statsmodels.genmod.generalized_linear_model.GLMResults.info_criteria¶

GLMResults.info_criteria(crit, scale=None, dk_params=0)[source]¶

Return an information criterion for the model.

Parameters:¶

critstr

One of ‘aic’, ‘bic’, or ‘qaic’.

scalefloat

The scale parameter estimated using the parent model, used only for qaic.

dk_paramsint or float

Correction to the number of parameters used in the information criterion. By default, only mean parameters are included, the scale parameter is not included in the parameter count. Use dk_params=1 to include scale in the parameter count.

Returns:¶

Value of information criterion.

Notes

The quasi-Akaike Information criterion (qaic) is -2 * llf/scale + 2 * (df_model + 1). It may not give meaningful results except for Poisson and related models.

The QAIC (ic_type=’qaic’) must be evaluated with a provided scale parameter. Two QAIC values are only comparable if they are calculated using the same scale parameter. The scale parameter should be estimated using the largest model among all models being compared.

References

Burnham KP, Anderson KR (2002). Model Selection and Multimodel Inference; Springer New York.