Chapter 12 – Visualising many responses –
Exercise solutions and Code Boxes
David Warton
2022-06-28
Exercise 12.1: Flower size and shape
How do these three species of Iris differ in terms of flower size
and shape?
What sort of graph should Edgar produce to visualise how species
differ in flower size and shape?
He has more than two response variables so I guess as well as
plotting each response one-at-a-time he could try plotting them
simultaneously using ordination (factor analysis, for example). In
either plot he would want to label different species differently to see
variation across species.
Exercise 12.2: Bush regeneration and invertebrate counts
Is there evidence of a change in invertebrate communities due to
bush regeneration efforts?
What sort of graph should Anthony produce to visualise the
effects of bush regeneration on invertebrate communities?
Anthony has a bunch of response variables so I guess as well as
plotting each response one-at-a-time he could try plotting them
simultaneously using ordination. He has counts so will need something
that works for non-Gaussian responses like generalised latent variable
models. In either plot he would want to label different species
differently to see variation between revegetated and control plots.
Code Box 12.1: Plotting the bush regeneration data of Exercise 12.1
using mvabund.
library(mvabund)
library(ecostats)
data(reveg)
reveg$abundMV=mvabund(reveg$abund) #to treat data as multivariate
plot(abundMV~treatment, data=reveg)
#> Overlapping points were shifted along the y-axis to make them visible.
#>
#> PIPING TO 2nd MVFACTOR
#> Only the variables Collembola, Acarina, Formicidae, Coleoptera, Diptera, Amphipoda, Isopoda, Larvae, Hemiptera, Soleolifera, Hymenoptera, Araneae were included in the plot
#> (the variables with highest total abundance).
Can you see any taxa that seem to be associated with bush
regeneration?
There seem to be less invertebrates in control plots for
Collembola, Acarina, Coloeptera,
Amphipoda and maybe a few Orders.
Code Box 12.2: A PCA of Edgar’s Iris data
data("iris")
pc = princomp(iris[,1:4],cor=TRUE)
pc
#> Call:
#> princomp(x = iris[, 1:4], cor = TRUE)
#>
#> Standard deviations:
#> Comp.1 Comp.2 Comp.3 Comp.4
#> 1.7083611 0.9560494 0.3830886 0.1439265
#>
#> 4 variables and 150 observations.
loadings(pc)
#>
#> Loadings:
#> Comp.1 Comp.2 Comp.3 Comp.4
#> Sepal.Length 0.521 0.377 0.720 0.261
#> Sepal.Width -0.269 0.923 -0.244 -0.124
#> Petal.Length 0.580 -0.142 -0.801
#> Petal.Width 0.565 -0.634 0.524
#>
#> Comp.1 Comp.2 Comp.3 Comp.4
#> SS loadings 1.00 1.00 1.00 1.00
#> Proportion Var 0.25 0.25 0.25 0.25
#> Cumulative Var 0.25 0.50 0.75 1.00
biplot( pc, xlabs=rep("\u00B0",dim(iris)[1]) )
Code Box 12.3: Factor analysis of the Iris data
library(psych)
fa_iris <- fa(iris[,1:4], nfactors=2, fm="ml", rotate="varimax")
loadings(fa_iris)
#>
#> Loadings:
#> ML1 ML2
#> Sepal.Length 0.997
#> Sepal.Width -0.115 -0.665
#> Petal.Length 0.871 0.486
#> Petal.Width 0.818 0.514
#>
#> ML1 ML2
#> SS loadings 2.436 0.942
#> Proportion Var 0.609 0.236
#> Cumulative Var 0.609 0.844
How do results compare to the principal components
analysis?
They look awfully similar. The second factor looks a little
different, and is flipped around the other way (so big vlues mean
narrow sepals), but it also has postive loadings for petal
variables. So this could be interpreted as a measure of how large petals
are relative to sepal width: big scores for large petals with narrow
sepals, low scores for small petals with wide sepals. Recall that
previously, the second PC was pretty much just a measure of how wide
sepals were, now it is relative to petal size.
