RPANDA (version 2.2)

fit_t_comp: Fits models of trait evolution incorporating competitive interactions

Description

Fits matching competition (MC), diversity dependent linear (DDlin), or diversity dependent exponential (DDexp) models of trait evolution to a given dataset and phylogeny.

Usage

fit_t_comp(phylo, data, error=NULL, model=c("MC","DDexp","DDlin"), pars=NULL, 
		geography.object=NULL, regime.map=NULL)

Value

a list with the following elements:

LH

maximum log-likelihood value

aic

Akaike Information Criterion value

aicc

AIC value corrected for small sample size

free.parameters

number of free parameters from the model

sig2

maximum-likelihood estimate of sig2 parameter

S

maximum-likelihood estimate of S parameter of matching competition model (see Note)

b

maximum-likelihood estimate of b parameter of linear diversity dependence model

r

maximum-likelihood estimate of r parameter of exponential diversity dependence model

z0

maximum-likelihood estimate of z0, the value at the root of the tree

nuisance

maximum-likelihood estimate of nuisance, the unknown, nuisance contribution to measurement error (see details)

convergence

convergence diagnostics from optim function (see optim documentation)

Arguments

phylo

an object of type 'phylo' (see ape documentation)

data

a named vector of trait values with names matching phylo$tip.label

error

if including measurement error, a named vector of error values (the standard error of the [intraspecific] mean) with names matching phylo$tip.label

model

model chosen to fit trait data, "MC" is the matching competition model of Nuismer & Harmon 2014, "DDlin" is the diversity-dependent linear model, and "DDexp" is the diversity-dependent exponential model of Weir & Mursleen 2013.

pars

vector specifying starting parameter values for maximum likelihood optimization. If unspecified, default values are used (see Details)

geography.object

if incorporating biogeography, a list of sympatry through time created using CreateGeoObject

regime.map

if running two-regime versions of models, a stochastic map of the two regimes stored as a simmap object output from make.simmap

Author

Jonathan Drury

Julien Clavel

Details

Note: if including known measurement error, the model fit incorporates this known error and, in addition, estimates an unknown, nuisance contribution to measurement error. The current implementation does not differentiate between the two, so, for instance, it is not possible to estimate the nuisance measurement error without providing the known, intraspecific error values.

For single-regime fits without measurement error, par takes the default values of var(data)/max(nodeHeights(phylo)) for sig2 and 0 for either S for the matching competition model, b for the linear diversity dependence model, or r for the exponential diversity dependence model. Values can be manually entered as a vector with the first element equal to the desired starting value for sig2 and the second value equal to the desired starting value for either S, b, or r. Note: since likelihood optimization uses sig rather than sig2, and since the starting value for is exponentiated to stabilize the likelihood search, if you input a par value, the first value specifying sig2 should be the log(sqrt()) of the desired sig2 starting value.

For two-regime fits without measurement error, the second and third values for par correspond to the first and second S, b, or r value (run trial fit to see which regime corresponds to each slope).

For fits including measurement error, the default starting value for sig2 is 0.95*var(data)/max(nodeHeights(phylo)), and nuisance values start at 0.05*var(data)/max(nodeHeights(phylo)). In all cases, the nuisance parameter is the last in the par vector, with the order of other variables as described above.

For two-regime fits, particularly under the matching competition model, we recommend fitting with several different starting values.

References

Drury, J., Clavel, J., Manceau, M., and Morlon, H. 2016. Estimating the effect of competition on trait evolution using maximum likelihood inference. Systematic Biology 65:700-710

Nuismer, S. & Harmon, L. 2015. Predicting rates of interspecific interaction from phylogenetic trees. Ecology Letters 18:17-27.

Weir, J. & Mursleen, S. 2012. Diversity-dependent cladogenesis and trait evolution in the adaptive radiation of the auks (Aves: Alcidae). Evolution 67:403-416.

See Also

sim_t_comp CreateGeoObject likelihood_t_MC likelihood_t_MC_geog likelihood_t_DD likelihood_t_DD_geog fit_t_comp_subgroup

Examples

data(Anolis.data)
geography.object<-Anolis.data$geography.object
pPC1<-Anolis.data$data
phylo<-Anolis.data$phylo

# \donttest{
#Fit three models without biogeography to pPC1 data
MC.fit<-fit_t_comp(phylo, pPC1, model="MC")
DDlin.fit<-fit_t_comp(phylo, pPC1, model="DDlin")
DDexp.fit<-fit_t_comp(phylo, pPC1, model="DDexp")

#Now fit models that incorporate biogeography, NOTE these models take longer to fit
MC.geo.fit<-fit_t_comp(phylo, pPC1, model="MC", geography.object=geography.object)
DDlin.geo.fit<-fit_t_comp(phylo, pPC1,model="DDlin", geography.object=geography.object)
DDexp.geo.fit<-fit_t_comp(phylo, pPC1, model="DDexp", geography.object=geography.object)
# }