saemix (version 3.4)

transformCatCov: Transform categorical covariates

Description

This function is used to automatically map categorical covariates to dummy binary covariates. It can also be used to regroup categorical covariates before the analysis, for example mapping a covariate initially in 5 categories to only 3 categories. For binary covariates, the covariate will simply be transformed to 0/1 with the reference category being 0.

Usage

transformCatCov(
  object,
  covariate,
  oldCat = NULL,
  newCat = NULL,
  newCatName = NULL,
  verbose = FALSE
)

Value

an object of class "SaemixData"

Arguments

object

saemixData object

covariate

name of the covariate

oldCat

a vector giving the initial categories of the covariate (if NULL, defaults to the unique values of the covariate)

newCat

a vector of consecutive integers 1:N, giving the N categories to which the initial values of the covariate should be mapped, with 1 denoting the reference class. If the resulting covariate is binary, it will be stored as 0/1. If it has more than 2 categories, N dummy covariates 0/1 will be created. If not given, the initial categories of the covariate will be mapped to 1:N and transformed into N dummy covariates 0/1.

newCatName

the name of the new group. If NULL (default), binary covariates will be renamed as covariate.mod, while for categorical covariates, the new covariates will be named covariate.ref, covariate.G2, covariate.G3, etc...

verbose

a boolean, prints messages during the execution of the function if TRUE. Defaults to FALSE.

Details

For a binary covariate, a dummy covariate with values 0 for the reference class and 1 for the other category will be created and will replace the original dataset in the covariate For covariates with 3 categories or more (categorical covariates), dummy variables with values 0/1 will be created for the reference class and for each contrast with the reference class (category 2 versus reference, category 3 versus reference, etc...) If these covariates have units, the dummy covariates will have the same unit (which may not be appropriate and can then be changed by the user)

Examples

data(cow.saemix)
saemix.data<-saemixData(name.data=cow.saemix,header=TRUE,name.group=c("cow"),
                   name.predictors=c("time"),name.response=c("weight"),
                   name.covariates=c("birthyear","twin","birthrank"),
                   units=list(x="days",y="kg",covariates=c("yr","-","-")))
unique(saemix.data@data$birthrank) # 5 categories, 3 4 5 6 7
# create 3 dummy variables regrouping 3 (reference), 4 and 5, and 6 and 7
cowt <- transformCatCov(saemix.data, covariate=birthrank, newCat=c(1,2,2,3,3), verbose=TRUE) 
head(saemix.data@data) # the original covariate is birthrank
head(cowt@data) 
# the new covariates are birthrank.ref (initially 3), birthrank.G2 (regrouping 4 and 5) and 
# birthrank.G3 (6 and 7)
# only birthrank.G2 and birthrank.G3 are included in the name.covariates slot of the object cowt
cowt <- transformCatCov(cowt, covariate=birthrank, newCat=c(1,2,2,3,3), 
   newCatName=c("ref","preg4-5","6-7"), verbose=TRUE)
head(cowt@data) 
# new names can be assigned to the dichotomised covariates through the newCatName argument
# the new covariates are now "ref", preg4-5" (regrouping 4 and 5) and "6-7" (6 and 7)

# Changing the reference for a binary variable
cowt<-transformCatCov(cowt, covariate=twin, newCat=c(2,1), newCatName="Singleton", verbose=TRUE)