docs: clarify DiD predict support contract (#580)

Author: shawcharles

diff_diff/estimators.py

@@ -999,26 +999,39 @@ def _validate_data(
 
     def predict(self, data: pd.DataFrame) -> np.ndarray:
         """
-        Predict outcomes using fitted model.
+        Predict outcomes using the fitted model.
+
+        Out-of-sample prediction is intentionally unsupported pending a broader
+        post-estimation design for estimator result objects. For fitted
+        training-data predictions, use ``results_.fitted_values`` after
+        :meth:`fit`.
 
         Parameters
         ----------
         data : pd.DataFrame
-            DataFrame with same structure as training data.
+            Candidate prediction data. Currently unused because out-of-sample
+            prediction is unsupported.
 
         Returns
         -------
         np.ndarray
             Predicted values.
+
+        Raises
+        ------
+        RuntimeError
+            If called before :meth:`fit`.
+        NotImplementedError
+            Always raised after fitting until the broader post-estimation
+            prediction contract is designed.
         """
         if not self.is_fitted_:
             raise RuntimeError("Model must be fitted before calling predict()")
 
-        # This is a placeholder - would need to store column names
-        # for full implementation
         raise NotImplementedError(
-            "predict() is not yet implemented. "
-            "Use results_.fitted_values for training data predictions."
+            "out-of-sample predict() is unsupported pending a broader "
+            "post-estimation design. Use results_.fitted_values for fitted "
+            "training-data predictions."
         )
 
     def get_params(self) -> Dict[str, Any]:

docs/api/estimators.rst

@@ -30,6 +30,11 @@ DifferenceInDifferences (alias: ``DiD``)
 
 Basic 2x2 DiD estimator.
 
+``DifferenceInDifferences.predict()`` is present for sklearn-like
+discoverability, but out-of-sample prediction is not currently supported. Use
+``results_.fitted_values`` for fitted training-data predictions until a broader
+post-estimation result-object contract is designed.
+
 .. autoclass:: diff_diff.DifferenceInDifferences
    :no-index:
    :members:
@@ -42,6 +47,7 @@ Basic 2x2 DiD estimator.
    .. autosummary::
 
       ~DifferenceInDifferences.fit
+      ~DifferenceInDifferences.predict
       ~DifferenceInDifferences.get_params
       ~DifferenceInDifferences.set_params
 
@@ -84,4 +90,3 @@ Synthetic control combined with DiD (Arkhangelsky et al. 2021).
    :undoc-members:
    :show-inheritance:
    :inherited-members:
-

tests/test_methodology_did.py

@@ -1550,6 +1550,18 @@ def test_residuals_and_fitted_values(self):
         assert np.allclose(reconstructed, original), \
             "Residuals + fitted should equal original outcome"
 
+    def test_predict_contract_points_to_fitted_values(self):
+        """predict() is intentionally unsupported until post-estimation is designed."""
+        data, _ = generate_hand_calculable_data()
+
+        did = DifferenceInDifferences()
+        did.fit(data, outcome='outcome', treatment='treated', time='post')
+
+        with pytest.raises(
+            NotImplementedError,
+            match="out-of-sample.*post-estimation.*results_\\.fitted_values",
+        ):
+            did.predict(data)
 
 # =============================================================================
 # Multi-absorb (N>1 FE) iterative alternating-projection demeaning