diffindiff 2.4.2__tar.gz → 2.5.1__tar.gz

{diffindiff-2.4.2 → diffindiff-2.5.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: diffindiff
-Version: 2.4.2
+Version: 2.5.1
 Summary: diffindiff: Python library for convenient Difference-in-Differences analyses
 Author: Thomas Wieland
 Author-email: geowieland@googlemail.com
@@ -29,7 +29,7 @@ A research note featuring a case study that utilizes the diffindiff library is a
 
 If you use this software, please cite:
 
-Wieland, T. (2026). diffindiff: A Python library for convenient difference-in-differences analyses (Version 2.4.2) [Computer software]. Zenodo. https://doi.org/10.5281/zenodo.18656820
+Wieland, T. (2026). diffindiff: A Python library for convenient difference-in-differences analyses (Version 2.5.1) [Computer software]. Zenodo. https://doi.org/10.5281/zenodo.18656820
 
 
 ## Installation
@@ -177,7 +177,9 @@ 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 (v2.4.2)
+## What's new (v2.5.1)
 
-- Bugfixes:
-  - DiffData.add_own_counterfactual() now works correctly in any case
+- Bugfixes
+  - In DiffData.create_treatment() and DiffData.create_data(), pre-post designs with after-treatment period (follow-up) now work correctly
+  - DiffModel.plot() nows supports pre-post designs with after-treatment period
+  - Some small stabilizations

{diffindiff-2.4.2 → diffindiff-2.5.1}/README.md

@@ -21,7 +21,7 @@ A research note featuring a case study that utilizes the diffindiff library is a
 
 If you use this software, please cite:
 
-Wieland, T. (2026). diffindiff: A Python library for convenient difference-in-differences analyses (Version 2.4.2) [Computer software]. Zenodo. https://doi.org/10.5281/zenodo.18656820
+Wieland, T. (2026). diffindiff: A Python library for convenient difference-in-differences analyses (Version 2.5.1) [Computer software]. Zenodo. https://doi.org/10.5281/zenodo.18656820
 
 
 ## Installation
@@ -169,7 +169,9 @@ 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 (v2.4.2)
+## What's new (v2.5.1)
 
-- Bugfixes:
-  - DiffData.add_own_counterfactual() now works correctly in any case
+- Bugfixes
+  - In DiffData.create_treatment() and DiffData.create_data(), pre-post designs with after-treatment period (follow-up) now work correctly
+  - DiffModel.plot() nows supports pre-post designs with after-treatment period
+  - Some small stabilizations

{diffindiff-2.4.2 → diffindiff-2.5.1}/diffindiff/config.py

@@ -4,15 +4,15 @@
 # Author:      Thomas Wieland 
 #              ORCID: 0000-0001-5168-9846
 #              mail: geowieland@googlemail.com              
-# Version:     1.0.20
-# Last update: 2026-06-28 12:31
+# Version:     1.0.22
+# Last update: 2026-07-10 11:36
 # Copyright (c) 2025-2026 Thomas Wieland
 #-----------------------------------------------------------------------
 
 # Basic config:
 
 PACKAGE_NAME = "diffindiff"
-PACKAGE_VERSION = "2.4.2"
+PACKAGE_VERSION = "2.5.1"
 
 VERBOSE = False
 
@@ -21,6 +21,11 @@ ROUND_PERCENT = 2
 
 AUTO_SWITCH_TO_PREPOST = True
 
+AUTO_SWITCH_TO_FIXED_EFFECTS = True
+FIXED_EFFECTS_THRESHOLD = 3
+
+AUTO_SKIP_CONSTANT_COLUMNS = True
+
 ACCEPT_CONTINUOUS_TREATMENTS = True
 
 # Description texts:
@@ -47,6 +52,7 @@ TREATMENT_PERIOD_DESCRIPTION = f"{TREATMENT_DESCRIPTION} period"
 STUDY_PERIOD_DESCRIPTION = "Study period"
 PREPOST_DESCRIPTION = "Pre-post"
 AFTER_TREATMENT_PERIOD_DESCRIPTION = "After-treatment period"
+FOLLOW_UP_DESCRIPTION = "Follow-up"
 
 UNITS_DESCRIPTION = "Units"
 
