Fig. 4: Association between Russia-Mueller coverage in the media and diversion, and association between diversion and subsequent Russia-Mueller coverage.
From: Using the president’s tweets to understand political diversion in the age of social media
The top row of panels shows the results for the NYT (panels (a–c)). The center (d–f) and bottom (g–i) row of panels show results for ABC News and the average of both media outlets, respectively. The left column of panels (a, d, g) shows results from two independent OLS models, the center column (b, e, h) is for a single 3SLS model in which suppression is modeled by relating yesterday’s tweets to today’s coverage, and the right column (c, f, i) is a 3SLS model in which suppression is modeled by relating today’s tweets to tomorrow’s coverage. In each panel, the axes show jittered t-values of the regression coefficients for diversion (X-axis) and suppression (Y-axis). Each point represents diversion and suppression for one pair of words in the Twitter vocabulary. Red vertical and horizontal lines denote significance thresholds (±1.96). Word pairs that are triggered by Mueller coverage (p < 0.05) and affect subsequent coverage (p < 0.05) are plotted in red. The gray contour lines in each panel show the distribution of points obtained if the timeline of tweets is randomized (red perimeter represents 95% cutoff, see “Methods”). The blue rugs represent univariate distributions.
