Much of contemporary social science is about constructing a counterfactual: what happens in the presence of a potential causal factor, and what happens to an arguably identical set of people (or cities or companies or schools) without it? Since people are never truly identical in characteristics or circumstances, any counterfactual can only be an approximation. RCTs can in theory provide a way around this– even if the individuals or groups participating in the trial are still all different from one another, differences (visible and invisible) are averaged out by the randomization process, to a degree. But aside from being formidably expensive and difficult to implement, RCTs can only address certain kinds of questions (you can’t get someone to randomly agree to get married, for example, though undoubtedly some people returning from Vegas felt that they did) and have their own analytic challenges. So researchers are left reaching for alternative methods of constructing counterfactuals, in particular by identifying allegedly quasi-random processes in the real world and using them as statistical instruments for the causal effect.
Almost any such identification strategy is susceptible to criticism. Any allegedly quasi-random processes may not be in practice all that random, and more importantly may not influence the variable of interest in exactly the way you think it does. Recently, a series of papers have attempted to use the timing of conception and birth to investigate the effects of intra-uterine environment on child outcomes.
Fetal Malnutrition And Academic Success: Evidence From Muslim Immigrants In Denmark — by Jane Greve, Marie Louise Schultz-Nielsen, Erdal Tekin
This paper examines the impact of potential fetal malnutrition on the academic proficiency of Muslim students in Denmark. We account for the endogeneity of fetal malnutrition by using the exposure to the month of Ramadan during time in utero as a natural experiment, under the assumption that some Muslim women might have fasted during Ramadan when they were pregnant. In some of our specifications, we use a sample of students from predominantly non-Muslim countries as an additional control group to address potential seasonality in cognitive outcomes in a difference-in-differences framework. Our outcome measures are the standardized test scores from the national exams on the subjects of Danish, English, Math, and Science administered by the Danish Ministry of Education. Our results indicate that fetal exposure to Ramadan has a negative impact on the achievement scores of Muslim students, especially females. Our analysis further reveals that most of these effects are concentrated on the children with low socioeconomic status (SES) background. These results indicate that fetal insults such as exposure to malnutrition may not only hamper the cognitive development of children subject to such conditions, but it may also complicate the efforts of policy-makers in improving the human capital, health, and labor market outcomes of low-SES individuals. Our findings highlight the importance of interventions designed to help economically disadvantaged women during pregnancy.
It is obvious at one level that what happens during fetal development is important to child outcomes. The current epidemic of the Zika virus in Latin America is a particularly tragic example of what can go wrong. But that does not mean that our presupposition should be that any change in maternal behavior or nutrition will have large effects, particularly on cognitive outcomes, which appear to be well-insulated against shocks. Thanks to the Dutch hunger winter of 44-45, the famine that took place in German-occupied but not Allied-occupied cities, we have several decades of research on what a much more plausibly exogenous decrease in calories did to the adult IQ of Dutch men and women who were intrauterine while their mothers were starving- approximately nothing ( http://www.lorentzcenter.nl/lc/web/2008/319/CD%20LORENTZ%20CENTER%20WORKSHOP/Dutch%20Recruits%20Mental%20Stein%20Science%201972.pdf http://m.ije.oxfordjournals.org/content/40/2/327.short ). On the other hand, schizophrenia, obesity, diabetes, and various neurodevelopmental disorders were more common in the starvation group. You are much more likely to influence those characteristics with early interventions than you are to influence IQ. So why do the researchers examining the effects of Ramadan on Danish Muslim children– and other studies using birth timing — find the opposite, that is, pervasive effects on later life outcomes from a reduction in calories or nutrients during pregnancy?
The first guess should be that birth timing is endogenous. At UW-Wisconsin up until the 80s, a story I was told went, if you asked a classroom of students how many of them were born in the Fall, most of the hands went up. Rural kids were born in the Fall because farmers had more time on their hands in the Winter. If you compared Fall births to non-Fall births for that population, you’d probably see large differences in educational attainment and later earnings, because you’d be comparing rural kids to urban kids. But it wouldn’t be intrauterine environment that was doing it: these were just different kids.
Similar issues could be present for a sample of Muslims in Denmark. For example, if agricultural or low-wage construction workers are away from their families during the summer and return to their families once the season is over, you’ll have an elevated number of fall conceptions (and summer births) who are the children of agricultural or low-wage construction workers and are predisposed towards lower socioeconomic outcomes as a result. Or it could be that observant Muslim woman planning to get pregnant might try to avoid being pregnant during Ramadan. The women who succeed in timing their pregnancy so that it does not overlap with fasting could be unusually smart or conscientious, and have children who are also smarter or more conscientious than average, but not primarily because of their mother’s diet during pregnancy.
This is made more plausible by the fact that the effects shown in the paper only emerge in the difference-in-differences specification. Look at Table 3- there’s no pattern of negative impacts at all in their Muslim sample. It’s only in their difference-in-difference specification (Table 4) that any pattern emerges. But so what? If reproductive timing is exogenous, then the effect should be identified within the Muslim population. The fact that the seasonality of births is different for their non-Muslim comparison population isn’t informative at all- you’re just adding noise to noise.
The key assumption for the whole paper is that seasonality, apart from Ramadan, operates similarly for Muslims and non-Muslims, and specifically for low-income Danish Muslims and non-Muslims. There is a long literature on seasonality in births that finds that that it is responsive to local culture and economics, and varies significantly in nearby populations: that is, to paraphrase William James– higenous, hogenous, birth timing’s endogenous.