Table 2.
Model Fit Statistics for Exploratory Factor Analysis, Latent Profile Analysis, and Structural Equation Mixture Models
| Profiles | P | LL | BIC | aBIC | cAIC | AIC | Entropy |
|---|---|---|---|---|---|---|---|
| LPA 1c | 15 | −8132.66 | 16353.10 | 16305.52 | 16368.10 | 16295.32 | -- |
| LPA 2c | 25 | −7921.73 | 15989.76 | 15910.45 | 16014.76 | 15893.45 | 0.94 |
| LPA 3c | 35 | −7826.97 | 15858.78 | 15747.74 | 15893.78 | 15723.95 | 0.94 |
|
| |||||||
| Factors | |||||||
|
| |||||||
| FA 1f | 24 | −8008.23 | 16156.91 | 16080.78 | 16180.91 | 16064.46 | -- |
| FA 2f | 32 | −7976.15 | 16139.58 | 16038.06 | 16171.58 | 16016.31 | -- |
| FA 3f | 39 | −7918.81 | 16065.86 | 15942.14 | 16104.86 | 15915.62 | -- |
|
| |||||||
| SEMMs | |||||||
|
| |||||||
| 1f, 1c | 24 | −8008.41 | 16157.27 | 16081.14 | 16181.27 | 16064.82 | -- |
| 1f, 2c | 26 | −7904.32 | 15960.81 | 15878.32 | 15986.81 | 15860.65 | 0.95 |
| 1f, 3c | 28 | −7836.75 | 15837.36 | 15748.53 | 15865.36 | 15729.50 | 0.95 |
| 3f, 1c | 27 | −7926.07 | 16010.15 | 15924.50 | 16037.15 | 15906.14 | -- |
| 3f, 2c | 31 | −7814.14 | 15809.71 | 15711.36 | 15840.71 | 15690.29 | 0.95 |
| 3f, 3c | 35 | −7726.11 | 15657.05 | 15546.01 | 15692.05 | 15522.22 | 0.96 |
Note. P=parameters; LL=Log-likelihood; BIC=Bayesian Information Criterion; aBIC=Sample-size adjusted Bayesian Information Criterion, calculated as −2LL + plog [(N+2)/24]; AIC=Akaike Information Criterion; cAIC=consistent Akaike Information Criterion, calculated as −2LL + p[log(N) + 1]. Smaller values of BIC, aBIC, AIC, and cAIC indicate better data-model fit. “--” was used to indicate that these data are not applicable to the model tested. BIC differences ≥ 6 provide strong evidence favoring the model with the smaller value (43). Higher posterior probabilities and entropy suggest better prediction of latent profile membership and clearer delineation of latent profiles, respectively. Although the three factor-three profile mixture model had the best fit to the data, this model was associated with substantial sparseness for the latent profile analysis, such that few participants were classified into LP3 (n=15) due to very extreme scores on latent profile indicators.