Table 2.

Relationship between reported aspect of video analyzed and correctly identifying video authenticity, by video deepfake status.

	Adults		Principals		Teachers		Middle school students		CMU students
	Real	Deep fake	Real	Deep fake	Real	Deep fake	Real	Deep fake	Real	Deep fake
Familiar with person's views	0.381**	0.217 +	0.479**	0.223**	0.389**	0.101	0.409**	− 0.136	0.455*	0.044
	(0.089)	(0.118)	(0.069)	(0.082)	(0.071)	(0.085)	(0.067)	(0.084)	(0.199)	(0.192)
Video quality	− 0.128	0.251*	− 0.062	0.261**	− 0.134*	0.210**	0.020	0.085 +	− 0.059	− 0.124
	(0.081)	(0.097)	(0.049)	(0.058)	(0.058)	(0.059)	(0.049)	(0.052)	(0.147)	(0.139)
Background	− 0.135	0.088	− 0.075	0.083	0.010	− 0.023	0.003	0.126*	− 0.135	− 0.075
	(0.125)	(0.102)	(0.053)	(0.070)	(0.062)	(0.076)	(0.046)	(0.055)	(0.152)	(0.156)
Facial features	− 0.033	0.152 +	0.063	0.154**	0.011	0.133*	0.090	0.197**	− 0.011	− 0.154
	(0.069)	(0.082)	(0.047)	(0.057)	(0.058)	(0.061)	(0.056)	(0.055)	(0.163)	(0.134)
Audio	0.222**	− 0.001	0.222**	− 0.085	0.240**	− 0.102	0.133*	− 0.110*	0.152	− 0.074
	(0.076)	(0.084)	(0.052)	(0.058)	(0.060)	(0.065)	(0.054)	(0.051)	(0.163)	(0.123)
Credibility of content	0.157*	− 0.043	0.179**	− 0.073	0.064	− 0.091	0.093 +	− 0.084	0.214	− 0.065
	(0.063)	(0.074)	(0.050)	(0.057)	(0.057)	(0.056)	(0.056)	(0.067)	(0.131)	(0.153)
N (Respondent-by-video)	1600	1444	1514	1446	1283	1285	1600	1620	169	179
N (Respondent)	727	693	692	685	604	596	757	755	82	84

Each column presents the results of a separate regression of video authenticity on aspects respondents reported analyzing when making their decisions. All models include speaker, video order, and respondent fixed effects. Regressions models used to analyze adult, principal, and teacher panels are weighted to retain national representation. Standard errors are clustered by respondent.

 + indicates p < 0.10; *p < 0.05; **p < 0.01.