diff-diff 2.1.6__tar.gz → 2.1.7__tar.gz

{diff_diff-2.1.6 → diff_diff-2.1.7}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: diff-diff
-Version: 2.1.6
+Version: 2.1.7
 Classifier: Development Status :: 5 - Production/Stable
 Classifier: Intended Audience :: Science/Research
 Classifier: Operating System :: OS Independent

{diff_diff-2.1.6 → diff_diff-2.1.7}/diff_diff/__init__.py

@@ -136,7 +136,7 @@ from diff_diff.datasets import (
     load_mpdta,
 )
 
-__version__ = "2.1.6"
+__version__ = "2.1.7"
 __all__ = [
     # Estimators
     "DifferenceInDifferences",

{diff_diff-2.1.6 → diff_diff-2.1.7}/diff_diff/visualization.py

@@ -73,8 +73,10 @@ def plot_event_study(
     periods : list, optional
         List of periods to plot. If None, uses all periods from results.
     reference_period : any, optional
-        The reference period (normalized to effect=0). Will be shown as a
-        hollow marker. If None, tries to infer from results.
+        The reference period to highlight. When explicitly provided, effects
+        are normalized (ref effect subtracted) and ref SE is set to NaN.
+        When None and auto-inferred from results, only hollow marker styling
+        is applied (no normalization). If None, tries to infer from results.
     pre_periods : list, optional
         List of pre-treatment periods. Used for shading.
     post_periods : list, optional
@@ -151,8 +153,9 @@ def plot_event_study(
        trends holds. Large pre-treatment effects suggest the assumption may
        be violated.
 
-    2. **Reference period**: Usually the last pre-treatment period (t=-1),
-       normalized to zero. This is the omitted category.
+    2. **Reference period**: Usually the last pre-treatment period (t=-1).
+       When explicitly specified via ``reference_period``, effects are normalized
+       to zero at this period. When auto-inferred, shown with hollow marker only.
 
     3. **Post-treatment periods**: The treatment effects of interest. These
        show how the outcome evolved after treatment.
@@ -170,10 +173,18 @@ def plot_event_study(
 
     from scipy import stats as scipy_stats
 
+    # Track if reference_period was explicitly provided by user
+    reference_period_explicit = reference_period is not None
+
     # Extract data from results if provided
     if results is not None:
-        effects, se, periods, pre_periods, post_periods, reference_period = \
-            _extract_plot_data(results, periods, pre_periods, post_periods, reference_period)
+        extracted = _extract_plot_data(
+            results, periods, pre_periods, post_periods, reference_period
+        )
+        effects, se, periods, pre_periods, post_periods, reference_period, reference_inferred = extracted
+        # If reference was inferred from results, it was NOT explicitly provided
+        if reference_inferred:
+            reference_period_explicit = False
     elif effects is None or se is None:
         raise ValueError(
             "Must provide either 'results' or both 'effects' and 'se'"
@@ -192,6 +203,19 @@ def plot_event_study(
     # Compute confidence intervals
     critical_value = scipy_stats.norm.ppf(1 - alpha / 2)
 
+    # Normalize effects to reference period ONLY if explicitly specified by user
+    # Auto-inferred reference periods (from CallawaySantAnna) just get hollow marker styling,
+    # NO normalization. This prevents unintended normalization when the reference period
+    # isn't a true identifying constraint (e.g., CallawaySantAnna with base_period="varying").
+    if (reference_period is not None and reference_period in effects and
+            reference_period_explicit):
+        ref_effect = effects[reference_period]
+        if np.isfinite(ref_effect):
+            effects = {p: e - ref_effect for p, e in effects.items()}
+            # Set reference SE to NaN (it's now a constraint, not an estimate)
+            # This follows fixest convention where the omitted category has no SE/CI
+            se = {p: (np.nan if p == reference_period else s) for p, s in se.items()}
+
     plot_data = []
     for period in periods:
         effect = effects.get(period, np.nan)
@@ -304,14 +328,17 @@ def _extract_plot_data(
     pre_periods: Optional[List[Any]],
     post_periods: Optional[List[Any]],
     reference_period: Optional[Any],
-) -> Tuple[Dict, Dict, List, List, List, Any]:
+) -> Tuple[Dict, Dict, List, List, List, Any, bool]:
     """
     Extract plotting data from various result types.
 