Code Box 12.4: Assumption checking for a factor analysis of the Iris
data
par(mfrow=c(2,2),mar=c(3,3,2,1),mgp=c(1.75,0.75,0))
for(iVar in 1:4)
{
irisIvar = iris[,iVar]
plotenvelope(lm(irisIvar~fa_iris$scores), which=1, col=iris$Species, main=print(names(iris)[iVar]), n.sim=99)
}
#> [1] "Sepal.Length"
#> Error in matrix(NA, nPred, n.sim): object 'n.sim' not found
plot(iris[,1:4],col=iris$Species)
(Note that plotenvelope was run with just
59 iterations, to speed up computation time.)
Exercise 12.4: A factor analysis for Anthony’s data(?)
Load Anthony’s revegetation data (stored as reveg in
the ecostats package) and do a factor analysis (with two
factors).
data(reveg)
library(psych)
fa_reveg <- try(fa(reveg$abund, nfactors=2, fm="ml", rotate="varimax"))
#> Warning in cor.smooth(R): Matrix was not positive definite, smoothing was done
#> In smc, smcs < 0 were set to .0
#> Warning in cor.smooth(R): Matrix was not positive definite, smoothing was done
#> In smc, smcs < 0 were set to .0
#> Warning in log(e): NaNs produced
#> Error in optim(start, FAfn, FAgr, method = "L-BFGS-B", lower = 0.005, :
#> L-BFGS-B needs finite values of 'fn'
You might not be able to get a solution when using maximum
likelihood estimation (fm=“ml”), in which case, try fitting using
without specifying the fm argument (which tries to minimise
residuals).
fa_reveg <- fa(reveg$abund, nfactors=2)
#> Warning in cor.smooth(R): Matrix was not positive definite, smoothing was done
#> In smc, smcs < 0 were set to .0
#> Warning in cor.smooth(R): Matrix was not positive definite, smoothing was done
#> In smc, smcs < 0 were set to .0
#> Warning in cor.smooth(R): Matrix was not positive definite, smoothing was done
#> In smc, smcs < 0 were set to .0
#> Warning in cor.smooth(r): Matrix was not positive definite, smoothing was done
#> Warning in fa.stats(r = r, f = f, phi = phi, n.obs = n.obs, np.obs = np.obs, : The estimated weights
#> for the factor scores are probably incorrect. Try a different factor score estimation method.
#> In factor.scores, the correlation matrix is singular, an approximation is used
#> Warning in cor.smooth(r): Matrix was not positive definite, smoothing was done
This returned some concerning warnings but did fit the model :)
Check some assumptions, by fitting a linear model to some of the
response variables, as a function of factor scores.
par(mfrow=c(3,3),mar=c(3,3,2,1),mgp=c(1.75,0.75,0))
for(iVar in 1:9)
{
y=reveg$abund[,iVar]
plotenvelope(lm(y~fa_reveg$scores), which=1,main=names(reveg$abund)[iVar], n.sim=99)
}
#> Error in matrix(NA, nPred, n.sim): object 'n.sim' not found
for(iVar in 1:9)
{
y=reveg$abund[,iVar]
plotenvelope(lm(y~fa_reveg$scores), which=2,main=names(reveg$abund)[iVar], n.sim=99)
}
Can you see any issues with factor analysis assumptions?
With only ten observations it is always hard to see issues, and
residual plots have huge error bars on them! But we have points outside
their normal quantile simulation envelopes in several of these plots,
and most have a suggestion of right-skew (the occasional large value).
Note also that fitted values go below zero for many species which is
most concerning considering we are modelling counts!
Code Box 12.5: Choosing the number of factors for the iris data
plot(fa_iris$values,type="l")
nFactors=3 # to compare models with up to 3 factors
BICs = rep(NA,nFactors) # define the vector that BIC values go in
names(BICs) = 1:nFactors # name its values according to #factors
for(iFactors in 1:nFactors) {
fa_iris <- fa(iris[,1:4], nfactors=iFactors, fm="ml", rotate="varimax")
BICs[iFactors] = fa_iris$objective - log(fa_iris$nh) * fa_iris$dof
}
BICs
#> 1 2 3
#> -9.436629 5.171006 15.031906
How many factors are supported by the data?
One, this has the smallest BIC.
