在收到@PhilipLeifeld的建议后,我将根据我的进度重新撰写此帖子(请参阅下面的评论部分)。
我试图使用clmm
将texreg
输出放到乳胶中。由于该软件包在默认模式下不支持clmm
,因此我尝试使用extract
函数扩展该软件包(请参见Print "beautiful" tables for h2o models in R的答案部分)。同时,我发现发布在https://gist.github.com/kjgarza/340201f6564ca941fe25上的代码可以用作我的起点;我将代码称为下面的基准代码。以下模型(结果)几乎可以代表我的实际代码。
library(ordinal)
library(texreg)
d<-data.frame(wine)
result<-clmm(rating~ 1+temp+contact+(1+temp|judge), data=d)
我想在乳胶表中显示的是随机效果组件,基线代码中省略了这些组件。以下是摘要输出的一部分。
summary(result)
Random effects:
Groups Name Variance Std.Dev. Corr
judge (Intercept) 1.15608 1.0752
tempwarm 0.02801 0.1674 0.649
Number of groups: judge 9
具体来说,我想显示方差(和组数);我不需要相关部分。在处理基线代码时,我还了解到“ texreg”仅允许乳胶显示器的参数有限,并且“ include.variance”选项与我的目标相关。因此,我尝试将随机效果组件添加到“ gof”参数中,包括在基准代码中包括“ include.variance”选项。
这是我所做的。首先,我在定义extract.clmm函数的部分添加“ include.variance”。
extract.clmm <- function(model, include.thresholds = TRUE, include.aic = TRUE,
include.bic = TRUE, include.loglik = TRUE, include.variance = TRUE, oddsratios = TRUE, conf.level= 0.95, include.nobs = TRUE, ...) {
s <- summary(model, ...)
tab <- s$coefficients
thresh <- tab[rownames(tab) %in% names(s$alpha), ]
threshold.names <- rownames(thresh)
threshold.coef <- thresh[, 1]
threshold.se <- thresh[, 2]
threshold.pval <- thresh[, 4]
beta <- tab[rownames(tab) %in% names(s$beta), ]
beta.names <- rownames(beta)
beta.coef <- beta[, 1]
beta.se <- beta[, 2]
beta.pval <- beta[, 4]
然后,我添加了以下三行。
### for random effect components###
rand<-s$ST[[1]]
rand.names<-rownames(rand)
rand.var<-rand[,1]
以下部分是我另外包含在基准代码中的内容(“ include.variance”)。
if (include.variance == TRUE) {
gof.names <- c(gof.names, rand.names)
gof <- c(gof, rand)
gof.decimal <- c(gof.decimal, TRUE)
}
运行extract.clmm函数后,我运行了以下命令。
test<-extract.clmm(result, include.variance=TRUE, oddsratios=FALSE)
然后,我收到一条错误消息:validationMethod(object)中的错误:gof.names和gof必须具有相同的长度!虽然我发现在“结果”的情况下,“ rand”和“ rand.names”的长度分别为4和2,但我不知道该如何处理。任何意见将不胜感激。提前致谢。
最佳答案
首先,让我们重写测试用例,使其包含来自clmm
包的具有随机效果的模型(clm
)和不具有随机效果的模型(ordinal
)。这将使我们能够检查要编写的extract.clmm
函数是否产生与extract.clm
包中现有的texreg
函数兼容的格式格式的结果:
library("ordinal")
library("texreg")
d <- data.frame(wine)
result.clmm <- clmm(rating ~ 1 + temp + contact + (1 + temp|judge), data = d)
result.clm <- clm(rating ~ 1 + temp + contact, data = d)
clm
中通用的extract
函数的现有texreg
方法如下所示,并且由于这两种对象类型的构造方式相似,我们可以将其用作编写clmm
方法的模板:# extension for clm objects (ordinal package)
extract.clm <- function(model, include.thresholds = TRUE, include.aic = TRUE,
include.bic = TRUE, include.loglik = TRUE, include.nobs = TRUE, ...) {
s <- summary(model, ...)
