The Effect of Maternal Drug Use on Birth Weight : Measurement Error in Binary Variables

This paper develops a method to correct for non-random measurement error in a binary indicator of illicit drugs. Our results suggest that estimates of the effect of self reported prenatal drug use on birth weight are biased upwards by measurement error -- a finding contrary to predictions of a model of random measurement error. We show that more accurate estimates of the true effect of drug use on birth weight can be obtained by using the predicted probability of falsely reporting drug use. This suggests that out-of-sample information on drug use may improve estimates of the effect of reported drug use in other settings