The effects of maternity leave legislation on mothers' labor supply after childbirth.

1. Introduction

Many more women work today compared with 25 years ago, and many of these women want to simultaneously raise children and work (Klerman and Leibowitz 1994; Leibowitz and Klerman 1995). Some of these women leave their jobs for a short period of time to give birth, only to return to the workforce soon after childbirth (Klerman 1994). To deal with the needs of these women, maternity leave legislation (MLL) has been passed in the United States, starting on the state level in 12 states in the late 1980s and early 1990s and culminating in 1993 with President Clinton's signing of the first federal maternity leave law, the Family and Medical Leave Act (FMLA). Like many of the 12 states' maternity leave laws passed before the FMLA, the FMLA allows eligible women to take up to 12 weeks of unpaid leave and guarantees that they can return to their old jobs. (1) Before passage of MLL, if women left the workforce to give birth, their employers were not required to give them their jobs back.

MLL may affect mothers' leave-taking and return-to-work decisions if the amount of employer-provided maternity leave is less than that mandated by the government and if a mother is eligible. Klerman and Leibowitz (1997) predict that MLL will have a positive effect on the incidence of leave taking and on the number of mothers who return to work at their prechildbirth jobs. They also predict that MLL will allow some mothers to take more maternity leave, delaying their return to work at their prechildbirth jobs. To arrive at these predictions, Klerman and Leibowitz first consider affected mothers who would have quit work in the absence of MLL. They predict that, with MLL, some of these mothers will now take maternity leave and return to their prechildbirth jobs. This occurs because MLL increases the amount of maternity leave offered by the employer, consequently increasing the value of returning to work at the prechildbirth job relative to quitting. Klerman and Leibowitz next consider affected mothers who would take a suboptimal amount of maternity leave and return to their prechildbirth jobs absent MLL. They predict that MLL will allow these mothers to get closer to (or actually attain) their optimal amount of maternity leave (and delay their return) by increasing the amount of employer-provided maternity leave.

The effects of MLL on leave-taking and return-to-work decisions are particularly important to employers. If MLL increases the incidence of leave taking or delays mothers' return to work, then MLL may increase the cost of production. The cost of production would increase because MLL requires employers to continue providing benefits such as health insurance coverage for employees on leave. Additionally, MLL would increase the cost of production by increasing employee absenteeism. If mothers take leave, then employers must hire and train temporary replacement workers to substitute for these mothers or else continue production without replacements. Certainly there are hiring and training costs associated with temporary workers. In addition, temporary workers should have less firm-specific human capital, making them less productive than these mothers would have been. Continuing to produce without temporary replacements for these mothers may increase demands for other workers to produce. Requiring remaining workers to cover absent employees' duties may harm worker morale.

Conversely, MLL may benefit employers by preserving employer--employee relationships if permanent separations are costly. Employers would not lose their investments in workers and they would not have to hire and train permanent replacements. However, Ruhm (1998) argues that employers and employees are free to privately negotiate maternity leave benefits in the absence of MLL. If maternity leave can be voluntarily negotiated, then optimal employer--employee relationships will be preserved without the legislation. (2) If this is the case, then the only workers whose behavior is affected by MLL are those without employer-provided maternity leave for whose employer's job turnover is not terribly costly. These may be low-skilled workers with low levels of firm-specific human capital, and employers may not find it optimal for these employees to be allowed to return to their prechildbirth jobs.

I examine three outcomes that are of interest to employers: how MLL affects (i) the incidence of leave taking, (ii) the probability that mothers eventually return to their prechildbirth jobs, and (iii) the timing of these mothers' return to work. (3) To address these outcomes, I first estimate a sequential discrete-outcome model that consists of a probit for the incidence of leave taking and, conditional on initially taking leave, a subsequent multinomial probit (MNP) for the probability of returning to work at the prechildbirth job and starting work at a new job. The equations are estimated simultaneously to control for cross-equation correlation in order to account for possible biases arising from unobserved heterogeneity. Then a second approach is taken where again the leave-taking equation is estimated but the subsequent equation becomes a dynamic MNP showing the timing of a mother's return to work. Conditional on taking leave, the dynamic MNP shows the monthly probability of returning to the prechildbirt h job and starting a new job for the first eight months after giving birth. The National Longitudinal Survey of Youth (NLSY) data set is used, which allows identifying each mother's state of residence and, consequently, whether state or federal leave mandates are in force. Because state-level maternity leave laws passed before the FMLA vary, the effects of MLL can be studied as a natural experiment: A "treatment" group of mothers for whom MLL applies is compared with a control group of mothers who were covered by neither the FMLA nor mandated leave from their state. The NLSY also allows determination of whether the data suggest that mothers are eligible for MLL. The results provide some evidence that MLL increases the probability that mothers eventually return to their prechildbirth jobs, but the results do not provide significant evidence that MLL impacts the incidence of leave taking. Further, the dynamic MNP model shows that MLL allows mothers to delay their return to work at their prechildbirth jobs. The results are somewhat stronger when account is taken of whether mothers are eligible for MLL. The remainder of the article is as follows: Section 2 reviews the literature and explains how this paper extends that literature; section 3 specifies the model; section 4 describes the data; section 5 presents the results; and section 6 discusses the results and concludes.

2. Literature Review

Two studies use multivariate regression analysis to explicitly determine the effects of government-mandated maternity leave in the United States (Klerman and Leibowitz 1997; Waldfogel 1999). Other studies (Ruhm and Teague 1997; Ondrich et al. 1998; Ruhm 1998) have examined the effects of European maternity leave mandates, but because European mandates guarantee leave that is both paid and longer in duration, the effects of MLL passed in the United States may be different.

Klerman and Leibowitz (1997) use data from the 1980 and 1990 censuses to see how state MLLs have affected mothers' employment status. In particular, they examine mothers' employment status at two points in time (1980 and 1990), where employment categories include employed, working, and on leave. Employing a difference-in-difference-in-difference technique, their preferred model shows that state MLL has not had significant effects.

Waldfogel (1999) examines the Bureau of Labor Statistics' Employee Benefits Survey and the 1994 Westat survey and finds that leave coverage offered by firms increased after the FMLA. In particular, she finds that two thirds of all employers reported having to change their leave policies due to the FMLA. Examining the March Current Population Survey (CPS), she finds that leave taking also increased after the FMLA. Then she employs a difference-in-difference-in-difference methodology with the CPS data to determine the effect of the FMLA on employment and wages. She finds that the FMLA has had a small, positive effect on employment and essentially no effect on wages.

In a related study, Averett and Whittington (2001) investigate other potential effects of MLL such as whether maternity leave affects fertility. Their results suggest that women with maternity leave are significantly more likely to give birth. Averett and Whittington also examine whether women who plan on giving birth are more likely to select jobs that offer maternity leave. Results suggest that fertility plans do not influence job sorting based on maternity leave benefits.

The literature has three primary shortcomings. First, the literature uses cross-sectional data rather than panel data and is therefore only able to determine the effects of MLL at a particular point in time. None of the existing studies examines the effect of MLL on the timing of the mother's return to work or on whether working mothers return to work for their prechildbirth employers. Second, the existing literature often fails to take advantage of both state variation in maternity leave laws and variation created by the FMLA. Third, none of the literature determines whether mothers are eligible for government-mandated maternity leave. (4) Many mothers who are living in states where MLL is in force are not actually eligible for the mandates either because they do not have the requisite work history or because they work for firms of insufficient size.

The literature is extended primarily by estimating a dynamic model that allows determining how MLL affects the timing of the mother's return to work as well as whether the mother actually returns to her prechildbirth employer. Additionally, both state variation in MLL and variation created by the FMLA are exploited. Because a panel data set is used, it is possible to determine whether the data suggest that the mother is eligible for MLL based on her work history and the employer's firm size.

3. Empirical Specification

The first approach estimates an equation for the effect of MLL on the probability that a mother takes leave from work and a subsequent equation for the probability that a mother who took leave eventually returns to work for her prechildbirth employer. The first-period outcome, which measures the incidence of maternity leave, is a static choice between taking time off from work after giving birth (temporarily or permanently, for any length of time) or not. (5) Let the expected present discounted value for mother i taking leave from work (L) after giving birth be [V.sup.L.sub.1i] and for not taking leave from work (NL) be [V.sup.NL.sub.1i], where

[V.sup.k.sub.1i] = [X.sub.1i][[beta].sub.k] + [[epsilon].sub.1ik]

for k = L and NL, where X is a vector of explanatory variables, 3 is a vector of coefficients, and [epsilon] is the disturbance. Mother i takes leave from work after giving birth if [V.sup.L.sub.1i] > [V.sup.NL.sub.1i]. It is assumed that [[epsilon].sub.1ik] follows the normal distribution, so the probability that the mother takes leave from work [[lambda].sup.L.sub.1i] takes the probit form.

The second period is represented by another static discrete choice equation made by mothers subsequent to their initial leave decision. This second-period equation measures the conditional probability that leave takers return to work for their prechildbirth employers (within eight months). (6) However, mothers who do not eventually return to their prechildbirth employers include mothers who start jobs for new employers and mothers who permanently leave the labor force. Thus, the second period is treated as a multinomial discrete-choice decision where three outcomes are modeled: the probability of returning to work for the prechildbirth employer, the probability of starting work for a new employer, and the probability of leaving the labor force permanently. Let the expected present discounted values for mother i returning to work for the prechildbirth employer (P), starting a new job (N), and permanently remaining out of the labor force (O) be given by [V.sup.P.sub.2i], [V.sup.N.sub.2i], and [V.sup.O.sub.2i], respectively, where

[V.sup.j.sub.2i] = [X.sub.2j][[beta].sub.j] + [[epsilon].sub.2ij]

for j = P, N, and O, where X, [beta], and [epsilon] are as defined above. Mother i is assumed to choose activity j from the set of J to maximize utility such that

[V.sup.j.sub.2i] = max {[V.sup.P.sub.2i], [V.sup.N.sub.2i], [V.sup.O.sub.2i]}.

It is assumed that the [epsilon]'s are jointly normally distributed, which yields the multinomial probit (MNP) functional form for the second-period probabilities of returning to the old job ([[lambda].sup.P.sub.2]) and switching to a new job ([[lambda].sup.N.sub.2i]).

The advantage of using a MNP instead of a multinomial logit (MNL) is that the MNP's error terms are correlated across the alternatives; conversely, the MNL assumes the outcomes' error terms are independent, which seems unrealistic in this context. Formally, the covariance matrix of the second period MNP's error terms ([SIGMA]) is

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII],

where [[sigma].sup.2.sub.j] is the variance of [[epsilon].sub.2ij] and [[rho].sub.jj], is the correlation between the alternatives for j = O, P, and N and j' = O, P, and N. Standard assumptions are made to identify the model. First, the utility associated with returning to the prechildbirth job and starting work at a new job relative to the value of leaving the labor force is measured. This yields a value of zero for [[sigma].sub.O]. To simplify the estimation, [[sigma].sub.P] and [[sigma].sub.N] are restricted to equal one (see Light and Strayer 1996).

