Pandas Passing a list to level param of unstack() unexpectedly changes dtype from numeric to object

I experience these strange behavior of unstack() when its level parameter gets a list instead of string. This changes dtype of some columns from numeric to object, which is obviously a worrying change. I am wondering if this a bug or there is something I am missing about dealing with this. I am on pandas 0.16.0, numpy 1.9.2.

Here is a code to illustrate the problem:

data = """a,b,c,d
2,101,A,11
3,101,A,12
1,201,B,21
2,201,B,22
3,201,B,23
1,301,C,33
2,301,C,32"""
df = pd.read_csv(pd.core.common.StringIO(data))
print df.dtypes
df_str = df.set_index(['a','b']).unstack(level='a')  # Note that level has a str
print df_str.dtypes
df_list = df.set_index(['a','b']).unstack(level=['a'])  # Note that level has a list
print df_list.dtypes

Which produces:

a     int64
b     int64
c    object
d     int64   <-- Note original dtype
dtype: object

   a
c  1     object
   2     object
   3     object
d  1    float64   <-- expected change of dtype due to NaN addition
   2    float64   <-- expected change of dtype due to NaN addition
   3    float64   <-- expected change of dtype due to NaN addition
dtype: object

   a
c  2    object
   3    object
   1    object
d  2    object   <-- UNexpected change of dtype, when level got list
   3    object   <-- UNexpected change of dtype, when level got list
   1    object   <-- UNexpected change of dtype, when level got list
dtype: object

Comment From: jreback

this is fixed in master. if someone can put the appropriate reference, let's close.

Comment From: gfyoung

@jreback : I believe this is a duplicate #11847 and patched in 0.19.2. Closing as a duplicate.

@wikiped : Could you try upgrading your pandas version? We are currently at 0.20.3.