Understanding Person Perceptions: Comparing Four Common Statistical Approaches to Third-Person Research

Abstract

This article addresses whether current methods of measurement are sufficient to reflect all nuances of the third-person effect. We do so by content analyzing all major third-person effect (3PE) articles, assessing measurement and analysis approaches employed by 3PE researchers. We revisit data from two published studies, comparing analytical models that emerged as commonly used by 3PE researchers (i.e., the standard subtractive measure, Whitt's Diamond model, first-and third-person estimates entered separately, and the subtractive measure with self-estimates as a control). Ultimately our analysis helps us Mike Schmierbach (Ph.D., University of Wisconsin) is an assistant professor of media studies in the College of Communications at Pennsylvania State University. His research interests include investigating the role media play in prompting issue interest and engagement and exploring the effects of non-narrative aspects of video games, particularly multiplayer elements. Michael P. Boyle (Ph.D., University of Wisconsin-Madison) is an assistant professor in Communication Studies at West Chester University. His research interests include news treatment of social protest and effects of news content on issue interest and political participation. Douglas M. McLeod 492understand past research and makes suggestions for research approaches in the future. First, researchers need to more carefully explore the role of firstand third-person perceptions in the behavioral hypothesis. Second, using only one analytical model does not tell the complete story. To that end, we propose a multimodel approach to analyses in third-person research be applied to extant and future work in this area.

Do difference scores make a difference on the third-person effect?

“…When a positive correlation components is present, a modest positive correlation (as a rule of thumb, less than .5 but greater than .1) is less harmful to the reliability of the difference score. Schmierbach et al (2008) once expressed their uncertainty about "whether the diamond model…conclusively addresses the potential problem that a significant subtractive term might really reflect variance due entirely to perceived effects on self or other in isolation." The present paper has discovered the problems associated with this model.…”

“…A number of past studies (Edwards, 2002;Gunther, 1998;Salwen, 1998) have specifically questioned the use of difference scores to test the perceived media effects on behavioral outcomes. Schmierbach et al (2008) held that the subtractive measurement of difference scores might mask the real influence on support for censorship and that it potentially confounds relative difference with the absolute level of the variables in question. Lo and Wei (2002) also suggested that the inconsistency was largely due to the inappropriate use of perceptual bias as a predictor.…”

“…Schmierbach et al (2008) attributed them to methodological inconsistencies, and they reported four methods commonly used by researchers into the TPE (i.e., the standard subtractive measurement, the diamond model, first-and third-person estimates entered separately, and the subtractive measurement, with self-estimates as a control). Specifically, 57 of 73 articles adopted a subtractive measurement (other minus self) or the simple difference score.…”

“…If the other component is also added, the equation after expansion will be: Peter et al (1993), Schmierbach, Boyle, and McLeod (2008), and Eeckhaut, Van de Putte, Gerris, and Vermulst (2013) considered Equation 9 to be more troublesome because multicollinearity among the terms can produce unstable parameter estimates and misleading results. The more dramatic finding of Hendrickx, De Graaf, Lammers, and Ultee (1993) with regard to Equation 9 was that the difference score is a linear transformation of the two components, and such a model would therefore be unidentifiable.…”

“…The diamond model is, unfortunately, problematic because Equation 10 explains exactly the same amount of variance of y as Equation 11 (Hendrickx et al, 1993;Schmierbach et al, 2008 …”