The only mothers who contribute to the second-period equation are those who take leave from work in the first period and are deciding whether to return. The two discrete choice equations are estimated simultaneously to allow cross-period correlation among the unobservables. if the same unobserved factors determine whether mothers take leave from work and whether leave takers return to their prechildbirth employers, then failure to control for cross-equation correlation will result in biased estimates. For example, suppose that, when MLL is unavailable, the only mothers who take maternity leave are those with very high preferences for leisure. When MLL is available, mothers with lower preferences for leisure will also take leave from work. Thus, among mothers who take leave, those who are covered by MLL systematically have different values for leisure than those who are not covered. If the same unobserved variables affect the subsequent decision to return to the prechildbirth employer, then self-selection into the leave-taker category must be taken into account in order to identify the true effect of MLL on the probability of eventually returning to work for the prechildbirth employer. To model this cross-equation correlation, it is assumed that the error terms include an independently and identically distributed component (v) and components representing the unobserved person-specific factors ([[micro].sub.1], ..., [[micro].sub.M]),

[[epsilon].sub.1ik] = [V.sub.1k] + [summation over (M/m=1)] [[gamma].sub.1km] [[micro].sub.im]

[[epsilon].sub.2ij] = [V.sub.2j] + [summation over (M/m=1)] [[gamma].sub.2jm] [[micro].sub.im],

where the [gamma]'s are factor loadings and M is the number of common factors for k = L and NL and j = P, N, and O. This structure assumes that the idiosyncratic disturbances (the v's) are uncorrelated with the unobserved factors (the [micro]s), but cross-equation correlation exists because the error structure contains the same unobserved variables (the [micro]s). This model's complete conditional likelihood (LL) function contribution for mother i is

[LL.sub.i]([[micro].sub.1],...,[[micro].sub.M]) = [[lambda].sup.L.sub.li]([L.sub.i] = 1 | [[micro].sub.1],...,[[micro].sub.M]){[summation over (2/j=1)] [d.sub.ij][[lambda].sup.j.sub.2i]([d.sub.ij] = 1 | [L.sub.i] = 1, [[micro].sub.1],...,[[micro].sub.M])}

for j = P and N, where [L.sub.i], is an indicator variable that equals one if the mother takes leave in the first period and where [d.sub.iP], and [d.sub.iN] are second-period indicator variables that equal one if mother i returns to her prechildbirth employer and switches to a new job in the second period.

As a second strategy, a second model is estimated that is a variant of the first. The second model continues to estimate the first-period probability that a mother takes leave from work as described above, but this model specifies the second period to become multiple periods so that examination of the timing of the mother's return to her prechildbirth employer can be examined. Essentially, this approach estimates the monthly probability that mothers who are on leave return to their old employer and start work for a new employer for each of the first eight months after giving birth. The only mothers who are "at risk" to return to work are those who are still on leave; once a mother returns to work at her prechildbirth job or starts work at a new job, she exits the sample. Assume the expected present discounted values for mother i returning to work for the prechildbirth employer (P), starting a new job (N), and remaining out of the labor force (O) in month t are given by [V.sup.P.sub.3it], [V.sup.N.sub.3it], an d [V.sup.O.sub.3it], respectively, with associated utility indexes of

[V.sup.j.sub.3it] = [X.sub.3j][[beta].sub.j] + [[epsilon].sub.3ij]

for j = P, N, and O, where X, [beta], and [epsilon] are as defined above. In month t, mother i is assumed to choose activity j from the set of J to maximize utility. That is, choose j if [V.sup.j.sub.3it] > [V.sup.j'.sub.3it] for all j [not equal to] j'. Let the probability of returning to the old job and switching to a new job in time period t be [[lambda].sup.P.sub.3it] and [[lambda].sup.N.sub.3it], respectively. I again assume the alternatives' errors are jointly normally distributed, which yields a dynamic MNP model for each month in which the mother has not yet returned to work. The error covariance matrix for this version of the second-period model is the same as that defined above for the first approach's second-period model. Many others have used the MNP functional form to estimate dynamic discrete choice models (e.g., see Blau 1997, 1998; Hotz et al. 1999).

Like the first approach, this second approach is estimated as a sequential dynamic discrete-choice model where the first-period incidence of leave equation is estimated simultaneously with the second-period dynamic multinomial probit model. This is done because the self-selection problem described above continues to apply. Let the error structure for the second-period discrete-choice model be specified as

[[epsilon].sub.3ijt] = [v.sub.3jt] + [summation over (M/m=1)] [[gamma].sub.3jm][[micro].sub.im],

where the [gamma]s and M are as defined above for j = P and N. This structure again assumes that the idiosyncratic disturbances (the vs) are uncorrelated with the unobserved factors (the [micro]s), but cross-equation correlation exists because the error structure contains the same unobserved variables (the [micro]s).

Failure to control for unobserved heterogeneity in the second-period dynamic MNP model also produces biased results due to dynamic self-selection. In particular, dynamic self-selection bias results because the monthly probabilities are estimated from a sample of mothers who have not yet returned to work. Thus, subsequent time periods may contain a sample of mothers with heterogeneous characteristics. Assume that some mothers have high preferences for market-place work and some mothers have low preferences for market-place work. Also, assume that tastes for market-place work are unobserved. Returning to work quickly after giving birth will be correlated with high preferences for work, and the "surviving" sample of mothers who have not yet returned to work will tend to have low preferences for market-place work. Lancaster (1979, 1985) shows that, when dynamic self-selection is present, unobserved heterogeneity will bias the effect of any included regressor toward zero. Conditional on the [micro]'s, the complete li kelihood function contribution for mother i in the second approach is

L[L.sub.i]([[micro].sub.1],...,[[micro].sub.M]) = [[lambda].sup.L.sub.li]([L.sub.i] = 1 | [[micro].sub.1],...,[[micro].sub.M])[[[PI].sup.T.sub.t=1]{[[PI].sup.2 .sub.j=1] [d.sub.itj][[lambda].sup.j.sub.3it]([d.sub.itj] = 1 | [L.sub.it-1] = 1, [[micro].sub.1],...,[[micro].sub.M])}]

for j = P and N, where [L.sub.i], [d.sub.itP], and [d.sub.itN] are indicator variables as defined above for time periods t = 1 to T, where T is the maximum number of months covered by the model (eight months).

To control for unobserved heterogeneity, a strategy is used similar to the one proposed by Heckman and Singer (1984) and used by many others (Gritz 1993; Ham and LaLonde 1996; Blau and Hagy 1998; Hotz et al. 1999; Mroz 1999), where a step function is used to approximate the distribution of the unobserved variables. In particular, the discrete values of the unobserved factors and their associated probabilities with the [beta]'s are jointly estimated. The Appendix describes the details of the nonparametric maximum likelihood estimation strategy more thoroughly, describing identification and determination of the number of factors and mass points to use.

4. Data

Data from the National Longitudinal Survey of Youth (NLSY) are used to estimate the effects of MLL. Beginning in 1979, the NLSY collects yearly information on the labor market experiences and background characteristics of people who were between the ages of 14 and 21 in 1979, and the survey remains in progress. The original NSLY sample contained 6283 women and an oversample of blacks, Hispanics, low-income whites, and military personnel. The military sample was dropped in 1984 and the low-income white sample was dropped in 1990; respondents from either sample are not considered. First, women were selected who gave birth from 1988 to l994. (7) Table 1 specifies the size and inclusion requirements for each sample and shows that 2101 NLSY women gave birth to 3037 children during this period. From this group, observations are excluded that do not provide the requisite information to be used in the estimation. This leaves 2796 observations (births). Also excluded are mothers who were not employed before giving bir th because mothers who are out of the labor force are not expected to take any maternity leave from work. Mothers are considered to be employed before giving birth if they worked for an employer within three months before giving birth. (8) This leaves 1712 births to be included in the sample. When weighted, the sample will be a nationally representative sample of children born from 1988 to 1994 to mothers who were between the ages of 23 and 30 in 1988. However, results from this sample will not necessarily be representative of other cohorts or children born to this cohort before 1988 or after 1994.

Because the NLSY collects extensive information on each mother's employment status, constructing a work history is possible for each mother that identifies whether she is employed in each month after giving birth. The NLSY identifies unpaid gaps in employment spells when the individual was employed but not working. However, prior to the 1988 survey, it was not possible to identify whether employed respondents were actually working or on paid leave. This distinction is important because many mothers who are employed after giving birth are not actually working but are instead on leave. Fortunately, beginning with the 1988 survey, the main NLSY questionnaire began identifying whether employed mothers were working or on paid leave. (9) Thus, disaggregating each mother's employment status in each month after giving birth into the following mutually exclusive and exhaustive categories is possible: working and not working (on leave). Only included are children who were born after 1987 in the sample so that identific ation is possible of the months in which their mothers worked after their births.

Unfortunately, the NLSY stopped annually interviewing respondents after the 1994 survey and began interviewing biennially. Though the work history file continues to track respondents on a weekly basis, biennial year-long gaps began during which it is not possible to identify paid leave because the main NLSY questionnaire was not administered on an annual basis. Therefore, the sample only contains births covered by the 1988-1994 surveys.

Each mother is followed for a maximum of eight months after giving birth because no mother in the sample returns to her prechildbirth job after this time. Once a mother returns to work after childbirth, there is no more interest in her behavior. Therefore, each woman's record once she returns to work is censored.

Finally, the last job at which a woman worked before giving birth and the first job at which she works after giving birth are identified to determine whether the jobs were for the same employer. Then the first job held after childbirth is identified as either an old or new job. An old job is defined as a job at which the mother worked within the three months preceding the child's birth.

Descriptive statistics indicate that the incidence of leave taking is not positively correlated with MLL: 83.9% of mothers with government-mandated maternity leave benefits take some amount of leave from work and 86.1% of mothers without MLL take leave. A mother with MLL is defined as a mother who gave birth in a state at a time when MLL is in force.

Figure 1 shows the percentage of mothers with and without MLL who have returned to work at their prechildbirth jobs and who have started work at new jobs for each of the 12 months following childbirth conditional on initially taking leave. In this figure, once a mother returns to work at either kind of job, she is assumed to remain working at that job for the duration of the 12-month period. In general, the percentage of mothers who return to work increases dramatically during the first five months after giving birth. By the sixth month, virtually all mothers who will eventually return to their old jobs have done so. More mothers without MLL have returned to work at their prechildbirth jobs by the second month, but a larger percent of mothers with MLL have returned to their prechildbirth jobs after the second month. By the 12th month, over 60% of mothers with MLL have returned to work at their old jobs, but only about 50% of mothers without MLL have done so.

