r - 计算与 'hetcor' 的多元相关时，关于 log(P) 的警告消息是什么意思？

Question

在 R 中计算多元相关时(library(polycor)，函数 hetcor)我收到警告消息 In log(P) : NaNs produced。我无法弄清楚此警告消息可能构成什么。我想这与用于测试双变量正态性的 p 值的计算有关。

因此我的问题是:

• 此数据集的哪些特征会导致此警告？
• 这个警告是什么意思？
• 就使用多元相关矩阵进行进一步分析而言，此警告是否有问题？

数据子集:

foo <- structure(list(item1 = structure(c(4L, 4L, 4L, 2L, 2L, 2L, 
2L, 2L, 4L, 2L, 2L, 3L, 2L, 3L, 2L, 2L, 2L, 3L, 2L, 2L, 3L, 1L, 
2L, 2L, 3L, 3L, 3L, 2L, 2L, 1L, 1L, 2L, 3L, 2L, 2L, 3L, 2L, 3L, 
2L, 2L, 2L, 2L, 3L, 2L, 2L, 2L, 2L, 2L, 2L, 3L, 2L, 2L, 2L, 2L, 
2L, 2L, 2L, 2L, 3L, 3L, 2L, 3L, 3L, 3L, 2L, 2L, 2L, 1L, 2L, 2L, 
2L, 2L, 2L, 2L, 2L, 2L, 1L, 3L, 2L, 2L, 1L, 2L, 2L, 2L, 1L, 2L, 
1L, 2L, 2L, 4L, 2L, 4L, 2L, 2L, 3L, 1L, 2L, 1L, 2L, 2L, 2L, 1L, 
2L, 2L, 3L, 2L, 2L, 2L, 3L, 1L, 2L, 2L, 2L, 2L, 4L, 2L, 2L, 2L, 
2L, 2L, 2L, 4L, 2L, 2L, 1L, 2L, 2L, 2L, 1L, 2L, 1L, 2L, 3L, 3L, 
1L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 3L, 2L, 3L, 3L, 3L
), .Label = c("0", "1", "2", "3"), class = c("ordered", "factor"
)), item2 = structure(c(1L, 1L, 1L, 1L, 1L, 1L, 1L, 2L, 2L, 
1L, 2L, 2L, 1L, 1L, 1L, 1L, 2L, 3L, 2L, 1L, 3L, 2L, 1L, 1L, 3L, 
1L, 3L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 2L, 1L, 3L, 2L, 2L, 1L, 
3L, 2L, 1L, 2L, 1L, 1L, 1L, 1L, 2L, 1L, 1L, 1L, 1L, 3L, 1L, 1L, 
2L, 3L, 2L, 1L, 2L, 2L, 3L, 1L, 2L, 2L, 2L, 2L, 2L, 1L, 2L, 1L, 
1L, 1L, 1L, 1L, 2L, 1L, 1L, 1L, 1L, 2L, 2L, 1L, 2L, 1L, 1L, 1L, 
2L, 2L, 3L, 1L, 2L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 2L, 1L, 2L, 
2L, 1L, 2L, 1L, 2L, 1L, 3L, 2L, 1L, 3L, 1L, 1L, 1L, 2L, 2L, 1L, 
2L, 1L, 3L, 1L, 1L, 1L, 2L, 1L, 1L, 1L, 1L, 1L, 4L, 1L, 1L, 1L, 
1L, 2L, 1L, 1L, 1L, 1L, 2L, 1L, 1L, 4L, 1L, 1L, 3L), .Label = c("0", 
"1", "2", "3"), class = c("ordered", "factor")), item3 = structure(c(4L, 
4L, 4L, 1L, 1L, 1L, 1L, 1L, 2L, 1L, 1L, 4L, 1L, 2L, 1L, 1L, 1L, 
1L, 2L, 1L, 4L, 2L, 2L, 1L, 3L, 2L, 2L, 1L, 1L, 1L, 1L, 1L, 1L, 
1L, 1L, 2L, 2L, 2L, 2L, 1L, 1L, 2L, 3L, 1L, 1L, 1L, 2L, 1L, 1L, 
2L, 1L, 2L, 1L, 1L, 1L, 1L, 1L, 1L, 2L, 2L, 1L, 1L, 2L, 2L, 1L, 
1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 2L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 
1L, 1L, 2L, 1L, 2L, 1L, 1L, 2L, 2L, 1L, 2L, 1L, 1L, 1L, 1L, 1L, 
1L, 3L, 1L, 1L, 1L, 2L, 1L, 1L, 1L, 2L, 1L, 2L, 2L, 1L, 2L, 3L, 
1L, 3L, 1L, 1L, 1L, 1L, 2L, 2L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 
1L, 1L, 1L, 2L, 3L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 
2L, 1L, 3L, 2L, 1L), .Label = c("0", "1", "2", "3"), class = c("ordered", 
"factor")), item4 = structure(c(1L, 1L, 1L, 1L, 1L, 1L, 1L, 
2L, 1L, 1L, 1L, 3L, 1L, 2L, 1L, 1L, 1L, 2L, 2L, 1L, 3L, 2L, 1L, 
1L, 3L, 1L, 3L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 2L, 1L, 2L, 2L, 
2L, 1L, 2L, 2L, 1L, 2L, 1L, 1L, 1L, 1L, 2L, 1L, 1L, 1L, 1L, 2L, 
1L, 1L, 1L, 1L, 1L, 1L, 2L, 2L, 3L, 1L, 2L, 3L, 2L, 1L, 1L, 1L, 
1L, 1L, 2L, 1L, 1L, 1L, 2L, 1L, 2L, 1L, 1L, 1L, 1L, 1L, 2L, 1L, 
1L, 2L, 1L, 2L, 3L, 1L, 2L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 2L, 
1L, 2L, 2L, 2L, 3L, 1L, 1L, 2L, 2L, 2L, 1L, 3L, 1L, 1L, 1L, 2L, 
2L, 1L, 1L, 1L, 2L, 1L, 3L, 1L, 1L, 1L, 1L, 1L, 1L, 2L, 4L, 1L, 
1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 2L, 4L, 1L, 2L, 3L), .Label = c("0", 
"1", "2", "3"), class = c("ordered", "factor")), item5 = structure(c(4L, 
4L, 4L, 1L, 1L, 1L, 1L, 2L, 3L, 2L, 2L, 4L, 2L, 3L, 2L, 1L, 1L, 
3L, 3L, 3L, 4L, 3L, 2L, 1L, 3L, 3L, 4L, 1L, 2L, 1L, 1L, 1L, 2L, 
2L, 2L, 3L, 3L, 3L, 3L, 1L, 1L, 3L, 4L, 2L, 1L, 2L, 2L, 2L, 2L, 
3L, 1L, 2L, 1L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 4L, 3L, 3L, 1L, 
2L, 1L, 1L, 3L, 1L, 2L, 2L, 1L, 3L, 2L, 1L, 2L, 2L, 1L, 1L, 2L, 
1L, 2L, 4L, 2L, 2L, 1L, 2L, 2L, 4L, 2L, 4L, 1L, 1L, 2L, 1L, 1L, 
1L, 2L, 2L, 2L, 2L, 3L, 2L, 3L, 2L, 1L, 3L, 2L, 1L, 1L, 3L, 3L, 
1L, 4L, 1L, 1L, 1L, 1L, 2L, 3L, 3L, 3L, 2L, 1L, 3L, 2L, 1L, 1L, 
1L, 1L, 2L, 3L, 4L, 1L, 1L, 1L, 1L, 2L, 1L, 2L, 1L, 1L, 3L, 1L, 
3L, 3L, 4L, 3L, 3L), .Label = c("0", "1", "2", "3"), class = c("ordered", 
"factor"))), .Names = c("item1", "item2", "item3", "item4", 
"item5"))

