Skip to main content
. Author manuscript; available in PMC: 2022 Sep 6.
Published in final edited form as: Drug Alcohol Depend. 2022 Jun 30;238:109550. doi: 10.1016/j.drugalcdep.2022.109550

Table 2.

Model Fit Indices and Model Comparison Statistics for Mixture Modeling of Tailored Programs for Substance Use Treatment.

Number of classes Log Likelihood Bayesian Information Criterion Akaike Information Criterion Sample-Size Adjusted Bayesian Information Criterion Number of Free Parameters Entropy Smallest Class Size (%)
1 −67086 134240 134186 134218 7 N/A 100
2 −41902 84036 83852 83960 24 0.99 29.9
3 −41946 84135 83943 84056 25 0.79 20.9
4a −41776 83863 83617 83761 32 0.76 13.7
5 −41768 83914 83615 83790 39 0.77 1.8
6 −41721 83897 83536 83747 47 0.73 2.1
7 −41706 83946 83523 83771 55 0.75 1.7
8 −41664 83939 83454 83738 63 0.68 2.1

Note.

a

Model selected as providing the best fit, as demonstrated by the relatively small Akaike Information Criterion, Bayesian Information Criterion, relatively high entropy, and relatively few parameters while keeping class size above 5%. Models 2 through 4 included bivariate residuals (i.e., addition of residual associations [local dependencies]) due to violation of the local independence assumption for LCA. Each criterion is based upon the Log-Likelihood.