causalis.scenarios.unconfoundedness.refutation.overlap.overlap_validation

Overlap diagnostics focused on positivity and propensity calibration.

Module Contents

Functions

run_overlap_diagnostics

Run overlap and calibration diagnostics for an estimated propensity model.

Data

__all__

API

causalis.scenarios.unconfoundedness.refutation.overlap.overlap_validation.run_overlap_diagnostics(data: causalis.dgp.causaldata.CausalData, estimate: causalis.data_contracts.causal_estimate.CausalEstimate, *, thresholds: Optional[Dict[str, float]] = None, n_bins: int = 10, use_hajek: Optional[bool] = None, return_summary: bool = True, auc_flip_margin: float = 0.05) → Dict[str, Any]

Run overlap and calibration diagnostics for an estimated propensity model.

The core overlap object is the propensity score

.. math::

    m(X) = \mathbb{P}(D=1 \mid X).

This diagnostic checks whether estimated propensities stay away from the
edges and whether the implied weights are stable. For example, ATE weights
use

.. math::

    w_i^{(1)} = 

rac{D_i}{m(X_i)}, \qquad w_i^{(0)} = rac{1-D_i}{1-m(X_i)},

so very small :math:`m(X_i)` or very large :math:`m(X_i)` can create large
leverage points. The report combines:

- edge mass near `0` and `1`,
- treated/control separation in propensity space (`KS`, `AUC`),
- effective sample size and tail diagnostics for weights,
- calibration summaries such as `ECE`, recalibration slope, and intercept.

Parameters
----------
data : CausalData
    Dataset used to fit the estimator.
estimate : CausalEstimate
    Effect estimate with ``diagnostic_data`` containing propensity-related
    arrays such as ``m_hat`` and ``d``.
thresholds : dict, optional
    Optional threshold overrides keyed by metric name.
n_bins : int, default 10
    Number of bins used for calibration summaries.
use_hajek : bool, optional
    Whether to evaluate normalized IPW identities. If omitted, the value is
    inferred from diagnostic metadata.
return_summary : bool, default True
    Include a compact tabular summary in the returned payload.
auc_flip_margin : float, default 0.05
    Margin around 0.5 used when flagging reversed treated/control ranking.

Returns
-------
Dict[str, Any]
    Diagnostic report containing edge-mass, calibration, weight-stability,
    and optional summary tables.

Raises
------
ValueError
    If required diagnostic arrays are missing or have incompatible shapes.

Examples
--------
>>> from sklearn.ensemble import RandomForestClassifier, RandomForestRegressor
>>> from causalis.dgp import obs_linear_26_dataset
>>> from causalis.scenarios.unconfoundedness.model import IRM
>>> data = obs_linear_26_dataset(
...     n=1000,
...     seed=3141,
...     include_oracle=False,
...     return_causal_data=True,
... )
>>> irm = IRM(
...     data=data,
...     ml_g=RandomForestRegressor(
...         n_estimators=200,
...         max_depth=6,
...         min_samples_leaf=5,
...         random_state=3141,
...     ),
...     ml_m=RandomForestClassifier(
...         n_estimators=200,
...         max_depth=6,
...         min_samples_leaf=5,
...         random_state=3141,
...     ),
...     n_folds=3,
...     random_state=3141,
... )
>>> estimate = irm.fit().estimate(score="ATE")
>>> report = run_overlap_diagnostics(data, estimate)
>>> report["summary"]  # doctest: +SKIP
>>> report["edge_mass"]  # doctest: +SKIP

causalis.scenarios.unconfoundedness.refutation.overlap.overlap_validation.__all__

[‘run_overlap_diagnostics’]