Table 4. Robustness tests.
| PropScore Matching | Exclude Lombardy | Exclude Touristic | Shorter Window | Arrival Time | No Lag | System GMM | Placebo | |||
|---|---|---|---|---|---|---|---|---|---|---|
| (1) | (2) | (3) | (4) | (5) | (6) | (7) | (8) | (9) | (10) | |
| d11t × Shutd11m | -0.0300*** | -0.0242*** | -0.320*** | -0.0337*** | -0.316*** | -0.0375*** | -0.0410*** | -0.0154 | -0.00020 | -0.00713 |
| (-3.26) | (-2.87) | (-3.69) | (-4.28) | (-3.31) | (-4.65) | (-4.54) | (-1.33) | (-0.02) | (0.49) | |
| d25t × Shutd25m | -0.0066 | 0.0075* | -0.00564 | -0.00613 | -0.0055 | -0.0169*** | 0.00051 | 0.0098 | ||
| (-0.97) | (1.72) | (-0.98) | (-1.15) | (-1.09) | (-2.88) | (0.10) | (1.50) | |||
| ym,t-1 | 0.0283*** | 0.00225 | 0.0387*** | 0.0225** | 0.0026 | 0.186*** | -0.0378*** | 0.364*** | 0.0112 | |
| (3.24) | (0.45) | (3.92) | (2.09) | (0.27) | (2.53) | (-3.19) | (3.97) | (0.75) | ||
| Interact. Controls | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ |
| Municipality FE | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | x | ✓ | ✓ | ✓ |
| ArrD × Time FE | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | x | ✓ | ✓ | ✓ |
| Observations | 61,885 | 79,281 | 79,722 | 85,520 | 70,608 | 102,320 | 100,657 | 48,957 | 93,410 | 7,039 |
| R2 | 0.188 | 0.141 | 0.341 | 0.300 | 0.264 | 0.268 | x | 0.291 | 0.278 | 0.273 |
| Adj. R2 | 0.184 | 0.101 | 0.245 | 0.213 | 0.258 | 0.227 | x | 0.289 | 0.236 | 0.268 |
The table presents robustness checks of our baseline specification (last column of Table 3). First column reports OLS estimates on a propensity score matched sample. The second and third column exclude the Lombardy region and winter-touristic areas, respectively. In column (4), the sample period is shortened to end on April 4th 2020. In column (5) the definition of “anomaly” in the virus arrival time definition is set to 2 (instead of 1) standard deviations of cumulative excess mortality rate. In column (6) the lagged dependent variable is excluded, while column (7) shows the results of the system GMM estimator, collapsing the instruments matrix (lag2–lag4) and using the two-step technique. Lastly, column (8) shifts the first policy treatment date backwards by ten days, column (9) swaps policy time dummies, and column (10) considers only municipalities in which the virus never circulated during our sample period. Interactions of d11t with PopDensm, IncIneqm, and IntMobm are included but suppressed for brevity. t statistics in parentheses. Standard Errors clustered at municipality- and day-level.
*, ** and *** represent statistical significance at the 10%, 5% and, 1% level, respectively. Variable definitions are in S1 Table.