0.260)
Nonrelated babysitter -0.076
 (-0.970)
Other child care -0.060
 (-0.620)
Overidentification test
 (test statistic and
 p-value)
Hausman test
 (test statistic and
 p-value)
F-test on instruments
 (test statistic and
 p-value)
N 1762

 IV

 (6)
 Model
 (5) plus
 (5) Occupation
 (4) Full and Child
 Basic Set of Care
 Covariates Covariates Variables

Number of weeks -0.062 -0.063 -0.072
 since birth when (-1.90) (-1.85) (-1.81)
 mother returned
 to work
Mother's age -0.004 -0.005 -0.003
 (-0.570) (-0.750) (-0.480)
High school dropout 0.183 0.136 0.108
 (1.58) (1.08) (0.840)
Some college -0.084 -0.080 -0.072
 (-1.39) (-1.32) (-1.10)
College graduate -0.125 -0.107 -0.080
 (-1.91) (-1.62) (-1.06)
Income in low-middle -0.274 -0.287 -0.283
 approximate (-3.38) (-3.69) (-3.56)
 quartile
Income in -0.195 -0.208 -0.197
 high-middle (-2.44) (-2.56) (-2.44)
 approximate
 quartile
Income in highest -0.374 -0.387 -0.358
 approximate (-4.11) (-4.14) (-3.47)
 quartile
Hispanic 0.321 0.318 0.324
 (2.76) (2.58) (2.54)
Black 0.344 0.341 0.354
 (4.79) (4.50) (4.02)
Asian 0.223 0.224 0.232
 (2.46) (2.26) (2.28)
Married -0.106 -0.088 -0.084
 (-1.54) (-1.30) (-1.19)
Number of children -0.057 -0.049 -0.056
 (-2.13) (-1.79) (-2.00)
Multiple birth 0.015 -0.049 -0.042
 (1.500) (-0.450) (-3.900)
Age of child -0.013 -0.012 -0.011
 in months (-2.31) (-2.16) (-1.84)
Welfare recipient 0.324 0.291
 (2.06) (1.86)
Prescribed bed rest 0.162 0.162
 (2.59) (2.54)
Premature infant -0.040 -0.041
 (-0.440) (-0.470)
Low birth weight 0.015 0.015
 (0.180) (0.160)
Prenatal care in -0.064 -0.058
 first trimester (-0.660) (-0.570)
Insured 0.096 0.110
 (1.55) (1.65)
Smoked daily -0.115 -0.120
 during pregnancy (-1.29) (-1.31)
Exercised before -0.008 -0.007
 pregnancy (-0.170) (-0.140)
Smoked daily 0.167 0.164
 before pregnancy (2.09) (2.06)
Works part time 0.086
 (1.12)
Manager -0.040
 (-0.400)
Technical -0.017
 (-0.230)
Service 0.050
 (0.570)
Relative babysitter 0.056
 (0.660)
Nonrelated babysitter -0.021
 (-0.220)
Other child care -0.077
 (-0.700)
Overidentification test 3.27 4.17 4.31
 (test statistic and (0.774) (0.654) (0.635)
 p-value)
Hausman test 3.13 2.99 2.93
 (test statistic and (0.077) (0.084) (0.087)
 p-value)
F-test on instruments 15.04 15.64 13.92
 (test statistic and (0.000) (0.000) (0.000)
 p-value)
N 1762

(a) t-statistics computed from Huber-White standard errors with
adjustment for clustering on state of residence. CES-D = Center
for Epidemiological Studies Depression Scale; OLS = ordinary
least squares.

