R学习如何进行配对T检验

T检验是小样本均值比较的正选手段。根据抽样数据的性质不同，T检验又分为单样本T检验、配对T检验和独立样本T检验，并有一些诸如总体正态分布、方差齐性等前置要求。这里我们只简单讲一下配对T检验的实现过程。

首先我们生成一组事前和事后的绩效样本：

>pretest<-c(85,78,82,79,73,68)
>aftertest<-c(86,78,80,82,75,70)

然后用T.TEST命令执行T检验并查看结果：

>t.test(pretest,aftertest,paired=T)

Pairedt-test

data:pretest and aftertest
t = -1.3693, df = 5, p-value =0.2292
alternative hypothesis: truedifference in means is not equal to 0
95 percent confidenceinterval:
-2.87728750.8772875
sample estimates:
mean of thedifferences
-1

由于默认是双尾检验，P值应该小于0.025才会推翻原假设，而检验P值结果为0.2292，故不能推翻原假设，即事前跟事后没有明显差异。

上面默认是双尾检验，下面看一下单尾（大于）检验：

>t.test(pretest,aftertest,paired=T,alternative="greater")

Pairedt-test

data:pretest and aftertest
t = -1.3693, df = 5, p-value =0.8854
alternative hypothesis: truedifference in means is greater than 0
95 percent confidenceinterval:
-2.471583Inf
sample estimates:
mean of thedifferences
-1

毫无意外，检验不通过，无明显差异。

T.TEST的默认置信度为0.95，如果调整置信度的话，只需加入conf.level参数：

>t.test(pretest,aftertest,paired=T,alternative="greater",conf.level=.99)

Pairedt-test数据分析师培训

data:pretest and aftertest
t = -1.3693, df = 5, p-value =0.8854
alternative hypothesis: truedifference in means is greater than 0
99 percent confidenceinterval:
-3.457397Inf
sample estimates:
mean of thedifferences
-1