Does Maternity Leave Induce Births?

Susan L. Averett [*]

Leslie A. Whittington [+]

Alleviating the tension between the conflicting responsibilities women may face as mothers and as workers is a topic of current policy interest. Expansion of guaranteed maternity leave to all employed women in the United States is suggested as one possible "family-friendly" solution. Controversy surrounding the issue of increased maternity leave centers around the potential cost to firms of widespread access to leave. One specific concern is that the availability of maternity leave will actually increase births among eligible working women. In this paper we use data from the National Longitudinal Survey of Youth to examine the impact of maternity leave on fertility. We explore two possible routes through which maternity leave may influence fertility. We first estimate the impact of desired fertility on the probability of being in a job offering maternity leave. We then estimate the impact of maternity leave and desired fertility on the probability of a birth. We find no evidence that women sort by fertility d esires into firms on the basis of their maternity leave policy. We do find that the probability of a birth increases as a result of maternity leave, and that the fertility effect of maternity leave increases with birth parity.

1. Introduction

The dramatic increase in U.S. female labor force participation in the past three decades has generated concern about the delicate balance between family and job that many women must achieve. In 1960, the labor force participation rate of all women in the United States was 37.7%. Thirty years later, the labor force participation rate of women reached 57.4%. Of perhaps even more policy interest is the fact that the participation rate of women in their prime childbearing years was actually higher than the overall female labor force participation rate by 1990. The labor force participation rates of mothers with preschool children have been rising steeply for several decades (Kamerman and Kahn 1991; Klerman and Leibowitz 1994; Olsen 1994). As a result of these trends there has been increased policy attention on how firms can accommodate the needs of women for both leave time after childbirth and stable job status. Public debate ultimately led to the adoption of the federal Family and Medical Leave Act (FMLA) in 19 93. Some firms did offer maternity leave (largely unpaid) before the passage of the FMLA, but in the United States it has generally been a benefit offered only to employees at large firms that pay relatively high wages or at government agencies (Kamerman and Kahn 1997). [1] The FMLA guarantees 12 weeks of unpaid parental (meaning both women and men are eligible) leave to most employees of relatively large firms. This offers substantial job protection to some parents after the birth of their children. It is, however, estimated that this legislation will only pertain to about half of U.S. workers because of coverage limitations, primarily because only firms employing over 50 persons are required to comply (Joesch 1995). Further, no firms are required by law to offer paid parental leave. [2]

The expansion of parental leave laws to provide coverage for all workers and the requirement that such leaves be compensated remain issues of national debate. Advocates emphasize that the United States is the only industrialized country that does not guarantee paid maternity leave (Kamerman and Kahn 1991). Critics argue that expanded leave will result in higher costs for employers as they have to hire replacement workers or deal with greater employee absenteeism. [3] Another cost concern is that women will be induced to have more births because maternity leave lowers the cost of a child, and that this increased fertility will exacerbate the financial burden on firms.

This final issue is the focus of this paper: Does employer-provided maternity leave affect fertility decisions of women in the United States? There is a growing literature exploring the impact of maternity leave on labor supply patterns and earnings in the United States (Waldfogel 1996, 1997; Klerman and Leibowitz 1997, 1998; Ruhm 1998). To date, however, no work specifically estimates the impact of maternity leave on births among U.S. women, although Winegarden and Bracy (1995) include aggregate U.S. data in a cross-national study on the topic.[4]

The impact of fertility incentives inherent in other U.S. policy vehicles has been previously explored. There is, for example, a large body of work exploring the relationship between Aid to Families with Dependent Children and childbearing (Moffitt 1992; Argys and Rees 1996; Robins and Fronstin 1996). Previous research has explored the role of tax exemptions for children (Whittington, Alm, and Peters 1990; Whittington 1992; Zhang, Quan, and Van Meerbergen 1994) and Medicaid benefits (Yelowitz 1994) on fertility, and both lines of work find that these policies have statistically significant, albeit small, effects on births. There is increasing empirical evidence supporting the theoretical supposition that economic policy can influence fertility, and we expand on this line of research by examining the impact of employer-provided maternity leave on the probability of a birth among working women.

Understanding the impact of maternity leave policies on fertility is crucial to fully estimating the costs of such programs. In the following section we present a conceptual framework of the relation between fertility and maternity leave, drawing on the standard neoclassical model of fertility. We then discuss the data and estimation strategies that we use, and our estimation results follow. Finally, we present our conclusions.

2. Conceptual Framework and Hypotheses

We consider two issues facing a firm considering the adoption of a maternity leave: Will the leave policy attract potential female employees most likely to have births? Will the maternity leave policy induce women in the firm to have more births? We hypothesize that the temporal ordering of events among working women is as follows. A woman first selects her job with or without maternity leave as a benefit. Then, she either has a birth or not. Because of the waiting period often required of benefits packages, a woman may be in a position for a year, or more, before having access to maternity benefits. It therefore seems unlikely, though not impossible, that a woman would move to a position with maternity leave because she is already pregnant, although we do recognize and discuss this possibility later.

Our conceptual outline is based on women who have already chosen to be in the labor market, as these are the only women for whom the maternity leave is currently relevant. The availability of maternity leave guarantees may also determine the probability of a woman participating in the labor force, either through initially attracting her to the labor market or through retention after a birth. We have no way of determining whether an individual woman who is not employed was, in fact, offered a position with maternity leave. Therefore, we cannot ultimately estimate the probability of entering the labor market as a function of being offered maternity leave. We do, however, acknowledge the importance of understanding the first sorting potentially caused by maternity leave--into or out of the labor market--as has already been carefully analyzed by other authors (Walker 1991; Winegarden and Bracy 1995; Klerman and Leibowitz 1998; Ruhm 1998). We necessarily start our analysis at the point where the woman has chosen t o participate in the labor market and look at the subsequent fertility impact.

