Changes in maternity leave coverage: Implications for fertility, labour force participation and child mortality

Highlights

• Longer duration of leave reduced infant mortality and demand for female labour.

• Maternity leave payments lead to higher fertility.

• We found significant interactions between the three development outcomes.

• Findings may not be generalisable to the whole of Africa and Asia.

Abstract

Analysing macro-panel data from 18 African and Asian countries over the period 1995–2016, this article investigates the effects of the level and duration of paid maternity leave on three dimensions of human development: fertility, female formal-sector employment and infant mortality. There is some evidence that, on average, extending the duration of leave leads to reductions in infant mortality and employment. However, there is no conclusive evidence that leave duration has a direct effect on fertility. In contrast, there is some evidence that higher maternity leave payments lead to higher fertility, but no evidence that payment levels have any effect on infant mortality or employment.

Introduction

There is now substantial evidence regarding the beneficial consequences of paid maternity leave (ML) in industrialised countries. These consequences include longer periods of breastfeeding (Berger et al., 2005; Baker and Milligan, 2008; Khanam et al., 2016; Albagli and Rau, 2018), lower mortality rates among infants and young children (Winegarden and Bracy, 1995; Ruhm, 2000; Tanaka, 2005; Fallon et al., 2017; Heymann et al., 2017), higher childhood vaccination rates (Daku et al., 2012; Khanam et al., 2016; Heymann et al., 2017) and, for mothers, better post-partum physical and mental health (Aitken et al., 2015, and literature cited therein). Research also suggests that ML has a positive effect on fertility (Winegarden and Bracy, 1995; Averett and Whittington, 2001; Risse, 2006; Luci-Greulich and Thévenon, 2013).

However, there is relatively little evidence linking ML with child and maternal well-being in low-income countries. There are a small number of non-statistical studies (Chang, 2004; İlkkaracan, 2012), but only Nandi et al. (2016) and Fallon et al. (2017) present detailed statistical evidence. Nandi et al. (2016) apply a difference-in-differences model to data from the Demographic and Health Surveys in 20 low- and middle-income countries for 2000–2008, but they caution that their findings may not be generalisable to other countries. Fallon et al. (2017) apply a cross-sectional time-series model to data from 121 low- and middle-income countries for 1999–2012.

Against this background, we assess the relationship between ML and key demographic indicators for women and children in 18 countries in Africa and Asia. In these countries, as in most others, there are two key dimensions to ML: the duration of the leave and the level of payment. The latter is usually expressed as a percentage of earnings, that is, as the wage replacement rate (WRR). Nearly all of the countries in our sample have had at least 12 weeks of ML for several decades, but many of them have increased the level or duration of payment in recent years (see Table 1). As we will see, substantial progress in terms of lower infant mortality rates (IMRs) and total fertility rates (TFRs) has also been seen during this time period, but there has been little or no increase in the rates of female formal-sector employment (FFSE). The reasons for these differences in trends are not well understood. Moreover, there is still little recent evidence on how changes in ML coverage are related to the development outcomes. This article updates earlier evidence by focusing on a period of substantial changes in IMRs and TFRs, examining the extent to which ML has contributed to these changes and suggesting explanations for the absence of any corresponding change in FFSE. We note as a caveat that our sample includes only a subset of countries in Asia and a small fraction of countries in Africa. Our results are relevant to policies in these countries, but we do not claim that they are relevant to the whole of Africa and Asia.

This article contributes to the literature in several ways. First, our sample includes a number of countries that have been absent from any previous data analysis (e.g., our Asian sample includes Afghanistan and the Maldives). Second, our statistical model allows for changes in laws relating to abortion. These changes reflect policy innovation in an area that still attracts a large amount of social stigma but has substantial heterogeneity across countries. In India, for example, abortion has been legal in a wide range of situations since the 1950s, while in Nepal it was not legalised until 2002 (Table 1). In many other countries, legal abortion is restricted to cases in which there is a serious threat to the mother's physical or mental health. To date, there has been no statistical study analysing the effects of both abortion policy and ML on IMR and TFR. Third, we follow the empirical design of Fallon et al. (2017) but improve on previous research by modelling IMR, TFR and FFSE simultaneously using three-stage least squares (3SLS) and allowing for interactions between the three development outcomes (see Winegarden and Bracy, 1995). We apply our model to a relatively lengthy series and control for a relatively wide set of institutional factors. Several different versions of the model are fitted to the data in order to assess the robustness of the results.

We present evidence that increasing the duration of ML leads to lower IMR and FFSE. This second result may reflect a negative effect on labour demand that exceeds any positive effect on labour supply. There is no robust evidence that ML duration has an effect on fertility. In contrast, there is some evidence that increases in WRR lead to higher TFR but no evidence that WRR has an effect on IMR or FFSE. There are some large and statistically significant interactions between the three development outcomes, which helps explain the persistently low rates of female employment. There are no statistically significant effects due to changes in abortion policy, on average.

The next section outlines the expected relationships between the variables of interest, based on the existing literature. Subsequent sections present the data and statistical results.