tab <- s$coefficients
thresh <- tab[rownames(tab) %in% names(s$aliased$alpha), , drop = FALSE]
threshold.names <- rownames(thresh)
threshold.coef <- thresh[, 1]
threshold.se <- thresh[, 2]
threshold.pval <- thresh[, 4]
beta <- tab[rownames(tab) %in% names(s$aliased$beta), , drop = FALSE]
beta.names <- rownames(beta)
beta.coef <- beta[, 1]
beta.se <- beta[, 2]
beta.pval <- beta[, 4]
if (include.thresholds == TRUE) {
names <- c(beta.names, threshold.names)
coef <- c(beta.coef, threshold.coef)
se <- c(beta.se, threshold.se)
pval <- c(beta.pval, threshold.pval)
} else {
names <- beta.names
coef <- beta.coef
se <- beta.se
pval <- beta.pval
}
n <- nobs(model)
lik <- logLik(model)[1]
aic <- AIC(model)
bic <- BIC(model)
gof <- numeric()
gof.names <- character()
gof.decimal <- logical()
if (include.aic == TRUE) {
gof <- c(gof, aic)
gof.names <- c(gof.names, "AIC")
gof.decimal <- c(gof.decimal, TRUE)
}
if (include.bic == TRUE) {
gof <- c(gof, bic)
gof.names <- c(gof.names, "BIC")
gof.decimal <- c(gof.decimal, TRUE)
}
if (include.loglik == TRUE) {
gof <- c(gof, lik)
gof.names <- c(gof.names, "Log Likelihood")
gof.decimal <- c(gof.decimal, TRUE)
}
if (include.nobs == TRUE) {
gof <- c(gof, n)
gof.names <- c(gof.names, "Num.\ obs.")
gof.decimal <- c(gof.decimal, FALSE)
}
tr <- createTexreg(
coef.names = names,
coef = coef,
se = se,
pvalues = pval,
gof.names = gof.names,
gof = gof,
gof.decimal = gof.decimal
)
return(tr)
}
setMethod("extract", signature = className("clm", "ordinal"),
definition = extract.clm)
clmm
对象的第一个区别是系数等不存储在summary(model)$aliased$alpha
和summary(model)$aliased$beta
下,而是直接存储在summary(model)$alpha
和summary(model)$beta
下。我们需要做的第二件事是为组数和随机方差添加拟合优度元素。
组的数量显然存储在
summary(model)$dims$nlev.gf
下,其中有多个条目用于不同的条件变量。这很容易。随机方差未存储在任何地方,因此我们需要在source code of the
ordinal
package中进行查找。我们在那里可以看到print.summary.clmm
函数使用一个称为formatVC
的内部帮助函数来打印差异。此函数包含在同一R
脚本中,并且基本上只是进行格式化,并调用另一个称为varcov
的内部帮助器函数(也包含在同一文件中)以计算差异。该函数依次计算model$ST
的转置叉积以获得方差。我们可以直接在extract.clmm
函数的GOF块中直接执行相同的操作,例如,将diag(s$ST[[1]] %*% t(s$ST[[1]]))
用于第一个随机效果。我们只需要确保对所有随机效果都这样做,这意味着我们需要将其放入循环中,并用[[1]]
之类的迭代器替换[[i]]
。clmm
函数的最终extract
方法如下所示:# extension for clmm objects (ordinal package)
extract.clmm <- function(model, include.thresholds = TRUE,
include.loglik = TRUE, include.aic = TRUE, include.bic = TRUE,
include.nobs = TRUE, include.groups = TRUE, include.variance = TRUE, ...) {
s <- summary(model, ...)
