Tree: Tree Weed Suppression over Time

In byandell/pda: Practial Data Analysis for Designed Experiments

Description

A student of plant pathology studied the ‘weed suppressiveness’ of tree soils. David Maxwell examined the sensitivity of ‘Rutgers’ variety tomato, known to be sensitive to biological factors, to several different soils. Three of the soils (vermix, promix and grnhous) are in a certain sense controls, the first two being artificial and the third coming from a controlled environment. A fourth soil contained only grass. The other four are of particular interest, as the scientist wants to know whether soil from tree nm-6 is better (lower score) than those of the other three trees. The score was the number of germinating tomato seeds out of ten planted. In addition, the mean height of seedlings was measured at different times.

For each type of soil, there were six soil samples (reps(soil)). These soil samples were inoculated with pathogen. Then these soil samples were split (acc), with one half autoclaved (sterilized) and the other half not. Thus there were 12 half-samples (acc*reps(soil)). Each of these 12 half-samples was placed in storage and scored on six different dates. In addition, the mean height of seedlings was measured at three of these times. The primary focus is on the half-samples that were not autoclaved.

Usage

data(TreeAve)
data(Tree)

Format

TreeAve data frame with 48 observations on 4 variables.

[,1]	soil	factor	soil type
[,2]	reps	factor	replicate number
[,3]	score	numeric	germinating tomatoes of 10
[,4]	height	numeric	mean plant height

Tree data frame with 96 observations on 12 variables.

[,1]	soil	factor	soil type
[,2]	reps	factor	replicate number
[,3]	acc	factor	autoclaved or not
[,4]	score1	numeric	germination at time 1
[,5]	score2	numeric	germination at time 2
[,6]	score3	numeric	germination at time 3
[,4]	ht1	numeric	mean plant height at time 3
[,6]	score4	numeric	germination at time 4
[,4]	ht2	numeric	mean plant height at time 4
[,6]	score5	numeric	germination at time 5
[,4]	ht3	numeric	mean plant height at time 5
[,6]	score6	numeric	germination at time 6

Source

David Maxwell, Plant Pathology Department, UW-Madison

References

Maxwell DL and Stanosz G (1996) 'Assay to determine if weed suppressiveness is an inherent property of soil in which NM-6 is growing', Department of Plant Pathology, University of Wisconsin-Madison.

Examples

# Analysis Averaged over Time
data( TreeAve )
# soil is automatically a factor() since it has characters
TreeAve.aov <- aov( score ~ soil, TreeAve)
summary( TreeAve.aov )
#plot( TreeAve.aov )
boxplot( split( TreeAve$score, TreeAve$soil))

# Analysis over Time
data( Tree )
Tree <- Tree[ Tree$acc == "notauto", ]
Tree$acc <- NULL
Tree$type <- factor( !is.na( match( Tree$soil,
   c("dtac2","nc11004","ne-332","nm-6") )))
Tree$reps <- factor( Tree$reps )

Trees <- Tree[ rep( seq( nrow( Tree ) ), 6 ), c("soil","reps","type") ]
Trees$date <- ordered( rep( 1:6, rep( nrow( Tree ), 6 ) ) )
Trees$score <- unlist( Tree[ , paste( "score", 1:6, sep="" ) ] )
row.names(Trees) <- interaction( Trees[, c("soil","reps","date") ] )

Trees.soil <- aov(score ~ soil * date + Error(reps/soil), Trees )
# combined (sort of)
Trees.all <- aov( score ~ soil * date, Trees )

# Figure I:25.1 Tree scores over time by soil type


Tree.lsm <- lsmean( Trees.all, factors = c("soil","date") )
Tree.lsm$type <- factor( !is.na( match( Tree.lsm$soil,
   c("dtac2","nc11004","ne-332","nm-6") ) ) )

Tree.lsm <- split( Tree.lsm, Tree.lsm$type )
# following values are SE for means taken for SAS output
Tree.lsm[[1]]$se <- rep( .84086871, nrow( Tree.lsm[[1]] ) )
Tree.lsm[[2]]$se <- rep( .63433993, nrow( Tree.lsm[[2]] ) )

