Font Size:
Log-mean linear models for causal inference
Last modified: 2017-05-23
Abstract
We discuss a log-mean linear regression approach to deal with causal inference when the interest is in assessing the effect of treatment on a set of multiple (binary) outcomes which might be not independent. We explore how the effect of treatment on joint outcomes can be decomposed considering the effect on single outcomes and the effect on their joint distribution. The method is illustrated through a randomized experiment concerning the effect of honey on nocturnal cough associated with childhood upper respiratory tract infections.