Ecological factors influencing primate vocal signaling:
a phylogenetic regression workflow for the
mmodely R-package (Version 0.2.5)
1 Introduction
The historical relationships between evolved traits of organisms and the ecological settings that shape these traits are complicated systems that can be challenging to untangle [2]. The origins of behavioral traits can particularly difficult to understand as they tend to also be mediated through the behaviors of other organisms, which are themselves constantly in flux and considerably labile [3]. A perfect example of such a trait is that of vocal signal complexity. Animals use complex calls to assert obscured position, unique identity, special status, or emotive state to conspecifics over interference from other calls or distortions from background noise [1]. Here, using the mmodely package on a primate vocalization dataset [15], I demonstrate how the origins of complex call structure, such as syllablic diversity [17], can be elucidated from a range of environmental and behavioral covariates from disparate datasets [10]. Model averaging [MA] [6] and model selection [MS] [7] results primarily highlight locomotion [16] and mating system [13] as important factors driving complex calling, as well as the trophic security [18] variables of mass, group size, and arboreality. The mmodely package enables implementation of a combination of phylogenetic controlled regression [8] and information theoretic [9] (MA and MS) examination to simultaneously compare (weighted) predictor coefficients across the numerous sub-datasets generated during exploration of all possible model combinations.
2 Licensing
The mmodely package is licensed under the Apache License v2.0: it is therefore free to use and redistribute, however, we, the copyright holders, wish to maintain primary control over any further development. Please be sure to cite mmodely if you use the package in presentations or work leading to publication.
3 Installation
This package largely depends upon the caper package, but most functions do not require any particular library. It is recommended that you have caper, ape, and the caroline package installed as a minimum.
> # wget https://cran.r-project.org/src/contrib/Archive/caroline/caroline_0.8.0.tar.gz> # wget https://cran.r-project.org/src/contrib/Archive/caper/caper_0.5.tar.gz> # wget https://cran.r-project.org/src/contrib/Archive/ape/ape_3.0-5.tar.gz> # R CMD INSTALL caroline_0.8.0.tar.gz> # R CMD INSTALL caper_0.5.tar.gz> # R CMD INSTALL ape_3.0-5.tar.gz
Building the mmodely package from source requires that you have the proper dependency packages, caroline, installed from CRAN. This can typically be accomplished via the following commands from within the R command line environment:
install.packages(c(’caroline’,’ape’,’caper’))
After a successful installation the mmodely package can be loaded in the normal way: by starting R and invoking the following library command:
> library(caper)> library(mmodely)
4 Reading in Data
Read in the tree [14] and datasets then merge them together.
> data.path <- system.file("extdata","primate-example.data.csv", package="mmodely")> data <- read.csv(data.path, row.names=1)> data$gn_sp <- rownames(data)> #multiply two vocalization metrics together to create "vocal complexity"> data$VC <- apply(data[,c('syllables_max','rhythm_max')], 1, prod)> data <- subset(data, !is.na(VC))> # merge data sets here if applicable>> tree.path <- system.file("extdata","primate-springer.2012.tre", package="mmodely")> phyl <- ape::read.tree(tree.path)[[5]]> #5. RAxML phylogram based on the 61199 bp concatenation of 69 nuclear and ten mitochondrial genes.>> phyl <- trim.phylo(phylo=phyl, gs.vect=data$gn_sp) # prune unused nodes and branches> comp <- comp.data(phylo=phyl, df=data)Typically there will be some missing data (species) in certain sources that do not occur in others. A merge of these will result in NA values for some cells. The more missing cells and merges there are, the more sub-datasets will be possible, due to case-wise deletion in the process of combinatorics underlying model iteration, averaging, and selection. The above example has little if any missing data, but the examples below introduce some artificially.
5 Basic Reporting
First, for illustration purposes, we perform a simple analysis of a single model using ’pgls’ directly from the caper package, then show-off the ’pgls.report’ functionality of the mmodely package. ANOVA, AIC, and one-line model reports can be output via this function.