@@ -96,6 +102,7 @@ LOG_PREFIX = "log"
 OBSERVED_SUFFIX = "observed"
 EXPECTED_SUFFIX = "expected"
 DEMEAN_SUFFIX = "demean"
+WEIGHTED_SUFFIX = "_weighted"
 PREDICTED_SUFFIX = "pred"
 CI_LOWER_SUFFIX = "CI_lower"
 CI_UPPER_SUFFIX = "CI_upper"
@@ -500,7 +507,11 @@ DATA_DIAGNOSTICS = {
         "description": N_DESCRIPTION,
         "show_in_summary": True
         },
-    }
+    "cols_constants": {
+        "description": "Columns with constant values",
+        "show_in_summary": False
+    },
+}
 
 DIAGNOSTICS_COLUMN = "Result"
 

{diffindiff-2.4.2 → diffindiff-2.5.1}/diffindiff/didanalysis.py

@@ -4,8 +4,8 @@
 # Author:      Thomas Wieland 
 #              ORCID: 0000-0001-5168-9846
 #              mail: geowieland@googlemail.com              
-# Version:     2.4.0
-# Last update: 2026-05-01 09:52
+# Version:     2.4.2
+# Last update: 2026-07-10 11:30
 # Copyright (c) 2024-2026 Thomas Wieland
 #-----------------------------------------------------------------------
 
@@ -735,16 +735,20 @@ class DiffModel:
         treatment_effects_df = self.treatment_effects()        
         
         width = treatment_effects_df[""].str.len().max()
-        treatment_effects_df[""] = treatment_effects_df[""].str.ljust(width)
-        total_width = (sum(treatment_effects_df.astype(str).map(len).max()) + len(treatment_effects_df.columns) * 2)
+        treatment_effects_df[""] = treatment_effects_df[""].str.ljust(width)        
         
-        print("=" * total_width)
+        total_width = (
+            treatment_effects_df.apply(lambda col: col.astype(str).str.len().max()).sum()
+            + len(treatment_effects_df.columns) * 2
+        )
+        
+        print("=" * int(total_width))
         print(model_config["analysis_description"]) 
-        print("-" * total_width)
+        print("-" * int(total_width))
 
         print(config.TREATMENT_EFFECTS_DESCRIPTION)
         print(treatment_effects_df.to_string(index=False))        
-        print("-" * total_width)
+        print("-" * int(total_width))
 
         covariates_effects_df = pd.DataFrame(
             [
@@ -761,7 +765,7 @@ class DiffModel:
         covariates_effects_df[config.COVARIATES_DESCRIPTION] = covariates_effects_df[config.COVARIATES_DESCRIPTION].str.ljust(width)
 
         print(covariates_effects_df.to_string(index=False, header=False))
-        print("-" * total_width)        
+        print("-" * int(total_width))        
 
         treatment_diagnostics = self.treatment_diagnostics()
         treatment_diagnostics_df = treatment_diagnostics[0]
@@ -782,7 +786,7 @@ class DiffModel:
             else:
                 print(f"NOTE: Treatments {', '.join(no_control_conditions)} have no control conditions.")  
 
-        print("-" * total_width)        
+        print("-" * int(total_width))        
 
         data_diagnostics_df = self.data_diagnostics()
 
@@ -792,7 +796,7 @@ class DiffModel:
         print(config.DATA_DIAGNOSTICS_DESCRIPTION)
         print(data_diagnostics_df.to_string(index=False, header=False))
 
-        print("-" * total_width)        
+        print("-" * int(total_width))        
 
         model_fit_metrics = self.fit_metrics()
 
@@ -804,7 +808,7 @@ class DiffModel:
         print(f"{config.MODEL_FIT_METRICS_DESCRIPTION}s")
         print(model_fit_metrics.to_string(index=False, header=False))
         
-        print("=" * total_width)
+        print("=" * int(total_width))
 
         return self
 
@@ -894,6 +898,49 @@ class DiffModel:
         
         treatment_effects = self.treatment_effects()
 
