Fig. 3.

Predictive margins: effect of “stay home” tweet by Trump vote share and governor party. Figure shows predicted values and 95% confidence intervals from a county-level regression of median time at home on the treatment indicator, its interaction with Donald Trump’s county-level vote share in the 2016 presidential election, county and day fixed effects, as well as day fixed effects interacted with control variables and Trump’s 2016 margin (see SI Appendix, Table SI-3, Panels A and B). The treatment is an indicator variable equaling 1 for all days after a governor issues their first recommendation encouraging citizens to stay at home. We estimate the model separately for states with Democratic (A) and Republican (B) governors. The fitted line shows the linear marginal effect of the treatment at different levels of Trump vote share for each state type. The points with 95% confidence intervals show semiparametric estimates of the marginal effect of the treatment at five different bins of Trump vote share. Bins are (−1, −0.25), (0.25, 0), (0, 0.25), and (0.25, 0.5). The histogram below the predicted margins displays the density of the county-level Trump vote margin by treatment status (red is treated, gray is untreated). Figure uses the INTERFLEX package (20).