XML output

Comparing REML models

The Akaike Information Criterion (AIC) and the Bayesian Information Criterion (BIC) are defined in equation 2.15 of the User Guide.

For comparing nested variance models we recommend the Likelihood Ratio test.

The AIC and BIC are provided for the convenience of users but without any recommendation from us. A problem in complex models is the counting of the number of parameters. The value used in calculating AIC and BIC is reported giving the opportunity for the user to verify/modify that value.

All these statistics, being based on the REML LogLikeihood statistics are only valid if the fixed effects model is unchanged between runs and, given it is usually over parameterised, is fitted in the same order (i.e. the same effects are aliased).

Return to index