diffindiff 2.3.5__tar.gz → 2.3.7__tar.gz

{diffindiff-2.3.5 → diffindiff-2.3.7}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: diffindiff
-Version: 2.3.5
+Version: 2.3.7
 Summary: diffindiff: Python library for convenient Difference-in-Differences analyses
 Author: Thomas Wieland
 Author-email: geowieland@googlemail.com
@@ -27,7 +27,7 @@ Thomas Wieland [ORCID](https://orcid.org/0000-0001-5168-9846) [EMail](mailto:geo
 
 If you use this software, please cite:
 
-Wieland, T. (2026). diffindiff: A Python library for convenient difference-in-differences analyses (Version 2.3.5) [Computer software]. Zenodo. https://doi.org/10.5281/zenodo.18656820
+Wieland, T. (2026). diffindiff: A Python library for convenient difference-in-differences analyses (Version 2.3.7) [Computer software]. Zenodo. https://doi.org/10.5281/zenodo.18656820
 
 
 ## Installation
@@ -173,10 +173,15 @@ See the /tests directory for usage examples of most of the included functions.
 This software was developed without the use of AI-generated code. The Continue Agent in Microsoft Visual Studio Code using the GPT-5 mini model (by OpenAI) was used solely to assist in drafting and refining docstrings for documentation. The corresponding guidelines and constraints defined by the author are documented in `AGENTS-docstrings.md` in the [public GitHub repository](https://github.com/geowieland/diffindiff_official).
 
 
-## What's new (v2.3.5)
+## What's new (v2.3.7)
 
+- Extensions:  
+  - DiffModel.counterfactual() and DiffModel.plot_counterfactual(): Re-transform log-transformed outcome via parameter retransform_log_outcome
+  - Difference between predicted and counterfactual-predicted in result df of DiffModel.counterfactual()  
+  - Extended parameters in didtools.model_wrapper() to be passed to sklearn.model_selection.train_test_split()
+  - Extension of documentations
 - Bugfixes:
-  - Test whether input data is panel data via didtools.is_panel() which is included in didanalysis_helper.data_diagnostics()
-  - Fixed false test results given continuous treatments are accepted in didtools.is_simultaneous()
-  - Fixed false test results given continuous treatments are accepted in didtools.is_prepost()
-  - Argument 'pre_post' is passed to is_simultaneous() in didanalysis_helper.treatment_diagnostics()
+  - diddata.create_counterfactual(): Fixed calculation error, additional check of input parameters
+  - didtools.is_prepost() now also checks number of observational units
+  - didtools.is_notreatment() now checks whether data is singular with respect to 0 and 1  
+  - didtools.fit_metrics() tries to convert observed and expected if not numeric before raising TypeError

{diffindiff-2.3.5 → diffindiff-2.3.7}/README.md

@@ -19,7 +19,7 @@ Thomas Wieland [ORCID](https://orcid.org/0000-0001-5168-9846) [EMail](mailto:geo
 
 If you use this software, please cite:
 
-Wieland, T. (2026). diffindiff: A Python library for convenient difference-in-differences analyses (Version 2.3.5) [Computer software]. Zenodo. https://doi.org/10.5281/zenodo.18656820
+Wieland, T. (2026). diffindiff: A Python library for convenient difference-in-differences analyses (Version 2.3.7) [Computer software]. Zenodo. https://doi.org/10.5281/zenodo.18656820
 
 
 ## Installation
@@ -165,10 +165,15 @@ See the /tests directory for usage examples of most of the included functions.
 This software was developed without the use of AI-generated code. The Continue Agent in Microsoft Visual Studio Code using the GPT-5 mini model (by OpenAI) was used solely to assist in drafting and refining docstrings for documentation. The corresponding guidelines and constraints defined by the author are documented in `AGENTS-docstrings.md` in the [public GitHub repository](https://github.com/geowieland/diffindiff_official).
 
 
-## What's new (v2.3.5)
+## What's new (v2.3.7)
 
+- Extensions:  
+  - DiffModel.counterfactual() and DiffModel.plot_counterfactual(): Re-transform log-transformed outcome via parameter retransform_log_outcome
+  - Difference between predicted and counterfactual-predicted in result df of DiffModel.counterfactual()  
+  - Extended parameters in didtools.model_wrapper() to be passed to sklearn.model_selection.train_test_split()
+  - Extension of documentations
 - Bugfixes:
-  - Test whether input data is panel data via didtools.is_panel() which is included in didanalysis_helper.data_diagnostics()
-  - Fixed false test results given continuous treatments are accepted in didtools.is_simultaneous()
-  - Fixed false test results given continuous treatments are accepted in didtools.is_prepost()
-  - Argument 'pre_post' is passed to is_simultaneous() in didanalysis_helper.treatment_diagnostics()
+  - diddata.create_counterfactual(): Fixed calculation error, additional check of input parameters
+  - didtools.is_prepost() now also checks number of observational units
+  - didtools.is_notreatment() now checks whether data is singular with respect to 0 and 1  
+  - didtools.fit_metrics() tries to convert observed and expected if not numeric before raising TypeError

{diffindiff-2.3.5 → diffindiff-2.3.7}/diffindiff/config.py

@@ -4,15 +4,15 @@
 # Author:      Thomas Wieland 
 #              ORCID: 0000-0001-5168-9846
 #              mail: geowieland@googlemail.com              
-# Version:     1.0.13
-# Last update: 2026-03-16 17:54
+# Version:     1.0.15
+# Last update: 2026-03-24 18:04
 # Copyright (c) 2025-2026 Thomas Wieland
 #-----------------------------------------------------------------------
 
