TABLE A3.
Regression discontinuity results, MTurk and Unacast Samples
| Panel 1: Mturk data | |||||
|---|---|---|---|---|---|
| Dependent variable | |||||
| Left home, non‐ess. | Work outside home | Work outside home, by choice | Wash hands | Stay home | |
| Opposing | 0.042 | 0.118** | 0.090*** | −0.049*** | −0.002 |
| (0.062) | (0.055) | (0.033) | (0.018) | (0.026) | |
| N | 1753 | 932 | 201 | 1753 | 1753 |
| Dependent variable | |||||
|---|---|---|---|---|---|
| Cancel travel | Limit contact | Wear PPE | Other | Number of behaviors | |
| Opposing | −0.076* | −0.144*** | −0.036 | 0.014** | −0.333*** |
| (0.045) | (0.021) | (0.051) | (0.005) | (0.083) | |
| N | 1753 | 1753 | 1753 | 1753 | 1753 |
| Panel 2: Unacast data | |||||
|---|---|---|---|---|---|
| Dependent variable | |||||
| % Change, daily distance | % Change, non‐ess. visits | ||||
| Opposing | −0.350 | 5.591* | |||
| (2.310) | (3.385) | ||||
| N | 1675 | 1132 | |||
Note: This table shows regression discontinuity estimates of the effect of an opposing party on preventative behavior, using the MTurk and Unacast data. The running variable is opposing party share, which by construction is symmetric and there shows no bunching at the cutoff. We implement these regressions using the Stata package rdrobust (Calonico et al., 2014), which optimally selects the bandwidth. There are no controls in the regression. Standard errors are clustered at the state by party level.
***p < 0.01, **p < 0.05, *p < 0.1.