stroup.nin: Nebraska Intrastate Nursery field experiment

Description

The yield data from an advanced Nebraska Intrastate Nursery (NIN) breeding trial conducted at Alliance, Nebraska, in 1988/89.

Arguments

Format

gen: Genotype factor, 56 levels
rep: Replicate factor, 4 levels
yield: The yield in bu/ac
x: Column ordinate
y: Row ordinate

Details

Four replicates of 19 released cultivars, 35 experimental wheat lines and 2 additional triticale lines were laid out in a 22 row by 11 column rectangular array of plots. The varieties were allocated to the plots using a randomised complete block (RCB) design. The blocks are not rectangular but partially overlap columns.

All plots with missing data are coded as being gen = "Lancer". (For ASREML, missing plots need to be included for spatial analysis and the level of 'gen' needs to be one that is already in the data.)

These data were first analyzed by Stroup et al (1994) and subsequently by Littell et al (1996, page 321), Pinheiro and Bates (2000, page 260), and Butler et al (2004).

This version of the data expresses the yield in bushels per acre. The results published in Stroup et al (1994) are expressed in kg/ha. For wheat, 1 bu/ac = 67.25 kg/ha.

Some of the gen names are different in Stroup et al (1994). (Sometimes an experimental genotype is given a new name when it is released for commercial use.) At a minimum, the following differences in gen names should be noted:

stroup.nin	Stroup et al
NE83498	Rawhide

Some published versions of the data use long/lat instead of x/y. To obtain the correct value of 'long', multiply 'x' by 1.2. To obtain the correct value of 'lat', multiply 'y' by 4.3.

Relatively low yields were clustered in the northwest corner, which is explained by a low rise in this part of the field, causing increased exposure to winter kill from wind damage and thus depressed yield. The genotype 'Buckskin' is a known superior variety, but was disadvantaged by assignment to unfavorable locations within the blocks.

References

Littell, R.C. and Milliken, G.A. and Stroup, W.W. and Wolfinger, R.D. 1996. SAS system for mixed models, SAS Institute, Cary, NC.

Jose Pinheiro and Douglas Bates, 2000, Mixed Effects Models in S and S-Plus, Springer.

Butler, D., B R Cullis, A R Gilmour, B J Goegel. (2004) Spatial Analysis Mixed Models for S language environments

Examples

Run this code

# NOT RUN {
data(stroup.nin)
dat <- stroup.nin
dat <- transform(dat, xf=factor(x), yf=factor(y))

# Show experiment layout
# Note: all "Buckskin" plots are near left side
desplot(yield~x*y, dat, out1="rep", num=gen, cex=1, main="stroup.nin")

require(nlme)
# Random block model
lme1 <- lme(yield ~ 0 + gen, random=~1|rep, data=dat, na.action=na.omit)

# Linear (Manhattan distance) correlation model
lme2 = gls(yield ~ 0 + gen, correlation = corLin(form = ~ x+y,nugget=TRUE), data=dat,
  na.action=na.omit)

# Compare the estimates from the two methods
eff = data.frame(ranblock=fixef(lme1), spat = coef(lme2))
rownames(eff) <- gsub("gen", "", rownames(eff))
plot(eff$ranblock, eff$spat, xlim=c(13,37), ylim=c(13,37),
  main="stroup.nin", xlab="RCB (random block)", ylab="corLin",type='n')
text(eff$ranblock, eff$spat, rownames(eff), cex=0.5)
abline(0,1)

# }
# NOT RUN {
require(asreml)

# RCB analysis
dat.rcb <- asreml(yield ~ gen, random = ~ rep, data=dat,
            na.method.X="omit")
pred.rcb <- predict(dat.rcb,classify="gen")$predictions

# Two-dimensional AR1xAR1 spatial model
dat <- dat[order(dat$xf, dat$yf),]
dat.sp <- asreml(yield~gen, rcov=~ar1(xf):ar1(yf),data=dat)
pred.sp <- predict(dat.sp,classify="gen")$predictions

# require(lucid)
# vc(dat.sp)
##     effect component std.error z.ratio constr
## R!variance   48.7      7.155       6.8    pos
##   R!xf.cor    0.6555   0.05638    12      unc
##   R!yf.cor    0.4375   0.0806      5.4    unc

# Compare the estimates from the two methods
plot(pred.rcb$pvals[,2],pred.sp$pvals[,2], xlim=c(16,37), ylim=c(16,37),
  xlab="RCB",ylab="AR1xAR1",type='n')
title("stroup.nin: Comparison of predicted values")
text(pred.rcb$pvals[,2],pred.sp$pvals[,2],
     as.character(pred.rcb$pvals[,1]),cex=0.5)
abline(0,1)

# }
# NOT RUN {
# }