Continuous partisanship scores are very left-skewed because majority are republicans, so I binarized scores to code for Republican or not - `repub` variable - if score > 0, repub (0.5) - if score <= 0, not repub (-0.5) - Users without party scores assigned -0.5 # exp 1 pilot - repubs shared more bad stuff? - negative condition-repub: treatment was more effective for repub? ![[count-winsorize-0.95-clustse 16.png]] count of good tweets (ignore title - should be count goodness): no. of non-bad domains shared, among domains we have ratings for ![[count-winsorize-0.95-clustse 25.png]] ![[frac-winsorize-0.95-clustse 14.png]] ![[sum-winsorize-0.95-clustse 17.png]] # exp 2 big one - weird results ![[count-winsorize-0.95-clustse 19.png]] count of good tweets (ignore title - should be count goodness): no. of non-bad domains shared, among domains we have ratings for ![[count-winsorize-0.95-clustse 28.png]] ![[frac-winsorize-0.95-clustse 17.png]] ![[sum-winsorize-0.95-clustse 20.png]] # exp 3 canadian trucker - weird results as well - maybe because users are Canadian ![[count-winsorize-0.95-clustse 20.png]] count of good tweets (ignore title - should be count goodness): no. of non-bad domains shared, among domains we have ratings for ![[count-winsorize-0.95-clustse 30.png]] ![[frac-winsorize-0.95-clustse 18.png]] ![[sum-winsorize-0.95-clustse 21.png]] # exp 4 deepstate - similar to exp 1 - repub shared more bad stuff - no condition-repub interaction ![[count-winsorize-0.95-clustse 21.png]] count of good tweets (ignore title - should be count goodness): no. of non-bad domains shared, among domains we have ratings for ![[count-winsorize-0.95-clustse 32.png]] ![[frac-winsorize-0.95-clustse 19.png]] ![[sum-winsorize-0.95-clustse 22.png]]