Can We Still Learn Something From the Relationship Between Fertility and Mother’s Employment? Evidence From Developing Countries

Although the authors did not find a clear pattern characterizing the distribution of women in the informal sector, they observed some trends. On the one hand, evidence indicates that those women are more likely to be engaged in nonwage employment (self-employed or unpaid family worker). On the other hand, when women are in wage employment, their numbers are disproportionally greater at the bottom of the distribution. Both elements are linked with a higher incidence of poverty among women in the informal sector.

Agüero and Marks, by the nature of the source of variation they used (infertility), captured the impact of fertility at lower parities. In a context of heterogeneity in the impact of family size and individuals behaving as a function of this heterogeneity, the parameters estimated by instruments can be interpreted as local average treatment effects (LATE) (Angrist and Imbens 1994). In that sense, the external validity of the estimates is compromised in a context where the population of interest for policy makers or development institutions promoting family planning programs are those families in the upper tail of the fertility distribution.

Ideally, as the sample size grows, I should allow that c* = c: that is, the impact of all the variables in the model could differ among countries. This is the equivalent of running a regression separately for each country. However, the relatively small sample size for each country and the data-intensive requirements of the identification strategy permit a point between a model fully flexible (with c* = c) and a model that assumes the impact that is constant across countries.

For the United States, the American Society for Reproductive Medicine (Martin and Park 1999) lists four such factors. First, the incidence is higher among African Americans. Second, women who are themselves fraternal twins give birth to twins at a rate of 1 set per 60 births, which is higher than the national rate of 1 set per 90 births. Third, women aged 35–40 with four or more children are three times more likely to have twins than women younger than 20 without children. Finally, multiple births are more common among women who use fertility medication (Martin and Park 1999).

Rosenzweig and Zhang (2009) used child birth weight as proxy of child endowment. For some country-year samples, DHS data have information on birth weight for some of the children in the household. This limitation in the sample and the high data requirements (sample size) when using multiple births as source of variation make this alternative not viable. Although I cannot directly address this concern in the main analysis, as a second best alternative and using auxiliary data constructed from census data for three developing countries, I check the robustness of the impact of fertility on mother’s labor employment to the addition of a measure of child endowment. Specifically, I use disability status as a proxy for child endowment, and I check the sensitivity of the estimates to the inclusion of this variable. The results for this analysis support the robustness of the results, and they can be found in a working paper that can be downloaded from the author’s webpage (http://​sites.​google.​com/​site/​caceresjulio/​home/​research).

The use of the concept of heterogeneity as synonymous with selection is often observed in the empirical literature. Here, I make reference to the fact that the parameter γ ^s cannot be summarized by scalar rather with a complete distribution (Heckman et al. 2006).

When γ ^s is homogenous, multiple births being a valid instrument (as well as any other valid instrument) will allow us to identify all the relevant parameters such as the average treatment effect (ATE), average treatment on the treated (ATT), or average treatment on the untreated (ATUT) because they all are the same (Heckman et al. 2006).

An extended discussion can be found in the author’s website (http://​sites.​google.​com/​site/​caceresjulio/​home/​research ).

Giving this sample design, all regressions in the present analysis incorporate sample weights. Also, the standard errors have been adjusted to allow for cluster correlation.

These three restrictions reduced the sample size, respectively, by 1.9%, 1.2%, and 38.6% for the sample 1+; by 1.1%, 1.1%, and 43.2% for the sample 2+; by 0.8%, 1.1%, and 52.9% for the sample 3+; and by 0.6%, 1.0%, and 60% for the sample 4+.

The same qualitative results are obtained when using number of surviving children as a measure of a woman’s fertility. I also used a measure of fertility adjusted by child spacing, and in general, the estimates for all regressions tend to be bigger. I opted for the most conservative measure.

A paid job is not necessarily a job for which a mother is paid in cash. Many jobs at a subsistence level are characterized by payment in kind or services.

Two educational levels are considered: mothers with no formal education (approximately 40% of all mothers), and mothers with some years of education.

To be consistent with the explanation of the estimated parameter, the sample must be defined such that sample 1+ contains families aiming for a first “birth,” some of whom go on to have just one child and others of whom have multiple births. In general, for sample s+, I keep families aiming for the s birth. Then, I restrict the sample to those with no multiple births in pregnancies prior to the pregnancy for which I study the unexpected change in family size. Specifically, I restrict sample 2+ to families that did not have multiple births in the first birth, but I restrict sample 3+ to families that did not have multiple births in either the first or the second birth. The results are not sensitive to this restriction because families who have two or more events of multiple births are extremely rare.

