I want to create a custom contrast function in emmeans which could remove a given list of levels from the input vector and apply the built-in contrast method ("trt.vs.ctrl") on the remaining levels. An example dataset is available here. I am using the following R code for computing ANOVA and post hoc comparisons:
options(contrasts=c("contr.sum", "contr.poly"))
my_lm <- lm(D1 ~ C*R, data=df)
Anova(my_lm, type = "III")
#show Interaction effects using emmeans
emmip(my_lm, C ~ R )
emm = emmeans(my_lm, ~ C * R)
emm
contrast(emmeans(my_lm, ~ C * R), "consec", by = "C")
#compare 1st with next 3 groups (how to remove other three levels?)
contrast(emmeans(my_lm, ~ C * R), "trt.vs.ctrl", by = "R")
The built-in contrast option ("trt.vs.ctrl") compares the first level with everything that follows it (there are 7 factor levels in C, and I want to remove last 3 of them and compute the contrasts for the remaining 4). An example is provided in the official documentation to write a custom contrast function.
skip_comp.emmc <- function(levels, skip = 1, reverse = FALSE) {
if((k <- length(levels)) < skip + 1)
stop("Need at least ", skip + 1, " levels")
coef <- data.frame()
coef <- as.data.frame(lapply(seq_len(k - skip - 1), function(i) {
sgn <- ifelse(reverse, -1, 1)
sgn * c(rep(0, i - 1), 1, rep(0, skip), -1, rep(0, k - i - skip - 1))
}))
names(coef) <- sapply(coef, function(x)
paste(which(x == 1), "-", which(x == -1)))
attr(coef, "adjust") = "fdr" # default adjustment method
coef
}
However due to my limited understanding I am not very sure where to apply the modifications that I need to to customise the example. Any ideas?
Is this something you are going to want to do lots of times in the future? My guess is not, that you only want to do this once, or a few times at most; in which case it is way too much trouble to write a custom contrast function. Just get the contrast coefficients you need, and use that as the second argument in contrast.
Now, consider these results:
> con <- emmeans:::trt.vs.ctrl.emmc(1:7)
> con
2 - 1 3 - 1 4 - 1 5 - 1 6 - 1 7 - 1
1 -1 -1 -1 -1 -1 -1
2 1 0 0 0 0 0
3 0 1 0 0 0 0
4 0 0 1 0 0 0
5 0 0 0 1 0 0
6 0 0 0 0 1 0
7 0 0 0 0 0 1
From the description, I think you just want the first 3 sets of contrast coefficients. So use those columns:
contrast(emm, con[, 1:3], by = "R")
StackOverflow can occasionally inspire developers to add software features. In this case, I decided it could be useful to add an exclude argument to most built-in .emmc functions in emmeans (all except poly.emmc()). This was fairly straightforward to do, and those features are now incorporated in the latest push to github -- https://github.com/rvlenth/emmeans. These features will be included in the next CRAN update as well.