Table 3. Depressive Case and Length of Maternal Leave Estimate
(t-Statistic) (a)

 Dependent Variable: Dummy Variable
 Indicating a Score of at
 Least 16 on CES-DOLS

 OLS (Linear Probability Model)

 (3)
 Model
 (2) plus
 (2) Occupation
 (1) Full and Child
 Basic Set of Care
 Covariates Covariates Variables

Number of weeks since -0.001 -0.001 -0.001
birth when mother (-0.350) (-0.300) (-0.320)
returned to work

Mother's age -0.002 -0.002 -0.002
 (-0.880) (-0.830) (-0.840)

High school dropout 0.061 0.054 0.047
 (0.930) (0.800) (0.690)

Some college -0.030 -0.028 -0.025
 (-1.13) (-1.02) (-0.900)

College graduate -0.029 -0.024 -0.020
 (-1.28) (-1.02) (-0.800)

Income in low-middle -0.038 -0.039 -0.036
approximate quartile (-1.17) (-1.18) (-1.05)

Income in high-middle -0.046 -0.047 -0.042
approximate quartile (-1.49) (-1.39) (-1.22)

Income in highest -0.095 -0.095 -0.089
approximate quartile (-2.42) (-2.28) (-1.94)

Hispanic 0.019 0.016 0.020
 (0.630) (0.490) (0.620)

Black 0.074 0.075 0.078
 (2.81) (2.70) (2.72)

Asian 0.019 0.021 0.019
 (0.330) (0.360) (0.320)

Married -0.082 -0.074 -0.073
 (-2.26) (-2.03) (-1.97)

Number of children 0.003 0.004 0.003
 (0.390) (0.450) (0.350)

Multiple birth 0.059 0.065 0.065
 (1.04) (1.07) (1.09)

Age of child in months -0.003 -0.003 -0.002
 (-1.26) (-1.11) (-1.02)

Welfare recipient 0.095 0.097
 (0.940) (0.980)

Prescribed bed rest 0.023 0.024
 (0.970) (1.03)

Premature infant -0.006 -0.005
 (-0.180) (-0.180)

Low birth weight -0.017 -0.017
 (-0.490) (-0.510)

Prenatal care in first -0.009 -0.006
trimester (-0.230) (-0.160)

Insured 0.003 0.003
 (0.080) (0.090)

Smoked daily during -0.041 -0.040
pregnancy (-0.980) (-0.960)

Exercised before -0.017 -0.016
pregnancy (-0.950) (-0.900)

Smoked daily before 0.055 0.053
pregnancy (1.64) (1.58)

Works part time 0.009
 (0.330)

Manager -0.032
 (-0.960)

Technical -0.031
 (-1.17)

Service -0.004
 (-0.140)

Relative babysitter 0.003
 (0.090)

Nonrelated babysitter 0.030
 (0.860)

Other child care 0.004
 (0.090)

Overidentification test
(test statistic and
p-value)

Hausman test
(test statistic and
p-value)

F-test on instruments
(test statistic and
p-value)

N 1762

 Dependent Variable: Dummy Variable
 Indicating a Score of at
 Least 16 on CES-DOLS

 IV

 (6)
 Model
 (5) plus
 (5) Occupation
 (4) Full and Child
 Basic Set of Care
 Covariates Covariates Variables

Number of weeks since -0.017 -0.018 -0.020
birth when mother (-1.23) (-1.22) (-1.24)
returned to work

Mother's age -0.0003 -0.0004 -0.0001
 (-0.120) (-0.160) (-0.040)

High school dropout 0.058 0.051 0.043
 (0.850) (0.740) (0.600)

Some college -0.026 -0.024 -0.022
 (-0.960) (-0.880) (-0.780)

College graduate -0.025 -0.021 -0.015
 (-1.14) (-0.930) (-0.620)

Income in low-middle -0.049 -0.054 -0.051
approximate quartile (-1.49) (-1.57) (-1.46)

Income in high-middle -0.044 -0.049 -0.044
approximate quartile (-1.45) (-1.52) (-1.35)

Income in highest -0.082 -0.087 -0.080
approximate quartile (-1.99) (-2.01) (-1.65)

Hispanic 0.026 0.024 0.029
 (0.810) (0.710) (0.840)

Black 0.091 0.091 0.098
 (3.12) (2.95) (2.84)

Asian 0.028 0.029 0.029
 (0.580) (0.580) (0.540)

Married -0.073 -0.068 -0.065
 (-1.96) (-1.82) (-1.71)