MLL is correlated with a lower probability of taking leave initially after giving birth and a higher probability of eventually returning to work at the prechildbirth job. However, these correlations do not necessarily imply a causal relationship. Instead, mothers in states with MLL in force may be less likely to take maternity leave and more likely to return to their prechildbirth jobs regardless of MLL due to state-specific characteristics. Or mothers may be increasingly more likely to continue working throughout the pregnancy or to return to their prechildbirth jobs over time regardless of MLL due to time trends. To determine the causal effects of MLL, multivariate regression analysis is used. Table 2 explains how each of these explanatory variables is measured, giving each variable's mean and standard deviation.

The variable of primary interest is the MLL variable, which is specified in two ways. The first is a MLL dummy variable that equals one if the mother gives birth in a state at a time when MLL is in force. The second MLL variable, MLL weeks mandated, equals the number of weeks of leave guaranteed under either state or federal law. If a mother gives birth in 1 of the 12 states that has MLL and the FMLA is in effect, then the MLL weeks mandated variable equals the number of weeks of leave guaranteed by the more generous legislation. If the mother gives birth before the FMLA went into effect and either lives in a state with no MLL or lives in a state at a time before that state's MLL was in force, then both MLL variables equal zero. Table 1 shows that, of the 1712 observations contained in the data set, 348 have MLL. (10) Table 3 lists the states that passed MLL, the length of leave that those states guarantee, and the date their legislation went into effect. Table 4 gives the number of observations (births) by s tate for three periods: before state MLL, after state MLL but preFMLA, and post-FMLA. (For non-MLL states, there are only two periods: pre- and post-FMLA.)

Note that the two MLL variable specifications do not account for the mother's eligibility status, which is determined by her work history and firm size. For example, to be eligible for maternity leave under the FMLA, a mother must have worked for the employer for at least one year, accumulating 1250 hours, and that employer must employ at least 50 workers. (11) These MLL variables do not account for the mother's eligibility status because it partially depends on the mother's previous employment decisions. If these decisions are influenced by the same unobserved factors that affect leave-talting and return-to-work behavior, then eligibility status is endogenous. Instead, the two MLL variables can be thought of as instruments for eligibility status because they exogenously assign the mother MLL based on her state of residence.

Strictly for comparison purposes, two additional MLL variables were created to pick up the effect of MLL on mothers who are eligible for the benefits and/or who work for covered employers. The first of these variables, eligible MLL, equals one if the data suggest that the mother is eligible for MLL. In particular, this variable will equal one if the mother gives birth in a state at a time when MLL was in force and the mother has the requisite work history to be eligible. As shown in Table 1, of the 348 mothers with MLL, 272 have the requisite work history to be eligible. Then, an eligible and covered MLL variable is created that equals one if the data suggest that the mother is eligible and works for an employer who is covered by MLL. Specifically, this variable will equal one if the mother gives birth in a state at a time when that state's MLL was in force, if she has the appropriate work history to be eligible, and if she works for an employer who employs the requisite number of workers to be covered by the legislation. Of the 348 mothers whose employer potentially could have been covered by MLL, 117 were known to be covered, the status of 131 observations was unknown, and 100 observations with MLL were known to work for employers who were not covered. (12) Of the 117 mothers whose employer is known to be covered, 98 mothers have the work history to be eligible. (Table 1 presents this information as well as these statistics for the leave-taking sample used in the second-period return-to-work model.) These MLL variables will reflect how MLL affects eligible and/or covered mothers. However, results from these variables will have to be interpreted cautiously; again, eligibility status may be endogenous and firm size is unknown for some mothers.

Duration terms are included to allow the monthly probability of returning to the old job and starting a new job to vary with the month since giving birth. These duration variables take the form of dummy variables representing the months since giving birth. To allow the effect of MLL to vary with duration, the MLL variable is interacted with the duration variables.

Also included are variables to control for the mother's characteristics such as race, marital status, age, education, weeks of work experience, and health. Household characteristics with variables that measure the number of other children aged 0-2, the number of children aged 3-5, and the spouse's education are controlled for. (13) The mother's previous wage rate and nonwage income are budget constraint variables that are included to control for financial constraints. (14) The mother's background characteristics are controlled for with the education level of the mother's mother and the mother's father. (15) Dummy variables indicating whether the mother's mother worked full time and part time when the mother was 14 are included to control for family preferences with respect to maternal work. To control for economic conditions, the local unemployment rate is included. State and year dummy variables are also included to control for state-specific effects and time trends.

5. Results

From the first approach, the effects of the MLL variables on the incidence of leave taking are presented in Table 5 and on the probability of eventually returning to work at the old job and on the probability of switching to a new job in Tables 6 and 7. The results of the MLL variables in the dynamic MNP models from the second approach are presented in Tables 8-11. Each of the models contains controls for unobserved heterogeneity. (16) A representative set of the other covariates' estimates is presented in the appendix table. Each table also contains simulations that show the magnitude of the effects of the MLL variables.

Incidence of Leave Taking

In this section, the effect of MLL on the probability of taking leave from work after giving birth is examined. As noted earlier, Klerman and Leibowitz's (1997) theory predicts that MLL will have a positive effect on the incidence of leave taking. In model 1 (Table 5), MLL's coefficient is positive but small and insignificant. Simulations indicate that MLL increases the incidence of leave taking from 86.0 to 88.1%. In model 2, the MLL variable is specified to equal the number of weeks of maternity leave mandated by MLL. Again, the results show that MLL's point estimate is positive, increasing the probability of taking leave from 86.1 to 87.7%. The results from a model (not shown) that allows MLL to have a nonlinear effect by including the MLL dummy variable and the MLL weeks mandated variable indicates that MLL's effect remains statistically insignificant.

Next, the MLL variable is specified to equal one if the mother is eligible for MLL. The results in model 3 are virtually unchanged, with MLL having a statistically insignificant effect. The MLL variable's point estimate suggests that MLL would increase the incidence of leave taking from 0.862 to 0.877. Models that specify the eligible MLL variable as the number of weeks of maternity leave mandated (model 4) and that allow MLL to have nonlinear effects by including the MLL dummy variable and the MLL weeks mandated variable (not shown) produce similar results. When MLL is specified to account for whether the mother is eligible and if the mother's employer is covered (models 5 and 6), the effect of MLL remains insignificant.

Probability of Eventually Returning to the Prechildbirth Job

Next, the effect of MLL on the probability of eventually returning to the prechildbirth job is examined. Recall that the theory predicts MLL will have a positive effect on this outcome. Model 1 in

Table 6 specifies MLL to be a dummy variable. In this model, MLL's coefficient is positive, with simulations indicating that MLL increases this probability from 0.492 to 0.609, though this effect is measured imprecisely. In model 2, MLL is specified to equal the weeks of leave mandated by the legislation, and in a model not shown, MLL is allowed to have a nonlinear effect by including the MLL dummy variable and the MLL weeks mandated variable. However, the effect of MLL remains positive but statistically insignificant in these models.

Model 3 includes the eligible MLL dummy variable and shows that MLL has a larger effect, significantly increasing the probability of eventually returning to the prechildbirth job from 49.3 to 64.9%. Similarly, when MLL is specified to equal the number of weeks of maternity leave mandated (model 4), MLL has a positive effect that is marginally significant. When eligible MLL is allowed to have a nonlinear effect (not shown), its effect becomes statistically insignificant. The mother's eligibility and whether the employer is covered in models 5 and 6 are accounted for, and there is again some evidence that MLL increases the probability of returning to the old job. For example, MLL has a marginally significant positive effect in model 5 that indicates MLL increases this probability from 0.497 to 0.617.

Probability of Switching to a New Job

Next, the probability of starting work at a new job after giving birth is examined. Table 7 shows the effect of various MLL specifications on this probability. The effect of MLL is statistically insignificant in model 1, with simulations indicating that MLL decreases this probability from 36.1 to 30.0%. Specifying MLL to equal the number of weeks of leave mandated by the legislation in model 2 and allowing MLL to have a nonlinear effect (model not shown) do not appreciably change the results.

Model 3 shows the effect of MLL on eligible mothers; the MLL coefficient is statistically significant, showing that MLL decreases the probability of eventually starting a new job from 36.5 to 27.8%. However, in models that specify MLL to equal the weeks of mandated leave (model 4) and to have a nonlinear effect (results not shown), MLL's effects become statistically insignificant. Last, MLL for mothers who are eligible and covered is examined. However, the effect of MLL is insignificant in these models (models 5 and 6).

Monthly Probability of Returning to the Prechildbirth Job

In this section, the effect of MLL on the monthly probability of returning to work at the prechildbirth job (conditional on having not yet returned to work) is examined. Recall that Klerman and Leibowitz's (1997) theory predicts MLL will allow some mothers (who would have quit without MLL) to return and MLL will allow other mothers (who without MLL would have returned to the prechildbirth job after a suboptimal period of leave) to postpone their return. Model 1 in Table 8 indicates that MLL has a negative effect on the probability of returning to the old job in the second month after giving birth but a positive effect in the remaining months. Specifically, MLL decreases the probability of returning to work at the prechildbirth job in month 2 from 28.4 to 28.2% and increases this probability in month 3 from 26.0 to 36.1%. Also, the MLL variable and the MLL interaction term for the second month are statistically significant at the 5% level. (17) Model 2 specifies MLL to equal the weeks of maternity leave mandat ed by MLL. The effect of MLL weeks mandated is somewhat similar to the effect of the MLL dummy variable; for example, mandating 12 weeks of MLL decreases the probability of returning in the second month from 30.3 to 23.9% but increases this probability in the third month from 26.3 to 33.5%. This is exactly what we would expect if MLL allows mothers to delay their return to their old jobs. A model (not shown) is examined that includes the MLL dummy variable, the MLL weeks mandated variable, and corresponding MLL dummy and MLL weeks mandated--duration interaction terms. This allows the marginal effect of an additional week of government-mandated maternity leave to differ. Allowing MLL to have a nonlinear effect does not change the results.

Model 3 in Table 9 examines the effect of MLL on eligible mothers and finds that MLL has a similar effect: MLL increases the probability of returning in month 1, decreases this probability in month 2, and increases the probability in the remaining months. Further, accounting for a mother's eligibility status increases the statistical significance of the results; the effects of MLL in the first three months are now statistically significant at the 5% level. This is seen again in model 4, which specifies the MLL variables to equal the weeks of leave mandated by MILL for eligible mothers. However, allowing the eligible MILL variables to have nonlinear effects (by including the eligible MILL dummy and weeks of eligible mandated leave variables) does not produce any MLL results that are statistically significant.

Evidence is again found that MLL delays a mother's return to work when the effect of MLL on mothers who are both eligible and covered is examined. In model 5, MLL significantly decreases the probability of returning in month 2 from 30.4 to 24.5%. MLL has a positive effect in months 3-8, with MLL significantly raising the probability of returning to the prechildbirth job from 27.1 to 35.1% in month 3. This indicates that some mothers, who would have returned to their prechildbirth job during the first two months absent MLL, postpone their return with MLL until a later month. Similarly, in model 6, which specifies the MLL variables to equal the weeks of leave mandated by MLL for eligible and covered mothers, MLL has a negative effect in the second month and a positive effect in remaining months.

