BUG: Dataframe.aggregate drops pyarrow backend for lambda aggregation functions

[flori-ko]

Pandas version checks

• I have checked that this issue has not already been reported.

• I have confirmed this bug exists on the latest version of pandas.

• I have confirmed this bug exists on the main branch of pandas.

Reproducible Example

import pandas as pd
import numpy as np

df = pd.DataFrame(data={"A": [np.nan, 1]}, dtype="double[pyarrow]")

df.aggregate(lambda x: x.mean()).dtypes

Issue Description

The input is a dataframe with a pyarrow backend but the output uses numpy float64.
This behaviour is very inconsistent especially considering that, if called like this: df.aggregate("mean").dtypes the output is a "double[pyarrow]".

Expected Behavior

I expect the returned dtypes to be double[pyarrow], since pyarrow type was given as input (based on #53831).

Installed Versions

INSTALLED VERSIONS ------------------ commit : c888af6 python : 3.11.9 python-bits : 64 OS : Windows OS-release : 10 Version : 10.0.22631 machine : AMD64 processor : Intel64 Family 6 Model 143 Stepping 8, GenuineIntel byteorder : little LC_ALL : None LANG : en_US.UTF-8 LOCALE : English_United States.1252

pandas : 2.3.1
numpy : 2.2.5
pytz : 2025.2
dateutil : 2.9.0.post0
pip : 25.1.1
Cython : None
sphinx : 8.2.3
IPython : 9.2.0
adbc-driver-postgresql: None
adbc-driver-sqlite : None
bs4 : 4.13.4
blosc : None
bottleneck : None
dataframe-api-compat : None
fastparquet : None
fsspec : None
html5lib : None
hypothesis : None
gcsfs : None
jinja2 : 3.1.6
lxml.etree : 5.4.0
matplotlib : None
numba : None
numexpr : None
odfpy : None
openpyxl : 3.1.5
pandas_gbq : None
psycopg2 : None
pymysql : None
pyarrow : 20.0.0
pyreadstat : None
pytest : 8.3.5
python-calamine : None
pyxlsb : None
s3fs : None
scipy : 1.15.3
sqlalchemy : 2.0.40
tables : None
tabulate : 0.9.0
xarray : 2025.4.0
xlrd : 2.0.1
xlsxwriter : 3.2.5
zstandard : None
tzdata : 2025.2
qtpy : None
pyqt5 : None

[AdrianoCLeao]

I've taken some time to verify it locally on my setup (pandas 2.3.1, pyarrow 20.0.0, Python 3.10.12).
I was able to test the return in some ways:

• df.aggregate(lambda x: x.mean()) → float64 (loses pyarrow dtype)
• df.aggregate("mean") → double[pyarrow] (preserves pyarrow dtype)
• df.aggregate(np.mean) → double[pyarrow] (preserves pyarrow dtype)
• df.mean() → double[pyarrow] (preserves pyarrow dtype)

So:

• Any aggregation using a lambda function does lose the extension dtype.
• Native string methods and numpy functions both preserve it.

I get the same fallback for df.apply(lambda x: x.mean())—so this isn't just limited to .aggregate() but is about how callables are dispatched internally.

[arthurlw]

Confirmed on main. PRs are welcome!

df.aggregate(np.mean) → double[pyarrow] (preserves pyarrow dtype)

In my testing, this also returns a float64 dtype, so investigations for this are welcome as well.

Thanks for raising this!