Choosing Between Causal Interpretations: An Experimental Study

Abstract

Good decision-making requires understanding the causal impact of our actions. Often, we only have access to correlational data that could stem from multiple causal mechanisms with divergent implications for choice. Our experiments comprehensively characterize choice when subjects face conflicting causal interpretations of such data. Behavior primarily reflects three types: following interpretations that make attractive promises, choosing cautiously, and assessing the fit of interpretations to the data. We characterize properties of interpretations that obscure bad fit to subjects. Preferences for more complex models are more common than those reflecting Occam’s razor. Implications extend to the Causal Narratives and Model Persuasion literatures.