Esther Mertens

| 55 Effectiveness of a Psychophysical Intervention To assess the effects of the intervention, we constructed three dummy variables (i.e., Light, Standard, and Plus condition) with the Control condition as a reference group and regressed the intercept and slope on these three dummy variables. Students’ age and ethnicity were added as covariates as the conditions differed significantly on these variables. If two or more intervention conditions appeared to be effective compared to the control group, we examined the effectiveness of those conditions compared to each other in a multigroup model by constraining the slopes of those conditions to be equal and by releasing this constraint. The model fits of the two nested models were compared using the Satorra-Bentler Scale Chi-Square test. This test applies a scaling correction to better approximate the chi-square distribution under non-normality (Satorra & Bentler, 2010). A significant Satorra-Bentler Scale Chi-Square test indicates that the unconstrained model fits better and, thus, that one intervention condition is more effective than the other. We calculated effect sizes by multiplying the rate of change by time span divided by the standard deviation of the concerned outcome ( d = (slope * duration) / SD; Feingold, 2013). We calculated effect sizes for the change between measurement points (i.e., change from T1 to T2, from T2 to T3, and from T3 to T4) and the overall change (i.e., change from T1 to T4). As there is no specific formula to calculate effect sizes for unspecified non-linear growth, the overall effect sizes were calculated using the formula for linear growth 1 . Results Descriptive Statistics Table 2 presents themeans and standard deviations of the outcomes for the conditions on each of the measurement points. The LGC models showed acceptable fit (see Table 3; RMSEA < .08, CFI > .90, SRMR < .10; Kline, 2005). The models for bullying and victimization showed a poor fit based on the CFI, but a good fit based on the RMSEA and SRMR. The standardized factor loadings of the time points on the slope, reflecting the average change in the observed variables, indicated that students showed generally the largest change on the outcomes from T1 to T2 (see Table 3). After T2 the average change leveled off. For instance, students changed on sexual autonomy from T1 to T2 with a rate of .69, from T2 to T3 with a rate of .15, and from T3 to T4 with a rate of .05. 1 To examine the robustness of the overall effect sizes of the unspecified growth models, we modeled linear LGC models and calculated the overall effect sizes. The overall effect sizes of the linear models were in general larger than the effect sizes based on the unspecified growth models indicating that the effect sizes of the unspecified growth models were more conservative. 3