The Prevention Effect of Third-Person Perception: A Study on the Perceived and Actual Influence of Polls

Abstract

This study ties the third-person effect phenomenon to social comparison research by positing that a downward comparison is made when people regard “others” as more influenced than themselves by persuasive messages. A likely consequence of the downward comparison is prevention behavior, which refers to the attempt to avoid undesirable outcomes. Hence we can expect a negative relationship between perceived influence of messages on others and actual influence of the messages on self. This negative relationship should be particularly likely to exist among less efficacious individuals. Utilizing a survey experiment (N = 800), which examined both the perceived and actual influence of opinion poll findings on people's issue opinions, this study shows that the prevention effect does not exist across the board, but there is enough evidence showing its existence among less efficacious people on certain issues.

Notes

¹For example, in Lockwood's (2002) study, students were either exposed to a target of downward comparison who experienced difficulties in adjusting to university life, or to no comparison target. In the analysis, Lockwood examined only the difference between the conditions. There was no attempt to calculate the actual distance between the comparison target and the student participants' evaluations of their own ease of adjusting to university life. Of course, in the experiment, the participants' evaluations of their own ease of adjusting to university life were controlled through randomization. The analysis, in other words, involved an examination of how the comparison target's ease of adjustment would affect an individual when the individual's own ease of adjustment was controlled.

²The lower age limit was set at 15 to include teenagers who, despite not having the right to vote, may have clear political judgments and can participate in non-institutionalized political activities such as protests. People older than 70 often have difficulties in completing a telephone interview. Setting an age ceiling can enhance the overall reliability of the results.

³For the four issues, the actual figures used were (a) minimum wage legislation: 60% vs. 20%, (b) rehabilitation of June: 55% vs. 20%, (c) satisfaction with press freedom: 55% vs. 10%, and (d) trust in Chinese government: 45% vs. 20%.

Note. Entries in the first two columns are mean scores on a 0-to-10 scale, with higher values representing stronger perceived influence. Paired-samples t tests were used to examine the significance of the differences between perceived influence on self and perceived influence on others.

***p < .001.

Note. Four demographics are controlled in the analysis. Cell entries are mean scores on a 5-point Likert scale derived from the multivariate analysis of variance with the demographic variables set at sex = 1.56, age = 4.76, education = 4.38, and income = 2.97. Numbers in parentheses are the standard errors of the mean scores. In the same row, entries sharing the same subscript differ from each other significantly at p < .05 in post hoc multiple independent samples t tests.

*p < .05.

Note. Entries are unstandardized coefficients. Numbers in parenthese are standard errors. The full model includes four demographics, attention to polls, media use, and self-efficacy. Coding of poll results: majority only or the majority-and-minority condition = 1, no results = 0. The main effects variables are centered around their respective means when constructing the interaction terms. Missing cases are deleted pairwise.

*p < .05. **p < .01. ***p < .001.

Note. Entries are unstandardized regression coefficients, and parenthetical numbers of standard errors. The full model is the one in Table 3 without self-efficacy. Pairs of coefficients sharing the same subscript differ from each other significantly at p < .05. Missing cases are deleted pairwise.

*p < .05. **p < .01. ***p < .001.

⁴In paired-samples t tests, the mean score of perceived influence of polls on self is lower than the mean scores of perceived influence of television on self, t(733) = 14.92, p < .001; perceived influence of newspaper on self, t(731) = 12.11, p < .001; and perceived influence of government on self, t(739) = 7.06, p < .001. At the same time, the mean score of perceived influence of polls on others is also lower than those of the three agents, t(720, 705, 702) = 15.00, 15.91, 5.78, For television, newspaper, and government, respectively (p < .001 in all cases).

Additional information

Notes on contributors

Francis L. F. Lee

Francis L. F. Lee (Ph.D., Stanford University, 2003) is Associate Professor at Chinese University of Hong Kong. His research interests include political communication, journalism studies, public opinion research, and media economics.