|
Distributional Reaction Time Data Analysis: Using Process Models with Tractable Complexity| old_uid | 15411 |
|---|
| title | Distributional Reaction Time Data Analysis: Using Process Models with Tractable Complexity |
|---|
| start_date | 2015/03/27 |
|---|
| schedule | 11h-12h |
|---|
| online | no |
|---|
| summary | The distributions of reaction time (RT) data are notoriously right-skewed, and classical descriptive statistics such as the mean and standard deviation, which correspond to a Gaussian distributional analysis of the data, provide a mis-matched distributional model. This incongruency can be problematic; and an appropriate distributional analysis approach can more effectively reveal experiment level differences. In this talk I will review important aspects of distributional RT analysis; and discuss a range of measurement models with tractable complexity that allow an appropriate quantification of the RT data. There are a number of measurement models that while accounting for the RT distribution, simultaneously model an underlying process leading to the observed response; and can thus provide a cognitive model for the experimental task. I will highlight my research with such a lesser-known, simple process measurement model, the shifted Wald distribution, which is similar to the Drift Diffusion model, but may be more easily generalizable to a number of experimental paradigms for RT data (due to its simplicity) where the other model is impractical. Such types of models with tractable complexity (easily measurable by data), such as these that model signal accumulation, will be contrasted and compared to models of less-tractable complexity (difficult to measure by data), such as neural network models, that pose a challenge to be truly validated on real data, despite being more informative of a richer underlying process. |
|---|
| responsibles | Pélissier |
|---|
| |
|