Causal Effect Estimation Using Instrumental Variables with Count Exposure and Dichotomous Outcome

• Activity Number: 294
• Type: Contributed
• Date/Time: Tuesday, July 31, 2012 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Qing Liu*+ and Dylan Small
• Companies: Temple University and The Wharton School
• Keywords: instrumental variables ; Count Exposure ; Dichotomous Outcome ; causal effects

Abstract

In econometrics, epidemiology, statistics, and other related research areas, the method of instrumental variables (IV) has been extensively used to estimate causal effects when controlled experiments are not feasible. However, most research has focused on situations when the outcomes are continuous, while less work has been done with dichotomous outcomes. In this work, we consider dichotomous outcomes with a count exposure. We consider several approaches - the generalized structural mean model, 2-stage logistic regression model and the Poisson-lognormal model - and develop implementations of them in R. We compare the approaches through a simulation study. We apply the methods to study the effect of number of cigarettes smoked per day by a pregnant woman on her child's probability of having very low birth weight (< 1500 grams).