Maartje Boer

CHAPTER 2 40 Table 2.1 EFA Eigenvalues, Parallel Analysis, and Velicer’s MAP Test (Calibration Sample, n = 3,310) Number of factors Empirical eigenvalues Parallel test: 95th percentile of random eigenvalues Velicer’s MAP test: Minimum average partial correlation 0 -- -- 0.196 1 4.572 1.103 0.027 2 0.819 1.070 0.048 3 0.746 1.048 0.071 4 0.630 1.028 0.127 5 0.599 1.010 0.222 6 0.562 0.995 0.314 7 0.456 0.978 0.461 8 0.349 0.960 1.000 Notes. EFA = Exploratory Factor Analysis; MAP = Minimum Average Partial. 0.0.016) solutions were all good, the one-factor solution showed the highest quality (Table 2.2). This is because in the one-factor solution, factor loadings of all items were higher than 0.5, while in the two- and three-factor solutions, there were multiple items with cross-loadings and factor loadings below 0.5. After removal of these items, the factors in the two- and three-factor solutions did not meet the requirement of having at least three items with loadings of 0.5 or higher per factor. Furthermore, the correlations between the factors in the two- and three-factor solutions were high ( r ≥ 0.59), which suggests that the additional factors strongly overlap and should not be considered as separate factors. The EFA obtained one-factor solution was also found by Velicer’s MAP test, because the one-factor solution showed the lowest average partial correlation (Table 2.1). The one-factor solution was further evaluated with a CFA using the validation sample. Model fit was good (CFI = 0.983, TLI = 0.977, RMSEA = 0.028, and SRMR = 0.040). Also, the quality of the factor was good, because all nine factor loadings exceeded 0.5 (Table 2.2). The one-factor solution was thus confirmed by the CFA using another, randomly selected sample. These results imply that all nine items contributed to one single dimension. Reliability and Item Performance The ordinal alpha of the one-factor solution was 0.87, which indicates that the internal consistency of the test scores was good. Reliability was further evaluated based on IRT item performance using the two-parameter logistic