Multivariate Analyses

As a first step in examining the issues of biasing effects of negative terminology and biased processing of nonnegative outcome information, two MANOVAs were performed, one on subjects' responses to the first seven items-i.e., those concerned with perceptions of the specific cases of sexual contacts between the adolescents and men described in the case studies - and the other on subjects' responses to the next seven items - i.e., those concerned with perceptions of general cases of sexual contacts between male adolescents and men.

Because the condensed articles concerned sexual relationships between men, it was of interest to examine gender differences in subjects' responses. Hence, the MANOVAs were performed on data from a 2 (Gender) X 2 (Terminology: neutral or negative) X 2 (Information: descriptive-only or descriptive plus long-term out- come) between-subjects factorial design.

The MANOVAs revealed no main effects or interactions involving the gender factor. Hence, the data were collapsed across this factor, and the MANOVAs were performed using the remaining two factors.

For the MANOVA for the specific items, a main effect occurred for terminology, Wilks' lambda = .83, F(7, 70) = 2.27, p < .04, and for information,

Wilks' lambda = .65, F(7, 70) = 5.32, p < .001.

The terminology X information interaction was nonsignificant,

Wilks' lambda = .92, F(7, 70) = .91, p > .10.

For the MANOVA for the general items, a main effect occurred for terminology,

Wi1ks' lambda = .71, F(7, 70} = 4.13. p < .001.

No main effect for information occurred,

Wilks' lambda = .87, F(7, 70) = 1.43, p > .10,

and the terminology X information interaction was also nonsignificant,

Wilks' lambda = .93, F(7, 70) = .79,p > .10.

Univariate Analyses

To determine the direction of the terminology and information effects, a series of univariate analyses (ANOVAs) involving the first 14 measures was performed. Because these analyses revealed no significant terminology X information interactions, the statistical results concerning the main effects for these two factors were considered separately, allowing for a clearer examination of the two experimental hypotheses.

Table 1 presents the results for terminology. The major hypothesis in this experiment was that the use of negative terminology would negatively bias judgments and perceptions. Support for this hypothesis was provided by subjects' judgments and perceptions of general cases of sexual encounters between male adolescents and male adults.

Significance was reached for six of the seven analyses. In the last column of the table, effect sizes are provided. The measure of effect size chosen was the product-moment r, a measure that has been used with focused tests (e.g., Fs with 1 df in the numerator) with increasing frequency because of its advantages over more traditional measures (Rosnow & Rosenthal, 1988).

Positive effect sizes indicate that the means were in the predicted direction, and negative effect sizes indicate the opposite. Effect sizes of .10, .30, and .50 can be interpreted as indicating small, medium, and large effects, respectively (Cohen, 1988).

For the general items, the effect sizes were all positive, and the average effect size was moderate (r = .31).

Although all pairs of means were in the predicted direction for the specific items - all effect sizes were positive - only one of the seven analyses yielded a significant main effect for terminology. Four of the remaining six main effects were marginally significant, however. The average effect size for these items (r = .20) was smaller than that for the general items.

These results indicate that negative terminology had consistent biasing effects on subjects' perceptions of general cases, but less reliable and weaker effects on their perceptions of the specific cases.

The second hypothesis was that nonnegative outcome information would be processed in a negatively biased manner. Relevant to this hypothesis are the results of the ANOVAs involving the information factor (see Table 2). Table 2 presents only the analyses of the specific items because the MANOVAs yielded a significant information effect only for the specific items.

Of the seven analyses presented in the table, only one produced a statistically significant result. Subjects exposed to long-term neutral and positive outcome information judged the long-term effects of the sexual contacts on the adolescents more positively than did subjects in the descriptive-only condition. Three of the remaining six statistics were marginally significant. The average effect size was positive. but small (r = .16).

 Of the seven measures for the specific cases, only two are directly relevant to the assessment of the hypothesis that the nonnegative outcome information would be processed in a biased manner. The first measure is perceptions of long-term effects of the sexual contacts on the adolescents. This measure is relevant because the additional information that half of the subjects read concerned long-term effects. If the hypothesis of biased processing was correct, then perceptions of

[Page 266]

long-term effects should have been the same whether subjects read the nonnegative outcome information or the descriptive-only information. As noted previously, subjects evaluated the long-term effects more positively when they were exposed to the nonnegative outcome information. 

This finding does not support the hypothesis of biased information processing. To test the hypothesis further, the mean evaluation of the long-term effects by subjects exposed to the nonnegative outcome information was compared with the neutral scale value for this measure (i.e., 7). If processing was unbiased, then it follows that this mean evaluation should have been nonnegative (i.e., greater than or equal to 7) because the outcome information was nonnegative.

Separate comparisons were made for subjects in the neutral and negative terminology conditions who were exposed to the nonnegative outcome information. The mean evaluation given by subjects in the neutral terminology condition (M = 6.95) was not significantly less than 7, t(19) = -.11, p > .10, one-tailed, failing to support the hypothesis of biased processing. The mean evaluation given by subjects exposed to negative terminology (M = 5.75) was significantly less than 7, t(19) = -2.70, p < .01, one-tailed, supporting the hypothesis of biased processing.

The second measure that is relevant to the hypothesis of biased processing of nonnegative outcome information is perceptions of therapy needed because the additional information addressed long-term psychological adjustment. If information processing was unbiased, then judgments of therapy needed should have been less among subjects who read the nonnegative outcome information than among those who did not because the nonnegative outcome information indicated no problems in adjustment and hence no need, or at least less of a need, for therapy. 

Subjects exposed to the nonnegative outcome information did not differ from subjects exposed to the descriptive-only information in their judgments, supporting the hypothesis of biased processing. 

Author Objectivity and Terminology Appropriateness

Two additional items were included among the dependent measures to assess how objective subjects thought the author of the condensed article was in terms of discussing the issue of adult- nonadult sexual relationships and how scientifically appropriate and proper the terminology employed by the author was.

A gender X terminology X information ANOVA performed on the author objectivity data yielded no significant effects. Thus, for example, the author's objectivity was judged to be statistically the same whether he used negative or neutral terms. The mean objectivity rating for all subjects was slightly above the neutral point on the 7-point scale (M = 4.90, SD = 1.74), indicating that subjects perceived the author to be objective, but only moderately so.

A gender X terminology X information ANOVA performed on the data regarding the appropriateness of the terminology used also yielded no significant effects. The mean appropriateness rating for all subjects indicated that subjects judged the terms to be appropriate (M = 5.66, SD = 1.30). Thus, subjects considered negative terms and neutral terms to be equally valid in describing sexual contacts between men and male adolescents.