+        try:
+            if model_config.get("pre_post", False) and len(model_config.get("after_treatment_col", [])) > 0:
+
+                labels = treatment_effects.iloc[:, 0].astype(str).values
+                has_after = any((config.AFTER_TREATMENT_PERIOD_DESCRIPTION in lab) or (config.ATT_COL in lab) for lab in labels)
+                if not has_after:
+                    model_results = self.data[0]
+
+                    aate_key = config.EFFECTS_TYPES["AATE"]["model_results_key"] if "AATE" in config.EFFECTS_TYPES else "average_after_treatment_effects"
+                    att_key = config.EFFECTS_TYPES["ATT"]["model_results_key"] if "ATT" in config.EFFECTS_TYPES else "after_treatment_time_effects"
+                    appended = False
+
+                    for key_try in [aate_key, att_key]:
+                        if key_try in model_results and isinstance(model_results[key_try], dict):
+                            for subkey, val in model_results[key_try].items():
+
+                                if isinstance(val, dict):
+
+                                    coef_k = config.OLS_MODEL_RESULTS["coef"]["model_results_key"]
+                                    se_k = config.OLS_MODEL_RESULTS["coef_standard_errors"]["model_results_key"]
+                                    t_k = config.OLS_MODEL_RESULTS["coef_teststatistic"]["model_results_key"]
+                                    p_k = config.OLS_MODEL_RESULTS["coef_p"]["model_results_key"]
+                                    cil_k = config.OLS_MODEL_RESULTS["coef_confint_lower"]["model_results_key"]
+                                    ciu_k = config.OLS_MODEL_RESULTS["coef_confint_upper"]["model_results_key"]
+
+                                    new_row = {
+                                        "": f"{config.EFFECTS_TYPES.get('AATE', {'description': 'After-treatment'})['description']} {subkey}",
+                                        "Estimate": val.get(coef_k, np.nan),
+                                        "SE": val.get(se_k, np.nan),
+                                        "t": val.get(t_k, np.nan),
+                                        "p": val.get(p_k, np.nan),
+                                        "CI lower": val.get(cil_k, np.nan),
+                                        "CI upper": val.get(ciu_k, np.nan),
+                                    }
+                                    treatment_effects = pd.concat([treatment_effects, pd.DataFrame([new_row])], ignore_index=True)
+                                    appended = True
+                                    break
+                        if appended:
+                            break
+        except Exception:
+
+            pass
+
         if sort_by_coef:
             treatment_effects = treatment_effects.sort_values(
                 by = treatment_effects.columns[1],
@@ -1803,14 +1850,14 @@ class DiffModel:
         model_data_TG_CG["t"] = pd.to_datetime(model_data_TG_CG["t"])
 
         if not model_config["pre_post"]:
-            pre_post_barplot = False        
+            pre_post_barplot = False
 
         if pre_post_barplot:
 
             x_pos_t1_TG = 0
             x_pos_t1_CG = x_pos_t1_TG + pre_post_bar_width  
             x_pos_t2_TG = 1.5  
-            x_pos_t2_CG = x_pos_t2_TG + pre_post_bar_width  
+            x_pos_t2_CG = x_pos_t2_TG + pre_post_bar_width
 
             plt.bar(
                 x = x_pos_t1_TG, 
@@ -1838,6 +1885,24 @@ class DiffModel:
                 color=lines_col[3], 
                 width = pre_post_bar_width
                 )
+            
+            if len(model_data_TG_CG) == 3:
+
+                x_pos_t3_TG = 3  
+                x_pos_t3_CG = x_pos_t3_TG + pre_post_bar_width
+
+                plt.bar(
+                    x = x_pos_t3_TG, 
+                    height = model_data_TG_CG[outcome_col_expected_TG][2],                 
+                    color = lines_col[2], 
+                    width = pre_post_bar_width
+                    )            
+                plt.bar(
+                    x = x_pos_t3_CG, 
+                    height = model_data_TG_CG[outcome_col_expected_CG][2],                 
+                    color=lines_col[3], 
+                    width = pre_post_bar_width
+                    )
 
             plt.xlabel(x_label)
             plt.ylabel(y_label)
@@ -1931,16 +1996,29 @@ class DiffModel:
             ax.xaxis.set_major_formatter(DateFormatter(model_config["date_format"]))
 
         if model_config["pre_post"]:
+
+            ticks_line = model_data_TG_CG["t"].unique()
+            ticks_bars = [0.25, 1.75]
+
+            if len(model_data_TG_CG) == 3:
+                
+                if len(pre_post_ticks) < 3:
+                    pre_post_ticks.append("Follow-up")
+
+                ticks_bars.append(3.25)
+
             if not pre_post_barplot:
+                
                 plt.xticks(
-                    model_data_TG_CG["t"].unique(), 
-                    labels = [pre_post_ticks[0], pre_post_ticks[1]]
+                    ticks_line, 
+                    labels = pre_post_ticks
                     )  
             else:
                 plt.xticks(
-                    [0.25, 1.75], 
-                    labels = [pre_post_ticks[0], pre_post_ticks[1]]
-                    )  
+                    ticks_bars, 
+                    labels = pre_post_ticks
+                    )
+                
         else:
             plt.xticks(rotation=90)
         