     Returns
     -------
     tuple
-        (effects, se, periods, pre_periods, post_periods, reference_period)
+        (effects, se, periods, pre_periods, post_periods, reference_period, reference_inferred)
+
+        reference_inferred is True if reference_period was auto-detected from results
+        rather than explicitly provided by the user.
     """
     # Handle DataFrame input
     if isinstance(results, pd.DataFrame):
@@ -328,7 +355,8 @@ def _extract_plot_data(
         if periods is None:
             periods = list(results['period'])
 
-        return effects, se, periods, pre_periods, post_periods, reference_period
+        # DataFrame input: reference_period was already set by caller, never inferred here
+        return effects, se, periods, pre_periods, post_periods, reference_period, False
 
     # Handle MultiPeriodDiDResults
     if hasattr(results, 'period_effects'):
@@ -348,7 +376,8 @@ def _extract_plot_data(
         if periods is None:
             periods = post_periods
 
-        return effects, se, periods, pre_periods, post_periods, reference_period
+        # MultiPeriodDiDResults: reference_period was already set by caller, never inferred here
+        return effects, se, periods, pre_periods, post_periods, reference_period, False
 
     # Handle CallawaySantAnnaResults (event study aggregation)
     if hasattr(results, 'event_study_effects') and results.event_study_effects is not None:
@@ -362,8 +391,12 @@ def _extract_plot_data(
         if periods is None:
             periods = sorted(effects.keys())
 
+        # Track if reference_period was explicitly provided vs auto-inferred
+        reference_inferred = False
+
         # Reference period is typically -1 for event study
         if reference_period is None:
+            reference_inferred = True  # We're about to infer it
             # Detect reference period from n_groups=0 marker (normalization constraint)
             # This handles anticipation > 0 where reference is at e = -1 - anticipation
             for period, effect_data in results.event_study_effects.items():
@@ -380,7 +413,7 @@ def _extract_plot_data(
         if post_periods is None:
             post_periods = [p for p in periods if p >= 0]
 
-        return effects, se, periods, pre_periods, post_periods, reference_period
+        return effects, se, periods, pre_periods, post_periods, reference_period, reference_inferred
 
     raise TypeError(
         f"Cannot extract plot data from {type(results).__name__}. "

{diff_diff-2.1.6 → diff_diff-2.1.7}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "maturin"
 
 [project]
 name = "diff-diff"
-version = "2.1.6"
+version = "2.1.7"
 description = "A library for Difference-in-Differences causal inference analysis"
 readme = "README.md"
 license = "MIT"

{diff_diff-2.1.6 → diff_diff-2.1.7}/rust/Cargo.lock

@@ -289,7 +289,7 @@ dependencies = [
 
 [[package]]
 name = "diff_diff_rust"
-version = "2.1.6"
+version = "2.1.7"
 dependencies = [
  "ndarray",
  "ndarray-linalg",
@@ -1846,9 +1846,9 @@ checksum = "b6c140620e7ffbb22c2dee59cafe6084a59b5ffc27a8859a5f0d494b5d52b6be"
 
 [[package]]
 name = "uuid"
-version = "1.19.0"
+version = "1.20.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "e2e054861b4bd027cd373e18e8d8d8e6548085000e41290d95ce0c373a654b4a"
+checksum = "ee48d38b119b0cd71fe4141b30f5ba9c7c5d9f4e7a3a8b4a674e4b6ef789976f"
 dependencies = [
  "getrandom 0.3.4",
  "js-sys",

{diff_diff-2.1.6 → diff_diff-2.1.7}/rust/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "diff_diff_rust"
-version = "2.1.6"
+version = "2.1.7"
 edition = "2021"
 description = "Rust backend for diff-diff DiD library"
 license = "MIT"