SharpBound: Inference of the Compliance Group ITT Effects Using...

In qkrcks0218/CRTBound: Analysis of Cluster Randomized Trials for Intent-to-Treat Effects and Network Effects

Description

Analyze the data (obtained from Data.Reform or Simulation.Reform functions) to infer the compliance group ITT effect using classifier-based approach; see Section 4 of Park and Kang (2021) for details.

Usage

SharpBound(Data, paraC=NULL, method="Linear", CIcalc=FALSE, SSsize=1000, level=0.95, seed=1, Input.Type="Data")

Arguments

Data	A list obtained from Data.Reform and Simulation.Reform functions.
paraC	(Optional) Indices of the pre-treatment covariate matrix Data$X that are used in inference of the compliance group ITT effect. The default option is to use all pre-treatment covariates.
method	(Optional) The type of the transformation function. "Linear" and "Logistic" are available where "Linear" and "Logistic" use the linear and logistic transformations. The default option is "Linear"; see Section 4 of Park and Kang (2021) for details.
CIcalc	(Optional) A logical value whether the bootstrap-based confidence intervals are obtained. The dafault option is TRUE.
SSsize	(Optional) The total number of the bootstrap resamples used for the construction of the confidence intervals. The dafault option is 1000.
level	(Optional) The confidence level of the confidence interval. The default option is 0.95.
seed	(Optional) A random seed to generate the bootstrap resamples used for the construction of the confidence intervals.
Input.Type	(Optional) A variable whether Data is from Data.Reform (Input.Type="Data") or Simulation.Reform (Input.Type="Sim") function. The default The default option is "Data".

Value

Bound	Bound estimates of the compliance group ITT effects. When Input.Type="Sim", the true bounds are available.
Violation	Summary whether the linear program was feasible.
Resample.Bound	Bootstrap bound estimates of the compliance group ITT effects.
BootCIBound	Confidence intervals of the bounds of the compliance group ITT effects.

References

Chan Park & Hyunseung Kang (2021+) Assumption-Lean Analysis of Cluster Randomized Trials in Infectious Diseases for Intent-to-Treat Effects and Network Effects, Journal of the American Statistical Association [Link]

Examples

#################################
# Generate Population
#################################

J <- 100 ; m <- 60
nc <- sample(c(3,4,5),J,replace=T)
N <- sum(nc)
C <- rep(1:J,nc)
n <- rep(nc,nc)
X1 <- rnorm(N)
X2 <- rbinom(N,1,0.3)
X3 <- rep(rnorm(J),nc)
A0 <- rbinom(N,1,logistic(-2+2*X3))
A1 <- apply(cbind(A0,rbinom(N,1,logistic(-2+3*X1+3*X2+2*X3))),1,max)
OtherA1 <- (rep(aggregate(A1~C,FUN="sum")[,2],nc)-A1)/(n-1)
NT <- which(A1==0 & A0==0)
AT <- which(A1==1 & A0==1)
CO <- which(A1==1 & A0==0)
Y0 <- Y1 <- rep(0,N)
for(jj in NT){
    Y0[jj] <- rbinom(1,1,logistic(-2+2*OtherA1[jj]))
    Y1[jj] <- max(Y0[jj],rbinom(1,1,logistic(-2+2*OtherA1[jj])))
}
for(jj in AT){
    Y0[jj] <- rbinom(1,1,logistic(2+X1[jj]+X2[jj]))
    Y1[jj] <- max(Y0[jj],rbinom(1,1,logistic(2+X1[jj]+X2[jj])))
}
for(jj in CO){
    Y0[jj] <- rbinom(1,1,logistic(-2+2*OtherA1[jj]))
    Y1[jj] <- max(Y0[jj],rbinom(1,1,logistic(2+X1[jj]+X2[jj])))
}
X <- cbind(1,X1,X2,X3,n)

Zc <- rep(0,J)
Zc[sort(sample(J,m))] <- 1
Z <- rep(Zc,nc)
A <- Z*A1 + (1-Z)*A0
Y <- Z*Y1 + (1-Z)*Y0

#################################
# Reform the Data
#################################

Reformed.Data <- Data.Reform(Y,Z,A,C,X,seed=1)
Simulated.Data <- Simulation.Reform(Y0,Y1,Z,A0,A1,C,X,seed=1)

#################################
# Bounds
#################################

Bound1 <- SharpBound(Reformed.Data,paraC=1:5,method="Logistic",CIcalc=TRUE,SSsize=100,level=0.95,seed=1,Input.Type="Data")
Bound2 <- LongHudgens(Reformed.Data,paraC=c(3),CIcalc=TRUE,SSsize=100,level=0.95,seed=1,Input.Type="Data")
Bound3 <- Bound.Intersect(Bound1,Bound2,level=0.95,Input.Type="Data")

Bound1.Sim <- SharpBound(Simulated.Data,paraC=1:5,method="Logistic",CIcalc=TRUE,SSsize=100,level=0.95,seed=1,Input.Type="Sim")
Bound2.Sim <- LongHudgens(Simulated.Data,paraC=c(3),CIcalc=TRUE,SSsize=100,level=0.95,seed=1,Input.Type="Sim")
Bound3.Sim <- Bound.Intersect(Bound1.Sim,Bound2.Sim,level=0.95,Input.Type="Sim")

qkrcks0218/CRTBound documentation built on Dec. 22, 2021, 10:56 a.m.