The effectiveness of this measure was evaluated based on a short survey directed at pupils before the lesson (base level) and around 2 3 months later (post-lesson) based on three variables which are considered as strong determinants of actual behaviour: behaviour intention, estimated dangerousness of the behaviour, and level of knowledge on the legality of the behaviour.
Theoretical range of scores is 0 to 3 for sub-domains, 0 to 9 for domains and 0 to 27 for intention estimate.
An " intention estimate coefficient" was therefore defined by simply calculating the sum of the three domains' mean values so that equal weight was given to attitude, subjective social norm and control beliefs (Table 1).
Regardless of the predictor set (intention estimate, 3 domains or six sub-domains) predictions based on PBMS after the programme were significantly better than before (before 34 to 56%, after 65 to 73% explained variance, p = <.01) (Table 2).
The PBMS explained 55% of the criterion's (intention to use or not use DMT) variance when calculated based on the intention estimate (one predictor) and even more, when intention was predicted based on the three domains (60%) or based on the six mean scores of the sub domains (69%) (Table 2).
Four weeks after an educational intervention, predictive power was higher in the intervention (IG) compared to the control group (CG) (intention estimate: CG 56% / IG 69%, p = .179; three domains CG 56% / IG 74%, p = .047; six sub-domains CG 64% / IG 78%, p = .073).073
All predictor sets showed higher percentages of explained variance in the intervention group (IG) than the control group (CG) (intention estimate: CG 56% / IG 69%, p = .179; three domains CG 56% / IG 74%, p = .047; six sub-domains CG 64% / IG 78%, p = .073) (Table 2).
Also, with the intention of estimating the effectiveness of the header design, thermal and hydraulic performance of heatsinks with optimized headers were also compared with the performance of the heatsinks with conventional header designs.
To examine whether determinants of intention choice estimated in Experiment 1 were nonetheless able to predict participants' responses, we then used the generated CART model to predict intention choice over the distribution and compared the actual classification accuracy by participants with the accuracy of classification predicted by the model.
With the intention of estimating severity distribution, the methodology presented in the section 'Severity distribution' was followed.