> model <- as.formula('VC ~ mass.Kg + group.size')> fit <- caper::pgls(formula=model, data=comp)> summary(fit)
Call:caper::pgls(formula = model, data = comp)Residuals: Min 1Q Median 3Q Max-8.7308 -3.1969 0.4275 3.3780 6.5978Branch length transformations:kappa [Fix] : 1.000lambda [Fix] : 1.000delta [Fix] : 1.000Coefficients: Estimate Std. Error t value Pr(>|t|)(Intercept) 2.2345483 1.1333734 1.9716 0.0662 .mass.Kg -0.0079678 0.0082070 -0.9709 0.3461group.size 0.0071381 0.0144792 0.4930 0.6287---Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1Residual standard error: 4.61 on 16 degrees of freedomMultiple R-squared: 0.1101,Adjusted R-squared: -0.001146F-statistic: 0.9897 on 2 and 16 DF, p-value: 0.3933
> pgls.report(comp, f=model, anova=TRUE, QC.plot=TRUE)
Call:pgls(formula = f, data = cd, lambda = l, kappa = k, delta = d, bounds = bounds)Residuals: Min 1Q Median 3Q Max-8.7308 -3.1969 0.4275 3.3780 6.5978Branch length transformations:kappa [Fix] : 1.000lambda [Fix] : 1.000delta [Fix] : 1.000Coefficients: Estimate Std. Error t value Pr(>|t|)(Intercept) 2.2345483 1.1333734 1.9716 0.0662 .mass.Kg -0.0079678 0.0082070 -0.9709 0.3461group.size 0.0071381 0.0144792 0.4930 0.6287---Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1Residual standard error: 4.61 on 16 degrees of freedomMultiple R-squared: 0.1101,Adjusted R-squared: -0.001146F-statistic: 0.9897 on 2 and 16 DF, p-value: 0.3933[1] "AIC = 58"Analysis of Variance TableSequential SS for pgls: lambda = 1.00, delta = 1.00, kappa = 1.00Response: VC Df Sum Sq Mean Sq F value Pr(>F)mass.Kg 1 36.89 36.894 1.7363 0.2062group.size 1 5.16 5.164 0.2430 0.6287Residuals 16 339.96 21.248 group(0.629) | mass(0.346)Call:pgls(formula = f, data = cd, lambda = l, kappa = k, delta = d, bounds = bounds)Coefficients:(Intercept) mass.Kg group.size 2.234548 -0.007968 0.007138
6 Multivariate Combinatoric Iteration
The mmodely package’s chief contribution is enabling approaches that utilize multi-model iteration averaging. Using a smaller subset of variables can speed up the (slower) maximum likelihood computation step and still achieve the desired result of fixed tree transformation parameters.
> pv0 <- c("mass.Kg","arboreal","home.range","monogamy") #"swing.pct"> est.mods <- get.model.combos(predictor.vars=pv0, outcome.var='VC', min.q=2)> ps <- get.phylo.stats(phylo=phyl, data=data, trait.clmn='VC');
$lambda[1] 0.2903945$logL[1] -55.25736$P[1] 0.7103404$K[1] 0.1886703$P[1] 0.378
> lambda <- ps$lambda$lambda ; print(lambda)
[1] 0.2903945
> PGLSi <- pgls.iter(models=est.mods, phylo=phyl, df=data, l=lambda, k='ML', d='ML')
1 VC~mass.Kg+arboreal+home.range+monogamy2 VC~mass.Kg+arboreal+home.range3 VC~mass.Kg+arboreal+monogamy4 VC~mass.Kg+home.range+monogamy5 VC~arboreal+home.range+monogamy6 VC~mass.Kg+arboreal7 VC~mass.Kg+home.range8 VC~mass.Kg+monogamy9 VC~arboreal+home.range10 VC~arboreal+monogamy11 VC~home.range+monogamy
7 Tree Transformation Averaging and Re-iteration
After running PGLS on a test-subset of predictor-variable combinations using maximum likelihood, we can average the tree transformation parameters [20]to obtain fixed values going forward. This approach can speed up computations for larger sets of modeling data and variable combinations.
> tt.avgs <- apply(PGLSi$params, 2, mean, na.rm=TRUE) # tree transformation averages> print(tt.avgs)
l k d0.2903945 0.5281747 1.2349216Next we use the full set of variables and our tree transform averages. For demonstration, we sprinkle in some missing values to our dataset so as to artificially boost the number of sub-datasets. The subsequent fixed tree parameter itteration run should now generate more diverse output upon which the mmodely can demonstrate it unique model averaging and model selection functionality.