The same robustness is observed for the subsamples defined by country-region, urban status, and mother’s education level. To save space, I did not include them in the table.

Angrist et al. (2010), for Israel, found an impact of multiple births that varies from 0.43 to 0.69, depending on the sample considered; Black et al. (2005), for Norway, found an impact that varies from 0.67 to 0.817. These point estimates are slightly lower than the ones reported in this article. This finding is explained by the differences in the period considered in each of these studies and the reported decrease of completed fertility among less-developed regions. My analysis uses data later than 1990 (Table 6 in the appendix), with 1972 as the median year of birth for the mothers considered in the analysis. In Angrist et al. (2010), the average birth cohort year of the mothers for which the impact of multiple births on the number of children is estimated as 1942. Angrist and Evans (1998) used 1980 U.S. census data with mothers born between 1948 and 1960. Therefore, by using older cohorts, these two studies obtained the impact of multiple births for individuals who had, on average, higher desired fertility and therefore studied a smaller group of compliers, which implies a lower observed impact on fertility.

This positive sign for some of the OLS estimates in Table 2 is consistent with the descriptive statistic and the omitted variable bias associated with OLS. Although many factors can be speculated as part of the omitted variables, usually a concept of “ability” is used to explain the sign in the bias. Nevertheless, independent of the factors responsible for the omitted variables, the descriptive statistics (Tables 7–9 in the appendix) show that those individuals with bigger family sizes not only have lower levels of human capital (education) but are also largely represented in job types with higher levels of informality, such as unpaid jobs, self-employment, or occasional jobs, for which a positive and significant impact is estimated.

One must be careful of reading these findings as evidence against a negative impact of childbearing on mother’s labor engagement. In addition to the loss of power associated with the use of instrumental variables, by defining narrow outcomes for female employment, I define a smaller group of compliers, making the estimates noisier.

Agüero and Marks (2008) argued, for example, that women with high career-based unobservables, such as “ambition or talent,” might be the ones choosing to have smaller families and that these mothers are overrepresented in the labor market.

Formally, the \( {\rm plim} ({{\gamma }_{{ols}}}) = {{\gamma }^s} + \rho \cdot {\rm cov} \,(ability,n)/V(n) \) with ρ as the impact of ability on the propensity to engage in labor market activities; and with a valid instrument, the \( {\rm plim} ({{\gamma }_{{IV}}}) = \gamma \). Therefore, while IV provides us with a consistent estimate of the negative impact of fertility, γ, the OLS estimator has asymptotic bias equal to \( \rho \cdot \frac{{{\rm cov} (ability,n)}}{{V(n)}} \lg 0 \) because cov(ability,n)  <  0 and ρ  ≶  0. Therefore, with household production in the model, \( (\rho < 0),\left| {{\rm plim} ({{\gamma }_{{ols}}})} \right| < \left| {{\rm plim} ({{\gamma }_{{IV}}})} \right| = \left| \gamma \right| \).

This lack of agreement is related to the fact that earlier definitions of informality were unable to capture an increasing proportion of unprotected jobs within the formal sector itself. Finally, the 17th International Conference of Labour Statisticians provided guidelines for the definition and measurement of informal employment (Jütting and Laiglesia 2009). Under these guidelines, the definition of informality should be related to characteristics of the job rather than the economic units to which they belong.

This proportion goes from approximately 70% for sample 1+ to approximately 80% for sample 4+. Jütting and Laiglesia (2009) reported for 76% for the sub-Saharan region (1990–1994), 54.2 for Latin America (1995–1999), and 70% for the South and Southeast Asia (1995–1999). These magnitudes are marginally smaller than the ones I report in Table 3. The differences are partially explained by the fact that I have not excluded jobs associated with agricultural activities and also that women in developing countries are overrepresented in informal activities (Blunch et al. 2001).

The same conclusions are obtained by using “currently working,” “currently informal,” and “currently formal.” The results are available in the working paper version of this article (http://​sites.​google.​com/​site/​caceresjulio/​home/​research).

This result should be read jointly with a negative impact on the likelihood of holding an informal job during the past 12 months, and a negative and nonsignificant impact of holding a job the last year. Therefore, the results speak for a compositional effect associated with an unexpected increase in fertility for this sample of mothers.

Nevertheless, they found that mothers with more education are more likely to adjust intensive margins—that is, by moving from full- to part-time employment—whereas mothers with lower levels of education are more likely to stop working. Therefore, controlling for wages, women with more education face larger negative effects associated with an increase in family size, which suggests, according to the authors, a combination of stronger preferences for time-intensive child quality, a higher marginal product of time spent on the production of child quality, and a higher marginal product of time relative to the marginal product of other inputs into the production of child quality.