rpt.binomGLMM.multi: GLMM-based Repeatability Using PQL Estimation for Binomial...

In rptR: Repeatability for Gaussian and non-Gaussian data

Description

Calculates repeatability from a generalised linear mixed-effects models fitted by PQL (penalized-quasi likelihood) estimation for binary and proportion data.

Usage

	rpt.binomGLMM.multi(y, groups, link=c("logit", "probit"), CI=0.95, nboot=1000, npermut=1000)

Arguments

y	Vector of a response values (for binary data) or a two-column matrix, array or data.frame with colums m, n-m, where m is the number of successes and n the number of trials.
groups	Vector of group identitites.
link	Link function, log and sqrt are allowed, defaults to log.
CI	Width of the confidence interval (defaults to 0.95).
nboot	Number of parametric bootstraps for interval estimation (defaults to 1000). Larger numbers of permutations give a better asymtotic CI, but may be very time-consuming.
npermut	Number of permutations for significance testing (defaults to 1000). Larger numbers of permutations give better asymtotic P values, but may be very time-consuming.

Details

Models are fitted using the glmmPQL function in MASS with the quasibinomial family (proportion data) or the binomial family (binary data).

Value

Returns an object of class rpt that is a a list with the following elements:

datatype	Type of response (here: "binomial").
method	Method used to calculate repeatability (here: "PQL").
link	Link function used (here: "logit" or "probit").
CI	Width of the confidence interval.
R.link	Point estimate for repeatability on the link scale.
se.link	Standard error (se) for repeatability on the link scale, i.e. the standard deviation of the parametric bootstrap runs. Note that the distribution might not be symmetrical, in which case the se is less informative.
CI.link	Confidence interval for repeatability on the link scale based on parametric-boostrapping of R.
P.link	Approximate P value from a significance test for the link scale repeatability based on randomisation.
R.org	Point estimate for repeatability R on the original scale.
se.org	Standard error (se) for repeatability on the original scale, i.e. the standard deviation of the parametric bootstrap runs. Note that the distribution might not be symmetrical, in which case se is less informative.
CI.org	Confidence interval for repeatability on the link scale based on parametric-boostrapping of R.
P.org	Approximate P value from a a significance test for the original scale repeatability based on randomisation.
omega	Multiplicative overdispersion parameter.
R.boot	Named list of parametric bootstap samples for R. R.link gives the samples for the link scale repeatability, R.org gives the samples for the original scale repeatability.
R.permut	Named list of permutation samples for R. R.link gives the samples for the link scale repeatability, R.org gives the samples for the original scale repeatability.

Note

Confidence intervals and standard errors are inappropriate at high repeatabilities (omega < 1), because parametric bootstrapping allows only omega greater than or equal to 1.

Author(s)

Holger Schielzeth ([e-mail]) & Shinichi Nakagawa ([e-mail])

References

Browne, W. J., Subramanian, S. V., et al. (2005). Variance partitioning in multilevel logistic models that exhibit overdispersion. Journal of the Royal Statistical Society A 168: 599-613.

Goldstein, H., Browne, W., et al. (2002). Partitioning variation in multilevel models Understanding Statistics 1: 223-231.

Nakagawa, S. and Schielzeth, H. (2011) Repeatability for Gaussian and non-Gaussian data: a practical guide for biologists. Biological Reviews 85: 935-956.

See Also

rpt.binomGLMM.add, rpt, print.rpt

Examples

## Not run: 
# repeatability estimations for egg dumping (binary data)
data(BroodParasitism)
attach(BroodParasitism)
(rpt.Host <- rpt.binomGLMM.multi(HostYN[OwnClutchesBothSeasons==1], FemaleID[OwnClutchesBothSeasons==1],
	nboot=10, npermut=10)) # low number of nboot and npermut to speed up error checking
(rpt.BroodPar <- rpt.binomGLMM.multi(cbpYN, FemaleID, nboot=10, npermut=10))  
	# low number of nboot and npermut to speed up error checking
detach(BroodParasitism)

# repeatability estimations for egg dumping (proportion data)
data(BroodParasitism)
attach(BroodParasitism)
ParasitisedOR <- cbind(HostClutches, OwnClutches-HostClutches)   
(rpt.Host <- rpt.binomGLMM.multi(ParasitisedOR[OwnClutchesBothSeasons==1,], 
	FemaleID[OwnClutchesBothSeasons==1], nboot=10, npermut=10))  # reduced number of npermut iterations
ParasitismOR <- cbind(cbpEggs, nEggs-cbpEggs)  
zz = which(ParasitismOR[,1]==0 & ParasitismOR[,2]==0) # some rows have entries 0,0 and need to be removed
(rpt.BroodPar <- rpt.binomGLMM.multi(ParasitismOR[-zz,], FemaleID[-zz], nboot=10, npermut=10))   
	# reduced number of npermut iterations
detach(BroodParasitism)

## End(Not run)

rptR documentation built on May 2, 2019, 5:01 p.m.