 # Basic config:
 
 PACKAGE_NAME = "diffindiff"
-PACKAGE_VERSION = "2.3.5"
+PACKAGE_VERSION = "2.3.7"
 
 VERBOSE = False
 
@@ -29,11 +29,12 @@ DID_DESCRIPTION = "Difference-in-Differences Analysis"
 DDD_DESCRIPTION = "Triple-Difference Analysis"
 
 TREATMENT_DESCRIPTION = "Treatment"
+CONTROL_DESCRIPTION = "Control"
 
 GROUP_DESCRIPTION = "Group"
 
 TREATMENT_GROUP_DESCRIPTION = f"{TREATMENT_DESCRIPTION} {GROUP_DESCRIPTION}"
-CONTROL_GROUP_DESCRIPTION = f"Control {GROUP_DESCRIPTION}"
+CONTROL_GROUP_DESCRIPTION = f"{CONTROL_DESCRIPTION} {GROUP_DESCRIPTION}"
 GROUPS_DESCRIPTION = f"{TREATMENT_DESCRIPTION} and {CONTROL_GROUP_DESCRIPTION}"
 
 TIME_PERIODS_DESCRIPTION = "Time periods"
@@ -96,6 +97,7 @@ PI_LOWER_SUFFIX = "PI_lower"
 PI_UPPER_SUFFIX = "PI_upper"
 SPILLOVER_PREFIX = "Spillover"
 SPILLOVER_UNIT_PREFIX = f"{SPILLOVER_PREFIX}{DELIMITER}{UNIT_COL}"
+DIFFERENCE_SUFFIX = "diff"
 
 # Modeling config:
 
@@ -362,6 +364,21 @@ MODEL_FIT_METRICS = {
         }
 }
 
+# Machine learning models:
+MODEL_WRAPPER_AVAILABLE = {
+    "ols": "Ordinary Least Squares",
+    "olsbg": "Ordinary Least Squares with Bagging",
+    "dtbg": "Decision Trees with Bagging",
+    "rf": "Random Forest",
+    "gb": "Gradient Boosted Trees",
+    "knn": "K-nearest neighbor",
+    "xgb": "Extreme Gradient Boosting",
+    "lgbm": "LightGBM",
+    "svr": "Support-vector regression",
+}
+MODEL_WRAPPER_AVAILABLE_LIST = list(MODEL_WRAPPER_AVAILABLE.keys())
+
+
 # Treatment diagnostics:
 
 TREATMENT_DIAGNOSTICS_DESCRIPTION = f"{TREATMENT_DESCRIPTION} diagnostics"

{diffindiff-2.3.5 → diffindiff-2.3.7}/diffindiff/didanalysis.py

@@ -4,8 +4,8 @@
 # Author:      Thomas Wieland 
 #              ORCID: 0000-0001-5168-9846
 #              mail: geowieland@googlemail.com              
-# Version:     2.3.4
-# Last update: 2026-03-16 17:39
+# Version:     2.3.7
+# Last update: 2026-03-24 18:37
 # Copyright (c) 2024-2026 Thomas Wieland
 #-----------------------------------------------------------------------
 
@@ -812,13 +812,13 @@ class DiffModel:
         self,
         colors = ["blue", "grey"],
         colors_by_signficance = ["red", "coral", "dimgray", "silver", "green", "palegreen"],
-        point_type = "s",
-        point_size = 8,
-        line_width = 6,
-        line_cap_size = 5, 
-        x_label = "Estimates with confidence intervals",
-        y_label = "Coefficient",
-        plot_title = "DiD effects",
+        point_type: str = "s",
+        point_size: int = 8,
+        line_width: int = 6,
+        line_cap_size: int = 5, 
+        x_label: str = "Estimates with confidence intervals",
+        y_label: str = "Coefficient",
+        plot_title: str = "DiD effects",
         plot_grid: bool = True,
         sort_by_coef: bool = False,
         sort_ascending: bool = True,
@@ -1042,11 +1042,13 @@ class DiffModel:
     def counterfactual(
         self,
         treatment = None,
-        after_treatment_col: str = None
+        after_treatment_col: str = None,
+        retransform_log_outcome: bool = False
         ):
         
         """
-        Compute counterfactual predictions for the treatment group by setting treatment to zero.
+        Compute counterfactual predictions for the treatment group by 
+        setting treatment to zero.
 
         Parameters
         ----------
@@ -1054,11 +1056,14 @@ class DiffModel:
             Treatment name to analyse. If None, the first treatment is chosen.
         after_treatment_col : str, optional
             After-treatment indicator column name.
+        retransform_log_outcome : bool, optional
+            If outcome was log-transformed, retransform to original scale for plotting. 
+            Default is False.
 
