geiger (version 1.0-91)

fit.continuous: Model fitting for continuous data

Description

Fits macroevolutionary models to phylogenetic trees

Usage

fit.continuous(phy, data, data.names=NULL, lambda=FALSE, kappa=FALSE, delta=FALSE, ou=FALSE, eb=FALSE, bounds=NULL, meserr=NULL)

Arguments

phy

object of type phylo

data

Data vector (one trait) or matrix (multiple traits)

data.names

Tip names for data vector that match tree species; ignored if data includes names

lambda

Estimate lambda?

kappa

Estimate kappa?

delta

Estimate delta?

ou

Estimate alpha for ornstein-uhlenbeck model?

eb

Estimate early burst model?

bounds

Range to constrain estimates; see below for details

meserr

Measurement error for each trait for each tip species; can also be a single vector (if so, error is assumed to be equal for all traits) or a single number (error assumed equal for all traits across all species).

Value

• Returns a matrix of parameter estimates, along with approximate standard errors and 95 Also returns the log-likelihood of the model (lnl).

Details

This function fits various likelihood models for continuous character evolution. The function returns parameter estimates, (approximate) confidence intervals based on the Hessian of the likelihood function, and the likelihood. Likelihood is maximized using the r function nlm. This is a purely univariate function at this point - if multivariate data are sent to the function, it will carry out calculations for each character independently. Possible models are as follows: lambda{ Pagel's lambda; multiplies all internal branches of the tree by lambda, leaving tip branches as their original length.} kappa{Pagel's kappa; raises all branch lengths to the power kappa. As kappa approaches zero, the model becomes speciational.} delta{Pagel's delta; raises all node depths to the power delta. If delta is less than one, evolution in concentrated early in the tree; delta > 1 concentrates evolution towards the tips.} Ornstein-Uhlenbeck{Set by the ou parameter; fits a model of random walk with a central tendency proportional to the parameter alpha. ' Also called the "hansen" model in ouch, although the way the parameters are fit is slightly different here.} Early burst{Set by the eb parameter; fits a model where the rate of evolution increases or decreases exponentially through time, under the model r(t) = ro * exp(r * t), where ro is the inital rate and r is the rate change parameter. The actual parameter estimated, endRate, is the proportion of the initial rate represented by the end rate.} Bounds for the parameters in the likelihood search are sometimes necessary. This is because in some cases, the likelihood surface can have long flat ridges that cause the search to get "stuck." This is particularly common, in my experience, for the OU model. You can set bounds with the "bounds" parameter. For example, to set bounds on alpha, use bounds=list(alpha=c(0.0001, 10)). One can also set multiple bounds at once: bounds=list(alpha=c(0.0001, 10), beta=c(1, 100)).

References

Lambda, kappa, and delta: Pagel, M. 1999. Inferring the historical patterns of biological evolution. Nature 401:877-884. OU: Butler, M.A. and A.A. King, 2004. Phylogenetic comparative analysis: a modeling approach for adaptive evolution. American Naturalist 164:683-695. Early burst: Paper in prep., L. J. Harmon et al.

Examples

data(geospiza)
attach(geospiza)


#---- PRINT RESULTS
fit.continuous(geospiza.tree, geospiza.data)
 
#---- STORE RESULTS 
brownFit <-  fit.continuous(geospiza.tree, geospiza.data)

aic.brown<-numeric(5)
for(i in 1:5) aic.brown[i]<-brownFit[[i]]$aic

#----------------------------------------------------
#   PHYLOGENETIC SIGNAL: FIT LAMBDA
#----------------------------------------------------

lambdaFit<-fit.continuous(geospiza.tree, geospiza.data, lambda=TRUE)

# Compare likelihoods:

d.lambda<-numeric(5)
for(i in 1:5) d.lambda[i]=2*(lambdaFit[[i]]$lnl-brownFit[[i]]$lnl)