Number of children -0.003 -0.002 -0.005
 (-0.350) (-0.250) (-0.460)

Multiple birth 0.057 0.060 0.060
 (1.05) (1.00) (1.03)

Age of child in months -0.001 -0.001 -0.001
 (-0.500) (-0.390) (-0.230)

Welfare recipient 0.085 0.080
 (-0.390) (0.800)

Prescribed bed rest 0.027 0.029
 (1.18) (1.26)

Premature infant -0.015 -0.015
 (-0.480) (-0.500)

Low birth weight -0.009 -0.010
 (-0.260) (-0.290)

Prenatal care in first 0.004 0.007
trimester (0.100) (0.160)

Insured 0.016 0.017
 (0.480) (0.510)

Smoked daily during -0.045 -0.045
pregnancy (-1.05) (-1.01)

Exercised before -0.017 -0.016
pregnancy (-0.900) (-0.840)

Smoked daily before 0.052 0.049
pregnancy (1.56) (1.50)

Works part time 0.021
 (0.720)

Manager -0.029
 (-0.850)

Technical -0.028
 (-0.960)

Service -0.001
 (-0.010)

Relative babysitter 0.025
 (0.640)

Nonrelated babysitter 0.046
 (1.20)

Other child care -0.001
 (-0.030)

Overidentification test 6.43 6.67 5.71
(test statistic and (0.377) (0.353) (0.457)
p-value)

Hausman test 1.47 1.47 1.51
(test statistic and (0.226) (0.225) (0.219)
p-value)

F-test on instruments 15.04 15.64 13.92
(test statistic and (0.000) (0.000) (0.000)
p-value)

N 1762

(a) t-statistics computed from Huber-White standard errors with
adjustment for clustering on state of residence. CES-D = Center
for Epidemiological Studies Depression Scale; OLS = ordinary
least squares.

Table 4. At Least Three Outpatient Visits and Length of Maternal
Leave Estimate (t-Statistic) (a)

 Dependent Variable: Dummy
 Variable Indicating at Least
 Three Outpatient Visits during
 Six Months after Childbirth

 OLS (Linear Probability Model)

 (3)
 Model
 (2) (2) plus
 (1) Full Occupation
 Basic Set of and Child Care
 Covariates Covariates Variables

Number of weeks since 0.004 0.004 0.003
birth when mother (2.15) (2.01) (1.96)
returned to work

Mother's age -0.001 -0.001 -0.001
 (-0.380) (-0.390) (-0.510)

High school dropout 0.042 0.021 0.030
 (0.770) (0.400) (0.560)

Some college 0.010 0.011 0.005
 (0.440) (0.480) (0.200)

College graduate 0.009 0.012 -0.004
 (0.340) (0.480) (-0.170)

Income in low-middle 0.026 0.032 0.031
approximate quartile (1.05) (1.26) (1.18)

Income in high-middle 0.017 0.022 0.016
approximate quartile (0.870) (1.02) (0.660)

Income in highest 0.053 0.058 0.044
approximate quartile (1.57) (1.56) (1.19)

Hispanic -0.030 -0.037 -0.027
 (-0.580) (-0.770) (-0.530)

Black -0.010 -0.014 -0.004
 (-0.560) (-0.680) (-0.180)

Asian 0.071 0.065 0.073
 (1.31) (1.14) (1.21)

Married -0.064 -0.057 -0.057
 (-2.83) (-2.39) (-2.39)

Number of children 0.003 0.009 0.009
 (0.310) (0.980) (1.07)

Multiple birth -0.038 -0.107 -0.110
 (-0.950) (-2.40) (-2.48)

Age of child in months -0.002 -0.002 -0.002
 (-1.05) (-0.910) (-1.05)

Welfare recipient 0.120 0.110
 (1.73) (1.58)

Prescribed bed rest 0.059 0.058
 (2.67) (2.62)

Premature infant -0.009 -0.007
 (-0.300) (-0.240)

Low birth weight 0.074 0.074
 (2.42) (2.41)