tab <- s$coefficients
thresh <- tab[rownames(tab) %in% names(s$alpha), ]
threshold.names <- rownames(thresh)
threshold.coef <- thresh[, 1]
threshold.se <- thresh[, 2]
threshold.pval <- thresh[, 4]
beta <- tab[rownames(tab) %in% names(s$beta), ]
beta.names <- rownames(beta)
beta.coef <- beta[, 1]
beta.se <- beta[, 2]
beta.pval <- beta[, 4]
if (include.thresholds == TRUE) {
cfnames <- c(beta.names, threshold.names)
coef <- c(beta.coef, threshold.coef)
se <- c(beta.se, threshold.se)
pval <- c(beta.pval, threshold.pval)
} else {
cfnames <- beta.names
coef <- beta.coef
se <- beta.se
pval <- beta.pval
}
gof <- numeric()
gof.names <- character()
gof.decimal <- logical()
if (include.loglik == TRUE) {
lik <- logLik(model)[1]
gof <- c(gof, lik)
gof.names <- c(gof.names, "Log Likelihood")
gof.decimal <- c(gof.decimal, TRUE)
}
if (include.aic == TRUE) {
aic <- AIC(model)
gof <- c(gof, aic)
gof.names <- c(gof.names, "AIC")
gof.decimal <- c(gof.decimal, TRUE)
}
if (include.bic == TRUE) {
bic <- BIC(model)
gof <- c(gof, bic)
gof.names <- c(gof.names, "BIC")
gof.decimal <- c(gof.decimal, TRUE)
}
if (include.nobs == TRUE) {
n <- nobs(model)
gof <- c(gof, n)
gof.names <- c(gof.names, "Num.\ obs.")
gof.decimal <- c(gof.decimal, FALSE)
}
if (include.groups == TRUE) {
grp <- s$dims$nlev.gf
grp.names <- paste0("Groups (", names(grp), ")")
gof <- c(gof, grp)
gof.names <- c(gof.names, grp.names)
gof.decimal <- c(gof.decimal, rep(FALSE, length(grp)))
}
if (include.variance == TRUE) {
var.names <- character()
var.values <- numeric()
for (i in 1:length(s$ST)) {
variances <- diag(s$ST[[i]] %*% t(s$ST[[i]]))
var.names <- c(var.names, paste0("Variance: ", names(s$ST)[[i]], ": ",
names(variances)))
var.values <- c(var.values, variances)
}
gof <- c(gof, var.values)
gof.names <- c(gof.names, var.names)
gof.decimal <- c(gof.decimal, rep(TRUE, length(var.values)))
}
tr <- createTexreg(
coef.names = cfnames,
coef = coef,
se = se,
pvalues = pval,
gof.names = gof.names,
gof = gof,
gof.decimal = gof.decimal
)
return(tr)
}
setMethod("extract", signature = className("clmm", "ordinal"),
definition = extract.clmm)
您可以只在运行时执行代码,并且
texreg
应该能够从clmm
对象创建表,包括随机变量。我会将这段代码添加到下一个texreg
版本中。您可以将其应用于您的示例,如下所示:
screenreg(list(result.clmm, result.clm), single.row = TRUE)
结果在
clmm
和clm
对象之间兼容,如您在输出中所见:==================================================================
Model 1 Model 2
------------------------------------------------------------------
tempwarm 3.07 (0.61) *** 2.50 (0.53) ***
contactyes 1.83 (0.52) *** 1.53 (0.48) **
1|2 -1.60 (0.69) * -1.34 (0.52) **
2|3 1.50 (0.60) * 1.25 (0.44) **
3|4 4.22 (0.82) *** 3.47 (0.60) ***
4|5 6.11 (1.02) *** 5.01 (0.73) ***
------------------------------------------------------------------
Log Likelihood -81.55 -86.49
AIC 181.09 184.98
BIC 201.58 198.64
Num. obs. 72 72
Groups (judge) 9
Variance: judge: (Intercept) 1.16
Variance: judge: tempwarm 0.03
==================================================================
*** p < 0.001, ** p < 0.01, * p < 0.05
如果需要,可以使用参数
include.variances == FALSE
和include.groups == FALSE
关闭差异和组大小的报告。