Identification of the Effects of Dynamic Treatments by Sequential Conditional Independence Assumptions

Abstract

This paper approaches the causal analysis of sequences of interventions from a potential outcome perspective. The identifying power of several different assumptions concerning the connection between the dynamic selection process and the outcomes of different sequences is discussed. The assumptions invoke different randomisation assumptions which are compatible with different selection regimes. Parametric forms are not involved. When participation in the sequences is decided every period depending on its success so far, the resulting endogeneity problem destroys nonparametric identification for many parameters of interest. However, some interesting dynamic forms of the average treatment effect are identified. As an empirical example for the application of this approach, we reexamine the effects of training programmes for the unemployed in West Germany.

Keywords: Dynamic treatment regimes, nonparametric identification, causal effects, sequential randomisation, programme evaluation, treatment effects, dynamic matching, panel data

JEL Classification: C21, C31