Fertility and Well-Being
At a Glance
Causal Inference Approaches to Fertility over the Life Course
Maarten Jacob Bijlsma, Mathias Lerch, Jessica Nisén; in Collaboration with Ben Wilson (Stockholm University, Sweden)
Demographers have shown considerable interest in understanding how socioeconomic processes collectively determine fertility. Most theories of fertility predict that a range of socioeconomic factors have an impact on the quantum and tempo of childbearing. In particular, a number of theories postulate that socioeconomic processes are interrelated and that they determine fertility collectively and simultaneously. However, it remains unclear (especially when examining the empirical literature) how different micro-level socioeconomic processes interact to collectively determine individual childbearing trajectories (tempo and quantum) and thus affect macro-level fertility. One reason for this lack of clarity is that traditional methods often struggle to disentangle path-dependent time-varying causal interrelationships between variables (e.g., socioeconomic factors).
We propose a new approach for studying the socioeconomic determinants of fertility, embedded in statistical theories of counterfactual causal inference. We focus in particular on mediation analysis with the parametric g-formula. The g-formula is a highly flexible statistical method specifically developed within counterfactual causal inference to account for the problematic aspects of causal interrelationships (selection bias, reverse causality, time-varying confounding, and intermediate confounding). It estimates effects specifically by generating counterfactual scenarios, which is in essence a “what-if” scenario such as: “What would women’s fertility trajectory look like if they had postponed childbearing by one year?” This method estimates the effects by using individual-level data, allowing for the control of various confounders and avoiding the ecological inference fallacy, but then allowing for individual-level effects to be generalized across heterogeneous populations to provide national-level effect estimates. Because it has these qualities, the g-formula seems to be a very promising technique for various demographic applications.
We used data from the 1970 British Birth Cohort to study the (near)-completed fertility trajectories of women. We tested the impact of four hypothetical scenarios on these trajectories: an increase in higher education attendance among women, a reduction in the preference for marriage, an increase in post-birth employment among women; and all women growing up as a single child (this provides insights into the intergenerational transfer of family-size preferences on fertility). We have found that all of these scenarios lead to lower fertility. The increased education scenario also shows some degree of catch-up of previously postponed births (though these had not yet been completely recuperated by age 38). The effect of the only-child scenario on fertility operates not just directly through child birth but also through changes in the mediating variables (on fertility) such as changes in employment, partnership, and education.
Another project that used the g-formula drew on Finnish register data to study the effect of parenthood postponement, to model male fertility, and to compare it with female fertility. We have found that a delay in parenthood increases educational enrollment and employment at young adult ages, and more so among women than among men. Later parenthood thus exacerbates the educational advantage of women and attenuates the income advantage of men. Furthermore, later parenthood strengthens the socioeconomic standing of both men and women when they become parents, essentially due to the accumulation of the above-mentioned effects.
Family Behavior, Fertility Development, Life Course, Statistics and Mathematics
Journal of the Royal Statistical Society/A 183:2, 493–513. (2020)
MPIDR Working Paper WP-2019-017. (2019)
MPIDR Working Paper WP-2017-013. (2017)