Merge pull request #46 from igerber/claude/plan-v1.2.1-release-pCTqu

Author: igerber

CHANGELOG.md

@@ -5,6 +5,20 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [1.2.1] - 2026-01-08
+
+### Added
+- **Expanded test coverage** for edge cases:
+  - Wild bootstrap with very few clusters (< 5), including 2-3 cluster scenarios
+  - Unbalanced panels with missing periods across units
+  - Single treated unit scenarios for DiD and Synthetic DiD
+  - Perfect collinearity detection (validates clear error messages)
+  - CallawaySantAnna with single treatment cohort
+  - SyntheticDiD with insufficient pre-treatment periods
+
+### Changed
+- **Refactored CallawaySantAnna bootstrap**: Extracted `_compute_effect_bootstrap_stats()` helper method for cleaner code and reduced duplication in bootstrap statistics computation.
+
 ## [1.2.0] - 2026-01-07
 
 ### Added
@@ -234,6 +248,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
   - `to_dict()` and `to_dataframe()` export methods
   - `is_significant` and `significance_stars` properties
 
+[1.2.1]: https://github.com/igerber/diff-diff/compare/v1.2.0...v1.2.1
 [1.2.0]: https://github.com/igerber/diff-diff/compare/v1.1.1...v1.2.0
 [1.1.1]: https://github.com/igerber/diff-diff/compare/v1.1.0...v1.1.1
 [1.1.0]: https://github.com/igerber/diff-diff/compare/v1.0.2...v1.1.0

TODO.md

@@ -74,7 +74,7 @@ Edge cases needing tests:
 
 ## Documentation Improvements
 
-- [ ] Troubleshooting section for common errors
+- [x] Troubleshooting section for common errors (added in v1.0.0)
 - [ ] Comparison of estimator outputs on same data
 - [ ] Real-world data examples (currently synthetic only)
 - [ ] Performance benchmarks vs. R packages

diff_diff/__init__.py

@@ -98,7 +98,7 @@
     plot_sensitivity,
 )
 
-__version__ = "1.2.0"
+__version__ = "1.2.1"
 __all__ = [
     # Estimators
     "DifferenceInDifferences",

diff_diff/staggered.py

@@ -1603,30 +1603,25 @@ def _run_multiplier_bootstrap(
                         weights * bootstrap_atts_gt[b, gt_indices]
                     )
 
-        # Compute bootstrap statistics
-        # ATT(g,t) statistics
+        # Compute bootstrap statistics for ATT(g,t)
         gt_ses = {}
         gt_cis = {}
         gt_p_values = {}
 
         for j, gt in enumerate(gt_pairs):
-            original_effect = original_atts[j]
-            boot_dist = bootstrap_atts_gt[:, j]
-
-            se = float(np.std(boot_dist, ddof=1))
-            ci = self._compute_percentile_ci(boot_dist, self.alpha)
-            p_value = self._compute_bootstrap_pvalue(original_effect, boot_dist)
-
+            se, ci, p_value = self._compute_effect_bootstrap_stats(
+                original_atts[j], bootstrap_atts_gt[:, j]
+            )
             gt_ses[gt] = se
             gt_cis[gt] = ci
             gt_p_values[gt] = p_value
 
-        # Overall ATT statistics
-        overall_se = float(np.std(bootstrap_overall, ddof=1))
-        overall_ci = self._compute_percentile_ci(bootstrap_overall, self.alpha)
-        overall_p_value = self._compute_bootstrap_pvalue(original_overall, bootstrap_overall)
+        # Compute bootstrap statistics for overall ATT
+        overall_se, overall_ci, overall_p_value = self._compute_effect_bootstrap_stats(
+            original_overall, bootstrap_overall
+        )
 
-        # Event study statistics
+        # Compute bootstrap statistics for event study effects
         event_study_ses = None
         event_study_cis = None
         event_study_p_values = None
@@ -1637,16 +1632,14 @@ def _run_multiplier_bootstrap(
             event_study_p_values = {}
 
             for e in rel_periods:
-                original_effect = event_study_info[e]['effect']
-                boot_dist = bootstrap_event_study[e]
-
-                event_study_ses[e] = float(np.std(boot_dist, ddof=1))
-                event_study_cis[e] = self._compute_percentile_ci(boot_dist, self.alpha)
-                event_study_p_values[e] = self._compute_bootstrap_pvalue(
-                    original_effect, boot_dist
+                se, ci, p_value = self._compute_effect_bootstrap_stats(
+                    event_study_info[e]['effect'], bootstrap_event_study[e]
                 )
+                event_study_ses[e] = se
+                event_study_cis[e] = ci
+                event_study_p_values[e] = p_value
 
-        # Group effect statistics
+        # Compute bootstrap statistics for group effects
         group_effect_ses = None
         group_effect_cis = None
         group_effect_p_values = None
@@ -1657,14 +1650,12 @@ def _run_multiplier_bootstrap(
             group_effect_p_values = {}
 
             for g in groups:
-                original_effect = group_agg_info[g]['effect']
-                boot_dist = bootstrap_group[g]
-
-                group_effect_ses[g] = float(np.std(boot_dist, ddof=1))
-                group_effect_cis[g] = self._compute_percentile_ci(boot_dist, self.alpha)
-                group_effect_p_values[g] = self._compute_bootstrap_pvalue(
-                    original_effect, boot_dist
+                se, ci, p_value = self._compute_effect_bootstrap_stats(
+                    group_agg_info[g]['effect'], bootstrap_group[g]
                 )
+                group_effect_ses[g] = se
+                group_effect_cis[g] = ci
+                group_effect_p_values[g] = p_value
 
         return CSBootstrapResults(
             n_bootstrap=self.n_bootstrap,
@@ -1817,6 +1808,35 @@ def _compute_bootstrap_pvalue(
 
         return float(p_value)
 
+    def _compute_effect_bootstrap_stats(
+        self,
+        original_effect: float,
+        boot_dist: np.ndarray,
+    ) -> Tuple[float, Tuple[float, float], float]:
+        """
+        Compute bootstrap statistics for a single effect.
+
+        Parameters
+        ----------
+        original_effect : float
+            Original point estimate.
+        boot_dist : np.ndarray
+            Bootstrap distribution of the effect.
+
+        Returns
+        -------
+        se : float
+            Bootstrap standard error.
+        ci : Tuple[float, float]
+            Percentile confidence interval.
+        p_value : float
+            Bootstrap p-value.
+        """
+        se = float(np.std(boot_dist, ddof=1))
+        ci = self._compute_percentile_ci(boot_dist, self.alpha)
+        p_value = self._compute_bootstrap_pvalue(original_effect, boot_dist)
+        return se, ci, p_value
+
     def get_params(self) -> Dict[str, Any]:
         """Get estimator parameters (sklearn-compatible)."""
         return {

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "diff-diff"
-version = "1.2.0"
+version = "1.2.1"
 description = "A library for Difference-in-Differences causal inference analysis"
 readme = "README.md"
 license = "MIT"