> pvs <- c("mass.Kg","group.size","arboreal","monogamy","leap.pct","swing.pct")> all.mods <- get.model.combos(predictor.vars=pvs, outcome.var='VC', min.q=2)> # randomly sprinkle in some missing values (for more interesting for model selection)> missing.value.ct <- 1> for(pv in pv0){ data[sample(x=1:nrow(data),size=missing.value.ct),pv] <- NA}> PGLSi <- pgls.iter(models=all.mods, phylo=phyl, df=data, l=lambda, k=tt.avgs['k'], d=tt.avgs['d'])
1 VC~mass.Kg+group.size+arboreal+monogamy+leap.pct+swing.pct2 VC~mass.Kg+group.size+arboreal+monogamy+leap.pct3 VC~mass.Kg+group.size+arboreal+monogamy+swing.pct4 VC~mass.Kg+group.size+arboreal+leap.pct+swing.pct5 VC~mass.Kg+group.size+monogamy+leap.pct+swing.pct6 VC~mass.Kg+arboreal+monogamy+leap.pct+swing.pct7 VC~group.size+arboreal+monogamy+leap.pct+swing.pct8 VC~mass.Kg+group.size+arboreal+monogamy9 VC~mass.Kg+group.size+arboreal+leap.pct10 VC~mass.Kg+group.size+arboreal+swing.pct11 VC~mass.Kg+group.size+monogamy+leap.pct12 VC~mass.Kg+group.size+monogamy+swing.pct13 VC~mass.Kg+group.size+leap.pct+swing.pct14 VC~mass.Kg+arboreal+monogamy+leap.pct15 VC~mass.Kg+arboreal+monogamy+swing.pct16 VC~mass.Kg+arboreal+leap.pct+swing.pct17 VC~mass.Kg+monogamy+leap.pct+swing.pct18 VC~group.size+arboreal+monogamy+leap.pct19 VC~group.size+arboreal+monogamy+swing.pct20 VC~group.size+arboreal+leap.pct+swing.pct21 VC~group.size+monogamy+leap.pct+swing.pct22 VC~arboreal+monogamy+leap.pct+swing.pct23 VC~mass.Kg+group.size+arboreal24 VC~mass.Kg+group.size+monogamy25 VC~mass.Kg+group.size+leap.pct26 VC~mass.Kg+group.size+swing.pct27 VC~mass.Kg+arboreal+monogamy28 VC~mass.Kg+arboreal+leap.pct29 VC~mass.Kg+arboreal+swing.pct30 VC~mass.Kg+monogamy+leap.pct31 VC~mass.Kg+monogamy+swing.pct32 VC~mass.Kg+leap.pct+swing.pct33 VC~group.size+arboreal+monogamy34 VC~group.size+arboreal+leap.pct35 VC~group.size+arboreal+swing.pct36 VC~group.size+monogamy+leap.pct37 VC~group.size+monogamy+swing.pct38 VC~group.size+leap.pct+swing.pct39 VC~arboreal+monogamy+leap.pct40 VC~arboreal+monogamy+swing.pct41 VC~arboreal+leap.pct+swing.pct42 VC~monogamy+leap.pct+swing.pct43 VC~mass.Kg+group.size44 VC~mass.Kg+arboreal45 VC~mass.Kg+monogamy46 VC~mass.Kg+leap.pct47 VC~mass.Kg+swing.pct48 VC~group.size+arboreal49 VC~group.size+monogamy50 VC~group.size+leap.pct51 VC~group.size+swing.pct52 VC~arboreal+monogamy53 VC~arboreal+leap.pct54 VC~arboreal+swing.pct55 VC~monogamy+leap.pct56 VC~monogamy+swing.pct57 VC~leap.pct+swing.pct
> pgls.iter.stats(PGLSi)
models: 57dimensions of sub-datasets: q n qXn rwGsm 5 4 13 8 q nMin. 2.000000 35.000001st Qu. 2.000000 36.00000Median 3.000000 36.00000Mean 3.263158 36.368423rd Qu. 4.000000 37.00000Max. 6.000000 38.00000tree transformation parameter averages: l k d0.2903945 0.5281747 1.2349216distributions of optimization parameters: n q rwGsm model.no R2 Min. :35.00 Min. :2.000 Min. :65394 Min. : 1 Min. :0.003976 1st Qu.:36.00 1st Qu.:2.000 1st Qu.:66995 1st Qu.:15 1st Qu.:0.079281 Median :36.00 Median :3.000 Median :67566 Median :29 Median :0.170974 Mean :36.37 Mean :3.263 Mean :68007 Mean :29 Mean :0.152660 3rd Qu.:37.00 3rd Qu.:4.000 3rd Qu.:69167 3rd Qu.:43 3rd Qu.:0.212376 Max. :38.00 Max. :6.000 Max. :71122 Max. :57 