Describe the bug

I hit this while feeding describe the output of an upstream filter that can
legitimately return zero rows. Instead of an empty/NaN summary, I get an error
coming from deep inside pandas/numpy, so at first it wasn't obvious the problem
was the empty input at all.

What makes it a bit worse is that the error depends on the column dtype, which
suggests there are actually two separate places where a 0-row frame isn't
handled:

• a numeric (e.g. float) column raises
  ValueError: Length of values (2) does not match length of index (1)
• a categorical column raises KeyError: 2

Neither message mentions that the input is empty, which is the real cause.

For what it's worth, pandas.DataFrame.describe() accepts the same 0-row inputs
happily and just returns a frame of NaNs, so this felt like something
statsmodels could handle more gracefully too.

Code Sample, a copy-pastable example if possible

import pandas as pd
from statsmodels.stats.descriptivestats import describe

# 1) numeric, 0 rows
#    -> ValueError: Length of values (2) does not match length of index (1)
describe(pd.DataFrame({"a": pd.Series([], dtype=float)}))

# 2) categorical, 0 rows
#    -> KeyError: 2
describe(pd.DataFrame({"a": pd.Series([], dtype="category")}))

# For comparison, pandas handles both without error (returns a frame of NaN):
pd.DataFrame({"a": pd.Series([], dtype=float)}).describe()
pd.DataFrame({"a": pd.Series([], dtype="category")}).describe()

I'm on the released 0.14.6, but I also checked the current main and the
relevant code is unchanged, so this reproduces there as well.

Expected Output

Either a frame of NaN statistics (matching pandas.DataFrame.describe()), or a
clear ValueError telling me the input needs at least one observation. Right now
I just get an internal error that doesn't point at the actual problem.

Root cause (as far as I can tell)

In statsmodels/stats/descriptivestats.py, Description.__init__ validates the
selected column types, ntop, percentiles and alpha, but there's no check
for an empty (0-row) input. The empty frame then reaches the per-dtype
statistics, which assume at least one row:

• the numeric path breaks in the Description.numeric property, where
  df.apply(_mode) on a 0-row frame makes pandas build a Series from the
  2-tuple returned by _mode against a length-1 index — hence the ValueError;
• the categorical path fails separately (the KeyError), so it looks like the
  cleanest place to fix this is a single guard in __init__, rather than
  patching _mode.

Two ways to fix it

I'd be happy to open a PR, but I think it's a design call, so I'd rather let you
decide between:

• (a) fail fast — raise an informative ValueError in Description.__init__
  when there are zero observations (right next to the existing "empty DataFrame"
  check that runs after select_dtypes);
• (b) graceful — return a frame of NaNs, matching
  pandas.DataFrame.describe().

Just let me know which you'd prefer and I'll send a PR with a test.

Small unrelated thing in the same spot

While reading that code I noticed the adjacent "empty DataFrame" error message is
missing a space and reads Selecting numericand categorical results in an empty DataFrame (should be numeric and categorical). It shows up with e.g.
Description(pd.DataFrame({"a": ["x", "y"]}), numeric=True, categorical=True).
Happy to fold the one-character fix into the same PR if that's alright.

Output of import statsmodels.api as sm; sm.show_versions()

INSTALLED VERSIONS
------------------
Python: 3.13.12.final.0

statsmodels
===========
Installed: 0.14.6

Required Dependencies
=====================
cython: Not installed
numpy: 2.4.4
scipy: 1.17.1
pandas: 2.3.3
    dateutil: 2.9.0.post0
patsy: 1.0.2

Optional Dependencies
=====================
matplotlib: 3.10.9 (backend: tkagg)
cvxopt: Not installed
joblib: 1.5.3

Developer Tools
================
IPython: Not installed
    jinja2: 3.1.6
sphinx: Not installed
    pygments: 2.19.2
pytest: 9.0.3
virtualenv: Not installed