Example of how to create custom contrasts to test hypotheses in lme4 models.

lme4_contrast_example.R

# Note: requires loading the "socsub" data frame (not a bundled R dataset)


# ------------------------------------------------------------------------------------
# pairwise comparisons including interactions


# use lm model to get design matrix
model1 <- lm(agro.rec.tot ~ sex*ageclass + loggrpmem, offset = logtimeage, data = socsub)

# get contrasts between sexes for each age class at median loggrpmem
cc <- contrast(model1,
    a = list(ageclass = levels(socsub$ageclass), sex = "F", loggrpmem = median(socsub$loggrpmem)),
    b = list(ageclass = levels(socsub$ageclass), sex = "M", loggrpmem = median(socsub$loggrpmem))
    )

# function to append new contrasts to existing contrast matrix
append.cc <- function(append, dn, cmat = cc$X) {
    # get attributes of cc matrix
    att <- attributes(cmat)
    # add new contrasts to cc matrix (strips attributes)
    cmat <- rbind(cmat, append)
    # change dimensions attributes of cc matrix
    att$dim[1] <- nrow(cmat)
    att$dimnames[[1]] <- dn
    # add the attributes back to the raw matrix
    attributes(cmat) <- att
    return(cmat)
}

# create matrix of new contrasts to append to contrast matrix
appMat <- matrix(ncol = 9, byrow = TRUE, data =
                                                 # females
                 c(0, 0, -1, 0, 0, 0, 0, 0, 0,   # Inf II - Inf I  == 0
                   0, 0, 0, -1, 0, 0, 0, 0, 0,   # Juv I  - Inf I  == 0
                   0, 0, 0, 0, -1, 0, 0, 0, 0,   # Juv II - Inf I  == 0
                   0, 0, -1, 1, 0, 0, 0, 0, 0,   # Juv I  - Inf II == 0
                   0, 0, -1, 0, 1, 0, 0, 0, 0,   # Juv II - Inf II == 0
                   0, 0, 0, -1, 1, 0, 0, 0, 0,   # Juv II - Juv I  == 0
                                                 # males
                   0, 0, -1, 0, 0, 0, -1, 0, 0,  # Inf II - Inf I  == 0
                   0, 0, 0, -1, 0, 0, 0, -1, 0,  # Juv I  - Inf I  == 0
                   0, 0, 0, 0, -1, 0, 0, 0, -1,  # Juv II - Inf I  == 0
                   0, 0, -1, 1, 0, 0, -1, 1, 0,  # Juv I  - Inf II == 0
                   0, 0, -1, 0, 1, 0, -1, 0, 1,  # Juv II - Inf II == 0
                   0, 0, 0, -1, 1, 0, 0, -1, 1)  # Juv II - Juv I  == 0
                 ) 

# create vector of dimnames that reflect contrasts
dnames <- c("Inf I: M - F",
            "Inf II: M - F",
            "Juv I: M - F",
            "Juv II: M - F",
            "F: Inf II - Inf I",
            "F: Juv I  - Inf I",
            "F: Juv II - Inf I",
            "F: Juv I  - Inf II",
            "F: Juv II - Inf II",
            "F: Juv II - Juv I",
            "M: Inf II - Inf I",
            "M: Juv I  - Inf I",
            "M: Juv II - Inf I",
            "M: Juv I  - Inf II",
            "M: Juv II - Inf II",
            "M: Juv II - Juv I")

# append the new contrasts to the matrix
cc2$X <- append.cc(append = appMat, dn = dnames)

# fit the same model in lme4
model2 <- glmer(agro.rec.tot ~ sex*ageclass + loggrpmem + (1|id),
                offset = logtimeage,
                family = poisson,
                data = socsub
                )

# plug in new appended contrast matrix 
summary(glht(model2, linfct = cc2$X), test = adjusted(type = "holm"))

Hi Steve

Where can I find the socsub dataset? I'd like to go through this example.

Cheers

Llew