Re: Please help with the design: 22 parameters influencing single outcome

In article <[EMAIL PROTECTED]>,
Pawel Kusmierek <[EMAIL PROTECTED]> wrote:
>I would appreciate your advice on statistical design.

>There are four subjects, each was tested with 171 stimuli. Each
>stimulus is described by 22 parameters. A test yields a single
>numerical outcome. I am interested whether the parameters of stimuli
>influenced test results.

>One idea is to average the subjects' test results for each stimulus
>and calculate partial correlation coefficients between the average
>test result and the 22 parameters.  However, this approach offers no
>view into the intersubject variability.

>Another idea is to categorize the stimuli according to each
>parameter's value: e.g. >=median vs. <median.  Then I would check
>whether test results obtained with stimuli which have high (>=median)
>parameter values differ from test results obtained with stimuli which
>have low (<median) values.

>I was thinking of a 23-factor ANOVA, with one repeated mesures factor
>(subject) and 22 factors=parameters.  In the design, only main factors
>and interactions (subject x a_parameter) would be analyzed.

>In such way, I would get a measure of whether any of the parameters
>influences the test result (high Fs for 22 factors-parameters) and a
>measure of across-subjets consistency of this influence (low Fs for
>subject x parameter).

I suggest instead you do a quantitative analysis.  A 23 parameter
ANOVA requires two many trials.


-- 
This address is for information only.  I do not claim that these views
are those of the Statistics Department or of Purdue University.
Herman Rubin, Department of Statistics, Purdue University
[EMAIL PROTECTED]         Phone: (765)494-6054   FAX: (765)494-0558