R help - Oct 2007 - Bootstrapping Contrasts for Repeated Measures ANOVA

I have executed a Repeated Measures ANOVA with one DV (latency) and
one within subject factor (acoustic condtion: 3 levels) by
bootstrapping my sampling distribution of F from the empirical sample
distribution. I chose to resample because the sample distribution
deviates from normality a lot.

The overall F is significant and now I wish to decompose this with
contrasts to ask if latencies to acoustic condition #1 are different
than acoustic condition #2 + #3.

However, I don't know where to begin to write a program to do
contrasts with a resampling technique.

Any ideas would be terrific.

Thanks,
Alex




-- 
Alexander T Baugh
Section of Integrative Biology
Univ. of Texas at Austin C0930
Austin, TX 78712
http://darktropic.blogspot.com/

On Mon, Oct 15, 2007 at 11:19:35PM -0500, Alex Baugh wrote:
> However, I don't know where to begin to write a program to do
> contrasts with a resampling technique.
select a random sample (with replacement) from the population with
factor level 1, and a random sample from the population with factor
level 2. Compute the difference in averages. This is a single sample
of the "difference of averages". Do this procedure several thousand
times. You will have a bootstrap distribution of the difference in
averages.

Am I missing some other issue?

-- 
Daniel Lakeland
dlakelan at street-artists.org
http://www.street-artists.org/~dlakelan

Two follow up questions from post regarding bootstrapping contrasts
for a RM ANOVA:

1. Because my data are repeated measures, isn't it true that I want to
shuffle my column heading individually for each subject (row) by
resampling WITHOUT replacement?

2. For my DV (latency) I have 3 columns of data (3 levels) for each
subject:

                  treatment 1     treatment 2    treatment 3
Subject 1      43 sec         110 sec           65 sec
Subject 2     400 sec         650 sec          430 sec
Subject 3      35 sec         40  sec           50 sec
  .                       .               .              .                 .
  .                       .               .              .                 .
  .                       .               .              .                 .


The data are nowhere near normally distributed, so I bootstrapped the RM ANOVA,
which was signficant. Now I want to do an orthogonal contrast asking if
(treatment 1) is different than (treatment 2 + treatment 3) because this is the
a priori sensible question to ask given the overall effect. If
this means I would average columns 2 and 3 I think this is
statistically wrong due to a difference in sampling?


Thanks,
Alex






Thanks


-- 
Alexander T Baugh
Section of Integrative Biology
Univ. of Texas at Austin C0930
Austin, TX 78712
http://darktropic.blogspot.com/