         Returns
         -------
         list
-            [modified_model_data (DataFrame), outcome_pred_col (str), outcome_pred_cf_col (str)]
+            [modified_model_data (DataFrame), outcome_pred_col (str), outcome_pred_cf_col (str), outcome_diff_col (str)]
 
         Raises
         ------
@@ -1106,7 +1111,7 @@ class DiffModel:
         didmodel = self.didmodel()
 
         predictions = self.predictions()
-        
+
         model_data = self.data[2]
         
         model_config = self.data[1]
@@ -1118,20 +1123,58 @@ class DiffModel:
 
         predictions_counterfac = didmodel.get_prediction(model_data_mod).summary_frame()
 
-        outcome_pred_col = f"{outcome_col}{config.PREDICTED_SUFFIX}"
+        outcome_pred_col = f"{outcome_col}{config.DELIMITER}{config.PREDICTED_SUFFIX}"
         if outcome_pred_col in model_data_mod.columns:
             outcome_pred_col = f"{config.DELIMITER}{outcome_pred_col}"
-        outcome_pred_cf_col = f"{outcome_col}{config.COUNTERFAC_SUFFIX_PRED_CF}"
+            
+        outcome_pred_cf_col = f"{outcome_col}{config.DELIMITER}{config.COUNTERFAC_SUFFIX_PRED_CF}"
         if outcome_pred_cf_col in model_data_mod.columns:
-            outcome_pred_cf_col = f"{config.DELIMITER}{outcome_pred_cf_col}"
+            outcome_pred_cf_col = f"{config.DELIMITER}{outcome_pred_cf_col}"        
 
         model_data_mod[outcome_pred_col] = predictions[config.PREDICTIONS_SUMMARY_FRAME_COLS_LIST[0]]
         model_data_mod[outcome_pred_cf_col] = predictions_counterfac[config.PREDICTIONS_SUMMARY_FRAME_COLS_LIST[0]]
+        
+        if retransform_log_outcome:
+            
+            if model_config["log_outcome"]:
+                
+                if outcome_col.startswith(f"{config.LOG_PREFIX}{config.DELIMITER}"):
+                    outcome_col = outcome_col[len(f"{config.LOG_PREFIX}{config.DELIMITER}"):]
+
+                if outcome_pred_col.startswith(f"{config.LOG_PREFIX}{config.DELIMITER}"):
+                    
+                    model_data_mod = model_data_mod.rename(
+                        columns = {
+                            outcome_pred_col: outcome_pred_col[len(f"{config.LOG_PREFIX}{config.DELIMITER}"):]
+                        }
+                    )
+                    
+                    outcome_pred_col = outcome_pred_col[len(f"{config.LOG_PREFIX}{config.DELIMITER}"):]
+
+                if outcome_pred_cf_col.startswith(f"{config.LOG_PREFIX}{config.DELIMITER}"):                    
+                    
+                    model_data_mod = model_data_mod.rename(
+                        columns = {
+                            outcome_pred_cf_col: outcome_pred_cf_col[len(f"{config.LOG_PREFIX}{config.DELIMITER}"):]
+                        }
+                    )
+                                        
+                    outcome_pred_cf_col = outcome_pred_cf_col[len(f"{config.LOG_PREFIX}{config.DELIMITER}"):]   
+
+                model_data_mod[outcome_pred_col] = np.exp(model_data_mod[outcome_pred_col])
+                model_data_mod[outcome_pred_cf_col] = np.exp(model_data_mod[outcome_pred_cf_col])                
+        
+        outcome_diff_col = f"{outcome_col}{config.DELIMITER}{config.DIFFERENCE_SUFFIX}"
+        if outcome_diff_col in model_data_mod.columns:
+            outcome_diff_col = f"{config.DELIMITER}{outcome_diff_col}"
+            
+        model_data_mod[outcome_diff_col] = model_data_mod[outcome_pred_col]-model_data_mod[outcome_pred_cf_col]
 
         return [
             model_data_mod,
             outcome_pred_col,
-            outcome_pred_cf_col
+            outcome_pred_cf_col,
+            outcome_diff_col
         ]
 
     def didmodel(self):
@@ -1292,11 +1335,13 @@ class DiffModel:
         TT_col_ = f"{config.TT_COL}{config.DELIMITER}{treatment}"
         TGxTT_ = f"Placebo{config.DELIMITER}{treatment}"        
         
+        treatment_col_errors = []
         if TG_col is None and TG_col_ not in model_config["TG_col"]:
-            raise ValueError(f"No treatment group identification variable for treatment {treatment}. Please state TG_col = your_treatment_group_dummy.")
-        
+            treatment_col_errors.append(f"No treatment group identification variable for treatment '{treatment}'. State parameter 'TG_col' = <<your_treatment_group_dummy>> (e.g., 'TG_{treatment}').")        
         if TT_col is None and TT_col_ not in model_config["TT_col"]:
-            raise ValueError(f"No treatment time variable for treatment {treatment}. Please state TG_col = your_treatment_time_dummy.")
+            treatment_col_errors.append(f"No treatment time variable for treatment '{treatment}'. State parameter 'TT_col' = <<your_treatment_time_dummy>> (e.g., 'TT_{treatment}.)")
+        if len(treatment_col_errors) > 0:
+            raise ValueError(f"Missing arguments in placebo analysis: {' '.join(treatment_col_errors)}")        
         
         if TG_col is not None:
             TG_col_ = TG_col
@@ -1327,6 +1372,8 @@ class DiffModel:
 
         model_data_c[TG_col_] = 0
         model_data_c.loc[(model_data_c[unit_col].isin(units_random_sample)), TG_col_] = 1
+        model_data_c[TT_col_] = 0
+        model_data_c.loc[((model_data_c[time_col] >= treatment_period_start) & (model_data_c[time_col] <= treatment_period_end)), TT_col_] = 1
         model_data_c[TGxTT_] = model_data_c[TG_col_] * model_data_c[TT_col_]
 
         model_data_c_analysis = did_analysis(
@@ -1364,7 +1411,7 @@ class DiffModel:
         y_lim = None,
         plot_title: str = "Treatment time",
         plot_symbol: str = "o",
-        treatment_group_only = True
+        treatment_group_only: bool = True
         ):
 