# Calculate p values assuming chi-squared distribution with 1 d.f.
p.lambda=pchisq(d.lambda, 1, lower.tail=FALSE)

aic.lambda<-numeric(5)
for(i in 1:5) aic.lambda[i]<-lambdaFit[[i]]$aic

#----------------------------------------------------
#    TIME PROPORTIONALITY: DELTA 
#---------------------------------------------------

deltaFit<-fit.continuous(geospiza.tree, geospiza.data, delta=TRUE)

# Compare likelihoods:

d.delta<-numeric(5)
for(i in 1:5) d.delta[i]=2*(deltaFit[[i]]$lnl-brownFit[[i]]$lnl)

# Calculate p values assuming chi-squared distribution with 1 d.f.
p.delta=pchisq(d.delta, 1, lower.tail=FALSE)

aic.delta<-numeric(5)
for(i in 1:5) aic.delta[i]<-deltaFit[[i]]$aic

#----------------------------------------------------
#   SPECIATIONAL MODEL: KAPPA 
#---------------------------------------------------

kappaFit<-fit.continuous(geospiza.tree, geospiza.data, kappa=TRUE)

# Compare likelihoods:

d.kappa<-numeric(5)
for(i in 1:5) d.kappa[i]=2*(kappaFit[[i]]$lnl-brownFit[[i]]$lnl)

# Calculate p values assuming chi-squared distribution with 1 d.f.
p.kappa=pchisq(d.kappa, 1, lower.tail=FALSE)

aic.kappa<-numeric(5)
for(i in 1:5) aic.kappa[i]<-kappaFit[[i]]$aic

#----------------------------------------------------
#   OU MODEL: ALPHA 
#---------------------------------------------------

ouFit<-fit.continuous(geospiza.tree, geospiza.data, ou=TRUE)

# Compare likelihoods:

d.ou<-numeric(5)
for(i in 1:5) d.ou[i]=2*(ouFit[[i]]$lnl-brownFit[[i]]$lnl)

# Calculate p values assuming chi-squared distribution with 1 d.f.
p.ou=pchisq(d.ou, 1, lower.tail=FALSE)

aic.ou<-numeric(5)
for(i in 1:5) aic.ou[i]<-ouFit[[i]]$aic

#----------------------------------------------------
#   EARLY BURST MODEL: R 
#---------------------------------------------------

ebFit<-fit.continuous(geospiza.tree, geospiza.data, eb=TRUE)

# Compare likelihoods:

d.eb<-numeric(5)
for(i in 1:5) d.eb[i]=2*(ebFit[[i]]$lnl-brownFit[[i]]$lnl)

# Calculate p values assuming chi-squared distribution with 1 d.f.
p.eb=pchisq(d.eb, 1, lower.tail=FALSE)

aic.eb<-numeric(5)
for(i in 1:5) aic.eb[i]<-ebFit[[i]]$aic

#----------------------------------------------------
#   COMPARE ALL MODELS
#---------------------------------------------------

# One way: use likelihood ratio test to compare all models to Brownian model

d.all<-cbind(d.lambda, d.delta, d.kappa, d.ou, d.eb)
p.all<-cbind(p.lambda, p.delta, p.kappa, p.ou, p.eb)

cat("Traittlambdatdeltatkappatouteb
")

for(i in 1:5) {
	cat("Tr", i, "t");
	for(j in 1:5) {
		cat(round(d.all[i,j],2));
		if(p.all[i,j]<0.05) cat("*");
		if(p.all[i,j]<0.01) cat("*");
		if(p.all[i,j]<0.001) cat("*");
		cat("t");
	}
	cat("");
}

# Another way: use AIC

aic.all<-cbind(aic.brown, aic.lambda, aic.delta, aic.kappa, aic.ou, aic.eb)
foo<-function(x) x-x[which(x==min(x))]
daic<-t(apply(aic.all, 1, foo))

rownames(daic)<-colnames(geospiza.data)
colnames(daic)<-c("Brownian", "Lambda", "Delta", "Kappa", "OU", "EB")

cat("Table of delta-aic values; zero - best model
")
print(daic, digits=2)