@@ -2389,16 +2467,37 @@ def did_analysis(
         verbose=verbose
     )
     treatment_diagnostics = treatment_diagnostics_results[0]
-    staggered_adoption = treatment_diagnostics_results[1]    
+    staggered_adoption = treatment_diagnostics_results[1]
+    unique_units = treatment_diagnostics_results[3]
+    unique_time_points = treatment_diagnostics_results[4]
     
     if no_treatments > 1:        
         
         intercept = False
         TG_col = []        
         
-        if not FE_unit:
+        if not FE_unit and config.AUTO_SWITCH_TO_FIXED_EFFECTS:
             FE_unit = True
-            print("NOTE: Quasi-experiment includes more than one treatment. Unit fixed effects are used instead of control group baseline and treatment group deviation.")
+            print(f"NOTE: Quasi-experiment includes more than one treatment ({no_treatments}). Unit fixed effects are used instead of control group baseline and treatment group deviation.")
+    
+    if unique_units < config.FIXED_EFFECTS_THRESHOLD:
+        
+        FE_unit = False
+        FE_group = False
+        ITE = False
+        ITT = False
+        intercept = True
+        
+        print(f"NOTE: Quasi-experiment includes less than {config.FIXED_EFFECTS_THRESHOLD} units. Unit or group fixed effects are skipped.")
+        
+    if unique_time_points < config.FIXED_EFFECTS_THRESHOLD:
+        
+        FE_time = False
+        ITT = False
+        GTT = False
+        intercept = True
+        
+        print(f"NOTE: Quasi-experiment includes less than {config.FIXED_EFFECTS_THRESHOLD} time points. Time fixed effects are skipped.")
     
     if demean and FE_unit:
         
@@ -2435,7 +2534,7 @@ def did_analysis(
             GTT = False
             print("NOTE: Both group and individual time trends were stated. Switching to individual time trends only.")
             
-    if staggered_adoption:
+    if staggered_adoption and config.AUTO_SWITCH_TO_FIXED_EFFECTS:
         
         if not FE_unit or not FE_time:
             print("NOTE: Quasi-experiment includes one or more staggered treatments. Two-way fixed effects model is used.")
@@ -2562,7 +2661,15 @@ def did_analysis(
         missing_replace_by_zero = missing_replace_by_zero,
         verbose = verbose
     )
+    
+    if len(data_diagnostics["cols_constants"]) > 0:
         
+        if config.AUTO_SKIP_CONSTANT_COLUMNS:
+        
+            print(f"NOTE: The following columns are constant and dropped from the analysis: {', '.join(data_diagnostics['cols_constants'])}.")
+            
+            covariates = [col for col in covariates if col not in data_diagnostics["cols_constants"]]
+
     if data_diagnostics["is_prepost"] and config.AUTO_SWITCH_TO_PREPOST:
         
         print(f"NOTE: Input is {config.PREPOST_PANELDATA_DESCRIPTION}. Data processing and model estimation will treat data as pre-post.")
@@ -2611,6 +2718,8 @@ def did_analysis(
     FE_time_vars = []
     dummy_time_original = []
     
+    drop_first = intercept
+
     if not demean:
 
         if FE_unit:
@@ -2619,7 +2728,7 @@ def did_analysis(
                 data = data,
                 col = unit_col,
                 type = "unit",
-                drop_first = intercept,
+                drop_first = drop_first,
                 verbose = verbose
             )
             
@@ -2634,7 +2743,7 @@ def did_analysis(
                 data = data,
                 col = time_col,
                 type = "time",
-                drop_first = intercept,
+                drop_first = drop_first,
                 verbose = verbose
             )
             
@@ -2658,7 +2767,7 @@ def did_analysis(
                 data = data,
                 col = group_by,
                 type = "group",
-                drop_first = intercept,
+                drop_first = drop_first,
                 verbose = verbose
                 )
         