An individual's decisions about work, marital status, and fertility are undoubtedly linked. In the analysis that follows, and in our empirical model, we make no attempt to simultaneously model all of these behaviors. Further, although the availability of parental leave is likely very important to some men, we confine our discussion and analysis to the impact of maternity leave on women.

Choosing Maternity Leave as a Benefit

Because maternity leave is not a benefit explicitly available with every job or firm, women may seek it out as a particular characteristic of their desired job just as people may search for other job benefits such as flexible schedules, tuition remission, or health insurance. Determining the impact of maternity leave on fertility, therefore, requires recognition of this potential sorting into jobs with maternity leave on the basis of anticipated fertility.

The probability of a woman having a job with maternity leave ([M.sub.i]) is a function of her desired fertility ([D.sub.i]); economic and social conditions in the area in which she resides (Z); and a vector of personal characteristics affecting her tastes, prices, and income ([P.sub.i]). This can be represented as:

[M.sub.i] = M([D.sub.i], Z, [P.sub.i]; [u.sub.i]),

where [u.sub.i] is an error term.

The impact of desired fertility is theoretically ambiguous. Women who expect to have children and remain attached to the labor market are likely to value maternity leave. Desired fertility will have a positive impact on the probability of these women having maternity leave coverage. Women who do not expect to remain in the labor force postpartum, on the other hand, or who expect to change working conditions by moving to a less stressful, time-consuming, or physically demanding job, may not value maternity leave in their current job. In fact, to the extent that provision of maternity leave replaces other fringe benefits and working conditions or reduces wages, desired fertility may actually have a negative impact on the probability of being in a job with maternity leave. Because of the competing effects, the expected sign of desired fertility on maternity leave-[partial][M.sub.i]/[partial][D.sub.i]--cannot be determined.

Maternity Leave and the Probability of a Birth

The probability of a birth can be represented as:

[B.sub.i] = B([M.sub.i], [D.sub.i], [P.sub.i], [W.sub.i], [J.sub.i]; [v.sub.i]),

where [W.sub.i] is the potential wage of the woman, [J.sub.i], is employment attributes, and [v.sub.i] is the error term. Maternity leave ([M.sub.i]) lowers the cost of a birth ([B.sub.i]) whether it is paid or unpaid leave, and lowering the cost of childbirth creates a fertility incentive. Hoem (1990) and Walker (1991), in analyses of maternity leave in Sweden, point out that this positive incentive may be dampened if there is any sort of minimum work period required to accrue full benefits. If workers are required to meet a minimum term of employment before becoming fully vested in maternity benefits, the existence of maternity leave might actually increase the time to birth (thereby decreasing the probability of a birth in early years). Further, a woman with maternity leave benefits may be a more highly valued employee of the firm, and may have a stronger, unobserved attachment to the labor force, making a birth less probable than for a woman with a lower labor force attachment (and no maternity leave). Th us, the direct effect of maternity leave on fertility,

d[B.sub.i]/d[M.sub.i] = ([partial][B.sub.i]/[partial][M.sub.i]),

cannot be determined a priori.

The impact of desired fertility on the probability of a birth is straightforward; we hypothesize that [partial][B.sub.i]/[partial][D.sub.i] [greater than] 0. The cumulative effect of desired fertility on the probability of a birth is the combination of the direct and indirect effects, through maternity leave, specifically

D[B.sub.i]/d[D.sub.i] = ([partial][B.sub.i]/[partial][D.sub.i]) + [([partial][B.sub.i]/[partial][M.sub.i]) ([partial][M.sub.i]/[partial][D.sub.i])].

Both the direct effect of maternity leave on births and the effect of desired children on the probability of having maternity leave are ambiguous, making it impossible to determine the direction of impact of the second term. Consequently, we cannot determine the expected analytically.

Fertility is also a function of a woman's opportunity cost, or wage rate. There is some ambiguity surrounding the potential impact of women's wages on fertility because of competing income and price effects. The general expectation, however, is that fertility will be negatively associated with a woman's wage rate, and this is supported by most empirical work (Schultz 1994). Thus, we expect [partial][B.sub.i]/[partial][W.sub.i] [less than] 0.

We expect that job tenure may play an important role in fertility decisions for working women. The receipt of full job benefits may not occur until an employee has been at a particular company for some minimum period of time. We expect that women will likely delay childbirth until they have established themselves sufficiently in the job to receive the full package of benefits. Further, the longer a woman has been associated with an employer before giving birth, the more likely it is that the employer has a commitment to retaining her as an employee. Therefore, women with longer job tenure are likely to be at lower risk of job loss or demotion as a result of childbirth. On the other hand, job tenure may signal a strong labor market attachment for women. This suggests that tenure is negatively correlated with the probability of having a birth. We expect that the relation between job tenure and fertility may be nonlinear, but we cannot predict the sign of either the primary effect or the quadratic term. We also suspect that job tenure may interact with the presence of maternity leave in determining fertility because women with longer job tenure are more likely to be eligible for full maternity benefits.