         """
@@ -1468,7 +1515,7 @@ class DiffModel:
 
     def plot(
         self,
-        treatment = None,
+        treatment: str = None,
         x_label: str = "Time",
         y_label: str = "Outcome",
         y_lim = None,
@@ -1563,7 +1610,7 @@ class DiffModel:
         treatment_diagnostics = model_config["treatment_diagnostics"]
         no_treatments = model_config["no_treatments"]
         outcome_col = model_config["outcome_col"]
-        outcome_col_predicted = f"{outcome_col}{config.PREDICTED_SUFFIX}"
+        outcome_col_predicted = f"{outcome_col}{config.DELIMITER}{config.PREDICTED_SUFFIX}"
 
         if TG_col is None and treatment is None:            
             if no_treatments == 1:
@@ -1620,6 +1667,26 @@ class DiffModel:
 
             if model_config["log_outcome"]:
 
+                if outcome_col.startswith(f"{config.LOG_PREFIX}{config.DELIMITER}"):                    
+                    
+                    model_data = model_data.rename(
+                        columns = {
+                            outcome_col: outcome_col[len(f"{config.LOG_PREFIX}{config.DELIMITER}"):]
+                        }
+                    )
+                    
+                    outcome_col = outcome_col[len(f"{config.LOG_PREFIX}{config.DELIMITER}"):]
+                
+                if outcome_col_predicted.startswith(f"{config.LOG_PREFIX}{config.DELIMITER}"):                    
+                    
+                    model_data = model_data.rename(
+                        columns = {
+                            outcome_col_predicted: outcome_col_predicted[len(f"{config.LOG_PREFIX}{config.DELIMITER}"):]
+                        }
+                    )
+                                        
+                    outcome_col_predicted = outcome_col_predicted[len(f"{config.LOG_PREFIX}{config.DELIMITER}"):]                    
+                
                 model_data[outcome_col] = np.exp(model_data[outcome_col])
                 model_data[outcome_col_predicted] = np.exp(model_data[outcome_col_predicted])
 
@@ -1897,7 +1964,8 @@ class DiffModel:
         lines_labels: list = ["TG", "TG counterfactual"],
         plot_legend: bool = True,
         plot_grid: bool = True,
-        plot_size: list = [12, 6]
+        plot_size: list = [12, 6],
+        retransform_log_outcome: bool = False
         ):
 
         """
@@ -1929,6 +1997,9 @@ class DiffModel:
             Show grid. Default is True.
         plot_size : list, optional
             Figure size as [width, height]. Default is [12, 6].
+        retransform_log_outcome : bool, optional
+            If outcome was log-transformed, retransform to original scale for plotting. 
+            Default is False.
 
         Returns
         -------
@@ -1965,11 +2036,12 @@ class DiffModel:
             if no_treatments == 1:
                 raise ValueError ("Model object has no column for treatment group with respect to one treatment. Set parameter treatment = [your_treatment].")
             else:
-                raise ValueError ("Model object has no column for treatment group with respect to ", str(no_treatments), " treatments. Choose one with parameter treatment.")
+                raise ValueError (f"Model object has no column for treatment group with respect to {no_treatments} treatments. Choose one with parameter treatment.")
 
         counterfac_results = self.counterfactual(
             treatment = treatment,
-            after_treatment_col = after_treatment_col
+            after_treatment_col = after_treatment_col,
+            retransform_log_outcome = retransform_log_outcome
             )
 
         model_data_mod = counterfac_results[0]

{diffindiff-2.3.5 → diffindiff-2.3.7}/diffindiff/didanalysis_helper.py

@@ -4,8 +4,8 @@
 # Author:      Thomas Wieland 
 #              ORCID: 0000-0001-5168-9846
 #              mail: geowieland@googlemail.com              
-# Version:     1.1.2
-# Last update: 2025-03-16 17:46
+# Version:     1.1.3
+# Last update: 2025-03-24 18:01
 # Copyright (c) 2025-2026 Thomas Wieland
 #-----------------------------------------------------------------------
 

{diffindiff-2.3.5 → diffindiff-2.3.7}/diffindiff/diddata.py

@@ -4,8 +4,8 @@
 # Author:      Thomas Wieland 
 #              ORCID: 0000-0001-5168-9846
 #              mail: geowieland@googlemail.com              
-# Version:     2.2.4
-# Last update: 2026-03-14 11:28
+# Version:     2.2.5
+# Last update: 2026-03-24 20:23
 # Copyright (c) 2024-2026 Thomas Wieland
 #-----------------------------------------------------------------------
 
@@ -2295,6 +2295,7 @@ def create_counterfactual(
     treatment_col: str,
     time_col: str,
     cf_for_unit: str,
+    use_treatment_col: bool = False,
     use_data: str = "both",
     model_type: str = "ols",
     test_size: float = 0.2,
@@ -2310,7 +2311,8 @@ def create_counterfactual(
     svr_kernel = "rbf",
     xgb_learning_rate = 0.1,
     lgbm_learning_rate = 0.1,
-    random_state = 71
+    random_state = 71,
+    verbose: bool = False
     ):
     