Monthly Probability of Starting a New Job

Finally, the effect of MLL on the monthly probability of starting work at a new job (conditional on having not yet returned to work) is examined. Model 1 in Table 10 shows that MLL has an insignificant negative effect on the probability of starting a new job in each month. In this model, the negative effect is largest in the fourth month, decreasing the probability of starting a new job from about 14% to 2%. To determine the robustness of these results, MLL is next specified as the number of weeks of leave mandated by MLL. Model 2's results remain similar: MLL has a negative effect on the probability of starting a new job in each month, with the largest negative effect occurring, again, in the fourth month. However, the effects of the MLL variables remain statistically insignificant. A model is also estimated that allows the marginal effect of MLL to differ, but results from this model (not shown) remain similar to those already reported.

Models 3 and 4 in Table 11 show the effect of MLL on eligible mothers. In these models, MLL again has a negative effect on the probability of starting a new job in most months. The largest negative effect remains in the fourth month. However, these effects are still statistically insignificant. The results remain virtually unchanged when examining the effect of MLL on eligible mothers whose employers are covered by MLL in models 5 and 6.

6. Discussion and Conclusions

The results do not provide any statistically significant evidence that MILL increases the incidence of leave taking. Furthermore, simulations using MLL's point estimates show that MLL's effect is uniformly small: MLL increases the incidence of leave taking by less than 5 percentage points in each of the models. This is similar to Klerman and Leibowitz (1997), who also found that MLL did not have a statistically significant effect on leave taking, but it differs from Waldfogel (1999), who found that the incidence of leave taking increased with MLL.

This article finds some evidence that MLL increases the probability that mothers who took leave eventually return to their prechildbirth jobs. For example, MLL significantly increases this probability in the models that identify whether mothers are eligible for MLL. MLL's effect is moderately large: in each of the models, MLL's point estimate indicates that MLL increases this probability between 10 and 17 percentage points. Klerman and Leibowitz (1997) thought that any increase in this percentage would be small because many mothers returned to their prechildbirth jobs before MLL. Correspondingly, simulations indicate the probability that leave-taking mothers start new jobs decreases with MLL, with this probability falling between 6 and 11 percentage points in all but one of the models. Thus, MLL may prompt some mothers who would otherwise have started a new job without MLL to instead return to their old jobs. We can also infer from these results that the probability of returning to work at all (for either typ e of employer) appears to increase by at least 3 or 4%.

The dynamic MNP framework shows that MLL significantly decreases the probability of returning to the old job in the second month and significantly increases this probability in the third month. This is exactly what we would expect if MLL allows mothers who otherwise would have returned to work sooner after giving birth (in the second month) to postpone their return to their prechildbirth jobs (until the third month). Further, this result is consistent with the finding that MLL does not have a significant effect on the incidence of leave taking: mothers who delay their return to their prechildbirth jobs do not increase the probability of leave taking (or add to the number of mothers who eventually return to their old jobs) because, without MLL, these mothers still would have taken leave and returned (albeit sooner).

From the second approach's results, calculation of the effect of MLL on the "survivor rate" can be done. Summarizing results from the models, the monthly probability of having not yet returned to work is higher without MLL than with MLL (in each month except the second). Simulations with MLL set to zero from model 1 indicate that 11.2% have not yet returned to work by the eighth month after giving birth; conversely, with MLL set to one, the survivor rate is 7.6% in the eighth month. (18) So MILL prompts more mothers to return to work. The cumulative probability of returning to the prechildbirth job (by the eighth month) and the cumulative probability of starting a new job (by the eighth month) with and without MLL are also calculated. Results are similar to those from the first approach. Thus, not only does MLL allow some mothers to delay their return, MLL allows some mothers who otherwise would have started a new job or not returned to work at all to return to their prechildbirth jobs.

These results should be of interest to employers because they provide evidence that MLL disrupts production by increasing employee absenteeism. Because these results suggest that some mothers may delay their return to work at their prechildbirth jobs as a result of MLL, employers will either have to retain temporary replacement workers longer or else continue production for a longer period without some of the mothers. This is a suboptimal outcome for employers if we assume that employers will only offer maternity leave voluntarily when it is in their best interest. Under this assumption, MLL should only have an effect on mothers whose interests diverge from that of their employers. That is, the mothers who delay their return to their prechildbirth jobs with MLL are presumably the mothers who prefer to be allowed more weeks of maternity leave than their employers are otherwise willing to offer. This reasoning suggests that, absent MLL, these employers would have been content to risk permanent separations rathe r than offer additional maternity leave.

Conversely, MLL may benefit employers if they were unable to offer optimal maternity leave benefits prior to the legislation due to some form of market failure. For example, suppose that, due to adverse selection, employers are hesitant to voluntarily offer maternity leave for fear of attracting employees who are disproportionately likely to take leave. If MLL mandates that all employers offer maternity leave benefits, then the problem of adverse selection is eliminated. MLL would then benefit those employers who were willing to assume the cost of maternity leave in exchange for the benefits of retaining that worker but were unwilling to bear other costs such as attracting a disproportionate number of leave takers.

Appendix

The effects of unobserved heterogeneity are controlled for using the simultaneous-equations version of an approach suggested by Heckman and Singer (1984). The sources of bias are cross-equation correlation captured by the [micro]'s. These factors are integrated out" by approximating the unobserved heterogeneity's distribution with a step function of mass points and probability weights jointly with the other parameters. For example, the distribution of each unobserved factor [micro] is Pr([micro] = [[micro].sub.n]) = [[theta].sub.n], with n = 1, ..., N and [summation over (n/n=1)] [[theta].sub.n] = 1, where N is the number of mass points in the distribution of [micro] and [theta] is the probability that [micro] equals a particular point of support. Identification is achieved by setting the first mass point equal to zero and the second mass point equal to one for each factor. The additional mass points and the probability weights are restricted to lie between zero and one, but the factor loadings are allowed to take any value. With M different factors of [micro], the unconditional likelihood function is given by

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

where N, [micro], and [theta] are as defined above and [THETA] the other parameters to be estimated.

Gritz (1987) and Heckman and Walker (1990) explain that there are no well-established rules for determining the number of factors and mass points to use in these types of models. Standard log-likelihood ratio tests are inappropriate in this instance because parameters that fall on the boundary space violate the chi-squared distribution conditions. In later work, Grits (1993), referring to Akaike's Information Criterion (Akaike 1973), suggests adding factors and points of support as long as the value of the likelihood function improves by at least one point for each additional parameter. Alternatively, Blau (1994) and Mroz (1999) continue adding factors and mass points to the model as long as they improve the value of the likelihood function. In this analysis, one common factor with three points of support is used. Specifications using additional factors or points of support did not improve the value of the likelihood function, with many of these specifications failing to converge. Using Grits's (1993) criteri a, rejection is not possible of the joint null hypothesis that additional factors and mass points are not warranted because the value of the likelihood function did not significantly improve with any combination of additional factors and mass points. As factors and mass points were added, the change in the coefficients was examined. Continuing to add factors and mass points (in addition to one common factor with three points of support) left the estimates virtually unchanged.

Table A1

Effect of Other Covariates


                                    Incidence of            Returned to
Model                               Leave Taking              Old Job

Constant                         0.595     (1.858) (a)     -2.222
Black dummy variable             0.257     (0.166)          0.548 ***
Hispanic dummy variable         -0.024     (0.154)          0.229
Marital status                   0.475     (0.411)          1.160 ***
Mother's age                    -0.029     (0.025)         -0.015
Mother's education level         0.011     (0.028)          0.093 ***
Weeks of work experience         0.002 **  (0.001)          0.003 ***
Poor health dummy                0.084     (0.156)         -0.421 **
Length of hospital stay          0.032     (0.037)         -0.006
Cesarean dummy variable         -0.001     (0.135)          0.080
Children aged 0-2               -0.305 **  (0.148)          0.116
Children aged 3-5               -0.348 **  (0.151)          0.295 ***
Spouse's education level        -0.036     (0.030)         -0.075 ***
Previous wage received           0.022 *   (0.013)          0.118 ***
Nonwage income                  -0.002     (0.003)         -0.013 ***
Mother's mother's education     -0.022     (0.022)          0.042 *
Mother's father's education      0.015     (0.018)         -0.038 *
Mother's education missing      -0.397     (0.341)          0.519
Father's education missing      -0.138     (0.235)         -0.571 *
Mother worked part time          0.143     (0.138)          0.321 **
Mother worked full time          0.062     (0.112)          0.003
Unemployment rate                0.005 *   (0.003)         -0.001
Factor loading                   0.672     (1.373)         -0.310

Log-likelihood value         -1410.1                    -1410.1


                              Returned           Started a
                                 to
Model                         Old Job             New Job

Constant                     (1.685)      -2.457      (1.945)
Black dummy variable         (0.163)       0.486 ***  (0.158)
Hispanic dummy variable      (0.190)       0.237      (0.191)
Marital status               (0.361)       1.025 ***  (0.326)
Mother's age                 (0.028)      -0.010      (0.025)
Mother's education level     (0.036)       0.063 **   (0.030)
Weeks of work experience     (0.001)       0.002 ***  (0.001)
Poor health dummy            (0.176)      -0.168      (0.176)
Length of hospital stay      (0.017)       0.001      (0.017)
Cesarean dummy variable      (0.135)       0.026      (0.141)
Children aged 0-2            (0.133)       0.131      (0.120)
Children aged 3-5            (0.095)       0.343 ***  (0.051)
Spouse's education level     (0.029)      -0.061 ***  (0.027)
Previous wage received       (0.017)       0.095 **   (0.042)
Nonwage income               (0.003)      -0.013 ***  (0.003)
Mother's mother's education  (0.026)       0.048 *    (0.026)
Mother's father's education  (0.022)      -0.028      (0.022)
Mother's education missing   (0.421)       0.552      (0.405)
Father's education missing   (0.312)      -0.419      (0.321)
Mother worked part time      (0.148)       0.341 **   (0.151)
Mother worked full time      (0.120)      -0.028      (0.124)
Unemployment rate            (0.003)       0.001      (0.003)
Factor loading               (1.574)       1.348      (1.824)

Log-likelihood value                   -1410.1

                                    Probability           Probability
                                   of Returning           of Starting
Model                               to Old Job              New Job