The number of previous births a woman has experienced is undoubtedly linked to her future fertility. We expect parity--the number of previous children--to have a negative impact on the birth probability. Maternity leave likely interacts with parity in the birth decision. Most women experience a first birth. Reducing the cost of a child at the margin (through maternity leave) will therefore likely have a stronger positive impact on a second or higher-order birth than on a first birth.

Is Maternity Leave Exogenous to Fertility?

Recall that we are not exploring whether maternity leave induces women into the labor force, and therefore are not directly examining the trade-off between labor force participation and fertility, a relation that economists do perceive as simultaneous. We focus our attention on the benefits-birth decision process among working women. The employment and benefit decision in our basic framework temporally precedes the observation of a birth. However, there is the possibility that firms with maternity leave appear more child-friendly to working mothers, and therefore attract some pregnant women regardless of their intentions regarding future maternity leave benefits. Further, it is possible that women in the firm pressed for maternity benefits precisely because they were having children. Thus, maternity leave may be endogenous to observed fertility. We test for endogeneity of maternity leave in our econometric work, as we discuss in the following sections. We also estimate separate models using a lagged maternity measure that is less likely to be endogenous to fertility.

3. Data and Estimation Strategies

National Longitudinal Survey of Youth (NLSY) Data

Our sample is drawn from the NLSY, which has been conducted annually since 1979 (CHRR 1994). At baseline (1979), respondents were ages 14 to 21. The NLSY contains very comprehensive labor force and demographic information, including the detailed fertility information necessary for this project. Most importantly, in 1979 the NLSY asked respondents the number of children they desired. Although this question was asked when the respondents were relatively young, it provides a control, albeit imperfect, for heterogeneity in generally unobserved fertility expectations. The NLSY also provides data on the number of siblings a woman has, and we use this as an alternative measure of desired family size.

Information on maternity leave was first collected by the NLSY in 1985, and this is one disadvantage of these data. Because the oldest women are 28 years old in 1985, it is possible that some have completed their childbearing. It is also possible that women who are at the upper end of the age range and have not yet experienced a birth have some unobserved characteristics that contribute negatively to the probability of having a birth. In both cases there may be some correlation with their fertility status and their job attachment and continuity over the previous years. This suggests that the error terms on the maternity leave and fertility equations could be correlated. At the lower end of the age range in our sample ([less than]22, for example), by selecting only working women we lose many young women who are pursuing a college education. This may introduce some sample selection bias. To determine if these characteristics of the NLSY introduce substantial bias we also use an age subsample in our model estima tion. Specifically, we estimate all models using the full age-range sample and again with a sample restricted to working women between the ages of 22 and 26.

There are other caveats that surround the use of the NLSY measure of maternity leave. First, although survey respondents are asked whether or not their employer provides job-protected maternity/paternity leave, we do not know if this is paid or unpaid leave. We do, however, know that explicit paid maternity leave was very rare in the United States during our sample period (Kamerman and Kahn 1997). Second, one may question what exactly the NLSY variable of maternity/paternity leave is really measuring. Does the firm provide an explicit maternity leave or is the employee simply aware that the firm makes some allowances for women surrounding pregnancy? Although it is possible that this is a noisy measure of maternity leave, the percentage reporting maternity leave in our sample is close to that reported nationally. At any rate, measurement error in the maternity leave variable biases the effect of maternity leave on births toward zero.

Choosing a Job With a Maternity Leave Benefit

The dependent variable in the maternity leave equation ([M.sub.i]) is whether or not a working woman chooses a job that offers maternity leave as a benefit, defined as

[M.sub.i] = {1 if the women reports that she has a maternity leave benefit 0 if the woman reports that she does not have a maternity leave benefit

The dependent variable is binary and so we estimate a logit transformation of the linear probability function. The standard logistic likelihood function is estimated for all observations.

We estimate the maternity leave equation using one year of data for each woman. Starting in 1985, we take the first year of data in which a women is working and reported whether or not she is in a job that offers her maternity leave. [5] Sample members are women who report working any hours during the survey week. Our sample includes 4679 women ranging in age from 20 to 28 years old. Almost 64% of the women have a job that offers some form of maternity leave. Sample means for the maternity leave equation are reported in Table 1.

The propensity of a worker to be in a job that offers maternity leave is specified as a linear function of individual characteristics that are expected to influence the preferences of the employee for the leave. Of primary interest in this research is the impact of fertility desires on the probability of observing a woman in a job with maternity leave, and so we use the desired number of children as a regressor. As additional determinants of the probability of being in a job with maternity leave, we include a set of regional dummy variables to allow for differences in tastes for maternity leave across the country; the northeastern United States is the omitted category. [6] We also include a measure of the woman's identification with traditional gender role attitudes. This variable is included to help control for "tastes" for maternity leave, for example to distinguish between those women who might be likely to use maternity leave versus those who would not use it even if it were available. [7] Controls for ra ce (omitted category is nonblack, non-Latina), age, marital status (omitted category is never married), Armed Forces Qualification Test (AFQT) score, work experience, and education are also included.

Probability of a Birth

We model fertility using a discrete time hazard model as an approximation of the continuous hazard (Allison 1984). The dependent variable ([B.sub.i]) is defined:

[B.sub.i] = {1 if the woman has a baby 0 if the woman does not have a baby

The unit of observation is the person-year. A woman contributes a person-year to the likelihood function each year from 1985 on that she is working, up to and including the year she has a birth or until 1992, the last year for which we have data. Because our focus is on working women, each sample member must have worked some positive number of hours in the survey week. We include a maximum of one "event" (birth) per woman, and so after a birth a woman contributes no additional person-years to the sample. Women who do not have full fertility histories are excluded from the sample, and years in which information on key variables of interest (i.e., availability of maternity leave) is missing are also excluded. These restrictions yield a sample of 3590 women who contribute a total of 14,895 person-years. [8] There are 1325 births over the period of observation, representing just under 9% of the person-years.

