Some simulations

library(gMCP)
library(flip)
library(mvtnorm)
source('../R/functions.R')
## set up correlation in terms of parallel group sampling model
## covariance matrix correlation of 1/3 between endpoints
S <- (kronecker(matrix(1/3,2,2),diag(3))+diag(2/3,6))
## correlation of z-scores
C <- kronecker(diag(2),cbind(-1,diag(2)))/sqrt(2)
R <- C %*% S %*% t(C)
## effect with only one treatment



## generate data from
## randomization model
S <- diag(2/3,2)+1/3

n <- 10
Y <- rmvnorm(3*n,sigma=S)
colnames(Y) <- c('E1','E2')

## treatment labels
X <- rep(0:2,each=10)
## treatment 1 is effective in endpoint 1
Y[X==1,'E1'] <- Y[X==1,'E1']+1


## simple Succesive test
G <- simpleSuccessiveI()

## z-scores & pvalues
z_scores <- pw_zstat(Y,X,control=0)
p_values <- as.numeric(pnorm(z_scores,lower=FALSE))


## graph based tests
gMCP(G,p_values)
gMCP(G,p_values,test='Simes')
gMCP(G,p_values,test='parametric',correlation = R)

test <- function(Y){
    as.numeric(pw_zstat(Y,X))
}

flip(Y,X,statTest=test)


floatofmath/resamplingMCP documentation built on May 16, 2019, 1:21 p.m.