Research Seminar
Theory testing with the prior predictiveWolf VanpaemelKU Leuven | |
| Abstract: | Existing model selection methods do not consider whether a good fit is meaningful. In this talk, I propose the prior predictive test, which is based on the prior predictive distribution. Upon observing data, three situations can occur. First, if the observed data are not among the central predictions of the model, the model is invalidated. Second, if the observed data are among the central predictions of the model, and a good fit is meaningful, the model is supported. Third, if the observed data are among the central predictions of the model, and a good fit is not meaningful, the model is neither supported nor invalidated. A good fit is considered meaningful and impressive if plausible outcomes exist that are not among the central predictions of the model. The prior predictive test does not only take the meaningfulness of a good fit into account, but, unlike most existing model selection methods, it is also sensitive to the prior. An application example focusing on category learning demonstrates the potential of the prior predictive test for testing psychological models. |
| Date: | Tue Oct 25, 12:00 pm - 1:00 pm |
| Place: | room 01.07 (Department of Psychology, Tiensestraat 102, 3000 Leuven) |