Using microsimulation, administrative data, and supercomputers to realistically model fertility behaviour: the case of fertility preferences and childlessness

Education is a strong driver of whether and when women become mothers. Many different and contradicting mechanisms have been proposed to explain why highly educated women are more likely to remain childless and become mothers at higher ages than less educated women. The demands on data to disentangle these mechanisms are extraordinary, and no dataset exists that allows for this. Microsimulation models can help in this situation by explicitly modelling the mechanisms and comparing the outcomes of the models to real-world outcomes. The simulation models presented here simulate fertility outcomes over the life courses of agents based on behavioural factors, such preferences and partnership trajectories, and biological factors that determine the ability to have children, such as the age at sterility, fecundability, and intrauterine mortality. To parametrise the models, we use administrative data from Social Statistics Netherlands, survey data from the LISS panel for the behavioural factors, and findings from reproductive medicine for the biological parameters. To estimate unknown parameters in the models (for which no data are available), we use Approximate Bayesian Computation. This is computationally rather demanding which is why we have used supercomputers. Our simulations reproduce the pattern of unintended childlessness strongly varying across women with different educational levels. Despite higher educated women preferring to have children at a later age, our simulations showed that these preferences hardly played a role in explaining childlessness. The higher age at cohabitation was the main explanation for the higher unintended childlessness among highly educated women. We discuss how these models can be used to explain the surprising reversal in gradient between education and fertility that is observed in Scandinavian countries. We end by discussing the advantages and drawbacks of our simulation approach and how it can contribute to family sociology

Summary

     ODISSEI conference

     Utrecht, the Netherlands

     Click here for website

     Download materials here

     There is a video!

Description

Education is a strong driver of whether and when women become mothers. Many different and contradicting mechanisms have been proposed to explain why highly educated women are more likely to remain childless and become mothers at higher ages than less educated women. The demands on data to disentangle these mechanisms are extraordinary, and no dataset exists that allows for this. Microsimulation models can help in this situation by explicitly modelling the mechanisms and comparing the outcomes of the models to real-world outcomes. The simulation models presented here simulate fertility outcomes over the life courses of agents based on behavioural factors, such preferences and partnership trajectories, and biological factors that determine the ability to have children, such as the age at sterility, fecundability, and intrauterine mortality. To parametrise the models, we use administrative data from Social Statistics Netherlands, survey data from the LISS panel for the behavioural factors, and findings from reproductive medicine for the biological parameters. To estimate unknown parameters in the models (for which no data are available), we use Approximate Bayesian Computation. This is computationally rather demanding which is why we have used supercomputers. Our simulations reproduce the pattern of unintended childlessness strongly varying across women with different educational levels. Despite higher educated women preferring to have children at a later age, our simulations showed that these preferences hardly played a role in explaining childlessness. The higher age at cohabitation was the main explanation for the higher unintended childlessness among highly educated women. We discuss how these models can be used to explain the surprising reversal in gradient between education and fertility that is observed in Scandinavian countries. We end by discussing the advantages and drawbacks of our simulation approach and how it can contribute to family sociology