Code Sample, a copy-pastable example if possible

import pandas as pd
from numpy import nan
d = {(0,
  'NNPDF31_nlo_as_0130_uncorr__combined',
  'chi2'): {('NMC', 'Total', 325, 394): nan, ('NMC',
   'Total',
   325,
   400): 906.5202301172686, ('SLAC', 'Total', 67, 394): nan, ('SLAC',
   'Total',
   67,
   400): 110.50940462648715},
 (1,
  'NNPDF31_nlo_as_0130_uncorr__combined',
  'chi2'): {('NMC', 'Total', 325, 394): 801.561757289261, ('NMC',
   'Total',
   325,
   400): 875.9830814675471, ('SLAC',
   'Total',
   67,
   394): 126.25943979466223, ('SLAC', 'Total', 67, 400): 101.69776217874313}}

df = pd.DataFrame.from_dict(d)
df.groupby(level=3).sum()

Problem description

The code above behaves differently in pandas 0.20 and in 0.22 (and also master) thereby causing incorrect results in code relying on the old behaviour. In pandas master I get:

                                       0                                    1
    NNPDF31_nlo_as_0130_uncorr__combined NNPDF31_nlo_as_0130_uncorr__combined
                                    chi2                                 chi2
394                             0.000000                           927.821197
400                          1017.029635                           977.680844

Expected Output

                                       0                                    1
    NNPDF31_nlo_as_0130_uncorr__combined NNPDF31_nlo_as_0130_uncorr__combined
                                    chi2                                 chi2
394                                  NaN                           927.821197
400                          1017.029635                           977.680844

Output of pd.show_versions()

INSTALLED VERSIONS ------------------ commit: None python: 3.6.4.final.0 python-bits: 64 OS: Linux OS-release: 4.13.0-32-generic machine: x86_64 processor: x86_64 byteorder: little LC_ALL: None LANG: en_US.UTF-8 LOCALE: en_US.UTF-8 pandas: 0.23.0.dev0+261.g6485a36 pytest: 3.3.2 pip: 9.0.1 setuptools: 38.4.0 Cython: 0.27.3 numpy: 1.13.3 scipy: 1.0.0 pyarrow: 0.8.0 xarray: 0.10.0 IPython: 6.2.1 sphinx: 1.6.6 patsy: 0.5.0 dateutil: 2.6.1 pytz: 2017.3 blosc: None bottleneck: 1.2.1 tables: 3.4.2 numexpr: 2.6.4 feather: 0.4.0 matplotlib: 2.1.2 openpyxl: 2.4.10 xlrd: 1.1.0 xlwt: 1.3.0 xlsxwriter: 1.0.2 lxml: 4.1.1 bs4: 4.6.0 html5lib: 1.0.1 sqlalchemy: 1.2.1 pymysql: 0.7.11.None psycopg2: None jinja2: 2.10 s3fs: 0.1.2 fastparquet: 0.1.4 pandas_gbq: None pandas_datareader: None

I understand the details of how this works have changed several times since 0.20 but I think it should not break by code if possible.

Comment From: jreback

pls read the whatsnew note:http://pandas.pydata.org/pandas-docs/stable/whatsnew.html

.sum() defaults have changed.