Days with hotter temperatures are becoming more and more common, as temperatures as well as the likelihood for extreme weather increase due to climate change. Hot temperatures affect human health and cognition negatively when the exposure is contemporaneous—students perform worse on tests and elderly die from it. It also turns out that exposure to heat can have consequences far later in time, as long as the exposure occurs during pregnancy or early childhood. Health, the risk of pre-term births, and educational attainment are all affected by prenatal exposure to heat.
In my Master’s thesis Frying the Brain: The Effect of Prenatal Exposure to Heat on Fetal Brain Development and Cognitive Abilities, I expand on the previous research on prenatal heat exposure by investigating the link between exposure to heat and cognitive abilities. In this post, I’ll explain the findings, without too many technical details.
Causal Theory
In short, without referencing a bunch of previous literature, my hypothesis is that prenatal heat exposure affects fetal brain development and, by extension, cognitive abilities through three main potential mechanisms.
(1) A direct effect on cell proliferation and migration in the brain of the fetus,
(2) a direct effect on prenatal stress,
and (3) an indirect effect on the behavior or income of the mother leading to either prenatal stress or nutritional deficiencies.
These paths all have support in the literature. The direct effect has been tested on animals, such as mice and guinea pigs—though one should be careful about generalizing too much from these studies. The prenatal stress effect has also been established on mice and in humans, although not with the direct connection to cognitive abilities later-in-life. Nutritional deficiencies and their effect on fetal brain development are also well studied, it seems like zinc, choline, and iron are among the nutrients that are important here.
These paths are summarized in the below picture, where a blue arrow indicates a direct path and a red arrow indicates an indirect path.

Setting
To understand this link, I use data from the Indonesian Family Life Surveys—a set of longitudinal studies collecting data on a wide range of behaviors and outcomes. Most importantly, they administer a couple of tests of cognitive ability
Fluid intelligence: The respondents answer Raven’s progressive matrices, which tests pattern-matching ability with as few cultural influences as possible.
Immediate recall: A list of 10 words is read to each respondent and the respondent tries to repeat back as many of them as possible.
Delayed recall: After completing another module of the survey, the respondent again tries to recall as many words as possible.
These are the main outcome variables I use in the paper. I combine this individual-level data with weather data from APHRODITE, which allows me to count the number of days in each temperature bin each individual is exposed to. The bins are defined as follows: Bin 1: <21.5°C, Bin 2: 21.5-23.5°C, Bin 3: 23.5-25.5°C, Bin 4: 25.5-27.5°C, Bin 5: 27.5-29.5°C, Bin 6: >29.5°C.
I also split the pregnancy into trimesters, to be able to test if the timing of the exposure matters. I do the same thing for another critical period of fetal brain development—corticogenesis—a period that happens approximately between week 7 and 18 of gestation.
Results
The main results, using the full set of bins during the entire pregnancy are shown in the below figure. They show that heat exposure is highly non-linear, we only find negative effects for the bin with the hottest temperatures. Further, only immediate and delayed recall seem to be affected negatively. The effect on fluid intelligence is small and not significant. This suggests that if there is interference with brain development, there’s only interference in very specific parts of the brain.

Further analysis on the trimester timing shows no conclusive evidence of exposure during one trimester being worse than any other. But, extending that analysis to corticogenesis shows that exposure to heat during corticogenesis makes the negative effects much larger. Compared to being exposed outside of corticogenesis, the effect is 4 times as large. The point estimate suggests that an additional week of exposure of days above 29.5°C leads to a 0.375 standard deviation decline in performance on immediate recall.

This set of results provides evidence in favor of an effect on fetal brain development. But they do not say much about the potential mechanisms.
Potential Mechanisms
To deepen the understanding of the results, I try to provide evidence of the potential mechanisms. I do this by using employment data from censuses and intra-censuses performed every 5 years in Indonesia. This allows me to construct variables of employment shares in agriculture, manufacturing, and services across time.
First, I look at if urban and rural populations are affected differently by heat. They are. Rural populations experience no effect at all. This is quite surprising, and contrary to previous research. One plausible explanation may be that heat interacts with pollution in urban areas, which creates a doubly negative situation.
It is better to look into why these populations differ through the data though. I do that by using the employment rates. It turns out that a higher employment rate in agriculture has a positive effect on the effect of prenatal heat exposure. Plausibly due to farmers getting higher incomes when it is warmer, due to their crops producing higher yields. Indonesia’s main crops are corn and rice—two crops that are quite good at resisting heat. On the other hand, a higher employment rate in manufacturing is associated with an increased negative effect. Perhaps due to it being hard to avoid heat while working in manufacturing. However, it is also true that manufacturing is more common in urban areas—thus this effect may only be due to urban areas being different.
I therefore complement the employment analysis by adding another level—the rural-urban distinction. The results are presented in the below figure.

The pattern is hard to miss, the employment effects are one or two orders of magnitude larger in rural areas. This suggests that rural populations are able to counteract or mitigate the negative effects of prenatal heat exposure through higher incomes. The large negative effects on manufacturing thus can be explained through manufacturing workers becoming poorer relatively to their peers in agriculture. These effects are negligible in urban areas suggesting that they experience the direct effect of prenatal heat without mediating factors.
Though these results are indicative, they should be interpreted with caution as I loose quite a lot of power in the final estimations, due to the census and the IFLS not covering the exact same populations. None of the estimated coefficients in the final picture are significant.
Conclusion
The picture that emerges is at the same time discouraging and encouraging. Already poor countries are those that are most affected by climate change—at the same time, I show that climate change and hotter temperatures affect one of the key ingredients to improve the situation for these countries; cognitive abilities and human capital formation. The encouraging part is that it seems possible to mitigate the effects, suggesting that policy and strategy can eliminate this negative effect.
If you want to read the full paper, then you can download it below.