- follow-up to [[220301_145044 user ECDFs#fact-checker CDFs 60 domains|fact-checker CDF pointwise proportion tests]] ## Threshold linear regressions (control for blocking) For each quality value (`q: [0...99]`; x-axis), we fit a model (`y ~ condition[-0.5/0.5] * mean_t0C | block`) where we assign the user's outcome `y = 100` if the user's mean quality during the campaign is above `q` (otherwise `y = 0`). - x-axis: each threshold value (one model per value) - y-axis: regression coefficients estimate (95% CI) - panels: different coefficients in the model ```r # for each quality thresold q in 0 to 99 dt1[, y := ifelse(mean_t1 > q, 100, 0)] m_temp <- feols(y ~ conditionC * mean_t0C | block, dt1[domain_type == "overall"]) ``` - first panel (**condition coefficient**): When `q` is between 15 and 45, condition coefficient is positive: Being in the treatment (vs. control) group makes it more likely for users to be belong to the "higher-quality" (`y = 100`) group. - second panel (**condition:mean_t0 interaction**): Condition coefficient depends on users' pre-campaign quality. The coefficient is more positive for those whose pre-campaign quality is lower (again, consistent with what we have found so far! see [[220303_135802 fact-checker analyses|fact-checker analyses]]). ![[dv_fc_thresholds_interact 5.png|800]]