In Western countries, rates of second and third births typically increase with educational attainment, a feature that usually disappears if unobserved heterogeneity is brought into the event-history analysis. By contrast, in a country like Romania, second and third birth rates have been found to decline when moving across groups with increasing education, and the decline becomes greater if unobserved heterogeneity is added to the analysis. The present paper demonstrates this pattern, and shows that, because this feature is retained in the presence of control variables, such as age at first birth and period effects, the selectivity is not produced by a failure to account for the control variables.
(ProQuest: ... denotes formulae omitted.)
In a tradition going back to the seminal book by Gary Becker (1981), economists typically hypothesize that fertility will decrease as women's educational levels increase, because childbearing intensities are dominated by differential opportunity costs. A long string of demographic contributions have shown the opposite pattern for countries in the West, namely, that in the populations studied in recent years, more highly educated women have had higher fertility at parities above zero than women with less education (Kravdal 2001, Hoem et al. 2001, Kreyenfeld 2002, Oláh 2003, Kreyenfeld and Zabel 2005, Köppen 2006). The explanation often given for this positive educational gradient at positive parities found in Western data is that the more highly educated women who become mothers constitute a select group of particularly childbearing-prone3 women, a development which should result in a reduction in the educational gradient at parities above zero when unobserved heterogeneity is brought into the picture. In fact, the gradient regularly disappears, i.e., it is essentially reduced to zero.
By contrast, a negative educational gradient has been found (as predicted by Becker's theory) for some countries in Eastern Europe. (For Albania, see Gjonca et al. 2008, Fig. 4; for the Ukraine, see Perelli-Harris 2008, Figure 5; for an investigation with particularly clear results, see Koytcheva 2006, who studied patterns for first and second births in Bulgaria in her Chapter 6.) In the present paper, we use event-history analysis with a number of control variables to show that negative educational gradients in fertility are present for Romania for all birth orders one through three. We suggest that Romanian educational gradients may be negative precisely for the reasons given in the Beckerian theory (i.e., opportunity cost considerations and quantity-quality trade-off), as we have little reason to suspect that a strong income effect will be found, especially during the socialist period, when education did not play any important role in wage differentials. We speculate that the rapid increase in income dispersion during the subsequent market-oriented period produced a growth in the return to education, which was likely to increase the opportunity costs of childbearing, especially for better educated women. In addition, we suggest that women with higher levels of education must have been more successful in avoiding the coercive measures of the pronatalist policies in the socialist period (simply because they were more talented in adapting to the circumstances), and that (for the same reason) they benefited to a greater extent when more childbearing-friendly conditions appeared. The lower childbearing intensities among women with higher levels of education appear to have persisted over the entire period of investigation. In Romania, the educational gradients become even more strongly negative when we control for unobserved personality characteristics. We suggest that the explanation may be the same as for Western countries: Women who become mothers are positively selected for childbearing-proneness, and, even though in Romania this selection is not strong enough to overcome other factors and produce a positive educational gradient, the gradient becomes smaller (in this case, more negative) when we account for unobserved heterogeneity. …