We are most interested in the impact of desired fertility and maternity leave on the probability of observing a birth. [9] Desired fertility is measured as in the maternity leave equation, and maternity leave is simply a dummy variable that equals one if a woman reports that she has a maternity leave benefit in that year. [10] We control for parity by including a variable equal to the number of children a woman has already. Because the effect of maternity leave may differ by parity, we also interact the two variables. Job tenure is measured in weeks and we include a quadratic tenure term to allow for expected nonlinearity. We also interact maternity leave with job tenure because of the potential for a phase-in period for full benefit receipt. The wage variable is the predicted log value. [11] We include a dummy variable for whether the individual has employer-provided health insurance. Additional control variables include age, urban residence, education, race, and unearned income.

4. Estimation Results

Maternity Leave

Results from estimation of the probability of maternity leave equation are presented in Table 2. The coefficients are presented as the partial derivatives evaluated at the means of the independent variables, or [delta]P/[delta]X = [beta]([P.sub.Mean][1 - [P.sub.Mean]]) from the logit P = 1/(1 + exp[-X]), where [P.sub.Mean] is the sample mean probability of having maternity leave and [beta] is the untransformed coefficient estimate.

Desired fertility does not approach statistical significance, and so appears to have no impact on the probability of a young woman choosing a job with maternity leave. In other model specifications, we use a variable measuring number of siblings of the women as an alternative measure of fertility plans. We likewise find that it has no significant impact on the sorting into jobs with maternity leave. [12] Our findings offer some support for the existence of the competing effects of desired fertility, as we discussed previously. Overall, we conclude that widespread selection into firms offering maternity leave by women seeking to utilize that leave (by having a child) was not a phenomenon occurring among the young women in our sample. This is consistent with the conclusions of Dalto (1989) based on the work of Gronau (1988) and Meyer (1978).

The probability of having maternity leave differs by race, with both black women and Latinas being significantly more likely to have leave than comparable white women. Work experience, AFQT score, and being married are also positive determinants of being in a job that offers maternity leave. Women residing in the western United States are significantly less likely to have maternity leave than residents of the Northeastern states, but we find no other significant regional differences. Finally, the probability of being in a job with maternity leave is negatively correlated with time (1985-1992), net of all the measured influences. In results not shown here, we enter separate dummy variables for each year and find that most are individually significant and all are negative. This finding is likely a result of the structure of our sample, however, rather than an indicator that selecting maternity leave has declined over the period.

There is the potential for selection bias by age in our sample, as we discussed previously. Although we anticipate that this bias would affect fertility more than the probability of being in a job with maternity leave, we reestimate the maternity leave model for a subsample of women ages 22-26 in 1985. Our findings largely substantiate those generated using the full sample of young women. One notable difference is that Latinas do not have a significantly higher probability of having maternity leave in this younger group, nor do married women. Of most importance, we again find that desired fertility has no significant impact on having a job that offers maternity leave. The same is true of number of siblings when it is substituted for desired fertility.

Fertility

In Table 3 we present the results of maximum likelihood estimation of the logit model of the probability of a birth. The first column presents the sample means of the person-year sample used in the fertility model. Column two (model 1) presents results from maximum likelihood estimation of a logit model of the probability of a birth using a contemporaneous measure of maternity leave. In the third column (model 2) we present the results from equations in which maternity leave is lagged one year, in both the direct and interaction effects. The coefficients in Table 3 are again presented as the partial derivatives at the means of the independent variables measuring the change in the probability of a birth with respect to each variable.

The coefficient on maternity leave measures the base effect of the availability of leave on fertility for women with no other children; adding the interaction term to the base effect captures the total impact of maternity leave by parity. Both the base effect and the interaction term are positive in model 1, but only the base effect is statistically significant at conventional levels. This does suggest that widespread adoption of guaranteed maternity leave policy will likely generate an increase in births among young, working women. Our finding is supported by the work of Winegarden and Bracy (1995), who conclude that national guaranteed maternity leave positively affects the general fertility rate of a country. They determine, however, that the direct, positive effect is almost completely offset by the indirect, negative effects on fertility caused by the impact of maternity leave on female labor force participation (increasing) and infant mortality (decreasing).

We argued earlier that the decision to be in a job with maternity leave temporally precedes the observed birth and so the behaviors are recursive. However, it can be argued that maternity leave may be endogenous to fertility in that the decisions may be simultaneous even if the behaviors are not temporally correlated. We test the exogeneity of maternity leave by generating a predicted value of having leave from the maternity leave equation, and then conducting a Hausman test in the fertility equation. Our results indicate that maternity leave is probably not endogenous in our sample (p = 0.16). Nevertheless, we also substitute a lagged value of maternity leave for the contemporaneous measure to test the sensitivity of our results. Some women may change jobs in the period, but generally, past maternity leave will be a reasonable instrument for current maternity leave among our sample of working women.

