Table 3.
Latent Class Analysis Model Fit Indices of Candidate Models
| Classes | Log-likelihood | Parameters in model | Adjusted χ2 LRT (df), p value | BIC | CAIC | LMR-LRT, p value | BLRT, p value |
|---|---|---|---|---|---|---|---|
| 1 | −2837.84 | 6 | 285.11 (57), p < 0.00001 | 5715.30 | 5721.30 | — | — |
| 2 | −2740.26 | 13 | 89.94 (50), p = 0.0005 | 5566.35 | 5579.34 | 191.04, p < 0.00001 | p < 0.00001 |
| 3 | −2719.52 | 20 | 48.47 (43), p = 0.26 | 5571.10 | 5591.10 | 40.59, p = 0.0002 | p < 0.00001 |
| 4 | −2713.03 | 27 | 35.49 (36), p = 0.49 | 5604.33 | 5631.33 | 12.71, p = 0.06 | p ≈ 1.00 |
| 5* | — | — | — | — | — | — | — |
| 6* | — | — | — | — | — | — | — |
Boldface indicates best-fitting model for each fit index measure. For the BIC, a difference <5 was not considered relevant and both two- and three-class solutions were considered adequate.34
*
Models with five and six classes were empirically underidentified and model fit indices are not shown.
BIC, Bayesian information criterion; BLRT, bootstrap likelihood ratio test; CAIC, consistent Akaike's information criterion; df, degrees of freedom; LMR-LRT, adjusted Lo–Mendell–Rubin likelihood ratio test; LRT, likelihood ratio test.