Code Box 12.6: A generalised latent variable model for Anthony’s
revegetation data
data(reveg)
library(gllvm)
#>
#> Attaching package: 'gllvm'
#> The following objects are masked from 'package:ecostats':
#>
#> coefplot, nobs, predict, simulate, vcov
#> The following objects are masked from 'package:VGAM':
#>
#> AICc, nobs, predict, vcov
#> The following objects are masked from 'package:stats4':
#>
#> nobs, vcov
#> The following object is masked from 'package:vegan':
#>
#> ordiplot
#> The following object is masked from 'package:mvabund':
#>
#> coefplot
#> The following objects are masked from 'package:stats':
#>
#> nobs, predict, simulate, vcov
reveg_LVM = gllvm(reveg$abund, num.lv=2, family="negative.binomial", trace=TRUE, jitter.var=0.2)
logLik(reveg_LVM)
#> 'log Lik.' -679.9619 (df=95)
Repeating this several times usually returns an answer of
-689.3, so we can be confident this is (close to) the
maximum likelihood solution. To get a biplot of this solution:
ordiplot(reveg_LVM, col=as.numeric(reveg$treatment), biplot=TRUE,
ind.spp=12)
par(mfrow=c(1,3),mar=c(3,3,1,1),mgp=c(1.75,0.75,0))
plot(reveg_LVM,which=c(1,2,5))
Exercise 12.5: Checking analysis decisions for Anthony’s
revegetation data
In Code Box 12.6, a negative binomial model was fitted, using two
latent variables. Are two latent variables needed, or should we use
more, or less? Fit a few models varying the number of latent variables.
Which model fits the data best, according to BIC?
reveg_LVM1 = gllvm(reveg$abund, num.lv=1, family="negative.binomial", trace=TRUE, jitter.var=0.2)
reveg_LVM2 = gllvm(reveg$abund, num.lv=2, family="negative.binomial", trace=TRUE, jitter.var=0.2)
reveg_LVM3 = gllvm(reveg$abund, num.lv=3, family="negative.binomial", trace=TRUE, jitter.var=0.2)
reveg_LVM4 = gllvm(reveg$abund, num.lv=4, family="negative.binomial", trace=TRUE, jitter.var=0.2)
reveg_LVM5 = gllvm(reveg$abund, num.lv=5, family="negative.binomial", trace=TRUE, jitter.var=0.2)
BIC(reveg_LVM1,reveg_LVM2,reveg_LVM3,reveg_LVM4,reveg_LVM5)
#> df BIC
#> reveg_LVM1 72 1557.821
#> reveg_LVM2 95 1610.781
#> reveg_LVM3 117 1575.276
#> reveg_LVM4 138 1629.856
#> reveg_LVM5 158 1669.682
For me two latent variable models was the winner!
Fit a Poisson model to the data and check assumptions. Are there
any signs of overdispersion?
I’ll go with two latent variable models, on account of this looking
the best in the above.
reveg_LVM1 = gllvm(reveg$abund, num.lv=2, family="poisson", trace=TRUE, jitter.var=0.2)
par(mfrow=c(1,3))
plot(reveg_LVM1,which=c(1,2,5))
Wow this does not look good! There is a clear fan-shape in the
residual vs fits plot, which also shows up as an increasing trend in the
scale-location plot. Points on the normal quantile plot are well outside
bounds on both sides, frequently falling below -5 or above 5 (when we
would expect most values between -3 and 3). These are all strong signs
of overdispersion.
Code Box 12.7: A non-metric multi-dimensional scaling ordination of
Anthony’s data
library(vegan)
ord_mds=metaMDS(reveg$abund)
#> Square root transformation
#> Wisconsin double standardization
#> Run 0 stress 0.1611237
#> Run 1 stress 0.1641162
#> Run 2 stress 0.1653221
#> Run 3 stress 0.1687788
#> Run 4 stress 0.177687
#> Run 5 stress 0.1805824
#> Run 6 stress 0.1745439
#> Run 7 stress 0.1880315
#> Run 8 stress 0.1611236
#> ... New best solution
#> ... Procrustes: rmse 0.0001157573 max resid 0.0002073088
#> ... Similar to previous best
#> Run 9 stress 0.3314815
#> Run 10 stress 0.1687788
#> Run 11 stress 0.2244987
#> Run 12 stress 0.164116
#> Run 13 stress 0.1641167
#> Run 14 stress 0.1611236
#> ... New best solution
#> ... Procrustes: rmse 0.0002265118 max resid 0.0004099737
#> ... Similar to previous best
#> Run 15 stress 0.1641163
#> Run 16 stress 0.1653221
#> Run 17 stress 0.1611236
#> ... Procrustes: rmse 0.0002056824 max resid 0.0003733336
#> ... Similar to previous best
#> Run 18 stress 0.2984
#> Run 19 stress 0.1611236
#> ... Procrustes: rmse 0.0001213268 max resid 0.0002138302
#> ... Similar to previous best
#> Run 20 stress 0.1611236
#> ... New best solution
#> ... Procrustes: rmse 5.352512e-05 max resid 9.438702e-05
#> ... Similar to previous best
#> *** Solution reached
plot(ord_mds$points,pch=as.numeric(reveg$treatment),col=reveg$treatment)
Exercise 12.6: MDS ordinations of coral data
library(mvabund)
data(tikus)
tikusAbund = tikus$abund[1:20,] # for 1981 and 1983 data only
tikusAbund = tikusAbund[,apply(tikusAbund,2,sum)>0] # remove zerotons
Construct an MDS plot of the data, using the Bray-Curtis distance
(default), and colour-code symbols by year of sampling.
