Question

I have a dataframe counts1 which looks like:

Factor            w-statin  wo-statin
AgeGroups Cancer                     
0-5       No           108       6575
          Yes            0        223
11-15     No             5       3669
          Yes            1        143
16-20     No            28       6174
          Yes            1        395
21-25     No            80       8173
          Yes            2        624
26-30     No           110       9143
          Yes            2        968
30-35     No           171       9046
          Yes            5       1225
35-40     No           338       8883
          Yes           21       1475

I wanted to calculate the oddsratio (w-statin/wo-statin). I did it old style like I would do it in paper:

counts1['sumwwoStatin']= counts1['w-statin']+counts1['wo-statin']

counts1['oddRatio']=((counts1['w-statin']/counts1['sumwwoStatin'])/(counts1['wo-statin']/counts1['sumwwoStatin']))

Is there a better way to calculate Odds-ratio, Relative risk, Contigency Table, & Chi-Square Tests in Pandas, just like in R? Any suggestions are appreciated. Oh by the way, I forgot to mention how my csv looks like:

    Frequency Cancer     Factor AgeGroups
0         223    Yes  wo-statin       0-5
1         112    Yes  wo-statin      6-10
2         143    Yes  wo-statin     11-15
3         395    Yes  wo-statin     16-20
4         624    Yes  wo-statin     21-25
5         968    Yes  wo-statin     26-30
6        1225    Yes  wo-statin     30-35
7        1475    Yes  wo-statin     35-40
8        2533    Yes  wo-statin     41-45
9        4268    Yes  wo-statin     46-50
10       5631    Yes  wo-statin     52-55
11       6656    Yes  wo-statin     56-60
12       7166    Yes  wo-statin     61-65
13       8573    Yes  wo-statin     66-70
14       8218    Yes  wo-statin     71-75
15       4614    Yes  wo-statin     76-80
16       1869    Yes  wo-statin     81-85
17        699    Yes  wo-statin     86-90
18        157    Yes  wo-statin     91-95
19         31    Yes  wo-statin    96-100
20          5    Yes  wo-statin      >100
21        108     No   w-statin       0-5
22          6     No   w-statin      6-10
23          5     No   w-statin     11-15
24         28     No   w-statin     16-20
25         80     No   w-statin     21-25
26        110     No   w-statin     26-30
27        171     No   w-statin     30-35
28        338     No   w-statin     35-40
29        782     No   w-statin     41-45
..

Answer 1

AFAIK熊猫不提供统计计算和测试，除了基本时刻，如均值，方差，相关等...

但是，您可以依赖scipy来满足此要求。你会发现你需要的大部分内容。例如，计算优势比：

import scipy.stats as stats

table = df.groupby(level="Cancer").sum().values
print(table)

>>> array([[  840, 51663],
           [   32,  5053]])

oddsratio, pvalue = stats.fisher_exact(table)
print("OddsR: ", oddsratio, "p-Value:", pvalue)

>>> OddsR:  2.56743220487 p-Value: 2.72418938361e-09

有关详情，请参阅here和here。

Answer 2

statsmodels 可用于估计优势比，如果两个变量都已经是同一 Pandas 数据框的一部分：

import statsmodels.api as sm

table = sm.stats.Table.from_data(df[['w-statin','wo-statin']])
rslt = table.test_nominal_association()
print(table.local_oddsratios)
print( rslt.pvalue )

Answer 3

我不知道在Pandas中执行此操作的方法...但是，您可以通过首先使用scikit-learn库找到相应的beta来计算Python中逻辑回归模型的优势比值，如下所述：

How to find beta values in Logistic Regression model with sklearn

该线程描述了如何从逻辑回归模型生成和提取Beta系数。然后，您可以使用Python NumPy包中的exp（）函数对Beta值取幂，从而计算出优势比。以这种方式计算出的几率将等于R通过glm（）函数提供的几率，指定二项式分布。

A Better Way to Calculate Odd Ratio in Pandas

3 个答案: