Causal Inference and Models for The Probablity of Treatment Given The
Past

James Robins

ABSTRACT

   In this talk I explore a number of issues in causal inference
including sensitivity analysis for unmeasured confounding, bounded
outcomes, time dependent treatments, and direct and indirect effects
all from the perspective of inference based on modelling the
probability of treatment given the past. I show the various problems
that can arise with this approach and the even greater problems that
arise with competing approaches. Analyses of  an AIDS observational
study will be described.