     """
@@ -2332,12 +2334,76 @@ def create_counterfactual(
         Time column name.
     cf_for_unit : str
         Unit id for which a counterfactual should be generated (excluded from training).
+    use_treatment_col : bool, optional
+        If True, treatment variable is included into counterfactual prediction.
     use_data : {'both','treatment','control'}, optional
         Subset used for training the predictive model (default 'both').
     model_type : str, optional
         Model type passed to ``tools.model_wrapper`` (e.g. 'ols','rf','xgb').
-    Other parameters
-        Passed to the model wrapper (hyperparameters, test/train split, random_state).
+    test_size : float or int, optional
+        Fraction or number of data to reserve for testing.
+        Passed to `sklearn.model_selection.train_test_split`;
+        see the corresponding documentation.
+    train_size : float or int, optional
+        Fraction or number of data to define for training.
+        Passed to `sklearn.model_selection.train_test_split`;
+        see the corresponding documentation.
+    shuffle : bool, optional
+        Whether or not to shuffle the data before splitting.
+        Passed to `sklearn.model_selection.train_test_split`;
+        see the corresponding documentation. 
+    stratify : array-like, default=None
+        If not None, data is split in a stratified fashion, 
+        using this as the class labels.
+        Passed to `sklearn.model_selection.train_test_split`;
+        see the corresponding documentation.
+    model_n_estimators : int, optional
+        The number of estimators in the ensemble.
+        Passed to `BaggingRegressor` and `RandomForestRegressor`;
+        see the corresponding documentation. 
+    model_max_features : int or float, optional
+        Passed to `BaggingRegressor`, `RandomForestRegressor`,
+        and `GradientBoostingRegressor`; see the corresponding documentation.
+    model_min_samples_split : int or float, optional
+        Minimum number of samples required to split an internal node.
+        Passed to `RandomForestRegressor`; 
+        see the corresponding documentation.
+    rf_max_depth : int, optional
+        Maximum depth of trees in Random Forest Regression.
+        Passed to `RandomForestRegressor`; see the corresponding documentation.
+    gb_iterations : int, optional
+        Passed to `GradientBoostingRegressor`, `XGBRegressor`, 
+        and `LGBMRegressor`; see the corresponding documentation.
+    gb_max_depth : int, optional
+        Maximum depth trees in Gradient Boosting Regression.
+        Passed to `GradientBoostingRegressor`; 
+        see the corresponding documentation.
+    gb_learning_rate : float, optional
+        Weighting with respect to the contribution of each tree
+        in the Gradient Boosting algorithm.
+        Passed to `GradientBoostingRegressor`; 
+        see the corresponding documentation.
+    knn_n_neighbors : int, optional
+        Number of neighbors to use in Nearest-neighbor algorithm.
+        Passed to `KNeighborsRegressor`; 
+        see the corresponding documentation.
+    svr_kernel : str or callable, optional
+        Kernel type to be used in the Support Vector Regression algorithm.
+        Passed to `SVR`; see the corresponding documentation.
+    xgb_learning_rate : float, optional
+        Weighting with respect to the contribution of each tree
+        in the Extreme Gradient Boosting algorithm.
+        Passed to `XGBRegressor`; see the corresponding documentation.
+    lgbm_learning_rate : float, optional
+        Weighting with respect to the contribution of each tree
+        in the Light Gradient Boosting algorithm.
+        Passed to `LGBMRegressor`; see the corresponding documentation.
+    random_state : int, optional
+        Random seed for reproducibility.
+        Passed to `sklearn.model_selection.train_test_split`
+        and regressors; see the corresponding documentations.
+    verbose : bool, optional
+        If True, print progress messages.
 
     Returns
     -------
@@ -2356,21 +2422,60 @@ def create_counterfactual(
     ...     cf_for_unit='counterfac'
     ...     )
     """
+    
+    if not isinstance(y, str):
+        if isinstance(y, list):
+            if len(y) == 1:
+                y = y[0]
+            elif len(y) == 0:
+                raise ValueError("Parameter y was stated as empty list")
+            else:
+                raise ValueError(f"Parameter y was stated as list with {len(y)} entries")
+        else:
+            raise TypeError(f"Parameter y must be stated as str, not: {y}")
+        
+    if not isinstance(X, list):
+        if isintance(X, str):
+            X = [X]
+        else:
+            raise TypeError(f"Parameter X must be stated as list of strings, not: {X}")
+    
+    cols = [
+        unit_col, 
+        time_col,
+        y
+        ]
+    cols.extend(X)
+    
+    if use_treatment_col:
+        cols.append(treatment_col)
             
-    data = data[[y] + X + [unit_col, treatment_col, time_col]].copy()
-
+    data = data[cols].copy()
+    
     data_len = len(data)
     data = data.dropna()
     if len(data) < data_len:
         print(f"NOTE: Because of NaN values, {data_len-len(data)} observations were skipped.")
 
