trendtesting: Performs multiple contrast test
In paulmanser/Trend-test-package: Analyzing Dose Finding Experiments

Description Usage Arguments Value Examples

View source: R/trendtest.R

This function performs a multiple contrast test. Optimal contrasts are derived based on the trend you choose.

trendtesting(muHat, S, df = NULL, doses = NULL,
  alternative = c("one.sided", "two.sided"), trend = c("CPL", "MCP Mod",
  "Aug Williams", "Williams", "Dunnett", "Linear", "Arbitrary"), model = NULL,
  muMat = NULL)

`muHat`	a numeric vector of expected values in all doses groups including placebo.
`S`	convariance matrix.
`df`	specify degree freedom for multivariate t. If this argument is missing, df=0 corresponding multivariate normal.
`doses`	a numeric vector of specified doses. When this argument is missing, doses are the object in the specified model if model is given for MCP Mod trend test, or doses are 0 to k by 1 for all other trend test.
`alternative`	a single character string, specifying the alternative for the multiple contrast trend test. If this argument is missing, alternative="one.sided".
`trend`	a single character string, determining the trend for the multiple contrast trend test, one of "CPL", "MCP Mod","Aug Williams", "Williams", "Dunnett", "Linear","Arbitrary". Note that when "MCP_Mod" is specified for trend, model needs to be specified. When "Arbitrary" is specified, muMat needs to be specified.
`model`	specify models if you choose trend to be MCP Mod trend test.
`muMat`	a numeric vector or matrix specified for mu when trend is specified to be "Arbitrary".

doses, optimal contrast, correlation contrast, test statisticis, p values

load(DoseFinding)
transformDoses<-function(doses,mindose,mult) {
log(1+(mult/mindose)*doses)
}

itransDoses<-function(doses,mindose,mult) {
 (exp(doses)-1)*mindose/mult
}

origDose<-c(0,2.5,5,10,20,50,100,200)

transDose<-transformDoses(doses=origDose,mindose=2.5,mult=5)

dat<-data.frame(dose=as.factor(origDose),n=c(133,32,44,63,63,65,59,58),
               respn=c(13,4,5,16,12,14,14,21))
dat$resprate<-dat$respn/dat$n

dat$txeff<-dat$resprate-dat$resprate[1]

logfit<-glm(cbind(respn,n-respn)~dose+0,family=binomial,data=dat)
muHat<-coef(logfit)
S<-vcov(logfit)

model<-Mods(
 logistic=c(log(1+2*10),0.5),
 sigEmax=c(log(1+2*10),4),
 emax=c(log(1+2*0.1),log(1+2*1.0)),
 linear=NULL,
 exponential=log(1+2*1.0),
 quadratic= -1/log(1+2*550),
 doses=transDose)

test_MCP<-trendtesting(muHat=muHat, S=S, doses=transDose, alternative="one.sided", trend="MCP Mod", model=model)

#with arbitrary trend
#arbitrary mu
mu<-mu.linear(7, doses=seq(0,7,1))
test_MCP<-trendtesting(muHat=muHat, S=S, doses=transDose, alternative="one.sided", trend="Arbitrary", muMat=mu)