StratifiedBalancing (version 0.2.0)

stratadisc: Stratadisc()

Description

This is the primary stratification function for discrete outcome variables.It locates naturally occurring strata in the data, weights them and returns a common odds ratio for the treatment variable which is adjusted for both confounding and interaction terms.Returns a matrix containing naturally occurring strata.

Usage

stratadisc(Treatment,Outcome,Matrix)

Arguments

Treatment

Column number of variable to be used as treatment.

Outcome

Column number of variable to be used as outcome.

Matrix

Name of matrix or data.frame where data is stored.

Details

This is the primary stratification function for discrete outcome variables.It locates naturally occurring strata in the data, weights them and returns a common odds ratio for the treatment variable which is adjusted for both confounding and interaction terms.

Examples

# NOT RUN {
## We will first begin by simulating data in 4 covariates and a discrete outcome, with
## significant interaction terms and correlations amongst covariates (to simulate a
## non-randomized experiment with a strongly non-linear underlying model).
## First, we will create a matrix with the input variables
m=matrix(nrow=1000,ncol=5)
for ( i in 1:ncol(m)){
m[,i]=rbinom(1000,1,0.5)
}
## Next, we will create correlations amongst covariates 2,1 and 3
for( i in 1:nrow(m)){
if(m[i,3] == 1)m[i,2]=rbinom(1,1,0.8)
if(m[i,3] == 0)m[i,2]=rbinom(1,1,0.2)
if(m[i,3] == 1)m[i,1]=rbinom(1,1,0.8)
if(m[i,3] == 0)m[i,1]=rbinom(1,1,0.2)
}
## Next, we will simulate the output variable and include interaction terms
for(i in 1:nrow(m)){
a=exp(2*m[i,4] + 0.5*m[i,1] - 4*m[i,2] + 2.3*m[i,3] + 2.3*m[i,3]*m[i,2]
+ 8*m[i,1]*m[i,2] + 2.1*m[i,2]*m[i,3] + 9*m[i,3]*m[i,1])/ (1 +exp(2*m[i,4]
+ 0.5*m[i,1] - 4*m[i,2] + 2.3*m[i,3] +
2.3*m[i,3]*m[i,2] + 8*m[i,1]*m[i,2] + 2.1*m[i,2]*m[i,3] + 9*m[i,3]*m[i,1]) )
m[i,5]=rbinom(1,1,a)
}
## We are interested in determining the coefficient of covariate 4 which is 2.
##The most straightforward way of doing this
##is to use logistic regression as follows
m=as.data.frame(m)
k=glm(m[,5]~.,data=m[,(1:4)],family=binomial)
## The value of the coefficient of variable 11 found by the
##logistic regression can be retrieved using
k$coeff[5]

## We can now use the stratadisc() function to find a
##more accurate estimation of the coefficient
g=stratadisc(4,5,m)
## We need to take the log() of the first
##number returned "Odds Ratio of Impact Of Treatment On Outcome"
## and the estimated value of the parameter should be more accurate

# }