pymargins.pool_imputations

pymargins.pool_imputations(results, *, label='pooled', complete_df=None)

Pool MarginsResult objects across imputations via Rubin’s rules.

All results must be the same estimand (matching labels) on imputed copies of the same data, computed under a shared posture. Pools on the inference scale; reports FMI / relative efficiency / df.

Parameters:

  • results (list of MarginsResult) – One result per imputation.

  • label (str, default "pooled") – Provenance tag for the pooled estimand, stored on the result as estimand_metadata["pooled_label"] for downstream bookkeeping. It does not rename the per-component row labels (those carry over from the input results) and is not shown in summary().

  • complete_df (float, optional) – Complete-data degrees of freedom for the Barnard–Rubin small-sample correction. If None, uses the classic Rubin (1987) df formula.

Returns:

A leaf result with method="pooled" and imputation_diagnostic set.

Return type:

MarginsResult