Maartje Boer

CHAPTER 2 42 model. The two-parameter logistic model, which allowed the discrimination parameters to vary, was selected because its fit was better than the one- parameter logistic model, which constrained the discrimination parameters to be equal ( χ ²(8) = 243.67, p < 0.001). IRT models showed that the difficulty parameters of all nine items ranged between 0.91 and 2.01, indicating high difficulty (Baker, 2001). This suggests that the criteria were most likely to be present among adolescents with higher levels of problematic SMU. Discrimination parameters were moderate (1.04 to 1.29; preoccupation, persistence, escape), high (1.55; displacement), or very high (1.80 to 2.40; withdrawal, problem, deception, tolerance, conflict) (Baker, 2001). This implies that the criteria had moderate to very high discriminative power to distinguish adolescents with high from those with low levels of problematic SMU. Figure 2.2A shows that for values at the mean of the latent trait ( ϴ = 0, corresponding to endorsement of ± one criterion), item ‘escape’ provided the most information. For values that were one standard deviation above the mean of the latent trait ( ϴ = 1.00, corresponding to endorsement of ± four criteria), item ‘problem’ provided the most information. For values two standard deviations above themean ( ϴ = 2.00, corresponding to endorsement of ± seven criteria), item ‘conflict’ provided the most information. Figure 2.2B shows the information function of the total scale. As can be seen, the scale provided most information on higher values of the latent trait, that is, higher than the mean ( ϴ = 0.00). These findings indicate that test scores were most reliable at moderate to high levels of the scale’s continuum. Total information was highest at ϴ = 1.68 (corresponding to endorsement of ± six criteria), which indicates that test scores were most reliable at this value. Measurement Invariance The configural multigroup CFAs all showed good model fit (gender: CFI = 0.983; TLI = 0.977; RMSEA = 0.027; SRMR = 0.039, educational level: CFI = 0.984; TLI = 0.978; RMSEA = 0.026; SRMR = 0.047, age category: CFI = 0.982; TLI = 0.975; RMSEA = 0.028; SRMR = 0.049, ethnic background: CFI = 0.983; TLI = 0.977; RMSEA = 0.027; SRMR = 0.042). All group comparisons showed scalar invariance (gender: Δ -CFI = -0.001; Δ -RMSEA = -0.001; Δ -SRMR = 0.001, educational level: Δ -CFI = -0.004; Δ -RMSEA = 0.001; Δ -SRMR = 0.004, age category: Δ -CFI = 0.001; Δ -RMSEA = -0.004; Δ -SRMR = 0.003, ethnic background: