Fix #216: Allow unequal sample sizes in multi-group paired wide-format data

Remove overly aggressive NaN filtering in _check_errors() that was causing
data truncation when using wide-format paired data with different group sizes.

Problem:
When loading wide-format paired data created by concatenating DataFrames of
different lengths (e.g., 20, 10, and 40 samples), the package was removing
ALL rows with ANY NaN value across ALL columns. This truncated all groups
to the size of the smallest group.

Root Cause:
In _check_errors() method, the code had:
elif x is None and y is None:
    self.__output_data.dropna(inplace=True)

This removed entire rows if they had NaN in ANY column, affecting all groups
even though NaN values were structural (from DataFrame concatenation) and not
actual missing data points.

Solution:
Removed the problematic elif block from _check_errors(). The downstream code
in _get_plot_data() already handles NaN values correctly by:
1. Using pd.melt() which preserves all non-NaN values
2. Calling dropna(subset=[self.__yvar]) which only removes rows with NaN in
    the measurement column, not across all columns

Testing:
- Added test_33_multi_paired_different_sizes() to verify groups with 20, 10,
and 40 samples are preserved correctly

Author: Jacobluke-

dabest/_dabest_object.py

@@ -559,14 +559,12 @@ def _check_errors(self, x, y, idx, experiment, experiment_label, x1_level):
         self.__x1_level = x1_level
 
         if self.__is_paired and self.__output_data.isnull().values.any():
-            warn1 = f"NaN values detected under paired setting and removed,"
+            warn1 = f"NaN values detected under paired setting,"
             warn2 = f" please check your data."
             warnings.warn(warn1 + warn2)
             if x is not None and y is not None:
                 rmname = self.__output_data[self.__output_data[y].isnull()][self.__id_col].tolist()
                 self.__output_data = self.__output_data[~self.__output_data[self.__id_col].isin(rmname)]
-            elif x is None and y is None:
-                self.__output_data.dropna(inplace=True)
 
         # Check if there is a typo on paired
         if self.__is_paired and self.__is_paired not in ("baseline", "sequential"):

nbs/API/dabest_object.ipynb

@@ -664,14 +664,12 @@
     "        self.__x1_level = x1_level\n",
     "\n",
     "        if self.__is_paired and self.__output_data.isnull().values.any():\n",
-    "            warn1 = f\"NaN values detected under paired setting and removed,\"\n",
+    "            warn1 = f\"NaN values detected under paired setting,\"\n",
     "            warn2 = f\" please check your data.\"\n",
     "            warnings.warn(warn1 + warn2)\n",
     "            if x is not None and y is not None:\n",
     "                rmname = self.__output_data[self.__output_data[y].isnull()][self.__id_col].tolist()\n",
     "                self.__output_data = self.__output_data[~self.__output_data[self.__id_col].isin(rmname)]\n",
-    "            elif x is None and y is None:\n",
-    "                self.__output_data.dropna(inplace=True)\n",
     "\n",
     "        # Check if there is a typo on paired\n",
     "        if self.__is_paired and self.__is_paired not in (\"baseline\", \"sequential\"):\n",

nbs/tests/mpl_image_tests/baseline_images/test_266_multigroups_prop_paired_meandiff_gridkey_userdefinedrows.png

@@ -451,6 +451,34 @@ def test_32_multigroups_baseline_change_palette():
     plt.rcdefaults()
     return multi_groups_baseline.mean_diff.plot(custom_palette="Dark2", delta_text=True)
 
+@pytest.mark.mpl_image_compare(tolerance=8)
+def test_33_multi_paired_different_sizes():
+    # Test for GitHub issue #216: multi-group paired data with different sample sizes
+    plt.rcdefaults()
+    np.random.seed(9999)
+
+    # Create three test pairs with different sample sizes (20, 10, 40)
+    c1DF = pd.DataFrame({'Test 1_pre': norm.rvs(loc=3, scale=0.4, size=20)})
+    t1DF = pd.DataFrame({'Test 1_post': norm.rvs(loc=3.5, scale=0.5, size=20)})
+    t2DF = pd.DataFrame({'Test 2_pre': norm.rvs(loc=2.5, scale=0.6, size=10)})
+    t3DF = pd.DataFrame({'Test 2_post': norm.rvs(loc=3, scale=0.75, size=10)})
+    t4DF = pd.DataFrame({'Test 3_pre': norm.rvs(loc=3.5, scale=0.75, size=40)})
+    t5DF = pd.DataFrame({'Test 3_post': norm.rvs(loc=3.25, scale=0.4, size=40)})
+
+    df = pd.concat([c1DF, t1DF, t2DF, t3DF, t4DF, t5DF], axis=1)
+    df["ID"] = pd.Series(range(1, len(df)+1))
+
+    multi_paired_diff_sizes = load(
+        df,
+        idx=(("Test 1_pre", "Test 1_post"),
+             ("Test 2_pre", "Test 2_post"),
+             ("Test 3_pre", "Test 3_post")),
+        paired="baseline",
+        id_col="ID"
+    )
+
+    return multi_paired_diff_sizes.mean_diff.plot()
+
 @pytest.mark.mpl_image_compare(tolerance=8)
 def test_99_style_sheets():
     # Perform this test last so we don't have to reset the plot style.