This question would be solved by also changing index to 'who' and removing margins parameter as not needed.

pd.pivot_table(df[df['who'] == 'child'], 
    index = 'who', 
    aggfunc = lambda x: x.isnull().sum())