Model: quality_time1 ~ condition x quality_time0 - each panel is one type of tweet - each row is one measure of domain quality: see [[different measures of domain quality]] - haven't accounted for blocks (models take forever to fit when I try to include block as dummy) - dealing with missing values - time0 (pre-campaign period): replace NAs with mean of *all users* during pre-campaign period - time1 (campaign period): replace NAs with mean of *control users* during campaign period # Interaction effect - Oct 4 to Oct 24 ![[coef_conditiont_combine-interact_Oct04-Oct24.png]] # Interaction effect - Oct 11 to Oct 24 ![[coef_conditiont_combine-interact_Oct11-Oct24.png]] ## Probe `quality04_tweet` ```r > summaryh(m) term results 1: (Intercept) b = 54.37, SE = 0.06, t(32884) = 956.93, p < .001, r = 0.98 2: time0C b = 0.42, SE = 0.005, t(32884) = 85.23, p < .001, r = 0.42 3: conditionC b = −0.05, SE = 0.11, t(32884) = −0.45, p = .649, r = 0 4: time0C:conditionC b = −0.03, SE = 0.01, t(32884) = −3.49, p < .001, r = −0.02 SIMPLE SLOPES ANALYSIS Slope of conditionC when time0C = -1.141909e+01 (- 1 SD): (users who share lower-quality stuff during pre-campaign) Est. S.E. t val. p ------ ------ -------- ------ 0.35 0.16 2.15 0.03 Slope of conditionC when time0C = 5.294287e-16 (Mean): Est. S.E. t val. p ------- ------ -------- ------ -0.05 0.11 -0.45 0.65 Slope of conditionC when time0C = 1.141909e+01 (+ 1 SD): (users who share higher-quality stuff during pre-campaign) Est. S.E. t val. p ------- ------ -------- ------ -0.45 0.16 -2.79 0.01 ``` ![[coef_conditiont_combine-interact_Oct11-Oct24_quality04_probe.png|600]] ![[coef_conditiont_combine-interact_Oct11-Oct24_quality04_probe2.png|600]]