[Repo Assist] fix: remove invalid [0] subscript on np.std() scalar in kappa auto-inference

dowhy/causal_refuters/add_unobserved_common_cause.py

@@ -266,7 +266,7 @@ def _infer_default_kappa_t(
     elif effect_on_t == "linear":
         # Estimating the regression coefficient from standardized features to t
         corrcoef_var_t = np.corrcoef(observed_common_causes, t, rowvar=False)[-1, :-1]
-        std_dev_t = np.std(t)[0]
+        std_dev_t = float(np.std(t))
         max_coeff = max(corrcoef_var_t) * std_dev_t
         min_coeff = min(corrcoef_var_t) * std_dev_t
     else:
@@ -318,7 +318,7 @@ def _infer_default_kappa_y(
         min_coeff, max_coeff = min(flips), max(flips)
     elif effect_on_y == "linear":
         corrcoef_var_y = np.corrcoef(observed_common_causes, y, rowvar=False)[-1, :-1]
-        std_dev_y = np.std(y)[0]
+        std_dev_y = float(np.std(y))
         max_coeff = max(corrcoef_var_y) * std_dev_y
         min_coeff = min(corrcoef_var_y) * std_dev_y
     else:

tests/causal_refuters/test_add_unobserved_common_cause.py

@@ -598,3 +598,33 @@ def test_evalue_logistic_regression(self, mock_fig, estimator_method):
         assert refute.stats["evalue_upper_ci"] is None
         assert refute.stats["evalue_lower_ci"] < refute.stats["evalue_estimate"]
         assert mock_fig.call_count > 0
+
+    def test_infer_default_kappa_linear_does_not_crash(self):
+        """Regression test: _infer_default_kappa_t/y crashed with IndexError when effect strengths
+        were omitted and confounders_effect_on_treatment/outcome were 'linear' because np.std()
+        returns a scalar that cannot be indexed with [0]."""
+        data = dowhy.datasets.linear_dataset(
+            beta=10,
+            num_common_causes=2,
+            num_samples=500,
+            treatment_is_binary=False,
+        )
+        model = CausalModel(
+            data=data["df"],
+            treatment=data["treatment_name"],
+            outcome=data["outcome_name"],
+            graph=data["gml_graph"],
+        )
+        identified_estimand = model.identify_effect()
+        estimate = model.estimate_effect(identified_estimand, method_name="backdoor.linear_regression")
+        # No effect_strength_on_treatment / effect_strength_on_outcome supplied — forces
+        # _infer_default_kappa_t and _infer_default_kappa_y to run the "linear" branch.
+        refute = model.refute_estimate(
+            identified_estimand,
+            estimate,
+            method_name="add_unobserved_common_cause",
+            confounders_effect_on_treatment="linear",
+            confounders_effect_on_outcome="linear",
+            simulation_method="direct-simulation",
+        )
+        assert refute is not None