If not now, when? The timing of childbirth and labor market outcomes

Abstract

Summary We study the effect of childbirth and birth timing on female labor market outcomes in Italy. The impact is traced up to 21 years since school completion by estimating a factor analytic model with dynamic selection into treatments. We find that childbirth, especially the first delivery, negatively affects female earnings and participation. Women having their first child soon after school exit catch up with childless women after 12–15 years. The negative consequences are smaller if the first child is delayed up to 7–9 years after school completion or if a second childbirth occurs within 6 years since school exit.

Off to a Bad Start: Youth Nonemployment and Labor Market Outcomes Later in Life

“…To account for time-varying unobserved heterogeneity, we set up a factor analytic model (Carneiro et al, 2003;Heckman and Navarro, 2007;Fruehwirth et al, 2016;Cockx et al, 2019;Picchio et al, 2021). 5 The unobserved terms of the equations of the outcomes and the treatment intensity are composed of a latent factor θ, which collects the timevarying unobserved differences among individuals and error terms that are conditionally independent given the latent factor:…”

“…Unobserved heterogeneity varies over time because of the factor distribution and a linear combination of the latent factor with time-varying coefficients α j t , the so-called factor loadings. 6 As in Picchio et al (2021), we adopt a one loading factor specification, i.e. we allow only for a single-dimensional time-varying unobserved determinant of the treatment intensity and of the outcomes.…”

“…Our identification approach of the causal effect fits into the factor-analytic dynamic models (FADM) (Carneiro et al, 2003;Heckman and Navarro, 2007), which has been more recently exploited by Fruehwirth et al (2016) and Cockx et al (2019) to study the impact of grade retention on subsequent school performances or by Picchio et al (2021) to investigate the effect of fertility on subsequent labor market outcomes. Our model is a simplified version of those in Fruehwirth et al (2016) and Picchio et al (2021). Compared to the former, we do not integrate essential heterogeneity (Heckman et al, 2006), i.e.…”

“…1 Compared to the latter, our treatment is unique and not multiple, being the fraction of time spent in nonemployment in the first three years after school completion. As such, the assumptions in Fruehwirth et al (2016) or Picchio et al (2021) are sufficient to attain the nonparametric identification of the treatment effect in our framework. The nonparametric identification is based on three main ingredients.…”

“…Second, since we can rebuild all the working history, we can take advantage of the longitudinal information in our data and observe multiple realizations over time of the endogenous variables. Third, as in Picchio et al (2021), we exploit measures of the latent factor which are free of selection into treatment (Carneiro et al, 2003), like the work experience before school completion and the number of siblings when the individual was 14. Because these measures are realized before the treatment occurs, i.e.…”