In the model for causal inference based on counterfactuals or potential outcomes, the assumption is usually made that the outcome in one individual is independent of the treatment assignment and outcomes in other individuals. This is known as no interference between units or one aspect of the stable unit treatment value assumption (SUTVA). However in many situations, the outcome in any individual is dependent on the outcomes in others, called dependent happenings by Sir Ronald Ross (1916). A particular example is infectious diseases. Whether one person gets infected is quite often dependent on the treatment assignment and outcomes of other individuals. The dependence of the outcomes of the individuals poses difficult challenges for the counterfactual approach to causal inference. We consider this approach to causal inference on the example of a preventive intervention in infectious diseases, in particular vaccination. Different kinds of effects, such direct and indirect effects are defined in light of the counterfactual model. We consider various approaches to solving the problem, such as expanded representation of outcomes, conditioning on exposure to infection or modeling the dependence. None is entirely satisfactory (so far). The assignment mechanism can influence the sampling mechanism when it determines who is exposed to infection, raising further unsolved problems.Dr. Halloran is Professor of Biostatistics at Emory and an internationally-recognized expert on infectious diseases. A relevant reference is Halloran and Struchiner, Epidemiology (1995) 6:142-151.
Return to Biostatistics Working Group