• No results found

2.6 Mechanisms at work

2.6.3 Sample selection effects

Last but not least, it could be that sample selection effects drive the esti- mate to zero. Section 2.4.2 discussed ‘culling’ of the weakest, which might occur when late-life outcomes are studied. Specifically, the individuals who are observed in the data are hypothesized to have better characteristics than those who are not. This is particularly important as the individuals who were most negatively affected/selected, may die prematurely and may thereby be no longer observed in the data. To illustrate, Van Ewijk and Lindeboom (2017) find no long-term health effects of prenatal exposure to WWII, if anything health is better, which they explain among other things with selective mortality. This section addresses potential selectivity into the sample.

First, parents with different characteristics may decide to conceive in different times, but it could also be that different parents are capable of con- ceiving in different times. Section 2.2.5 established that parental selection is not driven by available care. However, parental selection may be affected by mortality of men and women in childbearing age during WWII. If men and women who died are different from those who did not, children born in 1946 might be born to an even more selective set of parents. Figure 2.A7 shows mortality in the Netherlands from 1936 to 2015. Mortality during the war is especially high for individuals aged 45 to 80. In the group of men and women in childbearing age (aged 15 to 45) a peak of 7 deaths per 1000 inhabitants is observed in 1945. Mortality is even lower in other war

2.6. MECHANISMS AT WORK 67

years, and although there is no information available on selective mortality, the low mortality in the relevant age group suggests that it will only play a minor role. Second, there could be selection at birth. That is, conditional on getting pregnant there may be differences in which women brought their pregnancies to term. Induced abortion was not legal in 1946, implying that (legal) selective abortions will not affect which children are born. Likewise there are no trend-deviations in the number of stillbirths and perinatal deaths for 1946 (Figure 2.A6), and the sex ratio is more skewed to boys in 1946.68 Unfortunately, there is no data available on these outcomes on the

regional level.

The main analysis uses the restricted sample in which more informa- tion on family characteristics is available. Individuals are observed in the restricted sample if they themselves and their mother are still alive and registered in a Dutch municipality by 1995. The restricted sample con- tains 166, 475 individuals born in 1946, which is 58.5% of the total amount of births. The differences between the unrestricted and restricted sample were formalized in Section 2.4.2 and two processes underlie the survival from the unrestricted to the restricted sample: (1) individuals with a lower quality mother are less likely to be observed, and (2) individuals whose mother is older at giving birth have a lower probability of being observed. Table 2.11 shows the results from the difference-in-difference models for the unrestricted sample with an indicator for observing the mother in the restricted sample as the relevant outcome. As predicted, a higher maternal age at birth is associated with a lower probability of observing the moth-

68According to the fragile male hypothesis this does not raise concerns on bad pregnan-

cy circumstances Figure 2.A8. Bethmann and Kvasnicka (2014) argues that in many European countries, including the Netherlands, the sex ratios during and right after WWII were skewed towards boys.

er. The double-difference estimator, that captures the effect of being born March-September 1946 in the north, is not significantly different from zero for both the pre- and post-sample. CoBP are not more or less likely to be observed, which suggests that both processes outlined in Section 2.4.2 bal- ance each other. In line with this finding, when estimating the difference- in-difference model for the unrestricted sample (see Table 2.12) a similar pattern arises, suggesting that there are no large differences between the unrestricted and restricted sample.

The total number of births was 284, 456 in 1946, and 242, 196 of those births are observed in the unrestricted sample (alive and registered in a Dutch municipality by 1995), which is about 85.1% of the total number of births. The discrepancy can arise from both deaths and migration. It may be that children who are most negatively affected die before reaching adulthood, and thereby are not observed in the data. Lindeboom and Van Ewijk (2015) find that lower survival probabilities until age 55 for those born directly after the war. However, this relationship disappears after conditioning on survival up to age one and five. That lower survival prob- abilities are probably caused by higher child mortality, is consistent with Figure 2.A6 that shows that mortality after birth is higher for the 1946 cohort when comparing to later cohorts. Rau et al. (2017) argue that an increase in unanticipated conceptions caused by a large price increase in the price of contraceptives led to higher infant mortality, which could be co- herent with negative parental selection. Similarly, Gruber et al. (1999) find that access to abortion, a measure that can prevent unanticipated births is associated with lower infant mortality. Unfortunately such potential effects earlier in the life-cycle cannot be studied with the current data.