AIC and BIC in Mplus Workflows
Information criteria help compare fitted models. They are useful when models use the same data set. They also help when a strict likelihood ratio test is not suitable. In Mplus, users often review the loglikelihood, free parameters, sample size, AIC, BIC, and adjusted BIC. This calculator recreates those values from the reported output. It also compares two candidate models.
Why the Values Matter
AIC rewards fit but adds a lighter penalty for complexity. BIC adds a stronger penalty because it uses sample size. A lower value usually points to the preferred model. The result is not a proof of truth. It is a practical model selection signal. Use it with theory, residual checks, convergence status, and parameter meaning.
Using Mplus Output
Find the H0 loglikelihood value in the output. Then note the number of free parameters. Also confirm the number of observations used by the estimator. Enter those values exactly. Negative loglikelihood values are common. Do not remove the minus sign. If you compare two models, both should be estimated on the same cases. Otherwise, the comparison can be misleading.
Reading the Result
The page calculates AIC, BIC, adjusted BIC, and optional AICc. It also gives delta values. A delta near zero marks the best model in that criterion. Larger deltas show weaker support. Evidence weights give an easier comparison. They convert criterion differences into relative support scores. They are approximate, not absolute probabilities.
Good Practice
Always keep a record of model names. Save the estimator, sample size, and parameter count. Check warnings before trusting a final number. If Mplus reports a different value, inspect the input values first. Rounding, missing data handling, mixture settings, or alternative likelihood corrections can explain differences. This calculator is best for transparent checks, teaching, and fast reporting. It should not replace full statistical judgment. It helps reviewers see how each penalty changes the final model ranking. For publication, report the exact Mplus output and explain why the selected model is theoretically reasonable. When several models are close, describe the tradeoff rather than claiming one perfect answer. Small differences should be interpreted carefully. When sample sizes are large, BIC may favor simpler models more strongly than AIC does overall.