Constant                        -1.184 *    (0.649)      -0.752 ***
Black dummy variable             0.066      (0.098)       0.079
Hispanic dummy variable         -0.026      (0.106)       0.028
Marital status                   0.328      (0.231)       0.214
Mother's age                    -0.044 **   (0.019)      -0.035 **
Mother's education level         0.056 ***  (0.020)       0.002
Weeks of work experience         0.002 ***  (0.001)       0.001
Poor health dummy               -0.357 ***  (0.106)      -0.080
Length of hospital stay         -0.012      (0.010)      -0.006
Cesarean dummy variable          0.051      (0.073)      -0.051
Children aged 0-2               -0.014      (0.077)       0.056
Children aged 3-5                0.081      (0.073)       0.164 **
Spouse's education level        -0.030 *    (0.017)      -0.013
Previous wage received           0.046 ***  (0.009)       0.001
Nonwage income                  -0.005 **   (0.002)      -0.005 **
Mother's mother's education     -0.005      (0.016)       0.009
Mother's father's education     -0.017      (0.012)       0.001
Mother's education missing       0.110      (0.240)       0.058
Father's education missing      -0.288 *    (0.172)      -0.119
Mother worked part time          0.075      (0.092)       0.152 *
Mother worked full time          0.062      (0.075)      -0.015
Unemployment rate               -0.003 *    (0.002)       0.002
Factor loading                  -0.758      (0.498)       0.783 *

Log-likelihood value         -2737.6                  -2737.6

                             Probability
                                 of
                              Starting
Model                         New Job

Constant                     (0.264)
Black dummy variable         (0.093)
Hispanic dummy variable      (0.107)
Marital status               (0.220)
Mother's age                 (0.016)
Mother's education level     (0.020)
Weeks of work experience     (0.001)
Poor health dummy            (0.088)
Length of hospital stay      (0.008)
Cesarean dummy variable      (0.074)
Children aged 0-2            (0.076)
Children aged 3-5            (0.068)
Spouse's education level     (0.016)
Previous wage received       (0.010)
Nonwage income               (0.002)
Mother's mother's education  (0.016)
Mother's father's education  (0.013)
Mother's education missing   (0.250)
Father's education missing   (0.173)
Mother worked part time      (0.086)
Mother worked full time      (0.075)
Unemployment rate            (0.002)
Factor loading               (0.489)

Log-likelihood value

There are 1712 observations in the incidence of leave-taking equation.
These are the supplemented results from model 1 in Table 5. There are
1467 observations in the return-to-work equations. Thee results
supplement the results from model in Tables 6 and 7. The covariance
parameter estimate in [[rho].sub.PN] = 1.740 (0.955), the mass points
are [[micro].sub.11] = 0.00, [[micro].sub.12] = 0.508, and
[[micro].sub.13]  1.00, and the corresponding probability weights are
[[theta].sub.11] = 0.404, [[theta].sub.12] = 0.357, [[theta].sub.13] =
0.238. Finally, there are 1467 observations in the dynamic MNP model.
These results supplement the results from model 1 in Tables 8 and 10.
The covarianbce parameter estimate is [[rho].sub.PN] = -0.618 (1.492),
the mass points are [[micro].sub.11] = 0.00, [[micro].sub.12] = 0.604,
and [[micro].sub.13] = 1.00, and the corresponding probability weights
are [[theta].sub.11] = 0.251, [[theta].sub.12] = 0.321, and
[[theta].sub.13] = 0.427. Due to the large number of state and year
dummy variables, their estimated coefficients are not presented. These
results are available upon request.

(a) Standard errors are in parentheses.

* Indicates statistical significance at the 10% level

** at the 5% level

*** at the 1% level.

[FIGURE 1 OMITTED]

Table 1

Sample Selection Characteristics

Sample                                                 Number

  NLSY respondents                                     12,686
  Women in NLSY                                         6,283
  Women who gave birth from 1988 to 1994                2,101
  Number of births from 1988 to 1994                    3,037
Respondents without missing information
  Number of births from 1988 to 1994                    2,796
  Number of births from 1988 to 1994 to                 1,712
   employed women
    (i) With MLL                                          384
   (ii) With MLL and eligible                             272
  (iii) With MLL and known to be covered (eligibility     117
        not required)
   (iv) With MLL but coverage status unknown              131
        (eligibility not required)
    (v) With MLL and known not to be covered              100
        (eligibility not required
   (vi) With MLL, eligible, and                            98
        known to be covered
  Number of births from 1988 to 1994 to employed        1,467
    women who took leave
    (i) With MLL                                          292
   (ii) With MLL and eligible                             236
  (iii) With MLL and known to be covered (eligibility      97
        not required
   (iv) With MLL but coverage status unknown              105
        (eligibility not required)
    (v) With MLL and known not to be covered               90
        (eligibility not required
   (vi) With MLL, eligible, and known to be covered        86

Table 2

Sample Means and Standard Deviations

                                                               Standard
                                                      Mean    Deviation

Maternal characteristics
  Maternity leave legislation (= weeks mandated        1.795      4.303
   by legislation)
  Black dummy variable (= 1 if black)                  0.253      0.435
  Hispanic dummy variable (= 1 if Hispanic origin)     0.199      0.399
  Marital status (= 1 if married)                      0.776      0.416
  Mother's age (in years)                             28.497      2.769
  Mother's education level                            13.344      2.314
  Weeks of work experience (in weeks)                428.329    172.071
  Poor health dummy variable (= 1 if health            0.153      0.36
   limited work)
  Length of hospital stay (after giving birth,         3.074      3.876
   in days)
  Cesarean dummy variable                              0.274      0.446

Household characteristics
  Children aged 0-2                                    0.242      0.451
  Children aged 3-5                                    0.281      0.491
  Spouse's education level (1)                        13.535      2.582

Budget constraint variables
  Previous wage rate received                          7.829      5.112
  Nonwage income (in thousands)                       20.573     18.266

Family background variables
  Mother's mother's education (a)                     11.058      3.056
  Mother's father's education (a)                     11.291      3.763
  Mother's mother's education missing dummy            0.047      0.212
   variable
  Mother's father's education missing dummy            0.115      0.32
   variable
  Mother's mother worked part-time (= 1 if mother's    0.208      0.406
   mother worked part-time when the mother was 14)
  Mother's mother worked full-time (= 1 if mother's    0.399      0.489
   mom worked full-time when the mother was 14)
  Economic conditions
  Unemployment rate (percent)                         66.001     24.183

(a) Zero values are excluded from the descriptive statistics. Otherwise,
there are 1712 observations. Due to the large number of state and year
dummy variables, their means and standard deviations are not presented.

Table 3

Characteristics of Maternity Leave Legislation

                                      Employer Size
                      Weeks of Leave   (Number of     Tenure
State                   Guaranteed     Employees)    Required

California (a)              17         No minimum     1 year
Connecticut (b)             12             75         1 year
District of Columbia        16             50         1 year
Federal FMLA                12             50         1 year
Maine                        8             25         1 year
Minnesota                    6             21         1 year
Massachusetts                8              6        3 months
New Jersey                  12             75         1 year
Oregon                      12             25        90 days
Rhode Island                13             50         1 year
Tennessee                   16            100         1 year
Vermont                     12             10         1 year
Washington (c)              12            100         1 year
Wisconsin                    6             50         1 year


                        Date of
State                 Enforcement  Work Requirement

California (a)           1/92      No minimum
Connecticut (b)          7/90      1000 hours in prior year
District of Columbia     4/91      1000 hours in prior year
Federal FMLA             7/93      1250 hours in prior year
Maine                    4/88      No minimum
Minnesota                7/87      20 hours per week
Massachusetts           10/72      Full time
New Jersey               4/90      1000 hours in prior year
Oregon                   1/88      No minimum
Rhode Island             7/87      Full time
Tennessee                1/88      Full time
Vermont                  7/92      30 hours per week
Washington (c)           9/89      35 hours per week
Wisconsin                4/88      1000 hours in prior year

Source: Kane (1998), Klerman and Leibowitz (1997), the Women's Legal
Defense Fund (1994), Bond (1991), and the Bureau of National Affairs
(1987).

(a) California passed legislation mandating leave for disability in
1980.

(b) Connecticut passed legislation mandating leave for disability in
1973.

(c) Washington passed legislation mandating leave for disability in
1973. Eight additional states passed maternity leave provisions that did
not provide job reinstatement rights and 12 additional states passed
maternity leave provisions that only covered state employees. These
states' MLLs were considered to be nonbinding and the MLL variables were
set equal to zero for observations from these states.

Table 4

Observations (Births) by State

                                   Before  After State MLL
                                State MLL    (pre-FMLA)     Post-FMLA

Maternity leave
legislation states
 California                           141        36            14
 Connecticut                           27        26             8
 District of Columbia                   6         4             0
 Maine (a)                              0         0             0
 Minnesota                              6        31             2
 Massachusetts                          0        21             4
 New Jersey                            29        27             4
 Oregon                                 3         9             0
 Rhode Island (a)                       0         0             0
 Tennessee                              2        14             2
 Vermont (a)                            0         0             0
 Washington                             6        14             1
 Wisconsin                             11        51             5
 MLL states (total)                   231       233            40

Nonmaternity leave legislation   Pre-FMLA                   Post-FMLA
  states
 Non-MLL states (total)              1133                      75

All states
  Totals                             1364       233           115

There are 1712 onservations, 348 of whom have MLL.

(a) No births occurred in Maine, Rhode Island, and Vermont that meet the
conditions requisite to be included in my sample.

Table 5

Effect of Maternity Leave Legislation on the Incidence of Leave Taking

                                        Full Sample
                                   Model 1     Model 2

MLL dummy variable                     0.113     --
                                      (0.229)    --
MLL weeks mandated                   --            0.007
                                     --           (0.019)
Eligible MLL dummy variable          --          --
                                     --          --
Eligible MLL weeks mandated          --          --
                                     --          --
Eligible and covered MLL dummy       --          --
                                     --          --
Eligible and covered MLL weeks       --          --
                                     --          --
Incidence of leave taking
  With MLL = 0                         0.860       0.861
  With MLL = 1 (or 12 weeks)           0.881       0.877
  Log-likelihood value             -1410.1     -1410.3

                                        Eligible Sample
                                     Model 3        Model 4

MLL dummy variable                     --             --
                                       --             --
MLL weeks mandated                     --             --
                                       --             --
Eligible MLL dummy variable            0.079          --
                                      (0.220)         --
Eligible MLL weeks mandated            --               0.009
                                       --              (0.018)
Eligible and covered MLL dummy         --             --
                                       --             --
Eligible and covered MLL weeks         --             --
                                       --             --
Incidence of leave taking
  With MLL = 0                         0.862            0.862
  With MLL = 1 (or 12 weeks)           0.877            0.882
  Log-likelihood value             -1408.2          -1408.3

                                    Eligible and Covered Sample
                                       Model 5          Model 6

MLL dummy variable                        --              --
                                          --              --
MLL weeks mandated                        --              --
                                          --              --
Eligible MLL dummy variable               --              --
                                          --              --
Eligible MLL weeks mandated               --              --
                                          --              --
Eligible and covered MLL dummy             0.163          --
                                          (0.217)         --
Eligible and covered MLL weeks            --                0.021
                                          --               (0.020)
Incidence of leave taking
  With MLL = 0                             0.860            0.859
  With MLL = 1 (or 12 weeks)               0.891            0.904
  Log-likelihood value                 -1409.5          -1409.1

The dependent variable is the probability of taking leave, where leave
is defined as taking time off from work (temporarily or permanently, for
any length of time). Standard errors are in parentheses. There are 1712
observations.

