Latent profile count selection blends fit statistics with theory.
Evidence
As discussed in prior work42, the process of identifying the number of proļ¬les in LPA involves a combination of data and theory. Thus, we began by comparing model results from a one-proļ¬le model to a two-proļ¬le model, and so on until we reached a ļ¬ve-proļ¬le model. At that point, we felt we had a holistic sense of the number of proļ¬les that best balanced the model ļ¬t statistics with our theory. The model ļ¬t statistics we focused on most closely were the log-likelihood value, Akaikeās Information Criterion (AIC), the Bayesian Information Criterion (BIC), and the Sample-Adjusted BIC (SABIC). For all of these indices lower values indicate better model ļ¬t, although the magnitude of the difference should also be considered. We also considered the model entropy (i.e., a measure of classiļ¬cation uncertainty; ranges from 0 to 1 with higher values representing greater classiļ¬cation certainty), as well as the results of the Lo-Mendell Rubin (LMR) test and the bootstrapped log likelihood test (BLRT), both of which compare the ļ¬t of the selected model to a model with k-1 proļ¬les (i.e., a statistically non-signiļ¬cant result favors the more parsimonious model). Likewise, we aimed to avoid adding an additional proļ¬le if that additional proļ¬le accounted for less than 10% of the sample42 and/or did not ļ¬t with our theory. ā Erin Cooley, Jazmin L. Brown-Iannuzzi, Nava Caluori, Nicholas Elacqua, and William Cipolli, White Americans who perceive themselvesā¦, 37
Categorization
- DEFINE: Clarifies the standard criteria (AIC, BIC, SABIC, entropy, LMR, BLRT) for deciding how many latent profiles to retain in LPA.
- XREF: connects to: Combining exploratory and confirmatory factor analysis produces theory-driven results robust across samples ā Ruiqian Li and Paul Froese, The Duality of American Christianā¦, 94; Stereotyping functions as profile construction for information processing under data scarcity or overload; Empirically-selected dogmatism scale items fail to replicate across samples, performing at chance level ā Bob Altemeyer, The Authoritarian Specter, 4262
Broader Connections
- Empirically-selected dogmatism scale items fail to replicate across samples, performing at chance level
#complicates(0.76) ā Bob Altemeyer, The Authoritarian Specter, 4262 - Stereotyping functions as profile construction for information processing under data scarcity or overload
#expands(0.74)