Maternal Smoking, Misclassification, and Infant Health

Identifying the causal effect of prenatal maternal smoking on infant health is complicated by unobservable maternal characteristics and behaviors which are plausibly related to both birth outcomes and a mother’s propensity to smoke. Previous studies have addressed the omitted variables problem using instrumental variables (IV) techniques. However, with self-reported data on maternal smoking, misreporting can induce more severe biases in IV estimates than in OLS. This paper proposes an approach based on parametric methods for misclassified binary dependent variables that simultaneously addresses the endogeneity and measurement error problems. The relationship between infant health and maternal smoking is then re-examined using US Birth Records. I find that roughly 30% of smoking mothers are misclassified as non-smokers. As a result, conventional IV estimates deliver implausibly large birth weight losses (upwards of 1,000 grams among African Americans). Accounting for misclassification yields estimates that are considerably smaller in magnitude and more consistent with experimental evidence