问题
I would like to plot a model with ggplot2. I have estimated a robust variance-covariance matrix which I would like to use when estimating the confidence interval.
Can I tell ggplot2 to use my VCOV, or, alternatively, can I somehow force predict.lm to use my VCOV matrix? A dummy example:
source("http://people.su.se/~ma/clmclx.R")
df <- data.frame(x1 = rnorm(100), x2 = rnorm(100), y = rnorm(100), group = as.factor(sample(1:10, 100, replace=T)))
lm1 <- lm(y ~ x1 + x2, data = df)
coeftest(lm1)
## outputs coef.test, but can be modified to output VCOV
clx(lm1, 1, df$group)
It would be relatively easy to add to a ggplot, if I could get 'correct' predictions given my augmented VCOV-matrix.
回答1:
Only the standard errors, not the predictions, should change -- right?
getvcov <- function(fm,dfcw,cluster) {
library(sandwich);library(lmtest)
M <- length(unique(cluster))
N <- length(cluster)
K <- fm$rank
dfc <- (M/(M-1))*((N-1)/(N-K))
uj <- apply(estfun(fm),2, function(x) tapply(x, cluster, sum));
dfc*sandwich(fm, meat=crossprod(uj)/N)*dfcw
}
V <- getvcov(lm1,1,df$group)
X <- as.matrix(model.frame(lm1))
se <- predict(lm1,se=TRUE)$se.fit
se_robust <- sqrt(diag(X %*% V %*% t(X)))
来源:https://stackoverflow.com/questions/9260766/robust-standard-errors-in-ggplot2