Outcome: **summed badness** Identified 275 treatment users who retweeted our ads during the campaign. Can we find matching control users and then compare whether the treatment vs matched controls differ on the outcome? - greedy nearest neighbors (should be fine since we only have 275 treatment users and 40k control users?) # matching: all control users from the same block Each treatment user is matched with all control users from the same block. ![[_sum_winsorize-0.90-clustse 4.png]] # matching: time0 summed badness Each treatment user is matched with 10 control users with the most similar time0 summed badness. - 11 users (1 treatment + 10 control) per block (275 treatment users, 275 blocks) ![[_sum_winsorize-0.90-clustse 5.png]] # matching: time1 retweet count Each treatment user is matched with 10 control users with the most similar time1 retweet count. - 11 users (1 treatment + 10 control) per block (275 treatment users, 275 blocks) ![[_sum_winsorize-0.90-clustse 6.png]]