Appendix E

Latent profile analysis was used to create groupings of those who deidentified from the Church. Analyses were conducted in LatentGold 6.1 with full information maximum likelihood used to handle missing data. There was substantial improvement in model fit from 1–4 classes and then smaller improvements thereafter (table E1). The bootstrap difference test for k-1 classes was significant in every instance. Entropy was excellent throughout, ranging from .96 to .99, and therefore did not meaningfully distinguish classes. Substantively different classes emerged from within each additional class up to the fourth class. Adding a fifth and sixth class appeared to only add classes that were slight variants on already identified classes. We therefore determined the four-class solution to be the simplest without losing meaningful information.