Table 2.

Regression discontinuity estimates of changes in commercial electricity usage (log) due to COVID-19 mandates: global polynomial results

	(1)	(2)	(3)	(4)
	Arizona (Global)	Illinois (Global)	Arizona (Local)	Illinois (Local)
	School closure	School closure	School closure	School closure
COVID-19 Mandates
	−0.050 ∗∗∗	−0.068 ∗∗∗	−0.027 ∗∗∗	−0.088 ∗∗∗
	(0.002)	(0.002)	(0.002)	(0.002)
Weather-related variables	Yes	Yes	Yes	yes
Month FE	Yes	No	Yes	No
Day-of-week FE	Yes	yes	Yes	yes
Holiday FE	Yes	No	Yes	No
Hourly FE	Yes	yes	Yes	yes
Account FE	Yes	yes	Yes	yes
Observations	61,442,445	30,283,001	9,280,270	81,743,262
Number of businesses	14,271	40,757	14,097	40,757

Notes: 1. weather-related control variables include temperature (in a restricted cubic spline format), precipitation (linear and quadratic format), air pressure, relative humidity, and wind speed.

Standard errors, clustered by account id, are in parentheses.

2. Columns (3) and (4) are local linear approaches to validate the results from the polynomial approach. In this setting, a narrower bandwidth of 15 days before and after the policy effective dates (for Arizona) and 4 days (for Illinois) are adopted. The reason that we adopted different bandwidths for Arizona and Illinois is because of the different periods between the school closure day and stay-at-home order date. In Arizona, it is 15 days, and in IL, it is four days. That's why we use different bandwidths for the local linear regression because we do not want to include the impact of the stay-at-home order in the evaluation of the impact of the school-closure date. Additionally, we conducted robustness checks with different bandwidths of the local RD models in Table S11.

∗∗∗ Significant at the 1% level. ∗∗ Significant at the 5% level. ∗ Significant at the 10% level.