|
Logistic Regression – questions for empirical linguistics| old_uid | 13311 |
|---|
| title | Logistic Regression – questions for empirical linguistics |
|---|
| start_date | 2014/01/23 |
|---|
| schedule | 10h-12h |
|---|
| online | no |
|---|
| location_info | Bât. G E27, salle Modal'X |
|---|
| details | The discussion will be in English or in French as appropriate. |
|---|
| summary | Inherited from sociolinguistics, logistic regression has represented a mainstay quantitative technique in corpus-driven Cognitive and Functional Linguistics since the turn of the century. In recent years, the simple (and sometimes overly simple) use of fixed-effects multiple binary logistic regression has come to an end and the linguistics community is beginning to question certain assumptions and seeking to improve upon the status quo.
Instead of presenting results, this seminar is intended as a discussion forum, based on a series of questions. We will consider three relatively theoretical questions that have been discussed informally within the community for a few years. The questions concern: random effects; rare events; and the assumption of the categorical outcome. If time permits, we will move to three practical questions that currently face the community. These concern: diagnostics for ordinal logistic regression; the concordance statistic for predictive accuracy; and immediately possible advances in the field (exact regression, Bayesian probability, stochastic boosting, random forests, and non-parametric regression). |
|---|
| responsibles | Parisse |
|---|
| |
|