agridat (version 1.23)

digby.jointregression: Multi-environment trial of wheat

Description

Yield of 10 spring wheat varieties for 17 locations in 1976.

Format

A data frame with 134 observations on the following 3 variables.

gen

genotype, 10 levels

env

environment, 17 levels

yield

yield (t/ha)

Details

Yield of 10 spring wheat varieties for 17 locations in 1976.

Used to illustrate modified joint regression.

References

Hans-Pieter Piepho, 1997. Analyzing Genotype-Environment Data by Mixed-Models with Multiplicative Terms. Biometrics, 53, 761-766. https://doi.org/10.2307/2533976

RJOINT procedure in GenStat. https://www.vsni.co.uk/software/genstat/htmlhelp/server/RJOINT.htm

Examples

if (FALSE) {
  
  library(agridat)
  data(digby.jointregression)
  dat <- digby.jointregression
  
  # Simple gen means, ignoring unbalanced data.
  # Matches Digby table 2, Unadjusted Mean
  round(tapply(dat$yield, dat$gen, mean),3)
  
  # Two-way model. Matches Digby table 2, Fitting Constants
  m00 <- lm(yield ~ 0 + gen + env, dat)
  round(coef(m00)[1:10]-2.756078+3.272,3) # Adjust intercept
  # genG01 genG02 genG03 genG04 genG05 genG06 genG07 genG08 genG09 genG10 
  #  3.272  3.268  4.051  3.724  3.641  3.195  3.232  3.268  3.749  3.179 
  
  n.gen <- nlevels(dat$gen)
  n.env <- nlevels(dat$env)
  
  # Estimate theta (env eff)
  m0 <- lm(yield ~ -1 + env + gen, dat)
  thetas <- coef(m0)[1:n.env]
  thetas <- thetas-mean(thetas) # center env effects
  # Add env effects to the data
  dat$theta <- thetas[match(paste("env",dat$env,sep=""), names(thetas))]
  
  # Initialize beta (gen slopes) at 1
  betas <- rep(1, n.gen)
  
  done <- FALSE
  while(!done){
    
    betas0 <- betas
    
    # M1: Fix thetas (env effects), estimate beta (gen slope)
    m1 <- lm(yield ~ -1 + gen + gen:theta, data=dat)
    betas <- coef(m1)[-c(1:n.gen)]
    dat$beta <- betas[match(paste("gen",dat$gen,":theta",sep=""), names(betas))]
    # print(betas)

    # M2: Fix betas (gen slopes), estimate theta (env slope)
    m2 <- lm(yield ~ env:beta + gen -1, data=dat)
    thetas <- coef(m2)[-c(1:n.gen)]
    thetas[is.na(thetas)] <- 0  # Change last coefficient from NA to 0
    dat$theta <- thetas[match(paste("env",dat$env,":beta",sep=""), names(thetas))]
    # print(thetas)

    # Check convergence
    chg <- sum(((betas-betas0)/betas0)^2)
    cat("Relative change in betas",chg,"\n")
    if(chg < .0001) done <- TRUE
    
  }

  libs(lattice)
  xyplot(yield ~ theta|gen, data=dat, xlab="theta (environment effect)",
         main="digby.jointregression - stability plot")

  # Dibgy Table 2, modified joint regression
  
  # Genotype sensitivities (slopes)
  round(betas,3) # Match Digby table 2, Modified joint regression sensitivity
  # genG01 genG02 genG03 genG04 genG05 genG06 genG07 genG08 genG09 genG10
  #  0.953  0.739  1.082  1.024  1.142  0.877 1.089  0.914  1.196  0.947

  # Env effects. Match Digby table 3, Modified joint reg
  round(thetas,3)+1.164-.515 # Adjust intercept to match
  # envE01 envE02 envE03 envE04 envE05 envE06 envE07 envE08 envE09 envE10
  # -0.515 -0.578 -0.990 -1.186  1.811  1.696 -1.096  0.046  0.057  0.825
  # envE11 envE12 envE13 envE14 envE15 envE16 envE17
  # -0.576  1.568 -0.779 -0.692  0.836 -1.080  0.649

  # Using 'gnm' gives similar results.
  # libs(gnm)
  # m3 <- gnm(yield ~ gen + Mult(gen,env), data=dat) # slopes negated
  # round(coef(m3)[11:20],3)

  # Using 'mumm' gives similar results, though gen is random and the
  # coeffecients are shrunk toward 0 a bit.
  if(require("mumm", quietly=TRUE)) {
    libs(mumm)
    m1 <- mumm(yield ~ -1 + env + mp(gen, env), dat)
    round(1 + ranef(m1)$`mp gen:env`,2)
  }
  
}