Maartje Boer

VALIDATION OF THE SMD-SCALE 35 2 Analysis Strategies Structural Validity We explored the number of underlying factors measured by the SMD-scale by conducting an EFA using the calibration sample. A factor should consist of at least three items to be considered as a reliable factor (Costello & Osborne, 2005; Fabrigar et al., 1999). Therefore, with nine items on the scale, we decided a priori that a maximum of three factors should be extracted in the EFA. An oblique (goemin) rotation was applied to interpret the factor loadings, which assumed that factors in the multiple factor solution were correlated. The EFA- factor solutions were evaluated based on the empirical eigenvalues, Horn’s parallel analysis, model fit, and quality. The number of factors with empirical eigenvalues higher than one indicated the number of factors to extract (Ledesma & Valero-Mora, 2007). Parallel analysis evaluated this solution by comparing the empirical eigenvalues with 1000 randomly generated eigenvalues based on the same number of variables and sample size. The number of factors to retain was indicated by the number of factors where the 95 th percentile random data eigenvalues did not exceed the empirical eigenvalues (Ledesma & Valero-Mora, 2007). Model fit of the factor solution was assessed using the Comparative Fit Index (CFI), Tucker Lewis Index (TLI), Root Mean Square Error of Approximation (RMSEA), and Standardized Root Mean Square Residual (SRMR) (Schermelleh-Engel et al., 2003). We did not rely on the χ ²-statistic given its sensitivity to large sample sizes (Fabrigar et al., 1999). Quality of the factor solutions was considered poor when removal of items with factor loadings below 0.5 or with cross-loadings that differed by less than 0.2 yielded factors with less than three items (Costello & Osborne, 2005; Howard, 2016). To examine the robustness of the EFA results, we conducted Velicer’s minimum average partial (MAP) analysis using the calibration sample. This analysis evaluates multiple factor solutions based on principal component analysis by calculating the average partial correlation between items when the first component is partialled out, when the first two components are partialled out, and so on. The number of factors to retain was indicated by the number of components where the average partial correlation was at its minimum (Velicer, 1976). To examine the robustness and generalizability of the findings from the EFA and MAP analyses, the obtained factor solution was evaluated with a CFA using the validation sample.