Table 1. Meta-analysis of average treatment effects of advertisements on target candidate favorability and vote choice.

Observations are CATE estimates for each advertisement, conditional on subject partisanship and battleground residency. The signs of the outcomes are scaled with respect to the valence of the advertisement: Higher values indicate that promotional advertisements had positive effects on target candidate favorability or vote choice and that attack advertisements had negative effects. All meta-regressors have been demeaned so the intercept always refers to the estimate of the average treatment effect, but the coefficients still refer to the average difference in the effectiveness of the advertisement associated with a unit change in the regressor relative to the omitted category.

	Candidate favorability		Vote choice
Average effect	0.056*	0.062*	0.007	0.008
	(0.020)	(0.020)	(0.007)	(0.007)
Democratic respondent (versus Republican)	0.035	0.022	0.011	0.006
	(0.035)	(0.036)	(0.010)	(0.011)
Independent respondent (versus Republican)	0.023	0.015	0.009	0.007
	(0.051)	(0.052)	(0.020)	(0.020)
Battleground state (versus non-battleground)	−0.00	−0.007	−0.017	−0.017
	(0.033)	(0.033)	(0.010)	(0.010)
PAC sponsor (versus campaign sponsor)	−0.012	0.026	−0.023	−0.016
	(0.043)	(0.047)	(0.013)	(0.014)
Time (scaled in months)	−0.023	0.005	−0.009*	−0.008
	(0.014)	(0.010)	(0.004)	(0.004)
Attack advertisement (versus promotional advertisement)	−0.017		0.028
	(0.046)		(0.016)
General election (versus primary election)	0.123
	(0.067)
Pro-Trump advertisement (versus pro-Clinton advertisement)		−0.124		−0.016
		(0.101)		(0.034)
Anti-Clinton advertisement (versus pro-Clinton advertisement)		−0.105		0.012
		(0.070)		(0.023)
Anti-Trump advertisement (versus pro-Clinton advertisement)		−0.041		0.026
		(0.058)		(0.021)
Pro-Sanders advertisement (versus pro-Clinton advertisement)		−0.075
		(0.089)
Pro-Cruz advertisement (versus pro-Clinton advertisement)		0.047
		(0.116)
Pro-Kasich advertisement (versus pro-Clinton advertisement)		−0.182
		(0.145)
Number of observations	354	354	204	204

*P < 0.05.