cdowd
Comparisons to known null distributions currently difficult
Is this worth building?
The difference between just sampling from a known null and this is that we can gain power by comparing samples from the null against themselves.
Could just add a flag which allows all resampling to be from the left sample. Not quite the same as sampling from a distribution, but with a big left sample from a distribution, its difference is negligible. null = c("joint","left") rewrite joint to be left only when needed. Its one flag.
Could build a flag that can take a function input (i.e. rnorm) as well, but then I need to test that input. And build in the mean shift ability... seems like a pain.
Okay, even the easy one is a bit more than that, because I've not been sampling with replacement, and I've been keeping the exact joint sample nearly fixed (with implications for plot defaults at the moment too)
Easy version (simple flag) is worth doing. It is tremendous functionality -- comparing to a known null is a common problem, at relatively low cost. It is of course, not the 'two sample problem', but it is valuable. May be speed gains to consider here too. presorting the left sample should speed the internal sort?