Demographic consequences of maternal-leave programs in industrial countries: evidence from fixed-effects models.

• Author: Winegarden, C.R.

1. Introduction

   With the passage of the Family and Medical Leave Act of 1993, the U.S. became the last industrial country to establish an official program for maternity leave. Nearly all of these countries have maintained such programs for several decades and, unlike the U.S., have provided at least partial replacement of lost wages [30]. Economic theory strongly suggests that the various income-and-substitution effects embodied in these programs should have predictable demographic impacts. It appears, however, that very little systematic investigation has actually been undertaken along these lines, apart from a study mainly oriented to the Swedish experience with respect to female labor-force participation and fertility.(1)

   In the present paper, we try to measure the effects of paid maternity-leave on three demographic variables: infant mortality, labor-force participation of women in the prime childbearing ages, and fertility rates. For this purpose, we construct a simultaneous-equations model, using the individual fixed-effects method and a data set comprising 17 OECD countries and four time periods. The structural estimates provide substantial evidence in support of our predictions that lengthening the allowed duration of paid leave reduces infant mortality, while increasing both the labor-force participation of young women and the general fertility rate. However, the reduced-form analysis casts doubt on the long-run fertility effect.

   The next section of the paper briefly surveys maternal-leave policies in industrial countries. Section III deals with the data and estimation methods. Section IV discusses the theoretical basis of the empirical model. In section V, we describe our regression results. Section VI details a series of additional tests designed to check the validity of these regressions. Section VII presents the reduced-form results derived from the structural estimates. Conclusions are set forth in the final section of the paper. An Appendix provides the list of countries and data sources.

2. A Brief Survey of Maternal-Leave Programs in Industrial Countries

   Two kinds of benefits are included in official maternal-leave programs:

   1. The right of a mother (in some countries, fathers as well) to take a specified period of time off from her job upon the birth or adoption of a child, with assurance of a comparable job upon her return and the protection of her pension and seniority rights. Usually, six weeks of this leave may be taken prior to the birth of the child.

   2. In nearly all cases, the payment of a cash benefit that is equivalent to some proportion of the previous wage. This system is generally provided through the country's social insurance system and is typically funded through employee/employer contributions [14; 30].

   There are major variations among countries in terms of duration of leave granted and proportion of wages paid. Sweden, Finland, and Italy qualify as the most generous, both in terms of duration and rate of income replacement. Australia, while liberal in duration of leave, offers no income replacement. Among the seventeen countries in our sample, the number of weeks of leave allowed in 1989 ranged between a minimum of 12 and a maximum of 52; the minimum income replacement rate (excluding the U.S.) was 60% while the maximum was 100%.(2)

   Most maternal-leave programs in industrialized nations originally were established as part of an effort to arrest long-term declines in rates of population growth, coupled with rising concern for the health of working mothers and their children. During the 1930s, particularly, many European countries introduced maternal-leave benefits, or liberalized already existing policies, although female labor-force participation rates were still low. The most extreme pronatalist programs were undertaken by the fascist governments of Germany, Spain, and Italy. Nowhere did the implementation of these programs result in a reversal of the downward trend in birth rates. The continuing decline was especially dramatic in those very countries that were most extreme in their pronatal policies [3; 16].

   After World War II, although population concerns still were relevant, European countries' experience with Fascism led to a turning away from blatantly probirth policies. At the same time, increasing female labor-force participation fostered an interest in the labor market outcomes of maternal-leave policies, with special attention paid to minimizing tensions between women's work and family obligations. Other "quality of life" effects of family policy also played a role. The influence of these programs on child development and family formation was widely discussed. Feminists, who played an important role in focussing attention on the unique position of women in society (as both nurturers and producers), stressed the importance of gender equality [14].

   In keeping with the widening of concerns, there has been a strong tendency toward more generous leave programs, both in terms of duration and rate of wage replacement. In the 17 OECD countries comprising our sample, maximum leave duration rose from a mean of about 10 1/2 weeks in 1959 to nearly 17 weeks in 1989. Over this period, the average rate of wage replacement in these countries increased from less than half to more than three-quarters [30].

   [TABULAR DATA FOR TABLE I OMITTED]

3. Data and Estimation Methods

   We make use of a balanced cross-section/time-series sample comprising 17 OECD countries and four time periods, i.e., 1959, 1969, 1979, and 1989. OECD countries for which the necessary data were not available for all periods were dropped from the sample. The decade-long intervals are not only an accommodation to the availability of data, but also allow for the slowly-changing character of certain variables (notably, maternal leave programs, whose provisions are infrequently altered). The regression variables are defined and their sample means and standard deviations are shown in Table I.

   Fixed individual-effects estimation is our chosen econometric method. This approach in effect provides time-series estimates, utilizing the cross-sectional data only to increase the efficiency of estimation. (In the present case of t = 4, pure time-series estimation is obviously impossible). It has the further advantage of controlling for a variety of unobservable, time-invariant cross-national influences on demographic behavior, e.g., differences in religious observance, the psychic value of children, and the status of women. There are, of course, alternatives: (a) pooling all observations, which assumes identical intercepts across countries; (b) fixed time-effects, which assumes identical intercepts only across time periods; or (c) random-effects estimation. The first two were readily rejected by empirical testing. (See section VI below). Random-effects estimation must be ruled out, because this estimator would be impaired by omitted variable bias, i.e., the individual effects embodied in the random-effects error term are assumed to be correlated with one or more explanatory variables [13]. We believe that the individual effects are correlated with several regressors, including the duration of maternity leave, the incidence of divorce, and the frequency of recorded stillbirths. As in any time-series regression, time-varying unobservables may have biasing effects, but the rejection of fixed time-effects suggests that this may not be a serious problem in the present case. (See also the discussion of time trends and proxy variables in section IV below).

   For purposes of estimation, we use demeaned variables, i.e., each variable is measured as the deviation from its mean, computed across time-period observations for that country, which leads to the omission of a constant term. The results are identical to those from dummy-variable estimation except for the absence of direct estimates of the individual intercepts. It is evident from the inclusion of right-side endogenous variables that some form of instrumental variable estimation is required. (Exclusion restrictions serve to overidentify all equations). Further, the expected cross-equation correlation of the error terms justifies the use of 3SLS estimation.

4. Theoretical Basis of the Empirical Model

   In order to gauge the demographic impacts of maternal-leave programs, two sets of effects must be addressed. The first set consists of the direct effects on each of the variables of interest. The second requires taking into account indirect effects arising from the interrelationships of these variables. Both sets may be readily summarized in the generalized form of the linear model shown below:

   [Mathematical Expression Omitted]

   [Mathematical Expression Omitted]

   [Mathematical Expression Omitted]

   where IMR, LFP, and GFR are infant mortality rates, labor-force participation rates for women in the prime childbearing ages (20-34), and general fertility rates, respectively; MATLV gauges weeks of paid maternal leave; and the X's represent intersecting sets of "control" variables. The [Mu]'s are stochastic error terms, assumed to be homoscedastic, serially uncorrelated, and with non-zero covariance across equations. (Time and country subscripts are omitted). As stated above, all variables are demeaned. The hypothesized directions of these various effects are shown by the signs above the arguments of the functions.

   The expected, direct effects of maternal-leave programs are straightforward:

   1. We expect infant mortality to decline as paid maternal leave is lengthened [14]. Several factors should contribute: mothers should tend to remain at home with newborn babies for longer periods, which may in turn improve the quality of infant care and may also encourage or prolong breastfeeding, for which there is evidence of enhanced infant survival [8]. Moreover, there is a positive income effect on households with newborn infants insofar as the maternity-leave payments replace income otherwise...