tikus_mds=metaMDS(tikusAbund)
#> Square root transformation
#> Wisconsin double standardization
#> Run 0 stress 0.09863027
#> Run 1 stress 0.1318005
#> Run 2 stress 0.1323406
#> Run 3 stress 0.109769
#> Run 4 stress 0.1086391
#> Run 5 stress 0.1410027
#> Run 6 stress 0.1068254
#> Run 7 stress 0.1024098
#> Run 8 stress 0.1165249
#> Run 9 stress 0.09862977
#> ... New best solution
#> ... Procrustes: rmse 0.0003897009 max resid 0.0009384022
#> ... Similar to previous best
#> Run 10 stress 0.09863026
#> ... Procrustes: rmse 0.001910785 max resid 0.004769176
#> ... Similar to previous best
#> Run 11 stress 0.1120024
#> Run 12 stress 0.1215337
#> Run 13 stress 0.1042514
#> Run 14 stress 0.1063168
#> Run 15 stress 0.1024098
#> Run 16 stress 0.1024096
#> Run 17 stress 0.1097838
#> Run 18 stress 0.1209392
#> Run 19 stress 0.1010801
#> Run 20 stress 0.1100804
#> *** Solution reached
plot(tikus_mds$points,pch=as.numeric(tikus$x$time),col=tikus$x$time)
Does this plot agree with the Warwick et al. (1990)
interpretation? [Warwick et al. (1990) used this dataset and MDS
ordinations to argue that stress increases dispersion in coral
communities]
Yes it does, 1981 (before El Niño disturbance) the points are close
together in the middle of the ordination, 1983 (post disturbance) they
are spread out around the same point but way further apart, suggesting a
change in dispersion.
Construct another MDS plot using the Euclidean distance on
log(y+1)-transformed data.
tikus_mdsEuc=metaMDS(log(tikusAbund+1),distance="euclidean")
#> Run 0 stress 0.05741195
#> Run 1 stress 0.05741186
#> ... New best solution
#> ... Procrustes: rmse 0.0007236973 max resid 0.002122405
#> ... Similar to previous best
#> Run 2 stress 0.05741181
#> ... New best solution
#> ... Procrustes: rmse 0.0006510492 max resid 0.001915565
#> ... Similar to previous best
#> Run 3 stress 0.0574116
#> ... New best solution
#> ... Procrustes: rmse 0.000209974 max resid 0.0006023444
#> ... Similar to previous best
#> Run 4 stress 0.05741188
#> ... Procrustes: rmse 0.0002484388 max resid 0.0007128858
#> ... Similar to previous best
#> Run 5 stress 0.05741183
#> ... Procrustes: rmse 0.0002121153 max resid 0.0006080952
#> ... Similar to previous best
#> Run 6 stress 0.05741178
#> ... Procrustes: rmse 0.0003856582 max resid 0.001167658
#> ... Similar to previous best
#> Run 7 stress 0.05741211
#> ... Procrustes: rmse 0.000576628 max resid 0.001741823
#> ... Similar to previous best
#> Run 8 stress 0.05741227
#> ... Procrustes: rmse 0.0006196617 max resid 0.001878622
#> ... Similar to previous best
#> Run 9 stress 0.05741165
#> ... Procrustes: rmse 7.097593e-05 max resid 0.0002030925
#> ... Similar to previous best
#> Run 10 stress 0.05741164
#> ... Procrustes: rmse 0.0002544858 max resid 0.0007717277
#> ... Similar to previous best
#> Run 11 stress 0.05741192
#> ... Procrustes: rmse 0.0004851852 max resid 0.001463452
#> ... Similar to previous best
#> Run 12 stress 0.07425305
#> Run 13 stress 0.05741202
#> ... Procrustes: rmse 0.000318195 max resid 0.0009133398