Table 6

Effect of Maternity Leave Legislation on the Probability of Returning to
the Prechildbirth Job

                                       Full Sample
                                  Model 1       Model 2

MLL dummy variable                 0.436         --
                                  (0.316) (a)    --
MLL weeks mandated                --              0.026
                                  --             (0.023)
Eligible MLL dummy variable       --             --
                                  --             --
Eligible MLL weeks mandated       --             --
                                  --             --
Eligible and covered MLL dummy    --             --
                                  --             --
Eligible and covered MLL weeks    --             --
                                  --             --
Incidence of leave taking
 With MLL = 0                       0.492         0.488
 With MLL = 1 (or 12 weeks)         0.609         0.591
 Log-likelihood value           -1410.1        -1410.3

                                     Eligible Sample
                                Model 3            Model 4

MLL dummy variable                --                 --
                                  --                 --
MLL weeks mandated                --                 --
                                  --                 --
Eligible MLL dummy variable         0.625 **         --
                                   (0.272)           --
Eligible MLL weeks mandated       --                   0.043 *
                                  --                  (0.026)
Eligible and covered MLL dummy    --                 --
                                  --                 --
Eligible and covered MLL weeks    --                 --
                                  --                 --
Incidence of leave taking
 With MLL = 0                       0.493              0.494
 With MLL = 1 (or 12 weeks)         0.649              0.664
 Log-likelihood value           -1408.3            -1408.3

                                Eligible and Covered Sample
                                Model 5         Model 6

MLL dummy variable                --              --
                                  --              --
MLL weeks mandated                --              --
                                  --              --
Eligible MLL dummy variable       --              --
                                  --              --
Eligible MLL weeks mandated       --              --
                                  --              --
Eligible and covered MLL dummy      0.532 *       --
                                   (0.307)        --
Eligible and covered MLL weeks    --                0.035
                                  --               (0.027)
Incidence of leave taking
 With MLL = 0                       0.497           0.497
 With MLL = 1 (or 12 weeks)         0.617           0.635
 Log-likelihood value           -1409.5         -1409.1

The dependent variable is the probability of ultimately returning to
work at the predhildbirth job conditional on having taken leave from
work. There are 1467 observations.

(a) Standard errors are in parentheses.

* Indicates statistical significance at the 10% level,

** at the 5% level.

Table 7

Effect of Maternity Leave Legislation on the Probability of Starting a
New Job

                                       Full Sample
                                  Model 1        Model 2

MLL dummy variable                  0.290           --
                                   (0.362) (a)      --
MLL weeks mandated                  --               0.008
                                    --              (0.026)
Eligible MLL dummy variable         --              --
                                    --              --
Eligible MLL weeks mandated         --              --
                                    --              --
Eligible and covered MLL dummy      --              --
                                    --              --
                                    --              --
Eligible and covered MLL weeks      --              --
Incidence of leave taking
  With MLL = 0                      0.361            0.374
  With MLL = 1 (or 12 weeks)        0.300            0.306
  Log-likelihood value          -1410.1          -1410.3

                                      Eligible Sample
                                 Model 3           Model 4

MLL dummy variable                 --                --
                                   --                --
MLL weeks mandated                 --                --
                                   --                --
Eligible MLL dummy variable         -0.433 ***       --
                                    (0.065)          --
Eligible MLL weeks mandated        --                  0.021
                                   --                 (0.029)
Eligible and covered MLL dummy     --                --
                                   --                --
                                   --                --
Eligible and covered MLL weeks     --                --
Incidence of leave taking
  With MLL = 0                       0.361             0.364
  With MLL = 1 (or 12 weeks)         0.280             0.254
  Log-likelihood value           -1408.3           -1408.3

                                      Eligible and Covered Sample
                                           Model 5      Model 6

MLL dummy variable                           --           --
                                             --           --
MLL weeks mandated                           --           --
                                             --           --
Eligible MLL dummy variable                  --           --
                                             --           --
Eligible MLL weeks mandated                  --           --
                                             --           --
Eligible and covered MLL dummy                 0.393      --
                                              (0.278)     --
                                             --             0.019
Eligible and covered MLL weeks               --            (0.029)
Incidence of leave taking
  With MLL = 0                                 0.359        0.360
  With MLL = 1 (or 12 weeks)                   0.306        0.275
  Log-likelihood value                     -1409.5      -1409.1

The dependent variable is the probability of starting work at a new job
conditional on having taken leave from work. There are 1467
observations.

*** Indicates statistical significance at the 1% level.

Table 8

Effect of MLL on the Monthly Probability of Returning to Work at the
Prechildbirth Job

                                        Model 1                Model 2

MLL dummy                         1.235 **  (0.630) (a)        --
MLL Weeks mandated                 --          --            0.039
Month 1 dummy                     0.184     (0.471)         -0.393
Month 2 dummy                     1.873     (0.334)          1.126 ***
Month 3 dummy                     1.917     (0.321)          1.116 ***
Month 4 dummy                     1.102     (0.337)          0.391 *
Month 5 dummy                     0.315     (0.398)         -0.097
Month 6 dummy                     0.801     (0.360)          0.123
Month 7 dummy                     0.269     (0.380)         -0.168
MLL dummy-month 1 interaction    -0.699     (0.566)           --
MLL dummy-month 2 interaction    -1.236     (0.570)           --
MLL dummy-month 3 interaction    -0.871     (0.540)           --
MLL dummy-month 4 interaction    -0.283     (0.540)           --
MLL dummy-month 5 interaction     0.579     (0.568)           --
MLL dummy-month 6 interaction    -0.860     (0.711)           --
MLL weeks-month 1 interaction      --         --            -0.015
MLL weeks-month 2 interaction      --         --            -0.058 *
MLL weeks-month 3 interaction      --         --            -0.020
MLL weeks-month 4 interaction      --         --             0.004
MLL weeks-month 5 interaction      --         --             0.049
MLL weeks-month 6 interaction      --         --            -0.029
Simulated probability            MLL = 0    MLL = 1         MLL = 0
 Month 1                          0.029      0.071           0.033
 Month 2                          0.284      0.282           0.303
 Month 3                          0.260      0.361           0.263
 Month 4                          0.081      0.259           0.085
 Month 5                          0.019      0.266           0.032
 Month 6                          0.047      0.085           0.047
 Month 7                          0.015      0.124           0.025
 Month 8                          0.008      0.082           0.034
Log-likelihood value           -2737.6                   -2755.1

                                Model 2

MLL dummy                        --
MLL Weeks mandated             (0.033)
Month 1 dummy                  (0.273)
Month 2 dummy                  (0.225)
Month 3 dummy                  (0.229)
Month 4 dummy                  (0.263)
Month 5 dummy                  (0.275)
Month 6 dummy                  (0.263)
Month 7 dummy                  (0.291)
MLL dummy-month 1 interaction    --
MLL dummy-month 2 interaction    --
MLL dummy-month 3 interaction    --
MLL dummy-month 4 interaction    --
MLL dummy-month 5 interaction    --
MLL dummy-month 6 interaction    --
MLL weeks-month 1 interaction  (0.036)
MLL weeks-month 2 interaction  (0.034)
MLL weeks-month 3 interaction  (0.034)
MLL weeks-month 4 interaction  (0.039)
MLL weeks-month 5 interaction  (0.040)
MLL weeks-month 6 interaction  (0.051)
Simulated probability          MLL = 1
 Month 1                        0.057
 Month 2                        0.239
 Month 3                        0.335
 Month 4                        0.188
 Month 5                        0.186
 Month 6                        0.061
 Month 7                        0.063
 Month 8                        0.082
Log-likelihood value

The dependent variable is the month probability of returning to work at
the prechildbirth job conditional on having not yet returned to work.
There are 1467 observations.

(a) Standard errors are in parentheses.

* Indicates statistical significance at the 10% level

** at the 5% level

*** at the 1% level.

Table 9

Effect of MLL on the Monthly probability of Returning to Work  at the
Prechildbirth job

                                  Model 3                 Model 4

Eligible MLL dummy          1.567 ***  (0.560) (a)       --
Eligible MLL weeks           --          --             0.120 ***
Eligible/covered dummy       --          --              --
Eligible/covered weeks       --          --              --
Month 1 dummy              -0.120      (0.230)          0.151
Month 2 dummy               1.558 ***  (0.320)          1.714 ***
Month 3 dummy               1.666 ***  (0.346)          1.737 ***
Month 4 dummy               0.928 ***  (0.338)          1.035 ***
Month 5 dummy               0.387      (0.361)          0.498
Month 6 dummy               0.643      (0.354)          0.685
Month 7 dummy               0.133      (0.381)          0.225
MLL dummy-month             1.260 **   (0.597)           --
 1 interaction
MLL dummy-month            -1.586 ***  (0.567)           --
 2 interaction
MLL dummy-month             1.162 **   (0.547)           --
 3 interaction
MLL dummy-month            -0.458      (0.570)           --
 4 interaction
MLL dummy-month             0.019      (0.600)           --
 5 interaction
MLL dummy-month            -0.925      (0.767)           --
 6 interaction
MLL weeks-month              --          --            -0.090 *
 1 interaction
MLL weeks-month              --          --            -0.124 ***
 2 interaction
MLL weeks-month              --          --            -0.085 *
 3 interaction
MLL weeks-month              --          --            -0.045
 4 interaction
MLL weeks-month              --          --            -0.008
 5 interaction
MLL weeks-month              --          --            -0.059
 6 interaction
Simulated probability      MLL = 0     MLL = 1         MLL = 0
 Month 1                    0.035       0.059           0.034
 Month 2                    0.309       0.305           0.303
 Month 3                    0.303       0.419           0.275
 Month 4                    0.115       0.375           0.098
 Month 5                    0.047       0.343           0.036
 Month 6                    0.072       0.173           0.052
 Month 7                    0.027       0.255           0.019
 Month 8                    0.020       0.219           0.011
Log-likelihood value    -2745.0                     -2746.3

