R中Anova上的TurkeyHSD包含意外的比较

Question

我有一个部分显示在下面的数据集。我使用了方差分析来观察区域作为协变量的值中性别之间的差异。然而，当使用TukeyHSD进行多次比较校正时，我发现它显示了比必要的更多的比较（如下）。我唯一感兴趣的部分是第一行（即每个区域内的M与F），而不是区域之间。有没有办法在anova模型或Tukey中指定这一点？谢谢

ID  Age Sex Hemi Region value
1   62  M   R   a   1.81
2   62  M   R   a   1.90
3   72  M   R   a   2.25
1   61  M   L   a   1.58
2   57  F   L   a   2.66
3   62  M   L   a   2.19
1   72  M   R   b   1.93
2   64  F   R   b   1.07
3   65  F   R   b   1.37
1   64  M   L   b   0.97
2   77  F   L   b   1.59
3   27  M   L   b   1.84

model=aov(value~Sex*Region,data=dataTukeyHSD(model)

$Sex:as.factor(Region)
         diff          lwr        upr           p adj 
M:a-F:a  0.1125161943  -0.2989139  0.5239462408 1.0000000
F:b-F:a  -0.0383423077 -0.4866878  0.4100031910 1.0000000
M:b-F:a  0.1273767206  -0.2840533  0.5388067671 1.0000000

Answer 1

当然有。排除模型中的交互项。

> my.data <- data.frame(y = rnorm(100), region = rep(c("a", "b"), each = 50), gender = sample(c("M", "F"), 100, replace = TRUE))
> head(my.data)
           y region gender
1  0.3333316      a      M
2  1.6364059      a      F
3  0.6679500      a      F
4 -0.7460313      a      M
5  0.7327712      a      F
6 -0.8305134      a      M
> tail(my.data)
               y region gender
95  -1.387498267      b      F
96   0.124378046      b      M
97  -0.568782743      b      M
98  -0.004699849      b      M
99  -1.449213423      b      F
100 -1.146313313      b      F
> 
> mdl1 <- aov(y ~ region * gender, data = my.data)
> TukeyHSD(mdl1)
  Tukey multiple comparisons of means
    95% family-wise confidence level
Fit: aov(formula = y ~ region * gender, data = my.data)
$region
          diff        lwr       upr     p adj
b-a -0.0140727 -0.4041448 0.3759994 0.9430592
$gender
        diff        lwr       upr     p adj
M-F -0.13387 -0.5267813 0.2590414 0.5004702
$`region:gender`
               diff        lwr        upr     p adj
b:F-a:F -0.57419765 -1.3487805 0.20038521 0.2191452
a:M-a:F -0.63398152 -1.3658935 0.09793048 0.1136445
b:M-a:F -0.20795605 -0.9398681 0.52395596 0.8794703
a:M-b:F -0.05978387 -0.7916959 0.67212813 0.9965356
b:M-b:F  0.36624160 -0.3656704 1.09815361 0.5599334
b:M-a:M  0.42602548 -0.2605688 1.11261980 0.3710470
> 
> mdl2 <- update(mdl1, . ~ region + gender)
> TukeyHSD(mdl2)
  Tukey multiple comparisons of means
    95% family-wise confidence level
Fit: aov(formula = y ~ region + gender, data = my.data)
$region
          diff        lwr       upr     p adj
b-a -0.0140727 -0.4147704 0.3866251 0.9445724
$gender
        diff        lwr       upr     p adj
M-F -0.13387 -0.5374843 0.2697444 0.5119119

如果你想删除/保留特定的条款，你可以这样做。

> tk <- TukeyHSD(mdl1)
> str(tk)
List of 3
 $ region       : num [1, 1:4] -0.0141 -0.4041 0.376 0.9431
  ..- attr(*, "dimnames")=List of 2
  .. ..$ : chr "b-a"
  .. ..$ : chr [1:4] "diff" "lwr" "upr" "p adj"
 $ gender       : num [1, 1:4] -0.134 -0.527 0.259 0.5
  ..- attr(*, "dimnames")=List of 2
  .. ..$ : chr "M-F"
  .. ..$ : chr [1:4] "diff" "lwr" "upr" "p adj"
 $ region:gender: num [1:6, 1:4] -0.5742 -0.634 -0.208 -0.0598 0.3662 ...
  ..- attr(*, "dimnames")=List of 2
  .. ..$ : chr [1:6] "b:F-a:F" "a:M-a:F" "b:M-a:F" "a:M-b:F" ...
  .. ..$ : chr [1:4] "diff" "lwr" "upr" "p adj"
 - attr(*, "class")= chr [1:2] "TukeyHSD" "multicomp"
 - attr(*, "orig.call")= language aov(formula = y ~ region * gender, data = my.data)
 - attr(*, "conf.level")= num 0.95
 - attr(*, "ordered")= logi FALSE
> rntk <- rownames(tk[["region:gender"]])
> tk[["region:gender"]] <- tk[["region:gender"]][grepl("b:F-a:F|a:M-a:F", rntk), ]
> tk
  Tukey multiple comparisons of means
    95% family-wise confidence level
Fit: aov(formula = y ~ region * gender, data = my.data)
$region
          diff        lwr       upr     p adj
b-a -0.0140727 -0.4041448 0.3759994 0.9430592
$gender
        diff        lwr       upr     p adj
M-F -0.13387 -0.5267813 0.2590414 0.5004702
$`region:gender`
              diff       lwr        upr     p adj
b:F-a:F -0.5741976 -1.348781 0.20038521 0.2191452
a:M-a:F -0.6339815 -1.365894 0.09793048 0.1136445