Identification of Effects of Dynamic Treatments with a Difference-in-Differences Approach

This paper examines the power of a conditional difference-in-differences approach to nonparametrically identify the causal effects of sequences of interventions. In the classical difference-in-differences case, a period previous to the implementation of the intervention is used as a comparison period to get rid of common trends for treated and nontreated. When sequences of programs are studied, two different utilizations of the available information are possible to construct the comparison period. The first one only uses the information available in the period previous to the first possible intervention. The second one uses the information sequentially. Here, we use the information available between the interventions in addition to the period previous to the first intervention. Furthermore, we investigate if we still can achieve identification when participation to the sequences is decided sequentially on the intermediate results. Identification can be obtained in each case for all parameters of interest.