pn.mod.compare: Compare All Possible Positive-Negative Richards $nlslist$ Models

Description

This function performs model selection for nlsList models fitted using

SSposnegRichards.

Usage

pn.mod.compare(x,

y,

grp,

pn.options,

forcemod = 0,

existing = FALSE,

penaliz = "1/sqrt(n)",

taper.ends = 0.45,

mod.subset = c(NA),

Envir = .GlobalEnv,

...)

Value

A list object with two components: $'Model rank table' contains the

statistics from extraF ranked by the modified residual standard error,

and $'P values from pairwise extraF comparison' is a matrix of P values from

extraF for legitimate comparisons (nested and successfully fitted models).

The naming convention for models is a concatenation of 'richardsR', the modno and '.lis'

which is shortened in the matrix output, where the number of parameters has been

pasted in parentheses to allow users to easily distinguish the more general model from

the more reduced model

(see extraF and SSposnegRichards).

For extra flexibility, mod.subset can specify a vector of modno values (a number of different models) that

can be fitted in nlsList and then evaluated by model selection. This prevents fitting of unwanted models or

attempts to fit models that are known to fail. If the nlsList model already exists it will not be refitted

and thus existing models can be included in the ranking table without adding noticeably to processing time.

Arguments

x: a numeric vector of the primary predictor
y: a numeric vector of the response variable
grp: a factor of same length as x and y that distinguishes groups within
pn.options: required character string specifying name of
forcemod: optional numeric value to constrain model selection (see Details)
existing: optional logical value specifying whether some of the relevant models
penaliz: optional character value to determine how models are ranked (see Details)
taper.ends: numeric representing the proportion of the range of the x variable for which data are extended at
mod.subset: optional vector containing modno of models that the user desires to be estimated. If not NA, only
Envir: a valid R environment to find pn.options in and export any output to, by default this is the global
...: additional optional arguments to be passed to nlsList

Author

Stephen Oswald <steve.oswald@psu.edu>

Details

First, whether parameter M should be fixed

(see SSposnegRichards) is determined by fitting models 12 and 20 and comparing

their perfomance using extraF. Note that model 20 is identical to model 32.

If model 12 provides superior performance (variable values of M) then 16 models that estimate M

are run

(models 1 through 16), otherwise the models with fixed M are fitted (models 21 through 36).

Fitting these nlsList models can be time-consuming (2-4 hours using the dataset

posneg.data that encompasses 100 individuals) and if several of the relevant

models are already fitted the option existing=TRUE can be used to avoid refitting models that

already exist globally (note that a model object in which no grouping levels were successfully

parameterized will be refitted, as will objects that are not of class nlsList).

Specifying forcemod=3 will force model selection to only consider fixed M models and setting

forcemod=4 will force model selection to consider models with varying values of M only.

If fitting both models

12 and 20 fails, fixed M models will be used by default.

taper.ends can be used to speed up optimization as it extends the dataset at maximum and minimum extremes

of x by repeatedly pasting the y values at these extremes for a specified proportion of the range of x.

taper.ends is a numeric value representing the proportion of the range of x values are extended for and

defaults to 0.45 (45

tend towards a zero slope this is a suitable values. If tapered ends are not desirable then choose taper.ends = 0.

Models are ranked by modified pooled residual square error. By default residual standard error

is divided by the square root of sample size. This exponentially penalizes models for which very few

grouping levels (individuals) are successfully parameterized (the few individuals that are

parameterized in these models are fit unsuprisingly well) using a function based on the relationship

between standard error and sample size. However, different users may have different preferences

and these can be specified in the argument penaliz (which residual

standard error is multiplied by). This argument must be a character value

that contains the character n (sample size) and must be a valid right hand side (RHS) of a formula:

e.g. 1*(n), (n)^2. It cannot contain more than one n but could be a custom function, e.g. FUN(n).

Examples

Run this code



# \donttest{


#these examples will take a long while to run as they have to complete the 32 model comparison


#run model selection for posneg.data object (only first 3 group levels for example's sake)


   try(rm(myoptions),silent = TRUE)


   subdata <- subset(posneg.data, as.numeric(row.names (posneg.data) ) < 40)


   modseltable <- pn.mod.compare(subdata$age, subdata$mass,


      subdata$id, existing = FALSE, pn.options = "myoptions")


   modseltable


   


#fit nlsList model initially and then run model selection


#for posneg.data object when at least one model is already fit


#(only first 3 group levels for example's sake)


    richardsR22.lis <- nlsList(mass ~ SSposnegRichards(age, Asym = Asym, K = K,


      Infl = Infl, RAsym = RAsym, Rk = Rk, Ri = Ri , modno = 22, pn.options = "myoptions")


                        ,data = subdata)


   modseltable <- pn.mod.compare(subdata$age, subdata$mass,


      subdata$id, forcemod = 3, existing = TRUE, pn.options = "myoptions")


   modseltable





#run model selection ranked by residual standard error*(1/sample size)


    modseltable <- pn.mod.compare(subdata$age, subdata$mass,


      subdata$id, penaliz='1*(1/n)', existing = TRUE, pn.options = "myoptions")


    modseltable


    # }

Run the code above in your browser using DataLab

pn.mod.compare: Compare All Possible Positive-Negative Richards \(nlslist\) Models