Max. :0.286229 R2.adj AIC AICc BIC AICw Min. :-0.077460 Min. :104.4 Min. :105.2 Min. :109.2 Min. :0.0006034 1st Qu.: 0.005563 1st Qu.:106.0 1st Qu.:108.1 1st Qu.:113.8 1st Qu.:0.0033157 Median : 0.096241 Median :108.2 Median :109.8 Median :115.3 Median :0.0100692 Mean : 0.066899 Mean :108.6 Mean :110.2 Mean :115.4 Mean :0.0175439 3rd Qu.: 0.118400 3rd Qu.:111.2 3rd Qu.:112.0 3rd Qu.:117.1 3rd Qu.:0.0239646 Max. : 0.160892 Max. :114.7 Max. :115.4 Max. :120.6 Max. :0.1033613 BICw Min. :0.0005816 1st Qu.:0.0032266 Median :0.0079549 Mean :0.0175439 3rd Qu.:0.0171026 Max. :0.1732078
8 Fixed iteration run statistics
We should briefly inspect how this fixed iteration run performed and how many sub-datasets we need to investigate. It is recommended to try mmodely using ’rwGsm.’ This abbreviation stands for ’raw Genus species sums.’ It represents a sum of the (concatenated) raw character values of all species constituting the underlying dataset (which has all rows with any missing data removed) for a particular combination of model predictor variables. While this default is preferred, the number of species ’n’ [default] or number of model variables ’q’ can also be used.
> pgls.iter.stats(PGLSi)
models: 57dimensions of sub-datasets: q n qXn rwGsm 5 4 13 8 q nMin. 2.000000 35.000001st Qu. 2.000000 36.00000Median 3.000000 36.00000Mean 3.263158 36.368423rd Qu. 4.000000 37.00000Max. 6.000000 38.00000tree transformation parameter averages: l k d0.2903945 0.5281747 1.2349216distributions of optimization parameters: n q rwGsm model.no R2 Min. :35.00 Min. :2.000 Min. :65394 Min. : 1 Min. :0.003976 1st Qu.:36.00 1st Qu.:2.000 1st Qu.:66995 1st Qu.:15 1st Qu.:0.079281 Median :36.00 Median :3.000 Median :67566 Median :29 Median :0.170974 Mean :36.37 Mean :3.263 Mean :68007 Mean :29 Mean :0.152660 3rd Qu.:37.00 3rd Qu.:4.000 3rd Qu.:69167 3rd Qu.:43 3rd Qu.:0.212376 Max. :38.00 Max. :6.000 Max. :71122 Max. :57 Max. :0.286229 R2.adj AIC AICc BIC AICw Min. :-0.077460 Min. :104.4 Min. :105.2 Min. :109.2 Min. :0.0006034 1st Qu.: 0.005563 1st Qu.:106.0 1st Qu.:108.1 1st Qu.:113.8 1st Qu.:0.0033157 Median : 0.096241 Median :108.2 Median :109.8 Median :115.3 Median :0.0100692 Mean : 0.066899 Mean :108.6 Mean :110.2 Mean :115.4 Mean :0.0175439 3rd Qu.: 0.118400 3rd Qu.:111.2 3rd Qu.:112.0 3rd Qu.:117.1 3rd Qu.:0.0239646 Max. : 0.160892 Max. :114.7 Max. :115.4 Max. :120.6 Max. :0.1033613 BICw Min. :0.0005816 1st Qu.:0.0032266 Median :0.0079549 Mean :0.0175439 3rd Qu.:0.0171026 Max. :0.1732078
9 Model Averaging
Now we can estimate the predictor variable parameters by averaging over all possible fixed PGLS runs, using the AICc differences (from the lowest AICc) as weights. By default this AICw weighted average is performed per sub-dataset using ’rwGsm’ or ’n’ [default] as mentioned in the preceding section. While model averaging is not recommended under high multicolinearity, as denominators of regression coefficients change across models, it is possible to rescale these using ’standarize’ [12]. A slightly more conservative alternative to MA uses ’variable importance’ which is equivalent to an AIC-weighted MA of binary indicators of presence or absence of covariate model inclusion [11].