-    data = data[data[unit_col].astype(str) != cf_for_unit]
-    data_unit = data[data[unit_col].astype(str) == cf_for_unit]
+    isnotreatment = tools.is_notreatment(
+        data = data,
+        unit_col = unit_col,
+        treatment_col = treatment_col,
+        verbose = verbose
+        )
+    if not isnotreatment[0]:
+        print(f"NOTE: No {config.NO_TREATMENT_CG_DESCRIPTION}. Counterfactual will not cover full treatment time.")    
+    
+    data_unit = data.loc[data[unit_col].astype(str) == cf_for_unit]
+    assert len(data_unit) > 0, f"Observational unit '{cf_for_unit}' does not exist in model data"
+    
+    data = data.loc[data[unit_col].astype(str) != cf_for_unit]    
     
     isnotreatment = tools.is_notreatment(
         data = data,
         unit_col = unit_col,
-        treatment_col = treatment_col
+        treatment_col = treatment_col,
+        verbose = verbose
         )
     control_group = isnotreatment[2]
     
@@ -2378,12 +2483,16 @@ def create_counterfactual(
         data = data,
         unit_col = unit_col,
         time_col = time_col,
-        treatment_col = treatment_col
+        treatment_col = treatment_col,
+        verbose = verbose
         )[0]
     units = tools.unique(units_tt[unit_col])  
     
     if not isnotreatment[0]:
-        print(f"NOTE: No {config.NO_TREATMENT_CG_DESCRIPTION}. Counterfactual will not cover full treatment time.")    
+        print(f"NOTE: No {config.TREATMENT_GROUP_DESCRIPTION} units in model data left.")
+    
+    if verbose:
+        print(f"Creating data for counterfactual with {len(X)} independent variables", end = " ... ")
     
     data_TG = pd.DataFrame(columns = data.columns)
 
@@ -2411,6 +2520,9 @@ def create_counterfactual(
     
     data_cf[X] = data_cf[X].apply(pd.to_numeric, errors='coerce')
     
+    if verbose:
+        print("OK")
+    
     counterfactual_pred = tools.model_wrapper(
         y = data_cf[y],
         X = data_cf[X],
@@ -2428,9 +2540,19 @@ def create_counterfactual(
         svr_kernel = svr_kernel,
         xgb_learning_rate = xgb_learning_rate,
         lgbm_learning_rate = lgbm_learning_rate,
-        random_state = random_state
+        random_state = random_state,
+        verbose = verbose
         )
     
+    if verbose:
+        print(f"Counterfactual prediction for observational unit '{cf_for_unit}'", end = " ... ")
+    
+    data_unit[treatment_col] = 0
+    data_unit[y] = counterfactual_pred[1].predict(data_unit[X])
+    
+    if verbose:
+        print("OK")
+    
     return [
         counterfactual_pred, 
         data_cf, 

{diffindiff-2.3.5 → diffindiff-2.3.7}/diffindiff/didtools.py

@@ -4,8 +4,8 @@
 # Author:      Thomas Wieland 
 #              ORCID: 0000-0001-5168-9846
 #              mail: geowieland@googlemail.com              
-# Version:     2.2.2
-# Last update: 2026-03-16 18:04
+# Version:     2.2.3
+# Last update: 2026-03-24 20:21
 # Copyright (c) 2025-2026 Thomas Wieland
 #-----------------------------------------------------------------------
 
@@ -582,7 +582,7 @@ def is_notreatment(
     treatment_timepoints = treatment_timepoints.reset_index()
 
     no_treatment = (treatment_timepoints[treatment_col] == 0).any()
-    if (treatment_timepoints[treatment_col] == 0).all():
+    if (treatment_timepoints[treatment_col] == 0).all() or (treatment_timepoints[treatment_col] == 1).all():
         no_treatment = False
 
     treatment_group = treatment_timepoints.loc[treatment_timepoints[treatment_col] > 0, unit_col]
@@ -999,7 +999,7 @@ def is_prepost(
     
     prepost = False
     
-    if data[time_col].nunique() == 2:
+    if data[time_col].nunique() == 2 and data[unit_col].nunique() > 1:
         prepost = True        
     
