[R] OT: Strange bootstrap approach to significance testing

[Mike Lawrence]

Hi all,

I'm reviewing a paper that employs a strange (to me) approach to a
non-parametric significance testing. I'm familiar with permutation
tests and bootstrapping confidence intervals around parameter
estimates, but here they seem to be bootstrapping CIs for a
manufactured Null F-value. They have a simple 3 groups between-Ss
design and compute the F-value of the observed data. Then, they
re-center each group's mean to zero then repeatedly: randomly resample
values from each group (with replacement, doubling each group's size)
and re-compute the F-value. The distribution of re-centered/resampled
F-values is then used as a reference distribution within which the
percentile of the observed F-value is computed. They determine that
the observed F-value is outside the 95% confidence interval of the
re-centered/resampled F-values and conclude a significant effect of
group.

In my head this makes some sense (though I'd think a straight
permutation test would be a lot simpler), but having never heard of
anything like this before I thought I'd see what others on this list
think of the approach.

Mike

-- 
Mike Lawrence
Graduate Student
Department of Psychology
Dalhousie University

Looking to arrange a meeting? Check my public calendar:
http://tr.im/mikes_public_calendar

~ Certainty is folly... I think. ~