[RsR] Robust R for dummies?

[Jenifer Larson-Hall]

I have been reading through some past messages and some people apologize
in advance for not being in the loop, but I'm afraid I am not even
anywhere near them! I am just a linguist, but one who had ambitions to
write a book about how to use statistical programs to analyze data in
the field. I am really just looking at very elementary types of
statistical procedures (correlation, t-tests, ANOVA, chi-square,
regression), not getting into PCA or anything like that. But I have a
friend at my university who's a works in statistical research and he
turned me on to R. Although the majority of applied linguists in my
field use SPSS for statistical analysis, I really liked the idea of R
because it would make analysis available to those without big budgets
for SPSS, and because my friend convinced me that there were lots of
things R could do that SPSS couldn't, especially robust statistics. So
what I am trying to do in my book is to show how to do things in SPSS,
then show that R can do them just as well or better, and then take my
readers a step beyond classical statistics to do robust statistics. It
sounds really good in theory but I am floundering! I feel fairly
comfortable with R now and understand the help files pretty much (that
'foo' thing really threw me at the beginning!), but I still cannot
really understand what to do with R and bootstrapping. Is there anything
that could give me more background about how I can do bootstrapping with
the R libraries? Something like Robust R for dummies?

For example, I am working on t-tests right now. I read Wilcox 2003 and
saw that he recommended generally 20% means trimming along with
percentile bootstrapping for comparing means. But I looked through boot
and robustbase and couldn't figure out how to do a one-sided or
independent-samples or paired-samples t-test using anything in those
robust libraries, and I couldn't see that any incorporated 20% means
trimming. So I directed my readers to go with Wilcox's functions, which
aren't in R and have to be downloaded into R (no big deal, we can do
that, not so complicated). But I'm working on trying to figure out how
to do a one-sided t-test with Wilcox's functions, and it hits me--I
probably CAN do this in boot, I would just need to have the function
that would do 20% means trimming! I conceptually understand 20% means
trimming, but wouldn't really know how to write my own program for it.
And even if I did (say, I pulled a function for it out of Wilcox's
code), I couldn't be sure I was doing it correctly. Part of that also
relates to the fact that I am not a statistician and the actual
mathematical calculations for statistical tests are something I have to
remind myself of every time. In other words, I am not really a
statistician either.

Anyway, I'm not really asking so much for any specific help here as
general help. What you guys do is a really hard code to crack for an
outsider, but I've been convinced that robust methods will greatly
improve our accuracy and power in applied work, and I really want to
bring that to the people in my field. I really just don't understand how
bootstrapping works computationally (again, I understand it conceptually
on a basic level). I can't follow the examples in the boot library
examples area. I haven't found a book that really walks me through R
with bootstrapping methods. So . . . any ideas out there of a book on
robust functions in R for dummies?

Thanks for any help you can give.
Jenifer

Dr. Jenifer Larson-Hall
Assistant Professor of Linguistics
University of North Texas
(940)369-8950