> w.means.pds <- average.fit.models(vars=pvs, fits=PGLSi$fits, optims=PGLSi$optim, by='rwGsm', standardize=TRUE)> #> apply(w.means.pds, 2, mean, na.rm=T) #average of weighted means over all sub-datasets
mass.Kg group.size arboreal monogamy leap.pct swing.pct 0.01284500 0.01412875 -0.15867000 0.83971250 1.41151625 1.49308125
> w.means.pds # weighted means per sub-dataset
mass.Kg group.size arboreal monogamy leap.pct swing.pct65394 0.01894 0.01626 -0.07205 0.83271 1.23000 1.2076666995 0.00633 0.01119 -0.08612 NaN 1.85829 2.0856767349 NaN 0.01818 -0.18527 0.81311 1.06517 1.1404667566 0.01853 0.01572 NaN 0.85671 1.04488 0.9462568950 NaN 0.01148 -0.29124 NaN 1.90969 2.1227069167 0.00758 0.01041 NaN NaN 1.66818 1.8162869521 NaN 0.01814 NaN 0.85632 0.81973 0.8279771122 NaN 0.01165 NaN NaN 1.69619 1.79766attr(,"MSE") mass.Kg group.size arboreal monogamy leap.pct swing.pct65394 0.00082 0.00006 0.05202 0.06481 0.06517 0.0732566995 0.00013 0.00004 0.01051 NA 0.01362 0.0154067349 NA 0.00004 0.02726 0.03385 0.02927 0.0345867566 0.00134 0.00018 NA 0.08651 0.06483 0.0751968950 NA 0.00003 0.00830 NA 0.01188 0.0136169167 0.00017 0.00005 NA NA 0.01037 0.0114569521 NA 0.00010 NA 0.03192 0.02099 0.0259871122 NA 0.00003 NA NA 0.00383 0.00495
> w.import.pds <- variable.importance(vars=pvs, fits=PGLSi$fits, optims=PGLSi$optim, by='rwGsm')> #> apply(w.import.pds, 2, mean, na.rm=T) #average of weighted means over all sub-datasets
mass.Kg group.size arboreal monogamy leap.pct swing.pct 0.986185 1.000000 0.245325 1.000000 1.000000 1.000000
> w.import.pds # weighted means per sub-dataset
mass.Kg group.size arboreal monogamy leap.pct swing.pct65394 1.00000 1 0.25516 1 1 166995 1.00000 1 0.51715 NaN 1 167349 NaN 1 0.00000 1 1 167566 1.00000 1 NaN 1 1 168950 NaN 1 0.20899 NaN 1 169167 0.94474 1 NaN NaN 1 169521 NaN 1 NaN 1 1 171122 NaN 1 NaN NaN 1 1attr(,"MSE") mass.Kg group.size arboreal monogamy leap.pct swing.pct65394 0.0000 0 0.16708 0 0 066995 0.0000 0 0.02099 NA 0 067349 NA 0 0.00000 0 0 067566 0.0000 0 NA 0 0 068950 NA 0 0.01532 NA 0 069167 0.0106 0 NA NA 0 069521 NA 0 NA 0 0 071122 NA 0 NA NA 0 0
10 Model Selection
We can select the best model by sorting each subset (e.g. by AICc) or by using visualization methods.