When we lag maternity leave by one year (model 2), our sample size declines because we lose all observations from 1985, the first year for which the maternity leave was included in the NLSY. However, the sample characteristics remain very consistent with those of the full sample presented in Table 3. A larger share of the sample has a maternity leave benefit (80.7 vs. 76.5%) and health insurance (80.6 vs. 75.8%), but otherwise there are no notable differences.

The model 2 results show that the base effect of maternity leave is statistically insignificant. This indicates that maternity leave does not influence the probability of having the first child. Interestingly, though, the interaction effect with parity is positive and statistically significant at a convincing level (p [less than] 0.001). Thus, after the first birth, the probability of a higher-order birth does increase with maternity leave. This is a plausible result because the marginal benefit of additional children may be in some sense lower, making parents more responsive to decreases in marginal cost at higher birth orders (Leibenstein 1974).

We also test for an interaction between maternity leave and job tenure. As the results in Table 3 demonstrate, we do not find that the interaction between the two variables is statistically significant, although job tenure itself exerts a significant positive, nonlinear effect on the probability of having a birth.

The overall magnitude of the positive effect of maternity leave on births that we find is relatively large for women at higher parities. Using the results from model 2, in Table 4 we calculate the probabilities of having a birth for black and white women of different parities with and without maternity leave. We have included only women who have one or more children because maternity leave does not appear to influence the first birth probability in model 2. The probability of a white working woman with one child and sample mean values of all other characteristics having another birth if she does not have a maternity leave benefit is 7.81. If, however, she has maternity leave the probability rises to 11.68, an increase of almost 50%. The probability for a similar black woman rises from 5.97 without maternity leave to 9.02 with maternity leave, an increase of 3.05 percentage points (51%).

Desired fertility has, theoretically, both a direct and indirect effect on fertility. The direct path is captured by the coefficient on desired fertility. In our results desired fertility is positive and significant at the 10% level or better. The indirect effect of desired fertility on a birth comes through the effect of desired fertility on the probability of having maternity leave, and the subsequent impact of maternity leave on fertility. However, because desired fertility plays no discernable role in the choice of maternity leave as a benefit, we conclude that the indirect impact of desired fertility on births is not a key determinant of fertility for this sample.

Surprisingly, the predicted wage of the woman is also not statistically significant, despite the fact that many researchers find it to be an important determinant of fertility (Schultz 1994). Our finding may be due to the construction of our sample, as we are using only working women. Further, we control independently for many of the factors that contribute to a woman's wage rate, such as age, education, and job tenure. As an alternative approach to using a predicted wage variable, we also estimated our models with the actual log wage and with simply the variables considered to be determinants of the wage. Both alternative approaches produced nearly identical parameter estimates on the key variables related to maternity leave and desired fertility.

The provision of health insurance is negatively associated with the probability of a birth, a finding that is also somewhat surprising. Health insurance lowers the direct cost of childbirth and thus we initially expected it to be positively related to fertility. However, the lack of health insurance may be signaling temporary or low-status jobs. Women may be in these jobs, in part, because they anticipate cycling out of the labor market because of childbirth. It may also be the case that women who intend to leave the labor market while rearing children rely primarily on their husband's insurance. Both of these possibilities could result in the negative relation between health insurance and fertility that we discover.

We find that the birth probability increases with age up to about 30, at which point it begins to decline. Unearned income also increases the probability of a birth, but the effect is quite small. In the full sample (model 1), urban women have a lower birth probability than other women, but this finding is not statistically significant in model 2. Finally, we determine that net of controlled influences, the black women in our sample are less likely to have a birth than the white women.

We estimate the fertility model for a subsample of person-years, eliminating all years in which the respondent is less than age 22 or older than age 26. This reduces our sample size but we continue to find that maternity leave has a statistically significant impact on the probability of a birth. The coefficient on maternity leave is almost double that of the full sample (model 1), indicating that the women in this age range may be particularly responsive to the existence of a maternity leave benefit.

The single variable measuring years of education is not a significant determinant of fertility, net of other control variables. When, however, we estimate fertility using categorical measures of education (results not presented), we find that the least educated women in our sample (less than high school) are the most responsive to maternity leave, with a marginal effect of 0.0454. There is little difference between high school graduates (marginal effect of 0.024) and women with higher education (marginal effect of 0.028).

5. Conclusions

It is often suggested that mandatory maternity leave in the United States will be prohibitively costly to business, and particularly devastating to small firms. Some opponents suggest further that providing maternity leave will induce women to have additional births, and that this will substantially increase the burden placed on firms. Opponents also argue that actual costs combined with employers' fears of increasing fertility will harm the position of women in the labor force because employers will steer away from hiring women in their reproductive years. To deal rationally with these concerns, it is crucial to understand whether women increase births in response to employer-provided maternity leave.

In this paper we explore two paths through which maternity leave provision may influence the birth rate experienced at a firm: selectivity into firms by women with strong fertility desires, and the direct impact of a maternity leave benefit on fertility. We find no evidence that working women who desire children self-select into firms offering maternity leave. Once in a firm, however, maternity leave does appear to directly increase the probability of a birth for working women, at least for women with at least one child already, and the effect is actually quite substantial in magnitude. Our work suggests that maternity leave may have no effect on first births.

What then is the potential impact on maternity leave on the number of births in the United States? The Census Bureau reports that 2,034,000 working women ages 15 to 44 had births in 1995, a rate of about 5.65% among the roughly 36 million working women in that age group. If all working women were given access to guaranteed maternity leave, the birth probability would presumably rise only among the 23.5% (on the basis of our sample) who previously had not benefited from such a policy, and, again, likely only for higher-order births. Thus, increased coverage might result in approximately an additional 118,000 births. This would increase the overall rough birth probability among working women ages 15 to 44 to just under 6%, an increase of .4 percentage points. Any change in the fertility or labor market behavior of women not in the labor market resulting from expanded maternity leave policy is not included in this estimated birth increase.

