I have to perform a statistical test to confirm the difference of the mean of a certain variable between male and female species using bootstrap. There are 8 measuremets (16 data points in all) of each, and my idea was to bootstarp the sample of 16 measurements under the nul hypothesis, there is no difference, and calculate the probability finding a sample with the mean difference of the original sample (pooled permutation test)
1- Doing that in "R" and using boot.ci, I find confidence intervals somehow strange, the percentile CI is perfectly around 0 (as it should be) all other CI's, however, are shifted by 2*(mean of original sampple), which give still a correct result but does'nt seem correct, is there any reason for that, does bootstrap produce only normal, BCa etc CI which are greater than zero? if so how can I eventually get rid of this problem?
2- Does anybody has an Idea, how one can construct a Monte Carlo exact test of the hypothesis of no difference (which statistic is reasonable for this case, are there any simple pivots other than student t statistic)?
Thanks you very much in advance
Advertise on Talk Stats