Python statsmodels OLS和R的lm的区别

Question

我不确定为什么我的简单OLS会得到略微不同的结果,这取决于我是否通过panda的实验性rpy接口进行回归,R或者我是否在Python中使用statsmodel.

import pandas
from rpy2.robjects import r

from functools import partial

loadcsv = partial(pandas.DataFrame.from_csv,
                  index_col="seqn", parse_dates=False)

demoq = loadcsv("csv/DEMO.csv")
rxq = loadcsv("csv/quest/RXQ_RX.csv")

num_rx = {}
for seqn, num in rxq.rxd295.iteritems():
    try:
        val = int(num)
    except ValueError:
        val = 0
    num_rx[seqn] = val

series = pandas.Series(num_rx, name="num_rx")
demoq = demoq.join(series)

import pandas.rpy.common as com
df = com.convert_to_r_dataframe(demoq)
r.assign("demoq", df)
r('lmout <- lm(demoq$num_rx ~ demoq$ridageyr)')  # run the regression
r('print(summary(lmout))')  # print from R

从中R,我得到以下摘要:

Call:
lm(formula = demoq$num_rx ~ demoq$ridageyr)

Residuals:
    Min      1Q  Median      3Q     Max 
-2.9086 -0.6908 -0.2940  0.1358 15.7003 

Coefficients:
                 Estimate Std. Error t value Pr(>|t|)    
(Intercept)    -0.1358216  0.0241399  -5.626 1.89e-08 ***
demoq$ridageyr  0.0358161  0.0006232  57.469  < 2e-16 ***
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 

Residual standard error: 1.545 on 9963 degrees of freedom
Multiple R-squared: 0.249,  Adjusted R-squared: 0.2489 
F-statistic:  3303 on 1 and 9963 DF,  p-value: < 2.2e-16

用statsmodels.api做OLS:

import statsmodels.api as sm
results = sm.OLS(demoq.num_rx, demoq.ridageyr).fit()
results.summary()

结果类似于R的输出但不相同:

OLS Regression Results
Adj. R-squared:  0.247
Log-Likelihood:  -18488.
No. Observations:    9965    AIC:   3.698e+04
Df Residuals:    9964    BIC:   3.698e+04
             coef   std err  t     P>|t|    [95.0% Conf. Int.]
ridageyr     0.0331  0.000   82.787    0.000        0.032 0.034

安装过程有点麻烦.但是,有一个IPython的笔记本电脑 在这里,可以重现不一致.

Answer 1

看起来Python默认情况下不会向您的表达式添加拦截,而R会在您使用公式界面时执行此操作.

这意味着您确实适合两种不同的型号.尝试

lm( y ~ x - 1, data)

在R中排除拦截,或者在你的情况下,用一些更标准的符号

lm(num_rx ~ ridageyr - 1, data=demoq)

Answer 2

请注意,您仍然可以使用ols来自statsmodels.formula.api:

from statsmodels.formula.api import ols

results = ols('num_rx ~ ridageyr', demoq).fit()
results.summary()

我认为它patsy在后端使用来翻译公式表达式,并自动添加拦截.