The microeconomic data that we use are not suited for an exploration of whether maternity leave induces women into the labor force, although there is evidence from aggregate work that it does. This possibility has intriguing implications for the impact of maternity leave on fertility. If women are actually induced into the labor force as a result of available maternity leave, the possibility for a negative relation between maternity leave and fertility arises. In fact, Winegarden and Bracy (1995) find that a national program for paid maternity leave increases labor force participation, which subsequently reduces fertility. This possibility clearly indicates the need for additional study on the topic of the fertility effects of maternity leave.

(*.) Lafayette College, Easton, PA 18042, USA; E-mail [e-mail]; corresponding author.

(+.) Georgetown Public Policy Institute, Georgetown University, 3600 N Street, N.W., Suite 200, Washington, DC 20007, USA; E-mail [e-mail].

Professor Leslie A. Whittington died tragically on September 11, 2001, in the airplane crash at the Pentagon after a hijacking. She, her husband, and two daughters were heading to Australia where she was to be a visiting fellow at Australia National University. She was a noted expert on taxation and its relationship to family behavior.

Earlier versions of this paper were presented at the 1996 annual meetings of the Population Association of America and the 1997 meeting of the North American Econometric Society. We thank David Figlio, Julie Hotchkiss, Jens Ludwig, Jacob Klerman, and two anonymous reviewers for helpful comments. Seminar participants at the Center for Population, Gender, and Social Inequality of the University of Maryland also provided useful guidance.

(1.) Across all firm types, paid explicit maternity leave benefits (as opposed to disability or sick leave) were very rare in the United States before the FMLA. The Bureau of Labor Statistics reports that 3% of full-time workers in medium and large firms received explicit paid maternity leaves in 1993; a mere 1% of full-time workers in small firms received paid maternity leave (U.S. Department of Labor 1994a,b).

(2.) Ruhm (1997) provides a full discussion of the specific provisions of the FMLA.

(3.) The issue of cost has been secondary in many debates surrounding parental leave policy in European nations. The principal motivation for expansive leave policies in Europe has been concern over falling fertility rates, and so the issue of primary interest is whether parental leave policies may have contributed to halting the decline. In the United States, the focus has been almost entirely on cost.

(4.) Walker (1991) examines the impact of paid parental leave on the timing and spacing of births in Sweden, and finds that maternity leave likely does not significantly affect fertility behavior. Hoem (1990) concludes that tile current structure of parental leave in Sweden is most likely to affect the timing of births rather than completed family size.

(5.) We have data on maternity leave for the years 1985 to 1992, although not every woman works in every year. We use only one year of data per woman because observations on a woman across years would not be independent. A cross-year fixed effects conditional logit would help capture unobserved heterogeneity in the decision to provide/seek out maternity leave. However, any variables that do not change over time would provide no information. The variable that we are principally interested in, desired fertility, is fixed and we therefore do not want to use a differencing scheme to estimate the maternity leave equation. Further, women who were either in or not in maternity leave jobs all years of the sample period would drop out of the model because of the differencing. This would leave us with a very small sample size.

(6.) State-mandated maternity leave would be a more precise measure, perhaps, but only Massachusetts adopted an unconditional maternity leave mandate before 1987 (Kallman 1996).

(7.) The attitude variable is an index ranging from 5 to 20, with higher scores reflecting more traditional gender role attitudes. It was created from an eight-item attitude scale. Our index was created from summing five of the eight items. Each of the selected items deals with the potential conflict between work outside the home and the traditional roles of women in the home. The total five-item scale yields a reliability coefficient of 0.74. For more details, see Center for Human Resource Research (CHRR) 1994.

(8.) Before experiencing a birth, a woman can contribute a person-year for each year that she is working, but she does not have to work continuously over the years under observation. We therefore have women that cycle in and out of the sample. This accounts for the difference in the number of women contributing information to the maternity leave equation, which contains information on women who work in a single year and the fertility equation.

(9.) Note that in the discrete-time hazard framework the coefficients are interpreted as the effect of the regressor on the probability of the event, not on the length of time until the event occurs.

(10.) It is possible that a woman may change jobs during the year, so that at the time she is interviewed she has a maternity leave benefit and later in the year does not have one (or vice versa). Although this measurement error may exist, we believe the incidence to be very small, especially for pregnant women. Women do not appear to be frequently job shifting in large numbers in these data, and the likelihood of a pregnant woman making a job change seems (anecdotally) small.

(11.) Hausman tests indicate that the observed wage is not endogenous in our fertility equation. However, because economic theory suggests that wages are endogenous in a fertility equation we elected to use the predicted wage in our models.

(12.) All estimation results discussed but not presented are available from the authors on request.

References

Allison, Paul D. 1984. Event history analysis. Beverly Hills, CA: Sage Publications.

Argys, Laura M., and Daniel I. Rees. 1996. The impact of welfare generosity on the fertility behavior of recipients. Center for Research on Economic and Social Policy, University of Colorado, Denver, Working Paper No. 9601.

Center for Human Resource Research (CHRR). NLS handbook. Ohio State University, 1994.