> select.best.models(PGLSi, using='AICc')
n q qXn rwGsm model.no R2 R2.adj AIC AICc BIC AICw36 36 2 2X36 67566 45 0.1687772 0.11840009 104.4109 105.1609 109.1615 0.10336127635 35 3 3X35 65394 27 0.1623961 0.08133768 104.3567 105.6901 110.5781 0.07933307937 37 2 2X37 69521 49 0.2030251 0.15614418 106.1313 106.8586 110.9641 0.04422961138 38 2 2X38 71122 57 0.1551053 0.10682562 110.2290 110.9348 115.1417 0.005761936 BICw36 0.17320781335 0.08529926637 0.07032969638 0.008709162
Plotting the coefficients of determination versus the AIC values allows selection of high-performing models for inspection and reporting.
> plot.pgls.iters(PGLSi)
> sdevs.objs <- get.pgls.coefs(PGLSi$fits, est='t value')> coefs.objs <- get.pgls.coefs(PGLSi$fits, est='Estimate')
> report.vect <- sapply(1:length(PGLSi$fits), function(i) fit.1ln.rprt(PGLSi$fits[[i]], rtrn.line=FALSE, mn=i))
1 +leap(0.113) +swing(0.158) +monog(0.162) +group(0.185) +mass(0.195) | arbore(0.628) R2adj: 0.133 AICc: 108.912 ++mono(0.033) +group(0.21) +mass(0.229) leap(0.341) | arbore(0.972) R2adj: 0.1 AICc: 108.293 ++mono(0.022) +group(0.196) +mass(0.311) swing(0.546) arbore(0.954) | R2adj: 0.083 AICc: 108.964 ++leap(0.022) ++swin(0.034) +group(0.233) mass(0.509) | arbore(0.624) R2adj: 0.078 AICc: 111.095 +monog(0.115) +leap(0.129) +group(0.148) +mass(0.168) +swing(0.193) | R2adj: 0.161 AICc: 108.116 +leap(0.117) +mass(0.159) +swing(0.178) +monog(0.297) | arbore(0.48) R2adj: 0.108 AICc: 107.997 +group(0.125) +leap(0.142) +monog(0.165) +swing(0.165) | arbore(0.487) R2adj: 0.141 AICc: 109.848 ++mono(0.013) +group(0.2) +mass(0.29) arbore(0.874) | R2adj: 0.102 AICc: 106.489 +leap(0.147) group(0.47) mass(0.693) arbore(0.831) | R2adj: -0.038 AICc: 113.6610 +swing(0.26) group(0.499) arbore(0.74) mass(0.822) | R2adj: -0.067 AICc: 114.6411 ++mono(0.023) +group(0.187) +mass(0.211) +leap(0.321) | R2adj: 0.14 AICc: 107.2812 ++mono(0.014) +group(0.181) +mass(0.295) swing(0.552) | R2adj: 0.122 AICc: 108.0313 ++leap(0.02) ++swin(0.038) +group(0.215) mass(0.405) | R2adj: 0.096 AICc: 111.0114 +monog(0.068) +mass(0.188) leap(0.327) | arbore(0.783) R2adj: 0.081 AICc: 107.2915 ++mono(0.048) +mass(0.258) swing(0.592) | arbore(0.863) R2adj: 0.06 AICc: 108.0816 ++leap(0.035) +swing(0.055) mass(0.613) | arbore(0.504) R2adj: 0.064 AICc: 109.9317 +mass(0.126) +leap(0.156) +monog(0.236) +swing(0.249) | R2adj: 0.128 AICc: 107.7718 ++mono(0.031) +group(0.142) leap(0.407) | arbore(0.796) R2adj: 0.112 AICc: 109.2919 ++mono(0.019) +group(0.131) swing(0.505) | arbore(0.861) R2adj: 0.105 AICc: 109.5820 ++leap(0.015) ++swin(0.028) +group(0.224) | arbore(0.411) R2adj: 0.12 AICc: 110.9121 +group(0.086) +monog(0.116) +leap(0.19) +swing(0.229) | R2adj: 0.161 AICc: 109.6122 +leap(0.15) +swing(0.191) monoga(0.338) | -arbor(0.322) R2adj: 0.099 AICc: 109.8223 