@@ -3115,7 +3224,21 @@ def ddd_analysis(
         verbose=verbose
     )
     treatment_diagnostics = treatment_diagnostics_results[0]
-               
+    unique_units = treatment_diagnostics_results[3]
+    unique_time_points = treatment_diagnostics_results[4]
+
+    if unique_units < config.FIXED_EFFECTS_THRESHOLD:
+        
+        FE_unit = False
+          
+        print(f"NOTE: Quasi-experiment includes less than {config.FIXED_EFFECTS_THRESHOLD} units. Unit fixed effects are skipped.")
+        
+    if unique_time_points < config.FIXED_EFFECTS_THRESHOLD:
+        
+        FE_time = False
+        
+        print(f"NOTE: Quasi-experiment includes less than {config.FIXED_EFFECTS_THRESHOLD} time points. Time fixed effects are skipped.")
+                
     if FE_unit:
         TG_col = None
     if FE_time:
@@ -3140,7 +3263,15 @@ def ddd_analysis(
         missing_replace_by_zero = missing_replace_by_zero,
         verbose = verbose
     )
-      
+    
+    if len(data_diagnostics["cols_constants"]) > 0:
+        
+        if config.AUTO_SKIP_CONSTANT_COLUMNS:
+            
+            print(f"NOTE: The following columns are constant and dropped from the analysis: {', '.join(data_diagnostics['cols_constants'])}.")
+            
+            covariates = [col for col in covariates if col not in data_diagnostics["cols_constants"]]
+    
     if data_diagnostics["is_prepost"] and config.AUTO_SWITCH_TO_PREPOST:
         
         print("NOTE: Panel data is pre-post. Data processing and model estimation will treat data as pre-post")

{diffindiff-2.4.2 → diffindiff-2.5.1}/diffindiff/didanalysis_helper.py

@@ -4,8 +4,8 @@
 # Author:      Thomas Wieland 
 #              ORCID: 0000-0001-5168-9846
 #              mail: geowieland@googlemail.com              
-# Version:     1.2.1
-# Last update: 2025-06 28 13:03
+# Version:     1.2.2
+# Last update: 2026-07-02 20:18
 # Copyright (c) 2025-2026 Thomas Wieland
 #-----------------------------------------------------------------------
 
@@ -876,6 +876,12 @@ def data_diagnostics(
     outcome_mean = np.mean(data[outcome_col])
     outcome_sd = np.std(data[outcome_col])
     
+    cols_constants = []
+    if len(other_cols_relevant) > 0:
+        for col in other_cols_relevant:
+           if data[col].nunique() == 1:
+                cols_constants.append(col)
+    
     data_diagnostics_results = {
         list(config.DATA_DIAGNOSTICS)[0]: bool(modeldata_isbalanced),
         list(config.DATA_DIAGNOSTICS)[1]: bool(modeldata_ismissing[0]),
@@ -885,7 +891,8 @@ def data_diagnostics(
         list(config.DATA_DIAGNOSTICS)[5]: data_type,
         list(config.DATA_DIAGNOSTICS)[6]: outcome_col, 
         list(config.DATA_DIAGNOSTICS)[7]: f"Mean={round(outcome_mean, config.ROUND_STATISTIC)} SD={round(outcome_sd, config.ROUND_STATISTIC)}",
-        list(config.DATA_DIAGNOSTICS)[8]: observations,        
+        list(config.DATA_DIAGNOSTICS)[8]: observations,
+        list(config.DATA_DIAGNOSTICS)[9]: cols_constants
         }
     
     return data_diagnostics_results
@@ -927,7 +934,7 @@ def treatment_diagnostics(
     Returns
     -------
     list
-        [treatment_diagnostics_results (dict), staggered_adoption (bool), untreated (list)]
+        [treatment_diagnostics_results (dict), staggered_adoption (bool), untreated (list), unique_units (int), unique_timepoints (int)]
 
     Examples
     --------
@@ -1020,6 +1027,15 @@ def treatment_diagnostics(
         verbose = verbose
         )
     
+    groups_units_list = []
+    for key, value in treatment_diagnostics_results.items():
+        groups_units_list.extend(value["treatment_group"])
+        groups_units_list.extend(value["control_group"])
+    groups_units_list = list(set(groups_units_list))
+    unique_units = len(groups_units_list)
+    
+    unique_timepoints = data[time_col].nunique()
+    
     if verbose:
         
         if no_treatments > 1:
@@ -1030,7 +1046,9 @@ def treatment_diagnostics(
     return [
         treatment_diagnostics_results,
         staggered_adoption,
-        untreated
+        untreated,
+        unique_units,
+        unique_timepoints
     ]
 
 def ols_fit(