Model with most significant interaction - outcome: afm_bias - thresholds: 20, 80 - problem with this approach: different no. of domains in the bad/good bins - no. of bad domains: 4 - no. of good domains: 133 ```r # conditionC: -0.5, 0.5 (control, treatment) # qualityC: -0.5, 0.5 (bad, good) > m <- feglm(t1 ~ conditionC * qualityC | block, dt1, family = "quasipoisson", vcov = ~ screen_name) NOTE: 266 fixed-effects (2,910 observations) removed because of only 0 outcomes. > m GLM estimation, family = quasipoisson, Dep. Var.: t1 Observations: 62,866 Fixed-effects: block: 5,158 Standard-errors: Clustered (screen_name) Estimate Std. Error t value Pr(>|t|) conditionC 0.077447 0.039036 1.98401 0.0472631 * # treatment group shared more stuff??? qualityC 3.501977 0.038196 91.68348 < 2.2e-16 *** # because more good domains than bad domains conditionC:qualityC -0.200948 0.076393 -2.63046 0.0085312 ** # interaction ``` Re-parameterized ```r # quality1: bad coded as 0, good coded as 1 GLM estimation, family = quasipoisson, Dep. Var.: t1 Observations: 62,866 Fixed-effects: block: 5,158 Standard-errors: Clustered (screen_name) Estimate Std. Error t value Pr(>|t|) conditionC 0.177921 0.076283 2.33238 0.0196871 * # treatment effect for bad sources quality1 3.501977 0.038196 91.68348 < 2.2e-16 *** conditionC:quality1 -0.200948 0.076393 -2.63046 0.0085312 ** # quality2: bad coded as 1, good coded as 0 GLM estimation, family = quasipoisson, Dep. Var.: t1 Observations: 62,866 Fixed-effects: block: 5,158 Standard-errors: Clustered (screen_name) Estimate Std. Error t value Pr(>|t|) conditionC -0.023027 0.012099 -1.90328 0.0570136 . # treatment effect for good sources quality2 -3.501977 0.038196 -91.68348 < 2.2e-16 *** conditionC:quality2 0.200948 0.076393 2.63046 0.0085312 ** ```