Horvitz-Thompson & Hajek IPW estimators

causallib/estimation/base_weight.py

@@ -103,21 +103,22 @@ def compute_weight_matrix(self, X, a, use_stabilized=None, **kwargs):
         raise NotImplementedError
 
     @staticmethod
-    def _compute_stratified_weighted_aggregate(y, sample_weight=None,
+    def _compute_stratified_weighted_aggregate(y, sample_weight=None, normalize=True,
                                                stratify_by=None, treatment_values=None):
         """
         Calculates aggregation of `y` weighted by `sample_weights` stratified by `stratify_by` variable.
 
         Args:
             y (pd.Series): The variable to aggregate (num_subjects,).
             sample_weight (pd.Series|None): Individual (sample) weights calculated.
-                                            Used to achieved unbiased average outcome.
-                                            If not provided, gives equal weights to every example.
+                                          Used to achieved unbiased average outcome.
+                                          If not provided, gives equal weights to every example.
+            normalize (bool): Whether to normalize the weights to sum to 1 within each strata. 
             stratify_by (pd.Series|None): Categorical variable to stratify according to (num_subjects,).
                                           Namely, aggregate within subgroups sharing the same values.
                                           If not provided, the aggregation is on the entire
             treatment_values (Any): Subset of values to stratify on from `stratify_by`.
-                                    If not supplied, all available stratification values are used.
+                                          If not supplied, all available stratification values are used.
 
         Returns:
             pd.Series[Any, float]: Series which index are treatment values, and the values are numbers - the
@@ -133,11 +134,16 @@ def _compute_stratified_weighted_aggregate(y, sample_weight=None,
         res = {}
         for treatment_value in treatment_values:
             subgroup_mask = stratify_by == treatment_value
-            aggregated_value = np.average(y[subgroup_mask], weights=sample_weight[subgroup_mask])
+            if normalize:
+                aggregated_value = np.average(y[subgroup_mask], weights=sample_weight[subgroup_mask])
+            else:
+                aggregated_value = np.sum(y[subgroup_mask] * sample_weight[subgroup_mask])/len(y)
+
             res[treatment_value] = aggregated_value
         res = pd.Series(res)
         return res
 
+
     def evaluate_balancing(self, X, a, y, w):
         pass  # TODO: implement: (1) table one with smd (2) gather lots of metric (ks, kl, smd) (3) plot CDF of each feature
 

causallib/estimation/ipw.py

@@ -201,14 +201,17 @@ def compute_propensity_matrix(self, X, a=None, clip_min=None, clip_max=None):
 
         return probabilities
 
-    def estimate_population_outcome(self, X, a, y, w=None, treatment_values=None):
+    def estimate_population_outcome(self, X, a, y, ipw_estimator, w=None, treatment_values=None):
         """
         Calculates weighted population outcome for each subgroup stratified by treatment assignment.
 
         Args:
             X (pd.DataFrame): Covariate matrix of size (num_subjects, num_features).
             a (pd.Series): Treatment assignment of size (num_subjects,).
             y (pd.Series): Observed outcome of size (num_subjects,).
+            ipw_estimator (str): The type of estimator to use for the inverse probability weighting.
+                                   Horvitz-Thompson estimator is unbiased but high variance, while Hajek estimator
+                                   is biased but lower variance.
             w (pd.Series | None): Individual (sample) weights calculated. Used to achieved unbiased average outcome.
                                    If not provided, will be calculated on the data.
             treatment_values (Any): Desired treatment value/s to stratify upon.
@@ -221,8 +224,14 @@ def estimate_population_outcome(self, X, a, y, w=None, treatment_values=None):
         """
         if w is None:
             w = self.compute_weights(X, a)
-        res = self._compute_stratified_weighted_aggregate(y, sample_weight=w, stratify_by=a,
+        if ipw_estimator == 'Horvitz-Thompson':
+            res = self._compute_stratified_weighted_aggregate(y, sample_weight=w, stratify_by=a, normalize=False, treatment_values=treatment_values)
+        elif ipw_estimator == 'Hajek':
+            res = self._compute_stratified_weighted_aggregate(y, sample_weight=w, normalize=True, stratify_by=a,
                                                           treatment_values=treatment_values)
+        else:
+            raise ValueError(f"Unknown ipw_estimator: {ipw_estimator}, not implemented in the current package version.")
+
         return res
 
     @staticmethod