Research Seminar
Model selection for qualitative data patternsSara SteegenKU Leuven | |
| Abstract: | Most model selection methods that are currently used in psychology focus on the correspondence between quantitative aspects in the data and quantitative model output. Most psychological data, however, tend to be noisy, which implies that not the quantitative minutiae of the data but rather the qualitative patterns across conditions are often of interest. Model selection methods based on qualitative data patterns are lacking. In this talk, a criterion is proposed that selects between models by concentrating on the fit of qualitative patterns in the data. This criterion is based on Parameter Space Partitioning (PSP; Pitt, Kim, Navarro, & Myung, 2006), a recently developed method that partitions a model’s parameter space into different regions that correspond to different qualitative data patterns. The proposed criterion evaluates models by assessing the fit between an empirical data pattern and a model’s partitioned parameter space, thereby balancing goodness of fit and complexity. Focusing on category learning models, the criterion will be evaluated in a model recovery study. We will explore the conditions under which the criterion can be usefully applied and propose measures that can be used to indicate whether these conditions are met. |
| Date: | Tue May 24, 1:15 pm - 2:00 pm |
| Place: | room 00.98 (Department of Psychology, Tiensestraat 102, 3000 Leuven) |