andybega / isa-2018 Goto Github PK

Given country to country differences in the quality of torture allegations, try to control for:

• press freedom
• number of human rights NGOs
• political and societal globalization
• year dummies to account for increase in data quality over time and/or decrease in torture over time

Take out:

number of excluded groups
vdem suffrage
vdem elected officials
vdem clean elections

Does it make a difference to the results?

Danny, 4 Feb. 2019:

I want to add a discussion of the cross validation and the out of sample fit plot. Is there any way to label the models in the plot so we know which models are beating the baseline model? Also looks like one of the marginalized models ties (or comes close) XGBoost, which is interesting. No rush on this. We probably won't hear anything for at least one month.

Danny, 6 Feb.:

it would be nice to have labels for models that outperform the RE/controls only models, so we know which variables are appreciably improving fit. Once we have that we can add a paragraph for the CV stuff.

Andy, 7 Feb.:

I need to re-run the xgboost model again, but for now, how’s this for the OOS plot?

Danny, 7 Feb.:

Very nice! Compact and clear. Provides a bit more support for our expectations about criminal suspects and marginalized people, I think.

Andy, 12 Feb.:

Hello, I re-ran xgboost and updated things a few days ago, should be all good to go from my end (I think).

Danny, 15 Feb.:

I'll let you know when I add discussion of the OOS results.