                         Model 4           Model 5             Model 6

Eligible MLL dummy        --           --          --           --
Eligible MLL weeks      (0.047)        --          --           --
Eligible/covered dummy    --          1.259 **   (0.369)        --
Eligible/covered weeks    --           --          --          0.099 ***
Month 1 dummy           (0.291)       0.129      (0.291)       0.199
Month 2 dummy           (0.348)       1.619 ***  (0.284)       1.620 ***
Month 3 dummy           (0.384)       1.627 ***  (0.201)       1.600 ***
Month 4 dummy           (0.348)       0.880      (0.290)       0.886 ***
Month 5 dummy           (0.368)       0.352      (0.333)       0.374
Month 6 dummy           (0.364)       0.583 ***  (0.215)       0.544 *
Month 7 dummy           (0.386)       0.123      (0.334)       0.109
MLL dummy-month           --         -1.116 **   (0.498)        --
 1 interaction
MLL dummy-month           --         -1.467 ***  (0.248)        --
 2 interaction
MLL dummy-month           --         -0.992 **   (0.502)        --
 3 interaction
MLL dummy-month           --         -0.349      (0.549)        --
 4 interaction
MLL dummy-month           --          0.101      (0.598)        --
 5 interaction
MLL dummy-month           --         -0.833      (0.699)        --
 6 interaction
MLL weeks-month         (0.049)        --          --         -0.084 **
 1 interaction
MLL weeks-month         (0.048)        --          --         -0.115 ***
 2 interaction
MLL weeks-month         (0.044)        --          --         -0.076 **
 3 interaction
MLL weeks-month         (0.049)        --          --         -0.037
 4 interaction
MLL weeks-month         (0.051)        --          --         -0.002
 5 interaction
MLL weeks-month         (0.059)        --          --         -0.055
 6 interaction
Simulated probability   MLL = 1      MLL = 0     MLL = 1      MLL = 0
 Month 1                 0.065        0.036       0.047        0.034
 Month 2                 0.293        0.304       0.245        0.292
 Month 3                 0.401        0.271       0.351        0.252
 Month 4                 0.300        0.086       0.290        0.078
 Month 5                 0.263        0.030       0.254        0.026
 Month 6                 0.161        0.048       0.099        0.038
 Month 7                 0.204        0.017       0.160        0.013
 Month 8                 0.151        0.012       0.132        0.009
Log-likelihood value              -2746.7                  -2748.1

                         Model 6

Eligible MLL dummy        --
Eligible MLL weeks        --
Eligible/covered dummy    --
Eligible/covered weeks  (0.038)
Month 1 dummy           (0.291)
Month 2 dummy           (0.284)
Month 3 dummy           (0.295)
Month 4 dummy           (0.298)
Month 5 dummy           (0.325)
Month 6 dummy           (0.320)
Month 7 dummy           (0.354)
MLL dummy-month           --
 1 interaction
MLL dummy-month           --
 2 interaction
MLL dummy-month           --
 3 interaction
MLL dummy-month           --
 4 interaction
MLL dummy-month           --
 5 interaction
MLL dummy-month           --
 6 interaction
MLL weeks-month         (0.043)
 1 interaction
MLL weeks-month         (0.040)
 2 interaction
MLL weeks-month         (0.041)
 3 interaction
MLL weeks-month         (0.045)
 4 interaction
MLL weeks-month         (0.049)
 5 interaction
MLL weeks-month         (0.058)
 6 interaction
Simulated probability   MLL = 1
 Month 1                 0.049
 Month 2                 0.239
 Month 3                 0.339
 Month 4                 0.234
 Month 5                 0.197
 Month 6                 0.100
 Month 7                 0.132
 Month 8                 0.109
Log-likelihood value

The dependent variable is the monthly probability of returning to work
at the prechildbirth job conditional on having not yet returned to work.
There are 1467 observations.

(a) Standard errors are in parentheses.

* Indicates statistical significance at the 10% level

** at the 5% level

*** at the 1% level.

Table 10

Effect of MLL on the Monthly Probability of Starting Work at a New Job

                                         Model 1

MLL dummy                          -0.476       (0.540) (a)
MLL weeks mandated                   --           --
Month 1 dummy                      -1.067 **    (0.486)
Month 2 dummy                       0.044       (0.436)
Month 3 dummy                       0.447       (0.347)
Month 4 dummy                       0.081       (0.276)
Month 5 dummy                      -0.137       (0.279)
Month 6 dummy                       0.183       (0.235)
Month 7 dummy                       0.094       (0.236)
MLL dummy-month 1 interaction       0.335       (0.502)
MLL dummy-month 2 interaction       0.199       (0.447)
MLL dummy-month 3 interaction       0.138       (0.440)
MLL dummy-month 4 interaction      -0.740       (0.632)
MLL dummy-month 5 interaction       0.149       (0.535)
MLL dummy-month 6 interaction      -0.150       (0.532)
MLL weeks-month 1 interaction       --            --
MLL weeks-month 2 interaction       --            --
MLL weeks-month 3 interaction       --            --
MLL weeks-month 4 interaction       --            --
MLL weeks-month 5 interaction       --            --
MLL weeks-month 6 interaction       --            --
Simulated probability             MLL = 0       MLL = 1
 Month 1                            0.024        0.018
 Month 2                            0.136        0.095
 Month 3                            0.212        0.147
 Month 4                            0.144        0.021
 Month 5                            0.110        0.069
 Month 6                            0.160        0.071
 Month 7                            0.146        0.078
 Month 8                            0.132        0.069
Log-likelihood value           -2,737.60

                                        Model 2

MLL dummy                            --          --
MLL weeks mandated                 -0.006      (0.028)
Month 1 dummy                      -0.686 ***  (0.089)
Month 2 dummy                       0.303 *    (0.167)
Month 3 dummy                       0.660 ***  (0.197)
Month 4 dummy                       0.303 *    (0.176)
Month 5 dummy                       0.061      (0.191)
Month 6 dummy                       0.238      (0.188)
Month 7 dummy                       0.105      (0.194)
MLL dummy-month 1 interaction        --          --
MLL dummy-month 2 interaction        --          --
MLL dummy-month 3 interaction        --          --
MLL dummy-month 4 interaction        --          --
MLL dummy-month 5 interaction        --          --
MLL dummy-month 6 interaction        --          --
MLL weeks-month 1 interaction      -0.006      (0.037)
MLL weeks-month 2 interaction      -0.017      (0.031)
MLL weeks-month 3 interaction      -0.006      (0.031)
MLL weeks-month 4 interaction      -0.110      (0.071)
MLL weeks-month 5 interaction      -0.006      (0.041)
MLL weeks-month 6 interaction      -0.043      (0.053)
Simulated probability             MLL = 0      MLL = 1
 Month 1                            0.020       0.013
 Month 2                            0.123       0.076
 Month 3                            0.195       0.152
 Month 4                            0.118       0.005
 Month 5                            0.076       0.054
 Month 6                            0.102       0.032
 Month 7                            0.080       0.068
 Month 8                            0.064       0.055
Log-likelihood value           -2,755.1

The dependent variable is the monthly probability of starting work at a
new job conditional on having not yet returned to work.

There are 1467 observations.

(a) Standard errors are in parentheses.

* Indicates statistical significance at the 10% level

** at the 5% level, and

*** at the 1% level.

Table 11

Effect of MLL on the Monthly Probability of Starting Work at a New Job

                                  Model 3                 Model 4

Eligible MLL dummy         -0.279      (0.531) (a)       --
Eligible MLL weeks           --          --            -0.018
Eligible/covered dummy       --          --              --
Eligible/covered weeks       --          --              --
Month 1 dummy              -0.531 ***  (0.181)         -0.547 ***
Month 2 dummy               0.429 ***  (0.167)          0.385 ***
Month 3 dummy               0.725 ***  (0.162)          0.699 ***
Month 4 dummy               0.296 *    (0.184)          0.276
Month S dummy               0.116      (0.197)          0.103
Month 6 dummy               0.260      (0.192)          0.253
Month 7 dummy               0.162      (0.200)          0.136
MLL dummy-month             0.133      (0.615)           --
 1 interaction
MLL dummy-month             0.043      (0.545)           --
 2 interaction
MLL dummy-month             0.286      (0.552)           --
 3 interaction
MLL dummy-month            -0.341      (0.706)           --
 4 interaction
MLL dummy-month             0.038      (0.724)           --
 5 interaction
MLL dummy-month             0.019      (0.739)           --
 6 interaction
MLL weeks-month              --          --            -0.020
 1 interaction
MLL weeks-month              --          --            -0.008
 2 interaction
MLL weeks-month              --          --             0.019
 3 interaction
MLL weeks-month              --          --            -0.058
 4 interaction
MLL weeks-month              --          --            -0.015
  5 interaction
MLL weeks-month              --          --            -0.036
 6 interaction

Simulated probability      MLL = 0     MLL = 1         MLL = 0
 Month 1                    0.020       0.014           0.021
 Month 2                    0.117       0.080           0.119
 Month 3                    0.170       0.167           0.179
 Month 4                    0.088       0.027           0.093
 Month 5                    0.062       0.039           0.067
 Month 6                    0.079       0.049           0.086
 Month 7                    0.066       0.038           0.069
 Month 8                    0.048       0.027           0.052
Log-likelihood value    -2745.0                     -2746.3

                         Model 4           Model 5             Model 6

Eligible MLL dummy        --           --          --           --
Eligible MLL weeks      (0.043)        --          --           --
Eligible/covered dummy    --         -0.134      (0.425)        --
Eligible/covered weeks    --           --          --         -0.014
Month 1 dummy           (0.189)       0.559 ***  (0.181)      -0.573 ***
Month 2 dummy           (0.165)       0.368 ***  (0.145)       0.343 **
Month 3 dummy           (0.157)       0.706 ***  (0.180)       0.686 ***
Month 4 dummy           (0.186)       0.288      (0.169)       0.281
Month S dummy           (0.198)       0.116      (0.183)       0.110
Month 6 dummy           (0.192)       0.255      (0.178)       0.257
Month 7 dummy           (0.200)       0.151      (0.194)       0.138
MLL dummy-month           --          0.030      (0.424)        --
 1 interaction
MLL dummy-month           --         -0.022      (0.401)        --
 2 interaction
MLL dummy-month           --          0.160      (0.459)        --
 3 interaction
MLL dummy-month           --         -0.362      (0.664)        --
 4 interaction
MLL dummy-month           --         -0.022      (0.571)        --
 5 interaction
MLL dummy-month           --         -0.030      (0.683)        --
 6 interaction
MLL weeks-month         (0.058)        --          --         -0.018
 1 interaction
MLL weeks-month         (0.046)        --          --         -0.007
 2 interaction
MLL weeks-month         (0.047)        --          --          0.013
 3 interaction
MLL weeks-month         (0.073)        --          --         -0.060
 4 interaction
MLL weeks-month         (0.066)        --          --         -0.015
  5 interaction
MLL weeks-month         (0.076)        --          --         -0.033
 6 interaction

Simulated probability   MLL = 1      MLL = 0     MLL = 1      MLL = 0
 Month 1                 0.007        0.021       0.016        0.022
 Month 2                 0.072        0.119       0.094        0.122
 Month 3                 0.176        0.186       0.186        0.192
 Month 4                 0.014        0.099       0.037        0.103
 Month 5                 0.030        0.071       0.051        0.074
 Month 6                 0.024        0.089       0.066        0.095
 Month 7                 0.045        0.073       0.056        0.075
 Month 8                 0.034        0.054       0.041        0.057
Log-likelihood value              -2746.7                  -2748.1

                         Model 6

Eligible MLL dummy        --
Eligible MLL weeks        --
Eligible/covered dummy    --
Eligible/covered weeks  (0.042)
Month 1 dummy           (0.181)
Month 2 dummy           (0.172)
Month 3 dummy           (0.180)
Month 4 dummy           (0.179)
Month S dummy           (0.190)
Month 6 dummy           (0.189)
Month 7 dummy           (0.197)
MLL dummy-month           --
 1 interaction
MLL dummy-month           --
 2 interaction
MLL dummy-month           --
 3 interaction
MLL dummy-month           --
 4 interaction
MLL dummy-month           --
 5 interaction
MLL dummy-month           --
 6 interaction
MLL weeks-month         (0.057)
 1 interaction
MLL weeks-month         (0.046)
 2 interaction
MLL weeks-month         (0.045)
 3 interaction
MLL weeks-month         (0.072)
 4 interaction
MLL weeks-month         (0.065)
  5 interaction
MLL weeks-month         (0.074)
 6 interaction

Simulated probability   MLL = 1
 Month 1                 0.009
 Month 2                 0.082
 Month 3                 0.184
 Month 4                 0.017
 Month 5                 0.037
 Month 6                 0.032
 Month 7                 0.055
 Month 8                 0.041
Log-likelihood value

The dependent variable is the monthly probability of starting work at a
new job conditional on having not yet returned to work. There are 1467
observations.