     if verbose:
@@ -1305,8 +1305,14 @@ def model_wrapper(
     y,
     X,
     model_type: str,
-    test_size = 0.2,
-    train_size = None,
+    test_size: float = 0.2,
+    train_size: float = None,
+    shuffle: bool = True,
+    stratify = None,
+    X_train: list = None, 
+    X_test: list = None, 
+    y_train: list = None, 
+    y_test: list = None,
     model_n_estimators = 1000,
     model_max_features = 0.9,
     model_min_samples_split = 2,
@@ -1333,12 +1339,68 @@ def model_wrapper(
         Independent variables (features).
     model_type : str
         One of: 'ols', 'olsbg', 'dtbg', 'rf', 'gb', 'knn', 'svr', 'xgb', 'lgbm'.
-    test_size : float, optional
-        Fraction of data to reserve for testing.
+    test_size : float or int, optional
+        Fraction or number of data to reserve for testing.
+        Passed to `sklearn.model_selection.train_test_split`;
+        see the corresponding documentation.
+    train_size : float or int, optional
+        Fraction or number of data to define for training.
+        Passed to `sklearn.model_selection.train_test_split`;
+        see the corresponding documentation.
+    shuffle : bool, optional
+        Whether or not to shuffle the data before splitting.
+        Passed to `sklearn.model_selection.train_test_split`;
+        see the corresponding documentation. 
+    stratify : array-like, default=None
+        If not None, data is split in a stratified fashion, 
+        using this as the class labels.
+        Passed to `sklearn.model_selection.train_test_split`;
+        see the corresponding documentation.
+    model_n_estimators : int, optional
+        The number of estimators in the ensemble.
+        Passed to `BaggingRegressor` and `RandomForestRegressor`;
+        see the corresponding documentation. 
+    model_max_features : int or float, optional
+        Passed to `BaggingRegressor`, `RandomForestRegressor`,
+        and `GradientBoostingRegressor`; see the corresponding documentation.
+    model_min_samples_split : int or float, optional
+        Minimum number of samples required to split an internal node.
+        Passed to `RandomForestRegressor`; 
+        see the corresponding documentation.
+    rf_max_depth : int, optional
+        Maximum depth of trees in Random Forest Regression.
+        Passed to `RandomForestRegressor`; see the corresponding documentation.
+    gb_iterations : int, optional
+        Passed to `GradientBoostingRegressor`, `XGBRegressor`, 
+        and `LGBMRegressor`; see the corresponding documentation.
+    gb_max_depth : int, optional
+        Maximum depth trees in Gradient Boosting Regression.
+        Passed to `GradientBoostingRegressor`; 
+        see the corresponding documentation.
+    gb_learning_rate : float, optional
+        Weighting with respect to the contribution of each tree
+        in the Gradient Boosting algorithm.
+        Passed to `GradientBoostingRegressor`; 
+        see the corresponding documentation.
+    knn_n_neighbors : int, optional
+        Number of neighbors to use in Nearest-neighbor algorithm.
+        Passed to `KNeighborsRegressor`; 
+        see the corresponding documentation.
+    svr_kernel : str or callable, optional
+        Kernel type to be used in the Support Vector Regression algorithm.
+        Passed to `SVR`; see the corresponding documentation.
+    xgb_learning_rate : float, optional
+        Weighting with respect to the contribution of each tree
+        in the Extreme Gradient Boosting algorithm.
+        Passed to `XGBRegressor`; see the corresponding documentation.
+    lgbm_learning_rate : float, optional
+        Weighting with respect to the contribution of each tree
+        in the Light Gradient Boosting algorithm.
+        Passed to `LGBMRegressor`; see the corresponding documentation.
     random_state : int, optional
         Random seed for reproducibility.
-    Other parameters
-        Model-specific hyperparameters.
+        Passed to `sklearn.model_selection.train_test_split`
+        and regressors; see the corresponding documentations.
     verbose : bool, optional
         If True, print progress messages.
 
@@ -1357,22 +1419,52 @@ def model_wrapper(
     >>> model_wrapper(y, X, model_type='ols')
     """
 
-    if model_type not in ["ols", "olsbg", "dtbg", "rf", "gb", "knn", "svr", "xgb", "lgbm"]:
-        raise ValueError("Please enter a valid model type ('ols', 'olsbg', 'dtbg', 'rf', 'gb', 'knn', 'svr', 'xgb', 'lgbm')")
+    if model_type not in config.MODEL_WRAPPER_AVAILABLE_LIST:
+        raise ValueError(f"Please enter a valid model type: {', '.join(config.MODEL_WRAPPER_AVAILABLE_LIST)}.")
     
     if verbose:
         print("Setting up training and testing data", end = " ... ")
 
-    X_train, X_test, y_train, y_test = train_test_split(
-        X, 
-        y, 
-        test_size = test_size,
-        train_size = train_size,
-        random_state = random_state
-    )
+    if X_train is None:
+        X_train = []
+    if X_test is None:
+        X_test = []
+    if y_train is None:
+        y_train = [] 
+    if y_test is None:
+        y_test = []
+        
+    self_defined_split = False
+        
+    if len(X_train) > 0 and len(X_test) > 0 and len(y_train) > 0 and len(y_test) > 0:
+        
+        if len(X_train) != len(y_train) or len(X_test) != len(y_test):
+            raise ValueError(f"Train resp. tests subsets y and X must have the same length.")
+        
+        else:
+            self_defined_split = True
+        
+    else:
+
+        X_train, X_test, y_train, y_test = train_test_split(
+            X, 
+            y, 
+            test_size = test_size,
+            train_size = train_size,
+            shuffle = shuffle,
+            stratify = stratify,
+            random_state = random_state
+        )
 
     if verbose:
+        
         print("OK")
+        
+        if self_defined_split:
+            print(f"NOTE: Train and test subsets are built by user-defined splitting with {len(X_train)} training and {len(X_test)} test observations.")
+        else:
+            print(f"NOTE: Random train and test subsets were built with test_size = {test_size} and train_size = {train_size}.")
+            
         print(f"Training {model_type} model", end = " ... ")
     
     model = None
@@ -1441,8 +1533,11 @@ def model_wrapper(
     
     params = {
         "model_type": model_type,
+        "model_type_description": config.MODEL_WRAPPER_AVAILABLE[model_type],
+        "self_defined_split": self_defined_split,
         "test_size": test_size,
         "train_size": train_size,
+        "random_state": random_state,
         "model_n_estimators": model_n_estimators,
         "model_max_features": model_max_features,
         "model_min_samples_split": model_min_samples_split,
@@ -1453,8 +1548,7 @@ def model_wrapper(
         "knn_n_neighbors": knn_n_neighbors,
         "svr_kernel": svr_kernel,
         "xgb_learning_rate": xgb_learning_rate,
-        "lgbm_learning_rate": lgbm_learning_rate,
-        "random_state": random_state
+        "lgbm_learning_rate": lgbm_learning_rate        
         }
     
     return [
@@ -1501,7 +1595,7 @@ def fit_metrics(
     AssertionError
         If observed and expected differ in length.
     ValueError
-        If observed or expected are not numeric, or contain NaNs when remove_nan is False.
+        If observed and/or expected are not numeric, or contain NaNs when remove_nan is False.
 
