R:解读ANOVA和TukeyHSD分析的结果。

I have run an ANOVA and TukeyHSD over a dataframe containing anatomical regions in column 1 (region) and gene expression values in column 2 (S1). I normally would expect the p-value from the aov summary to be expressed as Pr(>F), so I'm a little fuzzy on the results I've obtained. Also, can someone help me understand the Tukey multiple comparisons of means results? I'm not totally clear on what the diff and p adj results indicate. The results shown here are an abridged version of what I'm actually working with, FYI.

我在第1列(区域)和第2列(S1)的基因表达值上运行了一个ANOVA和TukeyHSD。我通常期望从aov总结中得到的p值表示为Pr(>F)，所以我对得到的结果有点模糊。还有，有人能帮我理解一下多重比较的结果吗?我还不清楚diff和p的结果表明了什么。这里显示的结果是我实际工作的一个简化版本，FYI。

> aov.result = aov(S1 ~ region, data=raw.data)
> summary(aov.result)
             Df  Sum Sq Mean Sq F value    Pr(>F)    
region       60  61.713 1.02856  5.9246 < 2.2e-16 ***
Residuals   655 113.712 0.17361                      
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 
> TukeyHSD(aov.result)
Tukey multiple comparisons of means
    95% family-wise confidence level

Fit: aov(formula = S1 ~ region, data = raw.data)

$region
                     diff           lwr          upr     p adj
AB-AA        0.4118651583 -2.864195e-01  1.110149848 0.9847745
AHA-AA      -0.0468785098 -7.608569e-01  0.667099930 1.0000000
APir-AA      0.4419135565 -2.563711e-01  1.140198246 0.9502924
B-AA         0.5379787168 -1.603060e-01  1.236263406 0.5846356

1 个解决方案

#1

Lets start with some reproducible data, one factor and one continuous variable:

让我们从一些可重复的数据开始，一个因素和一个连续变量:

set.seed(1)
df1 <- data.frame(
    f1=as.factor(rep(seq(1:3),4)),
    c1=abs(rnorm(12)))
s1 <- stats::aov(df1$c1 ~ df1$f1)
summary(s1)

This gives output similar to yours.

这将输出类似于您的输出。

The P-value for your data appears correct and can be confirmed with e.g.:

你的数据的p值看起来是正确的，可以用例来确认。

1-stats::pf(q=5.92, df1=60, df2=655)
[1] 0

Now, looking at output from:

现在，看看输出:

s2 <- stats::TukeyHSD.aov(s1)

i.e.

即。

$`df1$f1`
           diff       lwr       upr     p adj
2-1 -0.06282377 -1.038236 0.9125887 0.9823655
3-1 -0.09820762 -1.073620 0.8772048 0.9575774
3-2 -0.03538385 -1.010796 0.9400286 0.9943641

The first column is the difference in the means. In my example:

第一列是均值之差。在我的例子:

m1 <- mean( df1$c1[df1$f1==1] )
m2 <- mean( df1$c1[df1$f1==2] )

Now m2-m1 is approximately equal to s2$"df1$f1"[1,1], here -0.068..

现在m2-m1近似等于s2$"df1$f1"[1,1]，这里-0.068。

This 'difference of means' has a confidence interval calculated from the studentized range (q) distribution. The mechanics can be found in the source code of stats::TukeyHSD.aov(). See also ?ptukey. Note also the rationale for 'correction for multiple comparisons' is controversial in certain contexts. This sort of question might be better suited to CrossValidated.

这种“均值差异”的置信区间是由研究范围(q)分布计算出来的。可以在stats的源代码中找到该机制:TukeyHSD.aov()。参见? ptukey。还请注意，在某些情况下，“多次比较”的理由是有争议的。这类问题可能更适合交叉验证。

#1