. 2023 Apr 5;119:102815. doi: 10.1016/j.jeem.2023.102815

Table 5.

Wind and PM 2.5 IV — Confirmed COVID-19 cases and deaths across 1, 3, 7, 10, and 14 day periods .

	(1)	(2)	(3)	(4)	(5)
	Same-day	3-day total	7-day total	10-day total	14-day total
Case incidence per 100,000
Weighted PM 2.5	0.146**	0.351**	0.393	0.656	0.316
	(0.0684)	(0.158)	(0.332)	(0.457)	(0.579)
F Stat	14.28	14.28	14.28	14.28	14.28
Dep Var Mean	7.875	23.787	56.541	81.911	116.792
Pct change mean	1.849	1.475	0.694	0.801	0.270
Controls	✓	✓	✓	✓	✓
County & State-by-week FEs	✓	✓	✓	✓	✓
Observations	97,885	97,885	97,885	97,885	97,885
Death incidence per 100,000
Weighted PM 2.5	0.00918	0.00785	0.0307**	0.0153	0.0141
	(0.00571)	(0.00874)	(0.0139)	(0.0168)	(0.0212)
F Stat	14.28	14.28	14.28	14.28	14.28
Dep Var Mean	0.216	0.649	1.537	2.218	3.150
Pct change mean	4.256	1.209	1.998	0.688	0.448
Controls	✓	✓	✓	✓	✓
County & State-by-week FEs	✓	✓	✓	✓	✓
Observations	97,885	97,885	97,885	97,885	97,885

$^{*}$ $p < 0.1$ , $^{* *}$ $p < 0.05$ , $^{* * *}$ $p < 0.01$ . Standard errors clustered at the county level in parentheses. Controls include state-level and county-level policy adoption, wind speed, minimum and maximum daily temperature, precipitation, prior two-week social distancing behavior, day-of-week, and two lagged wind direction-by-monitor cluster interactions. Displayed output of a two-stage least squares regression model with county and state-by-week fixed effects in which wind direction and air quality monitor cluster interactions are used to predict PM 2.5 levels in a county on a given day. The outcomes in column (1) are same-day incidence of cases or deaths. Outcomes in column (2) are incidence of cases or deaths over the same day and the following two days. Outcomes in column (3) are deaths over the same day and the following nine days. Outcomes in column (4) are incidence of cases or deaths over the same day and the following thirteen days. All regressions use same-day predicted fine particulate matter as dependent variable. F-statistics of the relevance test are computed assuming first-stage standard errors are not serially correlated.