MZES Social Science Data Lab: Models all the way down

Abstract: The Neyman-Rubin causal model characterizes how, through experimental (or quasi-experimental) manipulation of an intervention, researchers can make data-informed counterfactual claims about what would happen in the absence of that intervention. The Neyman-Rubin causal model is, nevertheless, just that: a model. In this talk, I will present some excerpts from a larger book project in which my collaborators and I describe the connections between the Neyman-Rubin causal model, the basic estimands of randomized control trials targeting respondents’ preferences, and the theoretical object that is traditionally described as a preference. After a brief reminder of the basic structure of the Neyman-Rubin causal model, I will explain how this framework has been applied to preference elicitation experiments. Then, I will proceed to show that, although this gives us well-defined counterfactuals, the corresponding causal quantities do not straightforwardly represent preferences, either at the individual level or in the aggregate. Finally, I will present a model-based alternative for preference elicitation, with a hands-on application using replication data from a survey experiment.

This hybrid event will take place in room A231 (A5, 6) and will be live-streamed on Zoom from 13:45-15:15. If you plan on attending remotely, please register for the live stream.