Is There a Reversal in the Educational Gradient in Second Birth Risks? a Cross-Country Comparison of Timing and Quantum Effects
Daniele Vignoli, University of Florence
Anna Matysiak, Wittgenstein Centre (IIASA, VID/ÖAW, WU), Vienna Institute of Demography/Austrian Academy of Sciences
This studycontributes to the discussion on the educational gradient in second birthsamong women who have not yet completed their reproductive careers.
Expansion of higher education among women is one of the importantfactors that contributed to a decline in fertility in the developed world (NiBhrolchain1986, Blossfeld and Huinink 1991, Basten et al. 2014). Highly educated womenhave less time to realise their fertility intentions as they spend more time ineducation (Ní Bhrolcháin and Beaujouan 2012) and right after finishingeducation postpone childbearing to establish their position in the labourmarket (Gustafsson 2001, Nicoletti and Tanturri 2008). Furthermore, women withuniversity degree face high opportunity costs of parenting (Mincer and Ofek1982, Ermisch 1990). Recently, it has been argued however, that highly educatedwomen are starting to have larger families than the low educated (e.g.Esping-Andersen and Billari 2015, Goldscheider et al. 2015) as womens economicresources increasingly improve conditions for family formation by increasingcouples financial security (e.g. Oppenheimer 1997). Furthermore, theincreasing welfare support for work and family reconciliation and a gradualincrease in mens involvement in childcare and housework make it easier forhighly educated women to reconcile paid work and care.
Studies on women who completed their reproductive careers donot provide support for the change in the educational gradient in fertility.Instead they show that highly educated women have lower fertility than the loweducated in most of the countries (Sobotka et al 2016) apart from some Nordiccountries where the educational differences in completed fertility have nearlydisappeared (see also Kravdal and Rindfuss 2008, Andersson et al. 2009).Micro-level studies, conducted on women who have not yet completed theirreproductive careers, report, in contrast, that highly educated women aremore likely to progress to the second child than the low educated (e.g. Olah2003, Vikat 2004, Kravdal 2007, Köppen 2006, Mathews and Sear 2013,Impicciatore and Zuanna 2016). These micro-level findings are sometimesinterpreted as a sign of the changes in the educational gradient in fertility.
This study demonstrates that the evidence provided by themicro-level studies does not imply a change in the educational gradient infertility. Event history models, commonly applied in micro-level research onwomens education and birth transitions, are designed for studying the effectsof covariates on the timing of events that will eventually occur. Neither thefirst nor the second births are such events. Ignoring this fact leads to a biasin the effect of the estimated coefficients on the timing of the event. Consequently,not only do event-history models not provide us with information on whether highlyeducated women are more likely to experience the birth, but they also givebiased information on the effect of the education on the timing of the secondbirth.
This study utilizes mixture cure models in order to investigate theeffects of womens education on (1) the probability of giving birth to thesecond child and (2) the timing of this event. These models were proposed as anextension of the conventional survival analysis for studying events that maynot occur to all individuals (McDonald and Rosina 2001). They allow forassessing the effect of explanatory covariates on the timing of the event andthe probability of its occurrence. They are well established in the medicalliterature, but have been rarely applied in family demography, although therehas been some exceptions (Gray et al 2011, Bremhorst et al 2016, Beaujouan andSolaz 2013, Locatelli et al. 2007, Rosina 2006 or Pinnelli et al. 2002),including the very recent study by Bremhorst et al (2016) looking at theeffects of womens education on birth risks.
In this study, we use parametric mixture cure models. We model theprobability of giving birth to the second child with the linear probabilitymodel and the time-to-the-second birth with the log-normal hazard model. Weadopt a comparative approach and study several countries. They offer differentfertility regimes, welfare states and cultural settings. To this end, we usedata from Harmonised Histories, which is a compilation of retrospectivedatabases for 14 European countries, US and Australia that provide fullfertility and partnership histories as well as information on the year ofobtaining highest education (Perelli-Harris et al 2010).
Our preliminary findings obtained for Italy provide rationale forusing mixture cure models for modelling the differential effects of womenseducation on second birth risks (Table 1). While standard log-normal hazardmodels indicate that highly educated women are more likely to give birth to thesecond child, the mixture models show that this finding is fully driven bytiming effects: Highly educated women have their second children more quicklyafter the first, but they are not more likely to give birth to the second childthan the low educated women.
Presented in Session 1159: Fertility