Prenatal care in first 0.011 0.008
trimester (0.350) (0.290)

Insured -0.010 -0.008
 (-0.340) (-0.270)

Smoked daily during -0.055 -0.052
pregnancy (-1.42) (-1.32)

Exercised before 0.014 0.013
pregnancy (0.680) (0.650)

Smoked daily before 0.037 0.038
pregnancy (1.17) (1.18)

Works part time 0.018
 (0.870)

Manager 0.062
 (1.92)

Technical 0.039
 (1.70)

Service 0.037
 (1.12)

Relative babysitter 0.004
 (0.160)

Nonrelated babysitter 0.029
 (1.04)

Other child care 0.022
 (0.640)

Overidentification
test (test statistic
and p-value)

Hausman test
(test statistic
and p-value)

F-test on instruments
(test statistic
and p-value)

N 1762

 Dependent Variable: Dummy
 Variable Indicating at Least
 Three Outpatient Visits during
 Six Months after Childbirth

 IV

 (6)
 Model
 (5) (5) plus
 (4) Full Occupation
 Basic Set of and Child Care
 Covariates Covariates Variables

Number of weeks since -0.019 -0.018 -0.022
birth when mother (-1.52) (-1.46) (-1.62)
returned to work

Mother's age 0.001 0.001 0.001
 (0.670) (0.490) (0.530)

High school dropout 0.037 0.018 0.023
 (0.700) (0.340) (0.450)

Some college 0.016 0.016 0.008
 (0.620) (0.610) (0.330)

College graduate 0.015 0.016 0.002
 (0.590) (0.620) (0.070)

Income in low-middle 0.011 0.014 0.011
approximate quartile (0.420) (0.510) (0.380)

Income in high-middle 0.020 0.019 0.013
approximate quartile (1.00) (0.880) (0.540)

Income in highest 0.071 0.068 0.057
approximate quartile (1.88) (1.74) (1.49)

Hispanic -0.022 -0.027 -0.016
 (-0.410) (-0.550) (-0.300)

Black 0.012 0.006 0.023
 (0.460) (0.230) (0.780)

Asian 0.084 0.075 0.085
 (1.40) (1.26) (1.29)

Married -0.051 -0.049 -0.047
 (-1.91) (-1.83) (-1.70)

Number of children -0.007 0.001 -0.001
 (-0.690) (-0.080) (-0.060)

Multiple birth -0.041 -0.113 -0.116
 (-0.880) (-2.27) (-2.30)

Age of child in months -0.0004 -0.0002 -0.0002
 (-0.140) (-0.090) (-0.050)

Welfare recipient 0.106 0.088
 (1.35) (1.07)

Prescribed bed rest 0.064 0.063
 (2.60) (2.51)

Premature infant -0.021 -0.020
 (-0.590) (-0.570)

Low birth weight 0.083 0.083
 (2.52) (2.53)

Prenatal care in first 0.027 0.026
trimester (0.910) (0.900)

Insured 0.006 0.011
 (0.190) (0.340)

Smoked daily during -0.061 -0.059
pregnancy (-1.68) (-1.57)

Exercised before 0.014 0.013
pregnancy (0.630) (0.580)

Smoked daily before 0.033 0.034
pregnancy (0.990) (0.970)

Works part time 0.035
 (1.46)

Manager 0.065
 (1.95)

Technical 0.044
 (1.90)

Service 0.042
 (1.17)

Relative babysitter 0.033
 (1.06)

Nonrelated babysitter 0.051
 (1.59)

Other child care 0.015
 (0.410)

Overidentification 5.29 4.71 4.39
test (test statistic (0.508) (0.581) (0.624)
and p-value)

Hausman test 3.68 3.42 3.94
(test statistic (0.055) (0.064) (0.047)
and p-value)

F-test on instruments 15.04 15.64 13.92
(test statistic (0.000) (0.000) (0.000)
and p-value)

N 1762

(a) t-statistics computed from Huber-White standard errors with
adjustment for clustering on state of residence. OLS = ordinary
least squares.

