This paper considers identification of treatment effects when the outcome variables and covari-ates are not observed in the same data sets. Ecological inference models, where aggregate out-come information is combined with ...