TABLE 2.
Latent class model fit statistics, by number of latent classes specified a priori.
| Model | s | Smallest Class | LL | χ2 LRT (p-value) | AIC | BIC | Entropy | LMR LRT (p-value) |
|---|---|---|---|---|---|---|---|---|
| 2-Class | 13 | 25.2% | −2925.176 | 380.100 (p<0.001) | 5876.352 | 5945.302 | 0.986 | 4720.759 (p<0.001) |
| 3-Class | 20 | 19.3% | –2574.770 | 118.306 (p<0.001) | 5189.540 | 5294.616 | 0.952 | 687.268 (p=0.222) |
| 4-Class | 27 | 6.4% | −2510.023 | 58.630 (p=0.010) | 5074.047 | 5217.250 | 0.963 | 127.009 (p=0.439) |
| 5-Class | 34 | 1.8% | −2491.300 | 40.421 (p=0.077) | 5050.601 | 5230.931 | 0.967 | 36.937 (p=0.545) |
Notes: s = number of free parameters estimated; LL = log likelihood; χ2 = Chi-square likelihood ratio test; AIC = Akaike Information Criterion; BIC = Bayesian Information Criterion; LMR LRT = Lo-Mendell-Rubin likelihood ratio test. Goodness-of-fit indices should be interpreted relative to each other, with a best-fitting LCA model supported by one or more of the following conditions: predicted prevalence of the smallest class >10%, the smallest relative AIC and BIC values, entropy > 0.8, and rejection of the null hypothesis (p<0.05) for the χ2 and LMR LRT.