gosh! a negative prevalence

stella2011

What's wrong with my winbugs code[s:16]

model{

for(i in 1:N){

r ~dbin(p,n)

logit(p)<-beta0+beta1*x

pre<-(p+spec-1)/(sen+spec+1)

}

beta0~ dnorm(0,1.0E-6)

beta1~ dnorm(0,1.0E-6)

sen~dbeta(165.20,9.64)

spec~dbeta(77.72,14.54)

}

list(

r=c(17,15,71,17,9,23),

n=c(272,87,322,176,94,387),

x=c(1,1,1,1,1,1),

N=6)

list(

beta0=0,beta1=0,sen=0.1,spec=0.1)

list(

beta0=0.01,beta1=0.01,sen=0.01,spec=0.01)

rtist

I don't see the code/model has done anything to protect the sign of pre values.

Moreover, beta0 and beta1 are not identifiable.

stella2011

回复第2楼的 rtist：

I'm stupid. I gave the wrong prior. The beta(165,9.6) should go for "spec", and the other for "sen".[s:12]

I don't know how to protect the sign ,but the common sense tell me a "prevalence" can never be negative[s:18]