The authors would like to express sincere thanks to Kareo Conway, Dhaval Dave, Margarita Alegria, Thomas McGuire, and two anonymous referees who helped to greatly improve this paper.

Received July 2004; accepted January 2005.

(1) Among these mothers who took leaves that were longer than six months, the median length of leave was 11 months. Therefore, it seems likely that many of these mothers who took long leaves may have returned to different employers.

(2) Maternal depression imposes costs not only on individuals but on employers as well. It is estimated that workers with major depression have between 1.5 and 3.2 more short-term disability days per 30 days than other workers (Kessler et al. 1999). Moreover, maternal depression is important to study as an outcome because it is associated with adverse outcomes for children, including insecure infant/mother attachment and children's behavior problems (Civic and Holt 2000; Martins and Gaffan 2000).

(3) In the NMIHS, two items of the CES-D were imputed using the hot-deck method because of item nonresponse in 5-10% of cases. These two items were "people were unfriendly" and "I talked less than usual."

(4) Because the exact timing of the outpatient visits within the six-month period is not known, we cannot be certain that the outpatient visits occurred before or after "the mother returned to work. This issue affects the interpretation of the results. Preparing for the return-to-work (both physically and emotionally) could affect maternal health--therefore, in these models, it is not clear whether it is the actual return to work or the preparation for returning to work that affects outcomes. Although this distinction may not matter from a policy perspective, this problem remains a limitation of the analysis.

(5) Income is reported in ranges and the distribution of ranges does not yield values corresponding exactly to the 25th, 50th, and 75th percentiles. Approximations are used as described in Table 1.

(6) We estimated the models in Tables 3 and 4 using a probit and an IV probit instead of a linear probability model and 2SLS. The results were not appreciably different from those presented here and are available oil request.

References

American Academy of Pediatrics and the American College of Obstetricians and Gynecologists. 1997. Guidelines for perinatal care. 4th edition. Elk Grove Village, IL: American Academy of Pediatrics and the American College of Obstetricians and Gynecologists.

American Psychiatric Association. 2000. Diagnostic and statistical manual of mental disorders. 4th edition, Text Revision. Washington, DC: American Psychiatric Press.

Baum, C. L. 2003. Does early maternal employment harm child development? An analysis of the potential benefits of leave taking. Journal of Labor Economics 21:409-48.

Beeghly, M., K. L. Olson, M. K. Weinberg, S. C. Pierre, N. Downey, and E. Z. Tronick. 2003. Prevalence, stability, and sociodemographic correlates of depressive symptoms in black mothers during the first 18 months postpartum. Maternal and Child Health Journal 7:157-68.

Beeghly, M., M. K. Weinberg, K. L. Olson, H. Kernan, J. Riley, and E. Z. Tronick. 2002. Stability and change in level of maternal depressive symptomatology during the first postpartum year. Journal of Affective Disorders 71:169-80.

Blau, F. D., and A. J. Grossberg. 1992. Maternal labor supply and children's cognitive development. Review of Economics and Statistics 74:474-81.

Brooks-Gunn, J., W. J. Han, and J. Waldfogel. 2002. Maternal employment and child cognitive outcomes in the first three years of life: The NICHD Study of Early Child Care. Child Development 73:1052-72.

Bureau of Labor Statistics. 2003. Employment characteristics of families in 2000. Accessed 7 November 2003. Available ftp://ftp.bls.gov/pub/news.release/History/famee.04192001.news.

Campbell, S. B., and J. F. Cohn. 1991. Prevalence and correlates of postpartum depression in first-time mothers. Journal of Abnormal Psychology 100:594-9.

Campbell, S. B., J. F. Cohn, C. Flanagan, S. Popper, and T. Meyers. 1992. Course and correlates of postpartum depression during the transition to parenthood. Developmental Psychopathology 4:29-47.

Cantor, David, Jane Waldfogel, Jeffrey Kerwin, Mareena McKinley Wright, Kerry Levin, John Rauch, Tracey Hagerty, and Martha Stapleton Kudela. 2001. Balancing the needs of families and employers: The family and medical leave surveys 2000 update. Rockville, MD: Westat.

