R中如何在GLM，具有交互作用的分类数据之后更改事后对比中的因素顺序

Question

I have a multiway contingency table with 2 predictor variables: tmt (2 levels) and year (4 levels) and a response variable, number (out of n plots) where a species is present (succ).

The data (testsumm):

  year tmt  n succ
1 2012   1 72   27
2 2012   0 68    6
3 2013   1 71   37
4 2013   0 71    8
5 2014   1 72   13
6 2014   0 75    9
7 2015   1 64   20
8 2015   0 67   16

After creating a vector of successes and failures

resp<-cbind(testsumm$succ, testsumm$n-testsumm$succ)

I analyzed the data (testsumm) in R using glm as follows:

model<-glm(resp~year*tmt, family=binomial,data=testsumm)

The result told me that there is an interaction between year and tmt. Now I am trying to use posthoc tests (with R's multcomp package) to determine whether the 2 levels of tmt differ significantly WITHIN EACH YEAR. Other questions on this site directed me to https://cran.r-project.org/web/packages/multcomp/vignettes/multcomp-examples.pdf These instructions worked perfectly for comparing years WITHIN TREATMENT LEVELS.

temp <- expand.grid(year = unique(testsumm$year),tmt = unique(testsumm$tmt))
X1 <- model.matrix(~ tmt * year, data = temp)
glht(model, linfct = X1)
Tukey <- contrMat(table(testsumm$year), "Tukey")
K1 <- cbind(Tukey, matrix(0, nrow = nrow(Tukey), ncol = ncol(Tukey)))
rownames(K1) <- paste(levels(testsumm$tmt)[1], rownames(K1), sep = ":")
K2 <- cbind(matrix(0, nrow = nrow(Tukey), ncol = ncol(Tukey)), Tukey)
rownames(K2) <- paste(levels(testsumm$tmt)[2], rownames(K2), sep = ":")
K <- rbind(K1, K2)
colnames(K) <- c(colnames(Tukey), colnames(Tukey))
summary(glht(modintILAQ, linfct = K %*% X1))

So since I wanted to compare treatment levels within years, I tried exchanging the positions of the 2 variables in the code:

model2 <- glm(resp ~ tmt * year, family=binomial,data = testsumm)
summary(model2)
temp2 <- expand.grid(tmt = unique(testsumm$tmt),year =unique(testsumm$year))
X12 <- model.matrix(~ tmt * year, data = temp2)
glht(model2, linfct = X12)
Tukey <- contrMat(table(testsumm$tmt), "Tukey")
    K1 <- cbind(Tukey, matrix(0, nrow = nrow(Tukey), ncol = ncol(Tukey)))
    rownames(K1) <- paste(levels(testsumm$year)[1], rownames(K1), sep = ":")
K2 <- cbind(matrix(0, nrow = nrow(Tukey), ncol = ncol(Tukey)), Tukey)
rownames(K2) <- paste(levels(testsumm$year)[2], rownames(K2), sep = ":")
K <- rbind(K1, K2)
colnames(K) <- c(colnames(Tukey), colnames(Tukey))
summary(glht(model2, linfct = K %*% X1))

but I got this error message

    Error in K %*% X1 : non-conformable arguments

It's pretty clear that these 2 matrices are not the same shape, so they can't be multiplied, but I can't figure out what they should look like instead. Can anyone help me to create contrasts between the 2 treatment levels within each year, instead of between years within each treatment level?

Answer 1

multcomp does not make it very easy when there is more than one factor. However, the lsmeans package provides an alternative way to specify what you need.

library(lsmeans)
glht(model2, lsm(~ tmt | year))

Certainly saves a lot of typing!