Theory Development

14. Problems with Casual Models

Outside of textbook examples matters are rarely so neat (Freedman, 2006), and there is a large literature on these issues and on the question of the extent to which questions of causal interpretation can be resolved on statistical grounds alone, or on the basis of very small and unproblematic kinds of prior causal knowledge, such as the knowledge that certain variables cannot be the cause of others.

 One can see the issues in the Abrams case: he knew what to partial for out of more general knowledge of banking and economics and about the effects of race. The question is whether the role of this kind of knowledge can be minimized and replaced by purely statistical considerations, or whether the statistical assumptions necessary to employ these considerations smuggle in casual considerations (Freedman, 1997).

To put it differently, the logic of these methods is eliminative. The issue is whether one can eliminate enough to arrive at casual conclusions without significant non-statistical information.

Judea Pearl has argued that the attempt to use statistical methods alone to deal with confounding comes close to success, but fails. He proposes a solution for the problem that employs statistical considerations, but adds a non-statistical one, that of considering the most stable causes, those whose correlations persist, regardless of the other variables in the setting, to be the genuine causes in cases of confounding (Pearl, 2000:268).

Freedman, D. A. (2006)  Statistical models: Theory and practice. Cambridge: Cambridge University Press: 2-13.
Freedman, D. A. (1997) From association to causation via regression. In V. R. McKim and S. P. Turner (Eds.) Causality in crisis?: Statistical methods and the search for causal knowledge in the social sciences. Notre Dame, IN: University of Notre Dame Press.
Pearl, J. (2000) Causality: Models, reasoning, and inference. Cambridge: Cambridge University Press.