delta.many2: Compare many (multilevel) Conger kappa coefficients using the...
In svanbelle/multiagree: Comparaison of dependent kappa coefficients

delta.many2

R Documentation

Compare many (multilevel) Conger kappa coefficients using the delta method

Description

This function performs Hotelling's T square test using a variance-covariance matrix based on the delta method to compare dependent Conger kappa coefficients

Usage

delta.many2(data, cluster_id, nrat = c(6, 6), multilevel = T,
  a.level = 0.05)

Arguments

`data`	a N x sum(Rg) matrix representing the classification of the N items by G groups of Rg observers (g=1,...,G).
`cluster_id`	a vector of lenght N with the identification number of the K clusters
`nrat`	a vector of lenght G indicating the number of observers in the G groups of observers
`multilevel`	a binary indicator equal to TRUE in the presence of multilevel data and FALSE otherwise
`a.level`	the significance level

Details

This function compare several (multilevel) dependent Conger kappa coefficients. It uses Hotelling's T square test with the variance-covariance matrix obtained by the delta method. If only a single Conger kappa coefficient is computed, the kappa coefficient and its standard error are returned.

Value

$kappa a G x 2 matrix with the G kappa coefficients to be compared in the first column and their corresponding standard error in the second column

$T_test a vector of length 2 with the value of Hotelling's T square test as first element and the p-value as second element

$confidence confidence intervals for the pairwise comparisons of kappa coefficients

$var the G x G correlation matrix of the kappa coefficients

Author(s)

Sophie Vanbelle sophie.vanbelle@maastrichtuniversity.nl

References

Vanbelle S. (2017) Comparing dependent agreement coefficients obtained on multilevel data. Biometrical Journal, 59 (5):1016-1034

Vanbelle S. (submitted) On the asymptotic variability of (multilevel) multirater kappa coefficients

Examples

 
#dataset (multilevel) (Vanbelle, 2008)

data(depression)
attach(depression)
delta.pair(data=cbind(diag,BDI,diag,GHQ),cluster_id=ID,weight='unweighted',multilevel=FALSE)


#dataset (multilevel) (Vanbelle, submitted)
data(CRACKLES)
attach(CRACKLES)

AGREEMENT<-matrix(NA,ncol=21,nrow=4)

for (i in 1:7){
 AGREEMENT[1,((i-1)*3+1)]<-mean((rowSums(CRACKLES[UP==1,((i-1)*4+1):(i*4)])*(rowSums((CRACKLES[UP==1,((i-1)*4+1):(i*4)]))-1)+(4-rowSums(CRACKLES[UP==1,((i-1)*4+1):(i*4)]))*((4-rowSums((CRACKLES[UP==1,((i-1)*4+1):(i*4)])))-1))/12)
 AGREEMENT[1,((i-1)*3+2):(i*3)]<-delta.many2(cluster_id=patient[UP==1],data=CRACKLES[UP==1,((i-1)*4+1):(i*4)],nrat=c(4))$kappa
 AGREEMENT[2,((i-1)*3+1)]<-mean((rowSums(CRACKLES[LO==1,((i-1)*4+1):(i*4)])*(rowSums((CRACKLES[LO==1,((i-1)*4+1):(i*4)]))-1)+(4-rowSums(CRACKLES[LO==1,((i-1)*4+1):(i*4)]))*((4-rowSums((CRACKLES[LO==1,((i-1)*4+1):(i*4)])))-1))/12)
 AGREEMENT[2,((i-1)*3+2):(i*3)]<-delta.many2(cluster_id=patient[LO==1],data=CRACKLES[LO==1,((i-1)*4+1):(i*4)],nrat=c(4))$kappa
 AGREEMENT[3,((i-1)*3+1)]<-mean((rowSums(CRACKLES[UP!=1 & LO!=1,((i-1)*4+1):(i*4)])*(rowSums((CRACKLES[UP!=1 & LO!=1,((i-1)*4+1):(i*4)]))-1)+(4-rowSums(CRACKLES[UP!=1 & LO!=1,((i-1)*4+1):(i*4)]))*((4-rowSums((CRACKLES[UP!=1 & LO!=1,((i-1)*4+1):(i*4)])))-1))/12)
 AGREEMENT[3,((i-1)*3+2):(i*3)]<-delta.many2(cluster_id=patient[UP!=1 & LO!=1],data=CRACKLES[UP!=1 & LO!=1,((i-1)*4+1):(i*4)],nrat=c(4))$kappa
 AGREEMENT[4,((i-1)*3+1)]<-mean((rowSums(CRACKLES[,((i-1)*4+1):(i*4)])*(rowSums((CRACKLES[,((i-1)*4+1):(i*4)]))-1)+(4-rowSums(CRACKLES[,((i-1)*4+1):(i*4)]))*((4-rowSums((CRACKLES[,((i-1)*4+1):(i*4)])))-1))/12)
 AGREEMENT[4,((i-1)*3+2):(i*3)]<-delta.many2(cluster_id=patient,data=CRACKLES[,((i-1)*4+1):(i*4)],nrat=c(4))$kappa
}

AGREEMENT2<-matrix(NA,ncol=14,nrow=4)

for (i in 1:7)
{
 AGREEMENT2[,((i-1)*2+2)]<-paste(paste(paste(round(as.numeric(AGREEMENT[,((i-1)*3+2)]),2),' ('),round(as.numeric(AGREEMENT[,((i-1)*3+3)]),2),')'))
 AGREEMENT2[,((i-1)*2+1)]<-round(as.numeric(AGREEMENT[,((i-1)*3+1)]),2)
}
AGREEMENT2 #(table 3)

svanbelle/multiagree documentation built on Feb. 9, 2023, 2:37 p.m.