Re: resampling methods are serious procedure?

I've used resampling methods to teach the concept of "probability" in
hypothesis testing.
In fact, I've written a collection of software routines (1) for
purposes of teaching and (2) for serious work where this applies. For
instance, in comparing the deciles of cumulative distribution curves
in life-testing.

However, there is a potential for misusing these methods even in
simple applications. For instance, in comparing averages there is
still the underlying assumption that the variances of the two (or
more) sources of data are "equal". The hazards and potentials for
misuse are as present in resampling as they are in classical methods.

That said, I find resampling to be a power device for teaching
"comparative data"... comparing averages, standard deviations,
declies, and a whole lot more. By going this route I can teach and
"make it stick" efficiently and effectively. (The students get copies
of the relevant software as part of the course.) In fact, I'd go so
far as to suggest that resampling should be taught early to
undergraduates simply because the concepts can be grasped and
appreciated a lot faster than the obscurities and fine print in the
classic methods... which are a major turnoff.

Just fyi, I have taught "statistics" and other subjects (such as
fitting data to nonlinear models via differential equations) to many
hundreds of students in industry (engineers, scientists, managers) and
have given lectures for thousands over a period of 45+ years.
Resampling is one of the few concepts I've found that actually "turns
them on". Most come to my sessions with a bad taste (even bitterness)
about courses they have taken at universities.

Also, I have run extensive simulations to test the validity of
resampling methods and they do perform as expected provided we do not
violate the usual classic assumptions. In short, done properly they
closely parallel the classic methods without the agony.

"It works for me".

Let the flaming begin.
.

Loading