Extended Data Fig. 5: Replicates Fig. 5 but with adjustment covariates.
From: Post-January 6th deplatforming reduced the reach of misinformation on Twitter
Corresponding regression tables are Supplementary Information Tables 1 to 3. Two-way fixed effect point estimates (dots) and 95% confidence intervals (bars) of the difference-in-differences for high, moderate, and low activity users, as well as all these levels combined (x-axis). P-values (stars) are from two-sided t-tests based on ordinary least squares estimates with clustered standard errors at user-level. Estimates compare followers (treated group) and not-followers (reference group) of deplatformed users after January 12th (post-treatment period) and before January 6th (pre-treatment period). No multiple test correction was used. See Supplementary Information Tables 1–3 for exact values with all activity level users combined. Total sample sizes of not-followers (reference) and Trump-only followers: combined: 306,089, high: 53,962, moderate: 219,375, low: 32,003; Followers: combined: 662,216, high: 156,941, moderate: 449,560, low: 53,442; Followers (4+): combined: 463,176, high: 115,264, moderate: 302,907, low: 43,218.