Tree.data <- split( Trees, Trees$type )
Tree.xlab <- c("(b) control","(a) Tree")
Tree.ylab <- c("","score means")
Tree.ylim = range( pretty( c( Tree.lsm[[1]]$pred, Tree.lsm[[2]]$pred )))

for( i in 2:1 ) { # "FALSE" comes before "TRUE"
   int.plot( Trees.all, Tree.data[[i]], c("date","soil"),
      Tree.lsm[[i]], bar.plot = "none", ylim = Tree.ylim,
      panelf = function(...) {
        se.bar( 4.5, 2, Tree.lsm[[i]]$se[1] )
        Tree.tmp <- Tree.lsm[[i]][Tree.lsm[[i]]$date==6,]
        panel.text( rep( 4.5, nrow( Tree.tmp ) ),
          0.25 + Tree.tmp$pred,
          as.character( Tree.tmp$soil ), adj = 0 )
      },
      xlab = paste( Tree.xlab[i], "soils by date" ),
      ylab = Tree.ylab[i],
      more = (i == 2), split = c(3-i,1,2,1) )
}
rm( Tree.data, Tree.xlab, Tree.ylab, Tree.lsm )

# Figure I:25.2 Tree whole plot experimental units

Tree.xlab <- c("(b) control","(a) Tree")
Tree.ylab <- c("","soil sample means")

library(lme4)
Trees.lme <- lmer(score ~ soil * date + (1|reps:soil), data = Trees )
VarCorr( Trees.lme )
anova( Trees.lme )

## Not run: 
# projections (*** CAUTION: SLOW ***)
# This works for now in Splus but not in R!
Trees.type <- aov(score ~ type/soil * date +
   Error(reps %in% soil %in% type), Trees, proj = TRUE )
Trees.proj <- proj ( Trees.type )
# whole plot
Trees.wp <- nested( Trees.proj, Trees, c("type","soil","reps"), "score",
   c(1,2), Trees.ref )
Trees.wpaov <- aov( score ~ soil, Trees.wp )
Tree.grmean <- lsmean( Trees.wpaov, fac = NULL )$pred
Tree.lsm <- lsmean( Trees.wpaov, factors = "soil" )
Tree.lsm$type <- factor( !is.na( match( Tree.lsm$soil,
   c("dtac2","nc11004","ne-332","nm-6") ) ) )
Tree.lsm <- split( Tree.lsm, Tree.lsm$type )
Tree.data <- split( Trees.wp, Trees.wp$type )
for ( i in 2:1 ) { # "FALSE" comes before "TRUE"
   Tree.lsm[[i]]$rank <- rank( -Tree.lsm[[i]]$pred )
   # have to be careful about levels of soil in type subsets
   Tree.data[[i]]$soil <- factor( Tree.data[[i]]$soil )
   attach( Tree.data[[i]] )
   names( Tree.lsm[[i]]$rank ) <- unique( soil )
   plot( jitter( Tree.lsm[[i]]$rank[soil] ), score, xaxt = "n",
      ylim = range( Trees.wp$score ),
      xlab = paste( Tree.xlab[i], "soil sample means" ),
      ylab = Tree.ylab[i] )
   detach()
   axis( 1, Tree.lsm[[i]]$rank, row.names( Tree.lsm[[i]] ) )
   abline( h = Tree.grmean, lty = 3 )
}
axis( 1, Tree.lsm[[1]]["grnhous","rank"], "grnhous" )
rm( Tree.data, Tree.xlab, Tree.ylab, Tree.grmean, Tree.lsm )

# Figure I:25.3 Tree repeated measures within soil sample

Trees.ref <- data.frame( Reference = c( Trees.proj[[1]] ))
for (i in names( Trees.proj )[-1])
   Trees.ref[[i]] <- apply( Trees.proj[[i]], 1, sum )

# split plot
Trees.sp <- nested( Trees.proj, Trees, c("type","soil","reps","date"),
   "score", c(1,3), Trees.ref )

Trees.spaov <- aov( score ~ reps*soil + date + date:soil, Trees.sp )

# split plot with time trend removed
Trees.sp$res <- resid( aov( score ~ reps*soil + date, Trees.sp ) ) +
   lsmean( Trees.spaov, fac=NULL )$pred

Trees.spres <- aov( res ~ reps*soil + date + date:soil, Trees.sp )