Chaudron, L. H., M. H. Klein, P. Remington. M. Palta, C. Allen, and M. J. Essex. 2001. Predictors, prodromes, and incidence of postpartum depression. Journal of Psychosomatic Obstetrics and Gynecology 22:103-12.

Civic, D., and V. L. Holt. 2000. Maternal depressive symptoms and child behavior problems in a nationally representative birthweight sample. Maternal and Child Health Journal 4:215-21.

Deal, L. W., and V. L. Holt. 1998. Young maternal age and depressive symptoms: results from the 1988 National Maternal and Infant Health Survey. American Journal of Public Health 88:266-70.

Department of Labor, Women's Bureau. 1990. State maternity/parental leave laws: Facts on working women No. 90-1. Washington, DC: Department of Labor.

DuMouchel, W. H., and G. J. Duncan. 1983. Using sample survey weights in multiple regression analyses of stratified samples. Journal of the American Statistical Association 78:535-43.

Eaton, W. W., C. Muntaner, C. Smith, A. Tien, and M. Ybarra. Center for epidemiologic studies depression scale: Review and revision (CESD and CESDR). In The use of psychological testing for treatment planning and outcomes assessment, edited by M. E. Maruish. Mahwah, NJ: Lawrence Erlbaum Associates. In press.

Family and Medical Leave Commission. 1995. A workable balance: Report to Congress on family and medical leave policies, 1995. Accessed November 2003. Available http://www.dol.gov/esa/regs/compliance/whd/fmla/family.htm.

Gjerdingen, D. K., and K. M. Chaloner. 1994. The relationship of women's postpartum mental health to employment, childbirth, and social support. Journal of Family Practice 38:465-72.

Gjerdingen, D. K., and D. Froberg. 1991. Predictors of health in new mothers. Social Science and Medicine 33:1399-407.

Gjerdingen, D. K., D. G. Froberg, K. M. Chaloner, and P. M. McGovern. 1993. Changes in women's physical health during the first postpartum year. Archives of Family Medicine 2:277-83.

Gjerdingen, D. K., P. M. McGovern, K. M. Chaloner, and H. B. Street. 1995. Women's postpartum maternity benefits and work experience. Family Medicine 27:592-8.

Husaini, B. A., D. A. Neff, J. B. Harrington, M. D. Hughes, and D. Segal. 1980. Depression in rural communities. Validating the CES-D scale. Journal of Community Psychology 8:20-7.

Hyde, J. S. 1995. Women and maternity leave: Empirical data and public policy. Psychology of Women Quarterly 19:299-313.

Hyde, J. S., M. H. Klein, M. J. Essex, and R. Clark. 1995. Maternity leave and women's mental health. Psychology of Women Quarterly 19:257-85.

Kessler, R. C., C. Barber, H. G. Birnbaum, R. G. Frank, P. E. Greenberg, R. M. Rose, G. E. Simon, and P. Wang. 1999. Depression in the workplace: Effects on short-term disability. Health Affairs 18:163-71.

Klerman, J. A., and A. Leibowitz. 1994. The work-employment distinction among new mothers. Journal of Human Resources 29:277-303.

Klerman, J. A., and A. Leibowitz. 1999. Job continuity among new mothers. Demography 36:145-55.

Lennon, M. C., J. Blome, and K. English. 2001. Depression and low income women: Challenges for TANF and welfare-to-work policies and programs. Research forum on children, families and the new federalism, national center for children in poverty, Mailman School of Public Health, Columbia University.

Maddala, G. S. 1983. Limited-dependent and qualitative variables in economics. Cambridge, UK: Cambridge University Press.

Mandl, K. D., E. Z. Tronick, T. A. Brennan, H. R. Alpert, and C. J. Homer. 1999. Infant health care use and maternal depression. Archives of Pediatrics and Adolescent Medicine 153:808-13.

Martins, C., and E. A. Gaffan. 2000. Effects of early mammal depression on patterns if infant-mother attachment: A metaanalytic investigation. Journal of Child Psychology and Psychiatry 41:737-46.

