Add bootstrap_weights tests for event-study/group aggregation paths, update docstrings for PR #165 round 5

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

Author: igerber

diff_diff/imputation_bootstrap.py

@@ -124,7 +124,8 @@ def _precompute_bootstrap_psi(
 
         For each aggregation target (overall, per-horizon, per-group), computes
         psi_i = sum_t v_it * epsilon_tilde_it for each cluster. The multiplier
-        bootstrap then perturbs these psi sums with Rademacher weights.
+        bootstrap then perturbs these psi sums with multiplier weights
+        (rademacher/mammen/webb; configurable via ``bootstrap_weights``).
 
         Computational cost scales with the number of aggregation targets, since
         each target requires its own v_untreated computation (weight-dependent).
@@ -218,7 +219,8 @@ def _run_bootstrap(
         """
         Run multiplier bootstrap on pre-computed influence function sums.
 
-        Uses T_b = sum_i w_b_i * psi_i where w_b_i are Rademacher weights
+        Uses T_b = sum_i w_b_i * psi_i where w_b_i are multiplier weights
+        (rademacher/mammen/webb; configurable via ``bootstrap_weights``)
         and psi_i are cluster-level influence function sums from Theorem 3.
         SE = std(T_b, ddof=1).
         """

tests/test_imputation.py

@@ -1119,6 +1119,46 @@ def test_bootstrap_weights_webb(self, ci_params):
         assert br.overall_att_se > 0
         assert np.isfinite(br.overall_att_p_value)
 
+    def test_bootstrap_weights_event_study(self, ci_params):
+        """Bootstrap with non-default weights should work for event study aggregation."""
+        data = generate_test_data()
+        n_boot = ci_params.bootstrap(50)
+        est = ImputationDiD(
+            n_bootstrap=n_boot, bootstrap_weights="mammen", seed=42
+        )
+        results = est.fit(
+            data, outcome="outcome", unit="unit", time="time",
+            first_treat="first_treat", aggregate="event_study",
+        )
+
+        br = results.bootstrap_results
+        assert br is not None
+        assert br.weight_type == "mammen"
+        assert br.event_study_ses is not None
+        assert len(br.event_study_ses) > 0
+        for h, se in br.event_study_ses.items():
+            assert se > 0, f"Non-positive SE at horizon {h}"
+
+    def test_bootstrap_weights_group(self, ci_params):
+        """Bootstrap with non-default weights should work for group aggregation."""
+        data = generate_test_data()
+        n_boot = ci_params.bootstrap(50)
+        est = ImputationDiD(
+            n_bootstrap=n_boot, bootstrap_weights="mammen", seed=42
+        )
+        results = est.fit(
+            data, outcome="outcome", unit="unit", time="time",
+            first_treat="first_treat", aggregate="group",
+        )
+
+        br = results.bootstrap_results
+        assert br is not None
+        assert br.weight_type == "mammen"
+        assert br.group_ses is not None
+        assert len(br.group_ses) > 0
+        for g, se in br.group_ses.items():
+            assert se > 0, f"Non-positive SE for group {g}"
+
     def test_bootstrap_with_covariates(self, ci_params):
         """Bootstrap should work with covariates."""
         data = generate_test_data()

tests/test_two_stage.py

@@ -1092,6 +1092,44 @@ def test_bootstrap_weights_webb(self, ci_params):
         assert br.overall_att_se > 0
         assert np.isfinite(br.overall_att_p_value)
 
+    def test_bootstrap_weights_event_study(self, ci_params):
+        """Bootstrap with non-default weights should work for event study aggregation."""
+        data = generate_test_data()
+        n_boot = ci_params.bootstrap(50)
+        results = TwoStageDiD(
+            n_bootstrap=n_boot, bootstrap_weights="mammen", seed=42
+        ).fit(
+            data, outcome="outcome", unit="unit", time="time",
+            first_treat="first_treat", aggregate="event_study",
+        )
+
+        br = results.bootstrap_results
+        assert br is not None
+        assert br.weight_type == "mammen"
+        assert br.event_study_ses is not None
+        assert len(br.event_study_ses) > 0
+        for h, se in br.event_study_ses.items():
+            assert se > 0, f"Non-positive SE at horizon {h}"
+
+    def test_bootstrap_weights_group(self, ci_params):
+        """Bootstrap with non-default weights should work for group aggregation."""
+        data = generate_test_data()
+        n_boot = ci_params.bootstrap(50)
+        results = TwoStageDiD(
+            n_bootstrap=n_boot, bootstrap_weights="mammen", seed=42
+        ).fit(
+            data, outcome="outcome", unit="unit", time="time",
+            first_treat="first_treat", aggregate="group",
+        )
+
+        br = results.bootstrap_results
+        assert br is not None
+        assert br.weight_type == "mammen"
+        assert br.group_ses is not None
+        assert len(br.group_ses) > 0
+        for g, se in br.group_ses.items():
+            assert se > 0, f"Non-positive SE for group {g}"
+
 
 # =============================================================================
 # TestTwoStageDiDConvenience