Tree.lsm <- lsmean( Trees.spaov, factors = c("soil","date") )
Tree.lsm$type <- factor( !is.na( match( Tree.lsm$soil,
   c("dtac2","nc11004","ne-332","nm-6") ) ) )

Tree.lsm <- split( Tree.lsm, Tree.lsm$type )
# following values are SE for contrasts taken from SAS output
Tree.lsm[[1]]$se <- rep( .42023472 / sqrt( 2 ), nrow( Tree.lsm[[1]] ) )
Tree.lsm[[2]]$se <- rep( .35818796 / sqrt( 2 ), nrow( Tree.lsm[[2]] ) )

Tree.data <- split( Trees.sp, Trees.sp$type )
Tree.xlab <- c("(b) control","(a) Tree")
Tree.ylab <- c("","pure split plot score")
for ( i in 2:1 ) { # "FALSE" comes before "TRUE"
   lsd.plot( Trees.spaov, Tree.data[[i]], c("date","soil"),
      Tree.lsm[[i]],
      xpos = 6, xlim = c(1,7.7), ylim = range( Tree.lsm$pred ),
      xaxt = "n", xlab = paste( Tree.xlab[i], "soils by date" ),
      ylab = Tree.ylab[i] )
   axis(1,1:6)
   Tree.tmp <- Tree.lsm[[i]][Tree.lsm[[i]]$date==6,]
   text( rep( 6.2, nrow( Tree.tmp ) ),
      uncollide( Tree.tmp$pred, c(.7,1.6)[i] ),
      as.character( Tree.tmp$soil ), adj = 0 )
}
rm( Tree.data, Tree.xlab, Tree.ylab, Tree.lsm, Tree.tmp )

# Figure I:25.4 Tree detrended repeated measures
Tree.lsm <- lsmean( Trees.spres, factors = c("soil","date") )
Tree.lsm$type <- factor( !is.na( match( Tree.lsm$soil,
   c("dtac2","nc11004","ne-332","nm-6") ) ) )

Tree.lsm <- split( Tree.lsm, Tree.lsm$type )
# following values are SE for contrasts take for SAS output
Tree.lsm[[1]]$se <- rep( .42023472 / sqrt( 2 ), nrow( Tree.lsm[[1]] ) )
Tree.lsm[[2]]$se <- rep( .35818796 / sqrt( 2 ), nrow( Tree.lsm[[2]] ) )

Tree.data <- split( Trees.sp, Trees.sp$type )
Tree.xlab <- c("(b) control","(a) Tree")
Tree.ylab <- c("","pure split plot score")
for ( i in 2:1 ) { # "FALSE" comes before "TRUE"
   lsd.plot( Trees.spres, Tree.data[[i]], c("date","soil"), Tree.lsm[[i]],
      xpos = 4.5, ypos = 4.25,
      xlim = c(1,7.7), ylim = range( Tree.lsm$pred ),
      xaxt = "n",
      xlab = paste( Tree.xlab[i], "soils by date" ),
      ylab = Tree.ylab[i] )
   axis(1,1:6)
   Tree.tmp <- Tree.lsm[[i]][ Tree.lsm[[i]]$date == 6, ]
   text( rep( 6.2, nrow( Tree.tmp ) ),
      uncollide( Tree.tmp$pred, c(.3,.6)[i] ),
      as.character( Tree.tmp$soil ), adj = 0 )
}
rm( Tree.data, Tree.xlab, Tree.ylab, Tree.lsm, Tree.tmp )

# Figure I:26.1 Tree partial correlations over time
Tree.tmpc <- length( unique( Trees.sp$date ))
Tree.tmpr <- nrow( Trees.sp ) / Tree.tmpc
Tree.tmp <- matrix( resid( Trees.spaov ), Tree.tmpr, Tree.tmpc )
Tree.tmpid <- c("C","T")[ codes( Trees.sp$type[1:Tree.tmpr] ) ]

splom( Tree.tmp, varnames = as.character( unique( Trees.sp$date )),
   groups = Tree.tmpid )
rm( Tree.tmp, Tree.tmpr, Tree.tmpc )

## End(Not run)

byandell/pda documentation built on May 13, 2019, 9:27 a.m.