Descriptive Statistics and Zero-Order Correlations
In the first wave, 248 adolescents (17.2%) reported having engaged, at least once, in one of the four risk behaviors. In the second wave, 224 adolescents (15.5%) reported having engaged in risky sexual online behaviors in the last 6 months. Table
2 provides the zero-order correlation matrix for the four-item risky sexual online behavior scale and the perceptions of peer involvement, risks, benefits, and vulnerability for the two waves. As Table
2 shows, all variables were significantly correlated with each other. Engagement in risky sexual online behavior had moderate stability over time (
r = .38,
p < .01). These online behaviors were moderately and positively related to perceived peer involvement at both waves (
r = .43,
p < .01 and
r = .45,
p < .01, respectively). As expected, risky sexual online behavior was negatively related to perceived risks at both waves (
r = −.28,
p < .01 and
r = −.24,
p < .01), and positively related to perceived benefits (
r = .31,
p < .01 and
r = .26,
p < .01). As expected, there was also a negative relationship between risky sexual online behavior and perceived vulnerability for Wave1 and Wave 2 (
r = −.25,
p < .01 and
r = −.22,
p < .01, respectively).
Table 2
Zero-order correlations between risky sexual online behavior and risk-related perceptions
Risky sex. onl. Behav |
Time 1 | – | | | | | | | | | |
Time 2 | .38** | – | | | | | | | | |
Peer involvement |
Time 1 | .43** | .26** | – | | | | | | | |
Time 2 | .28** | .45** | .46** | – | | | | | | |
Risks |
Time 1 | −.28** | −.17** | −.33** | −.26** | – | | | | | |
Time 2 | −.15** | −.24** | −.19** | −.37** | .52** | – | | | | |
Benefits |
Time 1 | .31** | .16** | .43** | .24** | −.47** | −.33** | – | | | |
Time 2 | .19** | .26** | .22** | .41** | −.35** | −.50** | .47** | – | | |
Vulnerability |
Time 1 | −.25** | −.15** | −.27** | −.24** | .72** | .46** | −.40** | −.30** | – | |
Time 2 | −.14** | −.22** | −.16** | −.32** | .41** | .75** | −.27** | −.40** | .45** | – |
Causal Relationships Between Risky Sexual Online Behavior and Perceptions
The correlations in Table
2 already demonstrate significant relationships between perceptions and risky sexual online behavior. To analyze the causality of these relationships, we tested the hypothesized model as shown in Fig.
1 for all perceptions. The coefficients of the cause and effect paths, and the indicators of model fit are presented in Table
3. The model fits for the four hypothesized models were good. The CFI’s of the four models were all above .95, and the RMSEA values were below .05.
Table 3
Indicators of the four autoregressive cross-lagged models
Peer involvement | .13* | .07 | 35.28** | .99 | .03 [.02; .05] |
Risks | −.06* | .03 | 26.83* | 1.00 | .03 [.01; .04] |
Benefits | .04 | .02 | 20.03 | 1.00 | .02 [.00; .04] |
Vulnerability | −.06* | −.01 | 17.01 | 1.00 | .02 [.00; .03] |
Our first hypothesis (H1a) stated that adolescents who perceive more friends to engage in risks are more likely to engage in risky sexual online behavior 6 months later. H1a was supported as the relationship between perceived peer involvement at Time 1 and risky sexual online behavior at Time 2 (= cause path) was significant, β = .13, B = .16, SE = .04, p < .05 (bootstrap bias-corrected 95% confidence interval [bc 95% CI]: .04/.33).
Hypothesis 1b predicted that the reverse relationship would also be significant. As the effect path was not significant, β = .07, B = .06, SE = .03, ns (bc 95% CI: −.02/.14) this hypothesis was not supported. Therefore, perceptions of peer involvement and engagement in risky sexual online behavior were not reciprocally related. Instead, perceptions of peer involvement at Time 1 influenced subsequent online risk behavior. Engagement in risky sexual online behavior, however, did not influence subsequent perceptions of peer behavior.
Hypothesis 2a, which predicted that perceived risks negatively influence engagement in risky sexual online behaviors, received support. As expected, the relationship between perceived risks at Time 1 and risky sexual online behavior at Time 2 was significant, β = −.06, B = −.07, SE = .03, p = .05 (bc 95% CI: −.15/.00). The reverse relationship, as stated in Hypothesis 2b, was not significant, β = .03, B = .06, SE = .03, ns (bc 95% CI: −.02/.14). Therefore, this hypothesis also failed to find support.
The model for perceived benefits was not supported as neither the effect, β = −.01, B = .04, SE = .03, ns (bc 95% CI: −.02/.09), nor the cause path were significant, β = −.01, B = .02, SE = .05, ns (bc 95% CI: −.05/.11). Thus, adolescents’ perceptions of the benefits of risky sexual behavior were not significantly related to risky sexual online behavior (H3a and H3b).
The influence of perceived vulnerability at Time 1 on risky sexual online behavior at Time 2 was significant, β = −.06, B = −.06, SE = .02, p < .05 (bc 95% CI: −.12/−.01), as stated in Hypotheses 4a. The reverse relationship was not significant. Therefore, H4b failed to find support, β = −.01, B = −.01, SE = .03, ns (bc 95% CI: −.08/.06).
In sum, three causal paths—those of perceived peer involvement, perceived risks, and perceived vulnerability at Time 1 to risky sexual online behavior at Time 2—were significant. However, none of the effect paths from risky sexual online behavior at Time 1 to perceptions of these behaviors at Time 2 were significant.
Relative Influences of Perceptions on Risky Sexual Online Behavior
The results of the structural equation models showed that peer involvement at Time 1 had the strongest influence on online sexual risk taking at Time 2 (see Table
3). To investigate whether the other predictors provided additional explanatory value over and above the effect of peer involvement, we conducted a linear OLS regression analysis predicting the engagement in risky sexual online behavior at Time 2. Because our variables are not normally distributed, homoscedasticity in the errors cannot be assumed. We, therefore, analyzed our regression model with heteroscedasticity-consistent standard errors (Long and Ervin
2000). Time 1 online sexual risk behavior, all control variables, and perceptions of peer involvement, risks, benefits, and vulnerability were entered into the regression. Overall the model accounted for 17% of the variance. Of the perception variables, only perceived peer involvement at Time 1 was a significant predictor of risky sexual online behavior at Time 2, β = .12, SE = .05,
t(1444) = 2.28,
p < .05. No additional variance was explained by perceived risks (β = −.02,
ns), benefits (β = −.01,
ns), and vulnerability (β = −.02,
ns). Of the control variables, only frequency of internet communication was a significant predictor of risky sexual online behavior, β = .02,
t(1444) = 2.97,
p < .01.