McGovern, P., B. Dowd, D. Gjerdingen, I. Moscovice, L. Kochevar, and W. Lohman. 1997. Time off work and the postpartum health of employed women. Medical Care 35:507-21.

McLennan, J. D., M. Kotelchuck, and H. Cho. 2001. Prevalence, persistence, and correlates of depressive symptoms in a national sample of mothers. Journal of the American Academy of Child and Adolescent Psychiatry 40:1316-23.

Miller, L. J. 2002. Postpartum depression. Journal of the American Medical Association 287:762-5.

Murray, C. J. L., and A. D. Lopez, eds. 1996. The global burden of disease and injury series, volume 1: A comprehensive assessment of mortality and disability from diseases, injuries, and risk factors in 1990 and projected to 2020. Cambridge, MA: Harvard School of Public Health on behalf of the World Health Organization and the World Bank, Harvard University Press.

National Partnership for Women and Families. 2004. State legislative round-up--State paid leave initiatives in 2004 and prior stale legislatures: making family leave more affordable. Accessed 21 January 2005. Available http://www.nationalpartnership.org/portals/ p3/library/PaidLeave/StateRoundUp2004.pdf.

O'Hara, M. W., E. M. Zekoski, L. H. Phillips, and E. J. Wright. 1990. Controlled prospective study of postpartum mood disorders: Comparison of childbearing and non-childbearing women. Journal of Abnormal Psychology 99:3-15.

Radloff, L. S. 1977. The CES-D scale: A self-report depression scale for research in the general population. Journal of Applied Psychological Measurement 1:385-401.

Rand Labor and Population Program Research Brief. 1995. Time-out for new mothers: Some issues for maternity leave policy. Accessed November 2003. Available http://www.rand.org/publications/RB/RB5009/RB5009.html.

Reading R., and S. Reynolds. 2001. Debt, social disadvantage and maternal depression. Social Science and Medicine 53:441-53.

Ruhm, C. J. 2000. Parental leave and child health. Journal of Health Economics 19:931-60.

Siefert K., P. J. Bowman, C. M. Heflin, S. Danzige, and D. R. Williams. 2000. Social and environmental predictors of maternal depression in current and recent welfare recipients. American Journal of Orthopsychiatry 70:510-22.

U.S. Department of Health and Human Services, National Center for Health Statistics. 1992. National Maternal and Infant Health Survey 1988 [computer file]. Hyattsville, MD: U.S. Department of Health and Human Services, National Center for Health Statistics [producer], 1991. Ann Arbor, MI: Inter-University Consortium for Political and Social Research [distributor].

Waldfogel, J. 1998. The family gap for young women in the U.S. and Britain: Can maternity leave make a difference? Journal of Labor Economics 16:505-45.

Waldfogel, J. 1999. Family leave coverage in the 1990s. Monthly Labor Review, October, pp. 13-21.

Waldfogel, J., W. J. Han, and J. Brooks-Gunn. 2002. The effects of early maternal employment on child cognitive development. Demography 39:369-92.

Weinberg, M. K., E. Z. Tronick, M. Beeghly, K. L. Olson, H. Kernan, and J. Riley. 2001. Subsyndromal depressive symptoms and major depression in postpartum women. American Journal of Orthopsychiatry 71:87-97.

Winegarden, C. R., and P. M. Bracy. 1995. Demographic consequences of maternal-leave programs in industrial countries: Evidence from fixed-effects models. Southern Economic Journal 61:1020-35.

Pinka Chatterji * and Sara Markowitz ([dagger])

* Center for Multicultural Mental Health Research at Cambridge Health Alliance/Harvard Medical School, 120 Beacon Street, Fourth Floor, Somerville, MA 02143, USA; E-mail: pchatterji@charesearch.org; corresponding author.

([dagger]) Department of Economics, Rutgers University, 360 Dr. Martin Luther King Jr. Boulevard, Newark, NJ 07102, USA: E-mail: smarkow@rutgers.edu.