(a) Standard errors are in parentheses.

* Indicates statistical significance at the 10% level

** at the 5% level

*** at the 1% level.

Received February 2002; accepted June 2002.

(1.) Eligibility is determined by the mother's work history and by firm size.

(2.) The cases in which workers and employers are not able to voluntarily implement maternity leave policies are those where there is a market failure. Such market failures include asymmetric information (where only the mother knows with certainty whether she will return to her prechildbirth job), imperfect information (where mothers do not appropriately value the benefits of maternity leave for themselves or their children), and adverse selection (where firms who offer maternity leave benefits without MLL attract a disproportionate number of potential leave takers).

(3.) Employers are potentially interested in other outcomes as well. For example, employers might wonder how MLL affects their employees' fertility decisions (see Averett and Whittington 2001).

(4.) "Waidfogel (1999) looks at firm size but not the employee's work history.

(5.) Given this specification, mothers who take leave in absence of MLL either separate permanently from their employers or use maternity leave provided by their employers. These mothers may be inherently different in that the former are not firmly attached to their employers and the latter are. Ideally, the availability of employer-provided leave would be controlled for as well as government-provided leave. However, data on whether the employer offers maternity leave benefits are not available for alt mothers. First, the question only pertains to the job at which the mother was employed at the time of the survey and the survey job is not always the same job that the mother held around the child's birth. Second, this question was not answered by all mothers, with a sizable portion providing invalid responses. Presumably, some mothers did not know whether their employer provided maternity leave. Third, it only indicates whether the employer provided leave; it does not identify the number of weeks of leave prov ided by the employer. Thus, it does not tell us what we really want to know--whether MLL has a binding (or nonbinding) effect.

(6.) It may sound arbitrary to pick an eight-month limit within which mothers must return to their prechildbirth employer in order to be counted as "returned," but the data show that no mothers return to their old job more than eight months after giving birth.

(7.) Only births in 1994 for which there are sufficient data to determine the mother's employment status in the first eight months after giving birth arc included.

(8.) Three months was chosen because it is within this period that most working pregnant women from the sample stop working.

(9.) For more information on identifying paid and unpaid maternity leave in the NLSY data set, see Klerman (1994).

(10.) Table 1 presents similar information for the leave-taking sample used in the second-period return-to-work model. In particular, the return-to-work model (which consists of 1467 women who took leave from work in the first period) contains 292 observations with MLL.

(11.) Table 3 lists the minimum number of employees that a firm must employ to be covered by a leave mandate for each state with MLL.

(12.) Identifying whether the employer is covered for all observations was not possible because firm size was not available for alt jobs. Instead, the NLSY only asked for firm size for the job at which the respondent was working at the time of the survey. Because the prechildbirth employer is different from the "survey week" employer for some mothers, it is not possible to identify firm size for all mothers.

(13.) The spouse's education level is set equal to zero if no spouse is present. The marital status dummy variable will control for those with no spouse and a zero value for spouse's education.

(14.) Nonwage income includes alimony, child support, wages, and salary earned by the mother's spouse, the spouse's income from farm and/or business after expenses, and income from savings accounts and assets. Invalid missing values for any of the five components of nonwage income are replaced with the sample average for that component from the appropriate year. Valid missing information is replaced with a zero.

(15.) These variables are set equal to zero if there is missing information and corresponding missing-information dummy variables are then set equal to one.

(16.) In the first approach, Akaike's (1973) information criterion (as interpreted by Gritz 1993) fails to reject the null hypothesis that unobserved heterogeneity does not have an effect: the six parameters that make up the unobserved heterogeneity's distribution decrease the log-likelihood function's value by only 0.6 points. Thus, it appears that controlling for unobserved heterogeneity is not an issue. However, in the second approach, Akaike's Information Criterion Test (as interpreted by Grits) marginally rejects the null hypothesis that unobserved heterogeneity has no effect: the six parameters that comprise the unobserved heterogeneity's distribution reduce the log-likelihood function's value by 6.1 points. Additionally, though controlling for unobserved heterogeneity does not substantively change MLL's effects, controlling for unobserved heterogeneity has a larger impact on other covariates' coefficients. For these reasons, estimates of the models were done controlling for unobserved heterogeneity.

(17.) In these models, the simulations indicate that MLL increases in the probability of returning to the prechildbirth employer in some months (months 4 and 5, for example) that are beyond the guarantee period. (Actually, the fourth month is not beyond the guarantee period--California mandates 17 weeks of maternity leave and the District of Columbia and Tennessee each mandate 16 weeks.) However, it is important to note that the statistically significant MLL effects are generally only in the second and third months. So we can only say with confidence that MLL decreases the probability of returning to the old job in the second month and increases this probability in the third month. Simulated positive effects using the point estimates for the remaining months are actually not statistically different from zero.

(18.) These estimates closely predict actual sample averages. Sample means from Figure 1 show that about 55% have returned to their prechildbirth employer by the eighth month and about 33% have started a new job by the eighth month. Taken together, these statistics imply that about 12% of these mothers (who were working within three months prior to giving birth) have not yet returned to work by the eighth month.

References

Akaike, Hirotugu. 1973. Information theory and an extension of the maximum likelihood principle. In 2nd International Symposium on Information Theory, edited by B. N. Petrov and C. Csaki. Budapest: Akademiai Kiado, pp. 267-81.

Averett, Susan L., and Leslie A. Whittington. 2(101. Does maternity leave induce births? Southern Economic Journal 68:403-17.

Blau, David M. 1994. Labor force dynamics of older men. Econometrica 62:117-56.

Blau, David M. 1997. Social Security and the labor supply of older married couples. Labour Economics 4:373-418.

Blau, David M. 1998. Labor force dynamics of older married couples. Journal of Labor Economics 16:595-629.

Blau, David M., and Alison Hagy. 1998, The demand for quality in child care. Journal of Political Economy 106:104-46.

Bond, James T. 1991. Beyond the parental leave debate: The impact of laws in four states. New York: Families and Work Institute.

Bureau of National Affairs. 1987. Pregnancy and employment: The complete handbook on discrimination, maternity leave, and health and safety. Rockville, MD: Bureau of National Affairs.

Gritz, R. Mark. 1987. An empirical analysis of the effect of training programs on employment. Dissertation, Stanford University. Stanford, CA.

Grits, R. Mark. 1993. The impact of training on the frequency and duration of employment. Journal of Econometrics 57:21-51.

Ham, John C., and Robert J. LaLonde. 1996. The effect of sample selection and initial conditions in duration models: Evidence from experimental data on training. Econometrica 64:175-205.

Heckman, James J., and Burton Singer. 1984. A method for minimizing the distributional assumptions in econometric models for duration data. Econometrica 52:271-320.

Heckman, James J., and James R. Walker. 1990. The relationship between wages and income and the timing and spacing of births: Evidence from Swedish longitudinal data. Econometrica 58:1411-41.

Hotz, V. Joseph, Lixin Xu, Marta Tienda, and Avner Ahituv. 1999. Are there returns to the wages of young men from working while in school? Unpublished paper, National Bureau of Economic Research.

Kane, Carol Kallman. 1998. State mandates for maternity leave: Impact on wages, employment, and access to leave. Dissertation, Boston College, Boston, MA.

Klerman, Jacob. 1994. Characterizing leave for maternity. Unpublished paper, The RAND Corporation.

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

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

Lancaster, Tony. 1979. Econometric methods for the duration of unemployment. Econometrica 47:939-56.

Lancaster, Tony. 1985. Generalized residuals and heterogeneous duration models with applications to the Weibull model. Journal of Econometrics 28:155-69.

Leibowitz, Arleen, and Jacob Klerman. 1995. Explaining changes in married mothers' employment over time. Demography 32:365-78.

Light, Audrey, and Wayne Strayer. 1996. Determinants of college completion: School quality or student ability. Journal of Human Resources 35:299-332.

Mroz, Thomas A. 1999. Discrete factor approximations in simultaneous equation models: Estimating the impact of a dummy endogenous variable on a continuous outcome. Journal of Econometrics 92:233-74.

Ondrich, Jan, C. Katherine Spiess, Qing Yang, and Gert G. Wagner. 1998. Full time or part time? German parental leave policy and the return to work after childbirth in Germany. Unpublished paper, Syracuse University Center for Policy Research.

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

Ruhm, Christopher J., and Jackqueline L. Teague. 1997. Parental leave policies in Europe and North America. In Gender and family issues in the workplace, edited by Francine Blau and Ronald Ehrenberg. New York: Russell Sage, pp. 133-57.

Waldfogel, Jane. 1999. The impact of the Family and Medical Leave Act. Journal of Policy Analysis and Management 18: 281-301.

Women's Legal Defense Fund. 1994. State laws and regulations guaranteeing employees their jobs after family and medical leaves. Washington, DC: Women's Legal Defense Fund.

Charles L. Baum II *

* Department of Economics and Finance, P.O. Box 27, Middle Tennessee State University, Murfreesboro, TN 37132, USA; E-mail [e-mail].

I would like to especially thank David M. Blau for a substantial amount of help. I would also like to thank Thomas Mroz and Wayne Strayer for helping with the Fortran code that estimates the discrete factor random effects specification, and two anonymous referees for their helpful comments. I also thank the Faculty Research and Creative Activity Committee (FRCAC) and Donald L. Curry at Middle Tennessee State University for financial support.