相关矩阵的计算:

hetcor(foo)

评论:真实数据集包含大约 2500 行(和更多变量)，但在评估列联表时，稀疏矩阵似乎不是问题。

Answer 1

对一个非常古老的问题的简短(和迟来的)回答。该警告是因为变量交叉表中的某些单元格(例如，变量 1 和 2)的单元格中的值为 0。这可能会导致估计问题。

多元(和四元)相关性是二元正态(和连续)数据转换为分类(四元二分，多元多元)数据时会发生什么的正态理论近似值。正态理论近似假设所有单元格都有一些值。但是，可以找到具有 0 个单元格值的相关性，但带有警告。由此产生的相关性是正确的，但不稳定，因为如果我们为连续性添加一个小的校正(即，将 .1 或 .5 添加到 0 单元格)，值会发生很大变化。 Gunther 和 Hofler 针对四项相关的情况讨论了这个问题，他们将解决方案与连续性校正进行了比较。

(参见 A. Gunther 和 M. Hofler 的文章。mplus 和 stata-stata 中四元相关性的不同结果宣布修改程序。Int J Methods Psychiatr Res, 15(3):157-66, 2006。用四项相关讨论这个问题。)

使用 psych 包中的 polychoric 函数，如果我们不应用连续性校正，我们会找到与 polycor 中的 hetcor 函数相同的答案，但如果我们校正连续性，则会得到一些不同的值。我建议更正。

有关此问题的更长时间讨论，请参阅 polychoric in psych 的帮助功能。