Dalto, Guy C. 1989. A structural approach to women's hometime and experience-earnings profiles: Maternity leave and public policy. Population Research and Policy Review 8:247-66.

Gronau, Reuben. 1988. Sex-related wage differentials and women's interrupted labor careers--The chicken or the egg. Journal of Labor Economics 6:277-301.

Hoem, Jan M. 1990. Social policy and recent fertility change in Sweden. Population and Development Review 16:735-48.

Joesch, Jutta. 1995. Paid leave and the timing of women's employment surrounding birth. Center for Studies in Demography and Ecology, University of Washington. Working Paper No. 95-10.

Kallman, Carol. 1996. Wage, hours and employment effects of state legislation mandating maternity leave. Unpublished paper, Boston College.

Kamerman, Sheila B., and Alfred J. Kahn. 1991. A U.S. policy challenge. In Child care, parental leave, and die under 3s, edited by Sheila B. Kamerman and Alfred J. Kahn. New York: Auburn House, pp. 1-22.

Kamerman, Sheila B., and Alfred J. Kahn. 1997. United States. In Family change and family policies in Great Britain, Canada, New Zealand and the United States, edited by Sheila B. Kamerman and Alfred J. Kahn. Oxford: Clarendon Press, pp. 305-421.

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

Klerman, Jacob A., and Arleen Leibowitz. 1997. Labor supply effects of state maternity leave legislation. In Gender and family issues in the workplace. edited by Francine D. Blau and Ronald C. Ehrenberg. New York: Russell Sage Foundation, pp. 65-85.

Klerman, Jacob A., and Arleen Leibowitz. 1998. FMLA and the labor supply of new mothers: Evidence from the June CPS. Unpublished paper, presented at the Annual Meeting of the Population Association of America.

Leibenstein, Harvey. 1974. An interpretation of the economic theory of fertility: Promising path or blind alley? Journal of Economic Literature 12:457-79.

Meyer. M. 1978. Women and employee benefit plans. New York: The Conference Board.

Moffitt, Robert. 1992. Incentive effects of the U.S. welfare system, Journal of Economic Literature 30:1-61.

Olsen, Randall J. 1994. Fertility and the size of the U.S. labor force. Journal of Economic Literature 32:60-100.

Robins, Philip K., and Paul Fronstin. 1996. Welfare benefits and birth decisions of never-married women. Population Research and Policy Review 15:21-43.

Ruhm, Christopher J. 1997. Policy watch: The Family and Medical Leave Act. Journal of Economic Perspectives 11:175-86.

Ruhm, Christopher J. 1998. The economic consequences of parental leave mandates: Lessons from Europe. The Quarterly Journal of Economics 113:285-318.

Schultz, T. Paul. 1994. Marital status and fertility in the United States. Journal of Human Resources 29:637-69.

U.S. Department of Labor, Bureau of Labor Statistics. 1994a. Employee benefits in medium and large private establishments, 1993. Bulletin No. 2456. Washington, DC: U.S. Government Printing Office.

U.S. Department of Labor, Bureau of Labor Statistics. 1994b. Employee benefits in small private establishments, 1992. Bulletin No. 2441. Washington, DC: U.S. Government Printing Office.

Waldfogel, Jane. 1997. Working mother's then and now: A cross-cohort analysis of the effects of maternity leave on women's pay. In Gender and family issues in the workplace, edited by Francine Blau and Ronald Ehrenberg. New York: Russell Sage Foundation, pp. 92-126.

Waldfogel, Jane. 1996. The impact of the Family and Medical Leave Act on coverage, leave-taking, employment, and earnings. Unpublished Paper, Columbia University.

Walker, James R. 1991. Maternity benefits, fertility, and female labor supply. Paper presented at the Fourth Annual Conference of the European Society for Population Economies, Pisa, Italy, June 6-8, 1991.

Whittington, Leslie A. 1992. Taxes and the family: The impact of the tax exemption for dependents on marital fertility. Demography 29:215-26.

Whittington, Leslie A.; James Alm; and H. Elizabeth Peters. 1990. Fertility and the personal exemption: Implicit pronatalist policy in the United States. American Economic Review 80:545-56.

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

Yelowitz, Aaron. 1994. Is health care coverage a pro-natal policy? UCLA Department of Economics Working Paper.

Zhang, Junsen, Jason Quan, and Peter Van Meerbergen. 1994. The effect of tax-transfer policies on fertility in Canada, 1921-1988. Journal of Human Resources 29:181-201.

Table 1.

Sample Means for Cross-Section Maternity Leave Model

Continuous variables

Variable                            Mean     (Standard deviation)

Number of children desired            2.492        (1.516)
Age in years                         24.841        (2.793)
Education in years                   12.822        (2.207)
Previous work experience in years     4.524        (2.338)
AFQT Score (percentile)              40.791       (26.789)
Traditional gender role attitudes    10.331        (2.765)
Number of siblings                    3.781        (2.765)

Dummy variables

Variable                            Mean     (Standard deviation)

Firm offers maternity leave           0.638
Black                                 0.253
Latina                                0.159
North Central                         0.225
South                                 0.401
West                                  0.185
Married                               0.393
Divorced/separated                    0.465
AFQT score missing                    0.042
N                                  4679
Table 2.

