Abstract
This study examined the relationship between self-enhancement and third-person perception. It also investigated the behavioral consequences of third-person perception within a theory of reasoned action framework. A survey on the issue of Internet pornography was administered to 462 undergraduate students. A positive relationship was found between self-enhancement and third-person perception. Behavioral attitude emerged as a key mediator in the relationship between third-person perception and intention to support Internet censorship. Subjective norm overall was not an important factor in the perception–intention relationship. The lack of impact for subjective norm, however, had causes that varied across gender.
Notes
1The original correlation reported in the study was .24. This is indeed a negative correlation because the optimistic bias measure and the third-person effect measure were not consistently valenced. For the sake of clarity, we have changed the sign of the correlation here.
2The author did not interpret his findings as positive evidence for the optimistic bias theory, however. When the author formulated his hypotheses, he predicted that optimistic bias and third-person perception would be positively correlated. This prediction is problematic because it failed to take into consideration the desirability of the messages.
3One may argue that such correspondence is in fact rather superficial. Getting AIDS is a very different issue from being affected by safe-sex media messages; being struck by Y2K problems, likewise, bears little similarity to being influenced by news about Y2K.
Note. N male = 148. N female = 308. When α is set at .05, correlations larger than .16 are statistically significant for male participants; correlations larger than .11 are statistically significant for female participants.
4β represents standardized regression coefficient.
5A multiple-group analysis typically involves three sequential steps of model estimation. The first step seeks to establish a baseline model across groups without imposing any cross-group constraints. The second step constrains the regression weights in the measurement part of the model to be equal across groups and then re-estimate the model to see if model fit decreases significantly compared to the baseline model. The third step further constrains the regression weights in the structural part of the model to be equal across groups. This model is then compared against the model established in the second step to see if model fit decreases significantly. Our interest is in the third step, but this step is only meaningful when (a) the baseline model is established in Step 1 and (b) the measurement regression weights are found to be group-invariant in Step 2. Both of these conditions were satisfied in our analysis. Our predictive model, when unconstrained, provided adequate fit to the pooled data, χ2(122, N = 457) = 237.70, p < .001, χ2/df = 1.95, TLI = .94, CFI = .96, RMSEA = .05. The factor loadings in our model as a set also did not vary significantly across the two gender groups, Δχ2(8, N = 457) = 14.8, p > .05.