Encounted a problem that the treatment is confounded with the blocking structure. In this situation, it is not possible to assess the interaction of the treatment and genotypes.
Here is an example of the trial:
set.seed(1234)
Rep <- rep(as.vector(replicate(2,sample(c("Rep1","Rep2","Rep3"),3))),each=5)
Gens <- paste0("Gen",1:5)
Geno <- as.vector(replicate(6, sample(Gens,5)))
TOS <- rep(c("TOS1","TOS2"),each=15)
Rows <- rep(1:5,6)
Cols <- rep(1:6,each=5)
dat <- data.frame(Rows,Cols,Rep,Geno,TOS)
ggplot(dat) + geom_tile(aes(Cols,Rows,fill=Rep))
ggplot(dat) + geom_tile(aes(Cols,Rows,fill=TOS))
ggplot(dat) + geom_tile(aes(Cols,Rows,fill=Geno))
dat$y <- 3+rnorm(length(Rep))
names(dat)
## [1] "Rows" "Cols" "Rep" "Geno" "TOS" "y"
nms <- c("Rows","Cols","Rep","Geno","TOS")
dat[nms] <- lapply(dat[nms], factor)
The researchers would like to assess the effects of different Time of Sowing, Genotypes and their interactions on a set of traits.
With the trial, the interaction is not possibly being assessed because the treatment is confounded with the blocking structure.
mod1 <- asreml(y ~ TOS*Geno,
random = ~ TOS:Rep,
data = dat)
## Online License checked out Thu Mar 4 16:57:37 2021
## Model fitted using the gamma parameterization.
## ASReml 4.1.0 Thu Mar 4 16:57:37 2021
## LogLik Sigma2 DF wall cpu
## 1 -12.0709 0.654877 20 16:57:37 0.0
## 2 -12.0484 0.644739 20 16:57:37 0.0
## 3 -12.0259 0.624656 20 16:57:37 0.0
## 4 -12.0256 0.622321 20 16:57:37 0.0
plot(mod1)
wald(mod1,denDF = "default",ssType = "conditional")
## Model fitted using the gamma parameterization.
## ASReml 4.1.0 Thu Mar 4 16:57:37 2021
## LogLik Sigma2 DF wall cpu
## 1 -12.0256 0.622286 20 16:57:37 0.0
## 2 -12.0256 0.622286 20 16:57:37 0.0
## 3 -12.0256 0.622286 20 16:57:37 0.0
## $Wald
## [0;34m
## Wald tests for fixed effects.[0m
## [0;34mResponse: y[0m
##
## Df denDF F.inc F.con Margin Pr
## (Intercept) 1 4 131.600 131.600 0.00033
## TOS 1 4 0.003 0.003 A 0.95740
## Geno 4 16 0.394 0.394 A 0.80986
## TOS:Geno 4 16 0.619 0.619 B 0.65513
##
## $stratumVariances
## df Variance TOS:Rep units!R
## TOS:Rep 4 1.1778601 5 1
## units!R 16 0.6222861 0 1
The two TOS blocks were treated as two different environments. As such, a MET analysis techniques were used assuming heterogeneous variance for both TOS.
mod2 <- asreml(y ~ TOS + Geno,
random = ~ at(TOS):Rep,
data = dat)
## Model fitted using the gamma parameterization.
## ASReml 4.1.0 Thu Mar 4 16:57:37 2021
## LogLik Sigma2 DF wall cpu
## 1 -12.3657 0.609972 24 16:57:37 0.0
## 2 -12.3039 0.599510 24 16:57:37 0.0
## 3 -12.2332 0.579837 24 16:57:37 0.0
## 4 -12.2288 0.575302 24 16:57:37 0.0
## Warning in asreml(y ~ TOS + Geno, random = ~at(TOS):Rep, data = dat): Some
## components changed by more than 1% on the last iteration.
plot(mod2)
wald(mod2,denDF = "default",ssType = "conditional")
## Model fitted using the gamma parameterization.
## ASReml 4.1.0 Thu Mar 4 16:57:38 2021
## LogLik Sigma2 DF wall cpu
## 1 -12.2287 0.574921 24 16:57:38 0.0
## 2 -12.2287 0.574920 24 16:57:38 0.0
## 3 -12.2287 0.574918 24 16:57:38 0.0
## $Wald
## [0;34m
## Wald tests for fixed effects.[0m
## [0;34mResponse: y[0m
##
## Df denDF F.inc F.con Margin Pr
## (Intercept) 1 3.8 138.100 138.100 0.00039
## TOS 1 3.8 0.003 0.003 A 0.95753
## Geno 4 20.0 0.427 0.427 A 0.78766
##
## $stratumVariances
## df Variance at(TOS, TOS1):Rep at(TOS, TOS2):Rep units!R
## at(TOS, TOS1):Rep 2 1.4385847 5 -2.297635e-17 1
## at(TOS, TOS2):Rep 2 0.9171355 0 5.000000e+00 1
## units!R 20 0.5749179 0 0.000000e+00 1
Hans-Peter has a paper discussing this issue and an example shows that there is complete confounding of area and treatment effects. So the model 1
in our case is not valid.