Uncovering Sources of Heterogeneity in the Effects of Maternal Smoking on Infants' Health at Birth

By combining machine learning and econometrics, we estimate the heterogeneous treatment effects using a causal forest, which we then decompose by counterfactual distributions to isolate differences in treatment effects that are only driven by a single variable, while keeping other characteristics comparable.

We find that advanced mother's age is a risk factor amplifying the effect of smoking on gestation length, birth weight, and Apgar score, a key measure of infants' health at birth. Weight-related features, such as high prepregnancy BMI and excessive weight gain, prove useful in mitigating the negative effects of smoking on birth weight but have no effect on Apgar score and mixed effects on gestation length.

These findings can be useful in identifying mothers most at need for intensified care and enhanced assistance with smoking cessation