arbore(0.576) group(0.593) mass(0.853) | R2adj: -0.077 AICc: 113.4324 ++mono(0.007) +group(0.19) +mass(0.283) | R2adj: 0.139 AICc: 105.7425 +leap(0.11) group(0.502) mass(0.741) | R2adj: -0.004 AICc: 113.3726 +swing(0.236) group(0.572) mass(0.948) | R2adj: -0.041 AICc: 114.6827 ++mono(0.028) +mass(0.242) | arbore(0.934) R2adj: 0.081 AICc: 105.6928 +leap(0.167) mass(0.75) arbore(0.961) | R2adj: -0.023 AICc: 111.5629 +swing(0.289) arbore(0.859) mass(0.872) | R2adj: -0.049 AICc: 112.4730 +monog(0.054) +mass(0.158) leap(0.332) | R2adj: 0.118 AICc: 106.6331 ++mono(0.033) +mass(0.222) swing(0.626) | R2adj: 0.098 AICc: 107.4332 ++leap(0.038) +swing(0.069) mass(0.47) | R2adj: 0.08 AICc: 110.1333 ++mono(0.011) +group(0.134) | arbore(0.94) R2adj: 0.121 AICc: 107.3934 +leap(0.112) group(0.44) arbore(0.991) | R2adj: 0.006 AICc: 113.8835 +swing(0.238) group(0.456) arbore(0.816) | R2adj: -0.03 AICc: 115.1736 ++mono(0.023) +group(0.11) leap(0.416) | R2adj: 0.148 AICc: 108.6337 ++mono(0.012) +group(0.106) swing(0.536) | R2adj: 0.141 AICc: 108.9438 ++leap(0.018) ++swin(0.041) +group(0.198) | R2adj: 0.125 AICc: 111.5639 +monog(0.079) leap(0.394) | arbore(0.559) R2adj: 0.077 AICc: 109.1340 ++mono(0.052) swing(0.556) | arbore(0.622) R2adj: 0.066 AICc: 109.5641 ++leap(0.023) ++swin(0.044) | arbore(0.329) R2adj: 0.106 AICc: 109.9642 +leap(0.247) +monog(0.279) +swing(0.316) | R2adj: 0.106 AICc: 110.443 group(0.72) | mass(0.936) R2adj: -0.055 AICc: 113.7644 arbore(0.662) mass(0.891) | R2adj: -0.054 AICc: 111.2145 ++mono(0.017) +mass(0.214) | R2adj: 0.118 AICc: 105.1646 +leap(0.13) mass(0.753) | R2adj: 0.012 AICc: 111.3647 +swing(0.265) mass(0.947) | R2adj: -0.02 AICc: 112.5348 group(0.547) arbore(0.609) | R2adj: -0.043 AICc: 114.2449 ++mono(0.006) +group(0.112) | R2adj: 0.156 AICc: 106.8650 +leap(0.081) group(0.442) | R2adj: 0.037 AICc: 113.7851 +swing(0.217) group(0.509) | R2adj: -0.006 AICc: 115.4552 ++mono(0.03) | arbore(0.685) R2adj: 0.084 AICc: 107.4253 +leap(0.125) | arbore(0.875) R2adj: 0.017 AICc: 112.0454 +swing(0.269) arbore(0.944) | R2adj: -0.017 AICc: 113.2855 +monog(0.074) leap(0.448) | R2adj: 0.105 AICc: 109.0256 ++mono(0.042) swing(0.628) | R2adj: 0.096 AICc: 109.457 ++leap(0.033) +swing(0.075) | R2adj: 0.107 AICc: 110.93
> par(mar=c(5,5,3,3))> plot.pgls.R2AIC(PGLSi$optim)
11 Coefficient Plotting
Finally, the resulting model fits from the PGLS runs can be be plotted out horizontally as distributions so the influence of each ecological predictor variable can be compared.
> par.old <- par(mar=c(5,8,1,4),mfrow=c(2,1))> sparge.modsel(sdevs.objs, R2x=7, xlab='t value')> sparge.modsel(coefs.objs, R2x=7, xlab='Estimate')
References
- [1] Rogers, L.J. and Kaplan, G. (2000) Songs, Roars, and Rituals: communication in birds, mammals, and other animals. Harvard University Press, Cambridge.
