如何在 nlme 和 lme4 中指定不同的随机效应?

How to specify different random effects in nlme vs. lme4?

我想使用 nlme::lme(底部的数据)在模型中指定不同的随机效应。随机效应是:1) interceptpositionsubject 内变化; 2) interceptcomparison 内变化。使用 lme4::lmer 很简单:

lmer(rating ~ 1 + position + 
     (1 + position | subject) + 
     (1 | comparison), data=d)

> ...
Random effects:
 Groups     Name        Std.Dev. Corr 
 comparison (Intercept) 0.31877       
 subject    (Intercept) 0.63289       
            position    0.06254  -1.00
 Residual               0.91458      
 ...

但是,我想坚持使用 lme,因为我还想对自相关结构建模(position 是一个时间变量)。 如何使用lme实现与上面相同的效果?我在下面尝试嵌套效果,这不是我想要的。

lme(rating ~ 1 + position,
random = list( ~ 1 + position | subject,
               ~ 1 | comparison), data=d)

> ...
Random effects:
 Formula: ~1 + position | subject
 Structure: General positive-definite, Log-Cholesky parametrization
            StdDev     Corr  
(Intercept) 0.53817955 (Intr)
position    0.04847635 -1    

 Formula: ~1 | comparison %in% subject    # NESTED :(
        (Intercept)     Residual
StdDev:   0.9707665 0.0002465237
...

注意: SO 和 CV 上有一些类似的问题 here, here, and here 但我要么不明白答案要么建议使用 lmer这不算在这里 ;)

示例中使用的数据

d <- structure(list(rating = c(2, 3, 4, 3, 2, 4, 4, 3, 2, 1, 3, 2, 
2, 2, 4, 2, 4, 3, 2, 2, 3, 5, 3, 4, 4, 4, 3, 2, 3, 5, 4, 5, 2, 
3, 4, 2, 4, 4, 1, 2, 4, 5, 4, 2, 3, 4, 3, 2, 2, 2, 4, 5, 4, 4, 
5, 2, 3, 4, 3, 2), subject = structure(c(1L, 1L, 1L, 1L, 1L, 
1L, 1L, 1L, 1L, 1L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 3L, 
3L, 3L, 3L, 3L, 3L, 3L, 3L, 3L, 3L, 4L, 4L, 4L, 4L, 4L, 4L, 4L, 
4L, 4L, 4L, 5L, 5L, 5L, 5L, 5L, 5L, 5L, 5L, 5L, 5L, 6L, 6L, 6L, 
6L, 6L, 6L, 6L, 6L, 6L, 6L), .Label = c("1", "2", "3", "4", "5", 
"6", "7", "8", "9", "10", "11", "12", "13", "14", "15", "16", 
"17", "18", "19", "20", "21", "22", "23", "24", "25", "26", "27", 
"28", "29", "30", "31", "32", "33", "34", "35", "36", "37", "38", 
"39", "40", "41", "42", "43", "44", "45", "46", "47", "48", "49", 
"50", "51", "52", "53", "54", "55", "56", "57", "58", "59", "60", 
"61", "62", "63"), class = "factor"), position = c(1, 2, 3, 4, 
5, 6, 7, 8, 9, 10, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 1, 2, 3, 4, 
5, 6, 7, 8, 9, 10, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 1, 2, 3, 4, 
5, 6, 7, 8, 9, 10, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10), comparison = structure(c(1L, 
7L, 9L, 8L, 3L, 4L, 10L, 2L, 5L, 6L, 2L, 6L, 4L, 5L, 8L, 10L, 
7L, 3L, 1L, 9L, 3L, 9L, 10L, 1L, 5L, 7L, 6L, 8L, 2L, 4L, 4L, 
2L, 8L, 6L, 7L, 5L, 1L, 10L, 9L, 3L, 5L, 10L, 6L, 3L, 2L, 9L, 
4L, 1L, 8L, 7L, 6L, 5L, 2L, 10L, 4L, 3L, 8L, 9L, 7L, 1L), contrasts = structure(c(1, 
0, 0, 0, 0, 0, 0, 0, 0, -1, 0, 1, 0, 0, 0, 0, 0, 0, 0, -1, 0, 
0, 1, 0, 0, 0, 0, 0, 0, -1, 0, 0, 0, 1, 0, 0, 0, 0, 0, -1, 0, 
0, 0, 0, 1, 0, 0, 0, 0, -1, 0, 0, 0, 0, 0, 1, 0, 0, 0, -1, 0, 
0, 0, 0, 0, 0, 1, 0, 0, -1, 0, 0, 0, 0, 0, 0, 0, 1, 0, -1, 0, 
0, 0, 0, 0, 0, 0, 0, 1, -1), .Dim = c(10L, 9L), .Dimnames = list(
    c("1", "2", "3", "4", "5", "6", "7", "8", "9", "10"), NULL)), .Label = c("1", 
"2", "3", "4", "5", "6", "7", "8", "9", "10"), class = "factor")), .Names = c("rating", 
"subject", "position", "comparison"), row.names = c(1L, 2L, 3L, 
4L, 5L, 6L, 7L, 8L, 9L, 10L, 111L, 112L, 113L, 114L, 115L, 116L, 
117L, 118L, 119L, 120L, 221L, 222L, 223L, 224L, 225L, 226L, 227L, 
228L, 229L, 230L, 331L, 332L, 333L, 334L, 335L, 336L, 337L, 338L, 
339L, 340L, 441L, 442L, 443L, 444L, 445L, 446L, 447L, 448L, 449L, 
450L, 551L, 552L, 553L, 554L, 555L, 556L, 557L, 558L, 559L, 560L
), class = "data.frame")

