34 Appendix D: Factor Analysis Results #2 The following specific methods were used for second factor analysis. First, in order to use separate but essentially equivalent datasets to estimate the models, the NACE Research Department randomly split dataset into two with a 67:33 ratio for CFA Tuning and CFA Testing, resulting in sample sizes of 4,036 and 2,019 respectively. After tuning the CFA model with the “Tune” dataset, NACE tested that specific model again with the “Test” dataset. To run these analyses, NACE ran CFA with lavaan in R. For CFA, NACE fixed variance of latent variables to 1, so each item’s loading could be estimated because NACE wanted to examine how all the items specifically load onto the latent variables in this case, so fixing the latent variable variance at 1 was appropriate. NACE ran models with FIML to handle missing data because FIML is regarded as better producing more stable estimates. NACE ran models with both outliers included and excluded and saw minor differences in their results. NACE decided to keep the outliers because all values were within the acceptable range, were not data input errors, and would likely occur in the field to some extent.
RkJQdWJsaXNoZXIy MTM2NjgzMA==