- [2] Pagel, M. (1999) ‘Inferring the historical patterns of biological evolution’ Nature, 401, pp. 877–884. Available at: https://doi.org/10.1038/44766.
- [3] Blomberg, S.P., Garland Jr, T., and Ives, A.R. (2003) ‘Testing for phylogenetic signal in comparative data: behavioral traits are more labile’ Evolution, 57(4) pp. 717–745. Available at: https://doi.org/10.1111/j.0014-3820.2003.tb00285.x.
- [4] Grueber, C.E., Nakagawa, .R.J., and Jamieson, I.G. (2011) ‘Multimodel inference in ecology and evolution: challenges and solutions’ Journal of Evolutionary Biology, 24, pp. 699-711. Available at: https://doi.org/10.1111/j.1420-9101.2010.02210.x
- [5] Stephens, P.A., Buskirk, S.W., and del Rio, C.M. (2007) ‘Inference in ecology and evolution’ Trends in Ecology and Evolution, 22(4). Available at: https://doi.org/10.1016/j.tree.2006.12.003.
- [6] Symonds, M.R.E., and Moussalli, A. (2011) ‘A brief guide to model selection, multimodel inference and model averaging in behavioral ecology using Akaike’s information criterion’ Behavioral Ecology and Sociobiology, 65, pp. 13–21. Available at: https://doi.org/10.1007/s00265-010-1037-6.
- [7] Johnson, J.B., Omland, K.S. (2004) ‘Model selection in ecology and evolution’ Trends in Ecology and Evolution, 19(2). Available at: https://doi.org/10.1016/j.tree.2003.10.013.
- [8] Revell, L.J. (2014) phytools: An R package for phylogenetic comparative biology (and other things). [CRAN]. Available at: http://cran.r-project.org/package=phytools.
- [9] Garamszegi L.Z. (2011) ‘Information-theoretic approaches to statistical analysis in behavioral ecology: an introduction’ Behavioral Ecology and Sociobiology, 65, pp. 1–11. Available at: https://doi.org/10.1007/s00265-010-1028-7.
- [10] Rowe, N. and Meyers, M. (2017) All the World’s Primates. Charlestown, RI: Pogonias Press.
- [11] Burnham, P.B. and Anderson, D.R. (2000) ‘Model Selection and Inference: A Practical Information-Theoretic Approach’
- [12] Cade, B.S. (2015) ‘Model averaging and muddled multimodel inferences’ Ecology, 96(9), pp. 2370–2382. Available at: https://doi.org/abs/10.1890/14-1639.1.
- [13] Feuntes, A. (1998) ‘Re-evaluating primate monogamy’ American Anthropology, 100(4) pp. 890-907.
- [14] Springer, M.S., et.al. (2012) ‘Re-evaluating primate monogamy’ PLoS ONE 7(11) p. e49521 http://doi.org/10.1371/journal.pone.0049521.
- [15] Schruth, D.M. (2019a) ‘Structural acoustic features of human musicality scored on primate vocalizations’ The Center for Open Science. Available at: https://doi.io/bvsfz/.
- [16] Schruth, D.M. (2019b) ‘Primate Locomotor Activity’ The Center for Open Science. Available at: https://osf.io/cd68q/.
- [17] Schruth, D.M., Templeton, C.N. and Holman, D.J. (2021a) ‘On reappearance and complexity in musical calling’ PLoS ONE. Available at: https://doi.org/10.1371/journal.pone.0218006.
- [18] Schruth, D.M. (2021b) Arboreal locomotion and trophic security at the dawn of Euprimate vision. EcoEvoRxiv. Available at: http://doi.org/10.32942/osf.io/d6wk2.
- [19] Schruth, D.M. (2021c) ‘A global variable-permutation based approach for estimating tree transformation parameters used in phylogenetically controlled multivariate regression’ Protocols.io. Available at: http://doi.org/10.17504/protocols.io.bzdhp236.
- [20] Schruth, D.M. (2021d) ‘Primates evolved spectrally complex calls in compensation for reduction in olfactory cognition’ Proceedings of the Annual Meeting of the Cognitive Science Society 43. Available at: https://escholarship.org/uc/item/0jw446s9.