vuong_test

superglm.vuong_test(

model_a: SuperGLM,

model_b: SuperGLM,

X,

y,

sample_weight=None,

offset=None,

*,

correction: str = 'aic',

) → VuongTestResult

Vuong (1989) likelihood-ratio test for non-nested model comparison.

Compares two models’ per-observation log-likelihoods. Under H0 (models are equally close to the true DGP), the test statistic is asymptotically N(0,1).

Parameters:

model_a, model_bSuperGLM

Two fitted SuperGLM models.

Xpandas or eager Polars DataFrame

Design matrix (same for both models).

yarray-like

Observed response values.

sample_weightarray-like or None

Observation weights, read under the contract both models declare. Under "frequency" integer values are exactly equivalent to row replication; under "prior" they enter each density through phi_i = phi / w_i or the equivalent scaling of the family’s own parameters. A weighted comparison of two models that declare different contracts is rejected, because a single vector cannot represent both.

offsetarray-like or None

Optional offset.

correctionstr

Complexity correction: "none", "aic", or "bic".

Returns:

VuongTestResult