Maximum Likelihood Estimates of the Probability of Maternity Leave

Variable                                  Probability of Maternity
                                                       Leave [a,b]

Desired number of children                                  -0.002
                                                           (0.107)
Black                                                        0.105 [***]
                                                          (24.063)
Latina                                                       0.054 [*]
                                                           (5.846)
AFQT score/100 [c]                                           0.190 [**]
                                                           (9.459)
Missing AFQT score                                           0.019
                                                           (0.257)
North Central                                               -0.022
                                                           (0.920)
South                                                        0.012
                                                           (0.316)
West                                                        -0.077 [**]
                                                          (10.017)
Age                                                          0.037
                                                           (0.895)
Age squared                                                 -0.001
                                                           (1.472)
Education                                                    0.001
                                                           (0.050)
Work experience                                              0.046 [***]
                                                          (142.41)
Married                                                      0.082 [*]
                                                           (5.772)
Divorced or separated                                        0.005
                                                           (0.041)
Desired children X married                                  -0.008
                                                           (0.655)
Traditional gender role attitudes                            0.004
                                                           (1.485) [***]
Year of observation (1 = 1985, 2 = 1986,                    -0.045
etc.)                                                     (52.080)
Chi-square (17 d.f.)                                       478.352
N                                                         4679

(a)Chi-squared statistics in parentheses.

(b)The coefficients are presented as [delta]P/[delta]X therefore the
intercept is not shown.

(c)Missing values of this variable were replaced with the mean value for
all respondents.

(*),p [less than or equal to] 0.05;

(**),p [less than or equal to] 0.01;

(***),p [less than or equal to] 0.001.
Table 3.

Maximum Likelihood Estimates of Probability of a Birth

                                  Sample Mean [a]
                                        (standard    Model l [b]
Variable                               deviation)   (chi-square)

Dependent variable = 1 if birth
 occurs, 0 otherwise                       0.089
Desired number of children                 2.465     0.003 [*]
                                          (1.471)   (4.316)
Maternity leave is available               0.765     0.037 [***]
                                                   (12.267)
Parity                                     0.628    -0.016 [***]
                                                    (7.620)
Maternity leave X parity                             0.010
                                                    (2.479)
Age                                       27.116     0.061 [***]
                                          (3.052)  (20.384)
Age squared                                         -0.001 [***]
                                                   (22.458)
Urban residence                            0.779    -0.013 [*]
                                                    (4.171)
Don't know if urban residence              0.052    -0.011
                                                    (0.832)
Predicted log wage                         2.089    -0.20
                                          (0.551)   (1.188)
Years of education                        13.301    -0.001
                                          (2.187)   (0.451)
Black                                      0.282    -0.022 [***]
                                                   (11.685)
Latina                                     0.164     0.002
                                                    (0.117)
Unearned income (1000s of $)              25.919     0.000 [**]
                                         (80.557)   (8.575)
Missing unearned income                    0.158    -0.004
                                                    (0.428)
Job tenure (years)                         2.945     0.014 [***]
                                          (2.901)  (14.207)
Job tenure squared                                  -0.001 [***]
                                                   (16.104)
Maternity leave X tenure                             0.000
                                                    (0.008)
Employer offers health insurance           0.758    -0.036 [***]
                                                   (27.015)
Number of siblings                         3.612
                                          (2.487)
N                                                          14895
Chi-square (18 d.f.)                               175.80


                                        Model 2
Variable                           (chi-square)

Dependent variable = 1 if birth
 occurs, 0 otherwise
Desired number of children          0.003 [+]
                                   (2.827)
Maternity leave is available        0.002
                                   (0.033)
Parity                             -0.031
                                  (15.526)
Maternity leave X parity            0.028 [***]
                                  (12.350)
Age                                 0.060 [***]
                                  (12.617)
Age squared                        -0.001 [***]
                                  (14.276)
Urban residence                    -0.007
                                   (0.950)
Don't know if urban residence      -0.005
                                   (0.112)
Predicted log wage                 -0.033
                                   (2.398)
Years of education                  0.000
                                   (0.027)
Black                              -0.024 [***]
                                  (10.279)
Latina                              0.001
                                   (0.023)
Unearned income (1000s of $)        0.000 [**]
                                   (6.826)
Missing unearned income            -0.012
                                   (2.013)
Job tenure (years)                  0.012 [***]
                                   (8.450)
Job tenure squared                 -0.001 [***]
                                  (14.984)
Maternity leave X tenure            0.003
                                   (1.121)
Employer offers health insurance   -0.027 [***]
                                  (11.752)
Number of siblings

N                                  11073
Chi-square (18 d.f.)              116.95

Model 1 uses the contemporaneous maternity leave variable; model 2 uses
the lagged maternity leave variable.

(a)sample means are from the person-year sample with the exception of
the variables Black and Hispanic, which are from the actual sample of
women since they do not change over time.

(b)The coefficients are presented as [delta]P/[delta]X therefore the
intercept is not shown.

(+)p [less than or equal to] 0.10;

(*)p [less than or equal to] 0.05;

(**)p [less than or equal to] 0.01;

(***)p [less than or equal to] 0.001.
Table 4.

Predicted Probabilities of a Birth

                             White Women                  Black Women
                No Maternity   Maternity              No Maternity
                   Leave         Leave    Difference     Leave

One child           7.81         11.68       3.87         5.97
Two children        5.54         11.34       5.80         4.21
Three children      3.91         11.01       7.10         2.96

                      Black Women
                Maternity
                  Leave    Difference

One child         9.02        3.05
Two children      8.75        4.54
Three children    8.48        5.52

Probabilities are calculated using the coefficients presented in Table
3, mode 2 (lagged maternity leave) and evaluated at the sample means.