rptR: Repeatability estimation for Gaussian and non-Gaussian data

A collection of functions for estimating the repeatability of measurements with a single grouping factor. Most functions use mixed-effects models for extracting variance components (between-group and within-group variances), but alternatives are available for Gaussian data.

Installation

rpt is the most general function that is suitable for all types of models. More specialized functions are listed in the rpt help file. All functions require at least a vector of measurements (y) and vector of group identities (groups). These two vectors have to be of equal length. For details see:
?rpt

rpt.aov, rpt.remlLMM and rpt.mcmcLMM are suitable for Gaussian (normally distributed) data.
rpt.binomGLMM.add and rpt.binomGLMM.multi are suitable for binary and proportion data.
rpt.poisGLMM.add and rpt.poisGLMM.multi are suitable for Poisson (count) data.

All functions can be call via the function rpt. Hence, the following calls are equivalent:

`rpt(y, groups, datatype="Gaussian", method="ANOVA")`	`rpt.aov(y, groups)`
`rpt(y, groups, datatype="Gaussian", method="REML")`	`rpt.remlLMM(y, groups)`
`rpt(y, groups, datatype="Gaussian", method="MCMC")`	`rpt.mcmcLMM(y, groups)`
`rpt(y, groups, datatype="binomial", method="GLMM.multi")`	`rpt.binomGLMM.multi(y, groups)`
`rpt(y, groups, datatype="binomial", method="GLMM.add")`	`rpt.binomGLMM.add(y, groups)`
`rpt(y, groups, datatype="count", method="GLMM.multi")`	`rpt.poisGLMM.multi(y, groups)`
`rpt(y, groups, datatype="count", method="GLMM.add")`	`rpt.poisGLMM.add(y, groups)`

Results are returned as a rpt object. The rpt object is a list that can be nicely displayed using print (the only method that is currently defined for rpt objects):
print(rpt.obj)
Not all elements of the object are displayed by print. For example, vectors with results for all bootstrapping iterations and permutations are contained in the rpt object, but not displayed by print. For details help files type:
str(rpt.obj)

Bootstrapping (for SE and CI) and randomization (for P) procedures are very slow, since they involve a large number of model fits (each iteration requires a new model fit). Hence, you might want to reduce the number of bootstrap and/or randomizations by changing the nboot and npermut arguments, at least for data exploration. nboot = 0 and npermut = 0 suppress bootstrapping and randomization, respectively (though the point estimate will be returned as the single permutation).

We have recently included a new function rpt.remlLMM.adj that allows the estimation of adjusted repeatabilites. The function requires a formula argument, a grname argument and a data argument. This allows much greater flexibility for estimating adjusted repeatabilites. The formula argument uses the same syntax as lmer and allows the specification of fixed and random effects. With only one random effect, the function will return the same repeatability as the function rpt.remlLMM. The argument grname requires a character string or a vector of character strings that match one or more of the random effects specified in the formula. The option to specify more than one random effect allows estimating multiple variance components and their standard errors simultaneously. The data argument requires a dataframe that contains the columns spefied in the fomula. The function rpt.remlLMM.adj can also be accessed via the wrapper function rpt.adj (again using the formula, grname, data specification).

We are planning to implement the estimation of adjusted repeatabilities also for non-Gaussian models and for Gaussian models using MCMC sampling. The repeatabilities for non-Gaussian models will depend on the intercept and hence on the values of the covariate. Covariates should therefore best be centered on a mean of zero. Perspectively, we are planning to integrate the rpt.adj family of functions into rpt, since this allows greater flexibility for estiamting standard and adjusted repeatabilites.