\(Tr\) is the threshold/truncation, can be measured as cutoff in a standard normalized distribution.
| Parameter | Case | Control |
|---|---|---|
| mean | \(\mu_{cs}\): mean of cases | \(\mu_{cl}\): mean of controls |
| sampling variance | \(\sigma_{cs}\): observed cases after truncation | \(\sigma_{cl}\): observed controls after truncation |
| counts | \(n_{cs}\): observed cases after truncation | \(n_{cl}\): observed controls after truncation |
| conditional mean | \(\hat{\mu}_{cs}\): truncated mean for \(n_{cs}\) cases | \(\hat{\mu}_{cl}\): truncated mean for \(n_{ctrl}\) control |
##A toy example for truncated distribution
##Assuming normal distribution
pcut=0.1
cutoff=qnorm(1-pcut)
N_cs=700
N_ctrl=700
u_cs=qnorm(1-0.05)
u_ctrl=qnorm(1-0.2)
TechCS=0.5##measuring error for mean
csSimu=rnorm(n=N_cs, mean=u_cs, sd = TechCS)
TechCtrl=0.7
ctrlSimu=rnorm(n=N_ctrl, mean=u_ctrl, sd = TechCtrl)
layout(matrix(1:2, 2, 1))
plot(density(csSimu), bty="l", xlim=c(-2, 4), main=paste(N_cs, "Case"))
rug(csSimu, side = 1, col="red")
n1=length(which(csSimu>cutoff))
bar_u_cs=mean(csSimu[csSimu>cutoff])
abline(v=c(u_cs, cutoff), lty=c(2), lwd=c(1, 3), col=c("red","black"))
points(bar_u_cs, 0, pch=1, cex=2, col="red")
plot(density(ctrlSimu), bty="l", xlim=c(-2, 4), main=paste(N_ctrl, "Control"))
rug(ctrlSimu, side = 1, col="blue")
n2=length(which(ctrlSimu>cutoff))
bar_u_ctrl=mean(ctrlSimu[ctrlSimu>cutoff])
abline(v=c(u_ctrl, cutoff), lty=c(2), lwd=c(1, 3), col=c("blue","black"))
points(bar_u_ctrl, 0, pch=1, cex=2, col="blue")We need a likelihood function here.