Table 2—

Treatment Effect Estimates

	Rating sender	Rating signal	Flu vaccine intent	COVID-19 vaccine intent
	(1)	(2)	(3)	(4)
Panel A. Layperson versus expert: Black respondents
Layperson treat	−0.540	−0.081	0.019	0.088
	(0.071)	(0.067)	(0.025)	(0.030)
	[0.000]	[0.231]	[0.455]	[0.003]
Mean in control	0.00	0.00	0.37	0.43
Observations	845	845	845	592
Panel B. Standard versus acknowledgement signal: Black respondents
Acknowledgement signal treat	0.100	0.142	0.027	0.054
	(0.068)	(0.069)	(0.025)	(0.031)
	[0.145]	[0.040]	[0.287]	[0.080]
Mean in control	0.00	0.00	0.34	0.40
Observations	827	827	827	581
p-value	0.000	0.022	0.819	0.433
Panel C. Concordant versus discordant expert sender: Black respondents
Concordance treat	0.183	0.139	0.026	0.035
	(0.067)	(0.070)	(0.025)	(0.031)
	[0.007]	[0.049]	[0.302]	[0.254]
Mean in control	0.00	0.00	0.34	0.40
Observations	832	832	832	587
Panel D. Concordant versus discordant expert sender: White respondents
Concordance treat	−0.075	−0.009	0.003	0.009
	(0.057)	(0.057)	(0.021)	(0.025)
	[0.189]	[0.876]	[0.868]	[0.719]
Mean in control	0.00	0.00	0.37	0.45
Observations	1,221	1,221	1,221	866
p-value	0.004	0.100	0.487	0.512

Notes: Table reports OLS estimates. Each dependent variable in columns 1 and 2 is an inverse-covariance-weighted index as described in Anderson (2008) and standardized to a mean of 0 and standard deviation of 1. Dependent variables in columns 3 and 4 are on a scale of 0 to 1. COVID-19 vaccine intent was asked during the 2020–2021 flu season only. Outcome variables are described in Section II and in online Appendix Section E. The p-value in panel B tests the null hypothesis that the acknowledgement signal treatment and layperson treatment effects are equal. The p-value in panel D tests the null hypothesis that the concordance treatment effects are the same across Black and White respondents. Stratifying variables (platform and season) are included as controls in the regression but not reported. Robust standard errors are in parentheses. p-values are in brackets.