[vrjdev]

Description

DoubleMLVector._get_data_for_model always appended the other treatment columns to the covariates of each single-treatment sub-model, ignoring DoubleMLData(use_other_treat_as_covariate=False). PLRVector therefore fitted a different model than requested, and a different one from the legacy DoubleMLPLR on the same data.

• doubleml/double_ml_vector.py: append the other treatments only if use_other_treat_as_covariate is True, as DoubleMLData.set_x_d does.
  • The default behaviour (True) is unchanged.
  • The docstring is updated. Its old rationale ("so that the DoubleMLScalar single-treatment check passes") didn't hold: the sub-models pass all checks without the extra columns.
• doubleml/plm/tests/test_plr_vector_vs_plr.py:
  • The comparison with the legacy DoubleMLPLR (coefficients, standard errors, per-repetition values, sensitivity elements) now runs for both values of use_other_treat_as_covariate.
  • A new test_submodel_covariates checks each sub-model's covariates.
  • The False cases (64 of 128) fail before the fix.

Reference to Issues or PRs

Fixes #442

Comments

This only affects the unreleased vector API from #394.

#441 (the fix for #440) also changes double_ml_vector.py and test_plr_vector_vs_plr.py, in different places. The two PRs merge cleanly in either order. With both applied, the benchmark test added in #441 also runs, and passes, for use_other_treat_as_covariate=False.

PR Checklist

• The title of the pull request summarizes the changes made.
• The PR contains a detailed description of all changes and additions.
• References to related issues or PRs are added.
• The code passes all (unit) tests.
• Enhancements or new feature are equipped with unit tests.
• The changes adhere to the PEP8 standards.