#> ... Similar to previous best
#> Run 14 stress 0.05741167
#> ... Procrustes: rmse 0.0003011839 max resid 0.0009098946
#> ... Similar to previous best
#> Run 15 stress 0.05741182
#> ... Procrustes: rmse 0.0002108956 max resid 0.0006049422
#> ... Similar to previous best
#> Run 16 stress 0.0574122
#> ... Procrustes: rmse 0.000557587 max resid 0.001671816
#> ... Similar to previous best
#> Run 17 stress 0.05741172
#> ... Procrustes: rmse 0.0003450418 max resid 0.001041247
#> ... Similar to previous best
#> Run 18 stress 0.05741177
#> ... Procrustes: rmse 0.0003800631 max resid 0.001149431
#> ... Similar to previous best
#> Run 19 stress 0.07425315
#> Run 20 stress 0.05741199
#> ... Procrustes: rmse 0.0004999004 max resid 0.001515361
#> ... Similar to previous best
#> *** Solution reached
plot(tikus_mdsEuc$points,pch=as.numeric(tikus$x$time),col=tikus$x$time)
Does this plot agree with the Warwick et al. (1990)
interpretation?
Nope – this says the opposite, with much lower dispersion post
disturbance. It is suggestive of a location effect as well, that is, a
change in mean abundance not just variability.
Use the plot.mvabund function to plot each coral
response variable as a function of time. What is the main pattern that
you see?
tikusMV = mvabund(tikusAbund)
plot(tikusMV~tikus$x$time[1:20])
#> Overlapping points were shifted along the y-axis to make them visible.
#>
#> PIPING TO 2nd MVFACTOR
#> Only the variables Heliopora.coerulea, Montipora.digitata, Favites.abdita, Favites.chinensis, Platygyra.daedalea, Montipora.foliosa, Pocillopora.damicornis, Acropora.cytherea, Acropora.hyacinthus, Acropora.formosa, Pocillopora.verrucosa, Acropora.pulchra were included in the plot
#> (the variables with highest total abundance).
Convert the data into presence-absence and use the gllvm package
to construct an ordination
tikusPA = tikusAbund
tikusPA[tikusPA>1]=1
tikus_LVM = gllvm(tikusPA, num.lv=2, family="binomial", trace=TRUE, jitter.var=0.2)
ordiplot.gllvm(tikus_LVM, s.col=as.numeric(tikus$x$time), biplot=TRUE, ind.spp=12)
Do assumptions appear reasonable? How would you interpret this
plot?
par(mfrow=c(1,3))
plot(tikus_LVM,which=c(1,2,5))
Code Box 12.8: Studying each observation separately for the Iris
data
by(iris, iris$Species, function(dat){ apply(dat[,1:4],2,mean) } )
#> iris$Species: setosa
#> Sepal.Length Sepal.Width Petal.Length Petal.Width
#> 5.006 3.428 1.462 0.246
#> ---------------------------------------------------------------------------
#> iris$Species: versicolor
#> Sepal.Length Sepal.Width Petal.Length Petal.Width
#> 5.936 2.770 4.260 1.326
#> ---------------------------------------------------------------------------
#> iris$Species: virginica
#> Sepal.Length Sepal.Width Petal.Length Petal.Width
#> 6.588 2.974 5.552 2.026
par(mfrow=c(2,2),mar=c(3,3,1,1),mgp=c(1.75,0.75,0))
plot(Sepal.Length~Species,data=iris,xlab="")
plot(Sepal.Width~Species,data=iris,xlab="")
plot(Petal.Length~Species,data=iris,xlab="")
plot(Petal.Width~Species,data=iris,xlab="")
#remove this chunk once gllvm has been updated on CRAN:
