Allow to pass contrasts for the fixed effects

[timflutre]

A minor, yet useful enhancement would be to allow the user to specify contrasts for the fixed effects.

By looking at the code of mmer, a contrasts argument could be used on line 261 baseX <- model.matrix(newfixed, mf) maybe? But I don't know how it would work with multivariate models...

[covaruber]

Could you elaborate a more complete example using another software like base lm or lmer to understand a bit more what do you mean? With some mock up data if possible.

Cheers,

[timflutre]

Sure, sorry, here it is:

library(lme4)
library(sommer)

## simulate some data with a LMM y = X beta + Z u + epsilon:
set.seed(12345)
p <- 3
q <- 20
n <- p * q
dat <- expand.grid(geno=paste0("geno", 1:q),
                   block=LETTERS[1:p],
                   KEEP.OUT.ATTRS=FALSE)
str(dat)
head(dat)

## listContr <- list("block"="contr.treatment")
listContr <- list("block"="contr.sum")
X <- model.matrix(~ block, dat, contrasts.arg=listContr)
beta <- rnorm(p)

Z <- model.matrix(~ 0 + geno, dat)
u <- rnorm(q)

y <- X %*% beta + Z %*% u + rnorm(n)
dat$yield <- y

## plot the data:
hist(dat$yield)
boxplot(yield ~ block, dat)

## fit with lmer
fit1 <- lmer(yield ~ block + (1|geno), dat, contrasts=listContr)
cbind("truth"=beta, "estim"=fixef(fit1))

## fit with mmer:
fit2 <- mmer(yield ~ block, ~ geno, data=dat)
cbind("truth"=beta, "estim"=fit2$Beta[,"Estimate"])

By default for unordered factor, R chooses the "treatment" contrasts. But I often uses the "sum" contrast instead for my simulations. It would be great if we could give listContr (see above) to mmer.

p.s.: see here for more details about coding and contrast matrices in R