Causality, confounding, and control
Michel Mouchart, Université Catholique de Louvain
Federica Russo, University of Kent
Guillaume Wunsch, Université Catholique de Louvain
In demography and in epidemiology for example, it is recommended to control for confounding variables which “screen off” (Reichenbach, 1956) the true relation between a possible cause and its effects. This paper analyses what is actually meant by the “screening-off” effect in philosophy of science and, following the work of Judea Pearl (2001) among others, improve upon the usual definition of the confounding effect, from a causal or structural modelling approach. The paper also examines exogeneity and its implications for causality in a structural model, putting the emphasis on the causal structure and on stability rather than on exogeneity per se. Finally, again in a structural modelling approach, it will lead to practical recommendations on when to control, and especially on when not to control, for possible confounders.
|