一段时间以来,我一直想弄清楚这个问题。如果没有更多的工作,我认为我无法获得与 lme4 中完全相同的模型,但我可以接近。

## source("SO36643713.dat")
library(nlme)
library(lme4)

这是您想要的模型,具有 subject 的完整随机斜率项(相关斜率和截距)和 comparison 的随机截距:

m1 <- lmer(rating ~ 1 + position + 
               (1 + position | subject) + 
               (1 | comparison), data=d)

这是我可以弄清楚如何在 lme 中复制的那个:独立的截距和斜率。 (我不是特别喜欢这些模型,但它们作为人们简化过于复杂的随机效应模型的一种方式相当普遍。)

m2 <- lmer(rating ~ 1 + position + 
               (1 + position || subject) + 
               (1 | comparison), data=d)

结果:

VarCorr(m2)
##  Groups     Name        Std.Dev.
##  comparison (Intercept) 0.28115 
##  subject    position    0.00000 
##  subject.1  (Intercept) 0.28015 
##  Residual               0.93905 

对于这个特定的数据集,估计随机斜率无论如何都具有零方差。

现在让我们为 lme 设置它。关键 (???) 见解是 pdBlocked() 矩阵中的所有项必须 嵌套在同一分组变量 中。例如,Pinheiro 和 Bates 第 163ff 页上的交叉随机效应示例将块、块内的行和块内的列作为随机效应。由于没有 comparisonsubject 都嵌套在其中的分组因子,我将组成一个 dummy "factor" 将整个数据集包含在一个单块:

d$dummy <- factor(1)

现在我们可以拟合模型了。

m3 <- lme(rating~1+position,
          random=list(dummy =
                pdBlocked(list(pdIdent(~subject-1),
                               pdIdent(~position:subject),
                               pdIdent(~comparison-1)))),
          data=d)

我们在随机效应方差-协方差矩阵中有三个块:一个用于 subject,一个用于 position-by-subject 交互,一个用于 comparison。除了定义全新的 pdMat class,我想不出一个简单的方法让每个斜率 (position:subjectXX) 与其对应的截距 (subjectXX). (您可能认为可以使用 pdBlocked 结构来设置它,但我看不到任何方法可以将方差估计值限制为在 pdBlocked 对象内的多个块之间相同。)

结果几乎相同,尽管报告不同。

vv <- VarCorr(m3)
vv2 <- vv[c("subject1","position:subject1","comparison1","Residual"),]
storage.mode(vv2) <- "numeric"
print(vv2,digits=4)
                   Variance    StdDev
subject1          7.849e-02 2.802e-01
position:subject1 4.681e-11 6.842e-06
comparison1       7.905e-02 2.812e-01
Residual          8.818e-01 9.390e-01