Fit logistic regression models predicting probability of revert including features for:
- is newcomer
- ORES is deployed
- is anon
- ... (all the other features we have in the models)
To do this we have to:
- Create automated badwords lists for the large-enough wikis that don't adopt ores for rcfilters.
- Build datasets including the ores features for each wiki that did or didn't adopt oress for rcfilters.
- Add features for newcomer, ores being deployed, and is_anon
- fit diff-in-diff models combining these features.