Fix describe() for ExtensionArrays with multiple internal dtypes

Author: kernelism

doc/source/whatsnew/v3.0.0.rst

@@ -909,6 +909,7 @@ Other
 - Bug in :meth:`Index.sort_values` when passing a key function that turns values into tuples, e.g. ``key=natsort.natsort_key``, would raise ``TypeError`` (:issue:`56081`)
 - Bug in :meth:`MultiIndex.fillna` error message was referring to ``isna`` instead of ``fillna`` (:issue:`60974`)
 - Bug in :meth:`Series.describe` where median percentile was always included when the ``percentiles`` argument was passed (:issue:`60550`).
+- Bug in :meth:`Series.describe` where statistics with multiple dtypes for ExtensionArrays were coerced to ``float64`` which raised a ``DimensionalityError``` (:issue:`61707`)
 - Bug in :meth:`Series.diff` allowing non-integer values for the ``periods`` argument. (:issue:`56607`)
 - Bug in :meth:`Series.dt` methods in :class:`ArrowDtype` that were returning incorrect values. (:issue:`57355`)
 - Bug in :meth:`Series.isin` raising ``TypeError`` when series is large (>10**6) and ``values`` contains NA (:issue:`60678`)

pandas/core/methods/describe.py

@@ -12,6 +12,7 @@
 )
 from typing import (
     TYPE_CHECKING,
+    Any,
     cast,
 )
 
@@ -215,6 +216,14 @@ def reorder_columns(ldesc: Sequence[Series]) -> list[Hashable]:
     return names
 
 
+def has_multiple_internal_dtypes(d: list[Any]) -> bool:
+    """Check if the sequence has multiple internal dtypes."""
+    if not d:
+        return False
+
+    return any(type(item) != type(d[0]) for item in d)
+
+
 def describe_numeric_1d(series: Series, percentiles: Sequence[float]) -> Series:
     """Describe series containing numerical data.
 
@@ -251,6 +260,10 @@ def describe_numeric_1d(series: Series, percentiles: Sequence[float]) -> Series:
                 import pyarrow as pa
 
                 dtype = ArrowDtype(pa.float64())
+        elif has_multiple_internal_dtypes(d):
+            # GH61707: describe() doesn't work on EAs
+            # with multiple internal dtypes, so return object dtype
+            dtype = None
         else:
             dtype = Float64Dtype()
     elif series.dtype.kind in "iufb":

pandas/tests/series/methods/test_describe.py

@@ -95,6 +95,32 @@ def test_describe_empty_object(self):
         assert np.isnan(result.iloc[2])
         assert np.isnan(result.iloc[3])
 
+    def test_describe_multiple_dtypes(self):
+        """
+        GH61707: describe() doesn't work on EAs which generate
+        statistics with multiple dtypes.
+        """
+        from decimal import Decimal
+
+        from pandas.tests.extension.decimal import to_decimal
+
+        s = Series(to_decimal([1, 2.5, 3]), dtype="decimal")
+
+        expected = Series(
+            [
+                3,
+                Decimal("2.166666666666666666666666667"),
+                Decimal("0.8498365855987974716713706849"),
+                Decimal("1"),
+                Decimal("3"),
+            ],
+            index=["count", "mean", "std", "min", "max"],
+            dtype="object",
+        )
+
+        result = s.describe(percentiles=[])
+        tm.assert_series_equal(result, expected)
+
     def test_describe_with_tz(self, tz_naive_fixture):
         # GH 21332
         tz = tz_naive_fixture