     Examples
     --------
@@ -1513,10 +1607,19 @@ def fit_metrics(
 
     assert observed_no == expected_no, "Error while calculating fit metrics: Observed and expected differ in length"
     
+    type_errors = []
     if not pd.api.types.is_numeric_dtype(observed) or not np.issubdtype(observed.dtype, np.number):
-        raise ValueError("Error while calculating fit metrics: Observed column is not numeric")
+        try:
+            observed = pd.to_numeric(observed)
+        except:
+            type_errors.append("Observed column is not numeric.")
     if not pd.api.types.is_numeric_dtype(expected) or not np.issubdtype(expected.dtype, np.number):
-        raise ValueError("Error while calculating fit metrics: Expected column is not numeric")
+        try:
+            expected = pd.to_numeric(expected)
+        except:
+            type_errors.append("Expected column is not numeric.")
+    if len(type_errors) > 0:
+        raise TypeError(f"Error(s) while calculating fit metrics: {', '.join(type_errors)}")
     
     if outcome_col is not None:
         outcome_observed_col = f"{outcome_col}{config.DELIMITER}{config.OBSERVED_SUFFIX}"
@@ -1549,10 +1652,13 @@ def fit_metrics(
     
     else:
         
+        value_errors = []        
         if np.isnan(observed).any():
-            raise ValueError("Error while calculating fit metrics: Vector with observed data contains NaNs and 'remove_nan' is False")
+            value_errors.append("Observed data contains NaNs.")
         if np.isnan(expected).any():
-            raise ValueError("Error while calculating fit metrics: Vector with expected data contains NaNs and 'remove_nan' is False")
+            value_errors.append("Expected data contains NaNs.")
+        if len(value_errors) > 0:
+            raise ValueError(f"Error(s) while calculating fit metrics: {' '.join(value_errors)} Parameter 'remove_nan' is False.")
     
     if verbose:
         print("Calculating model fit metrics", end = " ... ")

{diffindiff-2.3.5 → diffindiff-2.3.7}/diffindiff.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: diffindiff
-Version: 2.3.5
+Version: 2.3.7
 Summary: diffindiff: Python library for convenient Difference-in-Differences analyses
 Author: Thomas Wieland
 Author-email: geowieland@googlemail.com
@@ -27,7 +27,7 @@ Thomas Wieland [ORCID](https://orcid.org/0000-0001-5168-9846) [EMail](mailto:geo
 
 If you use this software, please cite:
 
-Wieland, T. (2026). diffindiff: A Python library for convenient difference-in-differences analyses (Version 2.3.5) [Computer software]. Zenodo. https://doi.org/10.5281/zenodo.18656820
+Wieland, T. (2026). diffindiff: A Python library for convenient difference-in-differences analyses (Version 2.3.7) [Computer software]. Zenodo. https://doi.org/10.5281/zenodo.18656820
 
 
 ## Installation
@@ -173,10 +173,15 @@ See the /tests directory for usage examples of most of the included functions.
 This software was developed without the use of AI-generated code. The Continue Agent in Microsoft Visual Studio Code using the GPT-5 mini model (by OpenAI) was used solely to assist in drafting and refining docstrings for documentation. The corresponding guidelines and constraints defined by the author are documented in `AGENTS-docstrings.md` in the [public GitHub repository](https://github.com/geowieland/diffindiff_official).
 
 
-## What's new (v2.3.5)
+## What's new (v2.3.7)
 
+- Extensions:  
+  - DiffModel.counterfactual() and DiffModel.plot_counterfactual(): Re-transform log-transformed outcome via parameter retransform_log_outcome
+  - Difference between predicted and counterfactual-predicted in result df of DiffModel.counterfactual()  
+  - Extended parameters in didtools.model_wrapper() to be passed to sklearn.model_selection.train_test_split()
+  - Extension of documentations
 - Bugfixes:
-  - Test whether input data is panel data via didtools.is_panel() which is included in didanalysis_helper.data_diagnostics()
-  - Fixed false test results given continuous treatments are accepted in didtools.is_simultaneous()
-  - Fixed false test results given continuous treatments are accepted in didtools.is_prepost()
-  - Argument 'pre_post' is passed to is_simultaneous() in didanalysis_helper.treatment_diagnostics()
+  - diddata.create_counterfactual(): Fixed calculation error, additional check of input parameters
+  - didtools.is_prepost() now also checks number of observational units
+  - didtools.is_notreatment() now checks whether data is singular with respect to 0 and 1  
+  - didtools.fit_metrics() tries to convert observed and expected if not numeric before raising TypeError

{diffindiff-2.3.5 → diffindiff-2.3.7}/setup.py

@@ -7,7 +7,7 @@ def read_README():
     
 setup(
     name='diffindiff',
-    version='2.3.5',
+    version='2.3.7',
     description='diffindiff: Python library for convenient Difference-in-Differences analyses',
     packages=find_packages(include=["diffindiff", "diffindiff.tests"]),
     include_package_data=True,