imbi-heidelberg / fisherboschloo Goto Github PK

• Compute unconditional p-value
• Compute 1-alpha confidence interval for odds ratio (should equal the exact Fisher-test CI with raised level)
  Open theoretical question: Is Fisher's noncentral hypergeometric distribution function at a fixed value monotonic decreasing in the odds ratio?

If sample sizes are very small, the rejection region is theoretically empty. But function Raise.level gives an error.

In function Raise.level.NI: help function Compute.s.prob.vec gives Inf or NaN for high numbers. Fix problem by computing on log scale, e.g.:
Compute.s.prob.vec <- function(p0){
p1 <- 1/(1+(1-p0)/(deltap0))
sapply(
s.area,
function(x) {
k <- max(x-n1, 0):min(x, n0)
help.val <- log(choose(n0, k)) + log(choose(n1, x-k)) - klog(delta)
help.val <- help.val + n0*log(1-p0) + (n1-x)log(1-p1) + xlog(p1)
sum(exp(help.val))
}
)
}

any plans of packaging this up for R?

p0 >= p1 has to be reformulated in OR(p1, p0) <= delta.