Use heterogeneous outcomes instead of noise injection in bootstrap test

Replace random noise approach with naturally heterogeneous outcomes
(different y_pre per unit) while keeping identical dose. This exercises
the rank-deficient bootstrap path with real sampling variance rather
than papering over zero-residual data.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: igerber

tests/test_methodology_continuous_did.py

@@ -191,32 +191,32 @@ def test_all_same_dose(self):
             atol=1e-10,
         )
 
-        # Verify bootstrap path also produces finite ATT SE (not NaN) for
-        # rank-deficient cells — regression test for P1 bootstrap fix.
-        # Use noisy outcomes so residuals are non-zero, giving the bootstrap
-        # actual variance to measure (identical outcomes → zero residuals →
-        # zero-variance bootstrap → SE=0 → NaN by design on all platforms).
+        # Verify bootstrap path produces finite ATT SE for rank-deficient
+        # cells — regression test for P1 bootstrap fix.  Use data with
+        # heterogeneous outcomes (natural sampling variance) but the same
+        # dose so the design matrix is still rank-deficient.
+        # ACRT SE is correctly NaN: zero dose variation → zero-variance
+        # bootstrap distribution → degenerate SE → NaN by design.
         rng = np.random.default_rng(123)
-        rows_noisy = []
+        rows_hetero = []
         for i in range(n_control):
-            noise = rng.normal(0, 0.1)
-            rows_noisy.append({"unit": i, "period": 1, "outcome": noise, "first_treat": 0, "dose": 0.0})
-            rows_noisy.append({"unit": i, "period": 2, "outcome": noise, "first_treat": 0, "dose": 0.0})
+            y_pre = rng.normal(0, 0.3)
+            y_post = rng.normal(0, 0.3)
+            rows_hetero.append({"unit": i, "period": 1, "outcome": y_pre, "first_treat": 0, "dose": 0.0})
+            rows_hetero.append({"unit": i, "period": 2, "outcome": y_post, "first_treat": 0, "dose": 0.0})
         for j in range(n_treated):
             uid = n_control + j
-            noise = rng.normal(0, 0.1)
-            rows_noisy.append({"unit": uid, "period": 1, "outcome": noise, "first_treat": 2, "dose": dose_val})
-            rows_noisy.append({"unit": uid, "period": 2, "outcome": 5.0 + noise, "first_treat": 2, "dose": dose_val})
-        data_noisy = pd.DataFrame(rows_noisy)
-        # ACRT SE is correctly NaN: zero dose variation → zero-variance
-        # bootstrap distribution → degenerate SE → NaN by design.
+            y_pre = rng.normal(0, 0.3)
+            rows_hetero.append({"unit": uid, "period": 1, "outcome": y_pre, "first_treat": 2, "dose": dose_val})
+            rows_hetero.append({"unit": uid, "period": 2, "outcome": y_pre + 5.0, "first_treat": 2, "dose": dose_val})
+        data_hetero = pd.DataFrame(rows_hetero)
         est_boot = ContinuousDiD(
             degree=1, num_knots=0, n_bootstrap=199,
             rank_deficient_action="silent", seed=42,
         )
         with pytest.warns(UserWarning, match="[Ii]dentical"):
             results_boot = est_boot.fit(
-                data_noisy, "outcome", "unit", "period", "first_treat", "dose"
+                data_hetero, "outcome", "unit", "period", "first_treat", "dose"
             )
         assert np.all(np.isfinite(results_boot.dose_response_att.se))