Page 116 - Secondary school students’ university readiness and their transition to university Els van Rooij

Self-e cacy in being successful at university
4.3.4 Procedure
e questionnaires were all paper-and-pencil tests, handed to the students during class by the researchers or a teacher who had received instruction. Informed consent from parents was obtained in advance. Students who did not have parental consent to participate (three students out of the whole sample) went to an empty classroom or another place in the school where they did some homework. Participation by students was voluntary and without compensation but strongly encouraged by teachers. None of the students who got consent from his or her parents refused to participate.
4.3.5 Statistical analyses
We sought to determine how well our theoretical model (Figure 4.1) t the data 4 provided by a sample of Dutch, grade 10 and 11, pre-university students. To avoid
including unnecessary pathways from the background variables in the model, we
rst conducted t-tests and an analysis of variance (ANOVA) to test for signi cant
di erences in need for cognition, out-of-school academic activities, engagement, academic interest, and self-e cacy, based on gender, coursework, and level of parental education. We also looked at the bivariate correlations across all included factors. A er conducting these exploratory analyses, we undertook structural equation modelling (SEM) with the statistical package available in Mplus, Version 7. Regarding the background variables, we only included pathways if we found a signi cant di erence in the t-test or ANOVA. For example, if we found a signi cant di erence between boys and girls regarding their need for cognition, we added a pathway from gender to need for cognition. To evaluate the goodness of t of the models, we considered the ratio of the chi-square to its degrees of freedom (χ2/ df), the root mean square error of approximation (RMSEA), standardised root mean square residual (SRMR), comparative t index (CFI), and Tucker-Lewis index (TLI), which is less susceptible to sample size (Tucker & Lewis, 1973). With our relatively large sample size (n > 400), the p-value of the sample size-sensitive chi-square test could be erroneously signi cant and thus may not adequately re ect whether our model provides a good t to the data (Schumacker & Lomax, 2004). Following established guidelines, we determined that the model o ered an appropriate re ection of the data if the χ2/df value was less than 3 (Kline, 2005), the RMSEA was less than .07, the SRMR was less than .08, and the CFI and TLI were greater than .90 (Chen, Curran, Bollen, Kirby, & Paxton, 2008; Hu & Bentler, 1999; Kline, 2005; Steiger, 2007; Tucker & Lewis, 1973).
115