Skip to main content
. 2020 Nov 10;11:5764. doi: 10.1038/s41467-020-19644-6

Table 2.

Three-stage least squares (3SLS) models to predict diversionary tweets (CJI) from threatening media coverage (Russia-Mueller; columns 1–3) and predicting suppression from diversionary tweets (columns 4–6) simultaneously.

Dependent variable CJI tweetst NYTa ABCa Averageb
(1) (2) (3) (4) (5) (6)
Panel A: yesterday’s tweets predict today’s coveragec
 NYT Russia/Muellert 0.018
(0.006) p = 0.007
 ABC Russia/Muellert 0.225
(0.190) p = 0.236
 Average Russia/Muellert 0.293
(0.111) p = 0.009
 CJI tweetst−1 o o o −0.399 −0.038 −0.050
(0.207) p = 0.054 (0.018) p = 0.037 (0.019) p = 0.010
 N 703 710 703 703 710 703
Panel B: today’s tweets predict tomorrow’s coveragec
 NYT Russia/Muellert 0.012
(0.005) p = 0.009
 ABC Russia/Muellert 0.220
(0.068) p = 0.001
 Average Russia/Muellert 0.228
(0.060) p = 0.000
 CJI tweetst o o o −0.732 −0.090 −0.094
(0.620) p = 0.238 (0.054) p = 0.095 (0.057) p = 0.099
 N 702 709 702 702 709 702

aCoverage of Russia-Mueller on day t (panel A) or day t + 1 (panel B).

bAverage of the standardized values of coverage of NYT and ABC.

cEach model included control variables for each week during the sampling period and long-term time trends as well as the appropriate number of lagged observations for the dependent variable (see “Methods”). Table entries are coefficients (standard errors).

oLagged predictor is shown only when of interest.