ComLikMulCom: Multiple Comparisons using Composite Likelihood

In m-azad/CLMC: Multiple comparisons using composite likelihood for clustered data

Description

Multiple hypothesis testing H_0 :Cβ = 0, based on composite likelihood inference for clustered data, with common covariance structure across the clusters. The critical value is based on multivariate normal quantile, using the R-package multcomp by Hothorn, et.al.

Usage

ComLikMulCom(data, yno, idno, y.type = c("normal", "probit", "quadratic exponential")[1], type = c("many_one", "all_pair")[1], f = 1)

Arguments

data	data set in long format containing response, covariates and one variable that identifies the cluster
yno	the response column number
idno	column number of the variable that identify the clusters
y.type	the model of the response variable, it could be normal, probit, or quadratic exponential
type	type of the test, it could be many-to-one or all-pair comparisons
f	the baseline variable for many to one type

Details

The current functions handle the clustered normal and probit data with equal size. For quadratic exponential data, clusters can have different sizes.

Value

m	number of individuals in each cluster, if equal
n	number of clusters
p	number of covariates
Cov_beta	covariance matrix of the covariates
T_stat	the test statistic
normal_quantile	the simultaneous normal quantile
MNQ	results of individual null hypotheses based on composite likelihood
Bon	results of individual null hypotheses based on Bonferroni approach
Sidak	results of individual null hypotheses based on Dunn-Sidak approach
Scheffe	results of individual null hypotheses based on Scheffe method

Author(s)

Mahdis Azadbakhsh, Xin Gao, Hanna Jankowski.

References

Azadbakhsh,M., Gao,X., Jankowski,H. (2016) Multiple Comparisons Using Composite Likelihood in Clustered Data. Submitted.

Hothorn, T., Bretz, F., Westfall, P., Heiberger, R. M., Schutzenmeister, A.(2010) multcomp: Simultaneous Inference for General Linear Hypotheses, R package version 1.1-7, http://CRAN.R-project.org/package=multcomp

Examples

### artificial data set (normal)
m <- 3
n <- 20
p <- 5

v <- sample(seq(-4,4,0.05),360,replace= TRUE)
cx <- matrix(v, nrow=(n*m))

ind <- rep(0,n*m)
for(i in 1: n) {ind[(m*(i-1)+1):(m*i)]<-   i}

cxx <- cbind(cx,ind)
colnames(cxx) <- c("res", "c1","c2","c3","c4", "c5", "ind")

idno <- which( colnames(cxx)=="ind" )
yno <- which( colnames(cxx)=="res" )

ComLikMulCom(cxx, yno , idno , y.type ="normal" ,  type = "many_one", f=3)
ComLikMulCom(cxx, yno , idno , y.type ="normal" ,  type = "all_pair")



### probit


v <- sample(seq(-4,4,0.05),300,replace= TRUE)
y <- sample(c(0,1),60,replace= TRUE)
cx <- matrix(v, nrow=60)

colnames(cx) <- c("c1","c2","c3","c4", "c5")


id <- rep(0,n*m)
for(i in 1: n) {id[(m*(i-1)+1):(m*i)]<-   i}


cxx <- as.data.frame(cbind(y,cx,id))
idno <- which( colnames(cxx)=="id" )
yno <- which( colnames(cxx)=="y" )


ComLikMulCom(cxx, yno , idno , y.type ="probit" ,  type = "all_pair")



### Normal
m <- 3
n <- 40
p <- 5
require(MASS)
library(MASS)
library(mvtnorm)
beta <-   c(0,2,0,0,1)
sigma  <- matrix(c(0.8, 0.4, 0.4, 0.4, 0.8, 0.4, 0.4, 0.4, 0.8), nrow=3)

desx      <-    array(dim=c(m,p,n))
y         <-    matrix(0,nrow=m,ncol=n)
mu        <-    matrix(0,nrow=m,ncol=n)

for (k in 1:n){
  for(j in 1:p){
    desx[,j,k] <-   rnorm(m,1,3.5)
  }
  mu[,k]      <-   tcrossprod(desx[,,k], t(beta))
  y[,k]        <-   mvrnorm(1,mu[,k], sigma, tol = 1e-6, empirical = FALSE)
}
sss    <-    array(dim=c(m,(p+1),n))


for(i in 1:n){
  ind <- rep(i,m)
  sss[,,i]  <- cbind(ind,desx[,,i])
}

yyy         <-   rep(0,120)

for( i in 1:40){
  yyy[((i-1)*3+1):(i*3)]   <-   y[,i]
}

xxx       <-   matrix(0,nrow=120, ncol=6)

for( i in 1:40){
  xxx[((i-1)*3+1):(i*3),] 	<- 	sss[,,i]
}
dat <- cbind(yyy,xxx)
colnames(dat) <- c("response","id", "c1","c2","c3","c4", "c5")

idno <- which( colnames(dat)=="id" )
yno <- which( colnames(dat)=="response" )
ComLikMulCom(dat, yno , idno , y.type ="normal" ,  type = "many_one", f=3)
ComLikMulCom(dat, yno , idno , y.type ="normal" ,  type = "all_pair")

m-azad/CLMC documentation built on May 21, 2019, 9:17 a.m.