Newsgroup sci.stat.consult 21627

Hello all !
Here is the problem I have relating to a repeated measures analysis of 
variance design (consider bilogical variables which are measured over three 
points in time).
First here is a summary of my initial text for the review:
« For each variable we used a repeated measures analysis of variance design 
and two linear combinations of the differences between values at the three 
periods [contrasts]...
Error bars  [means +- sem] describes the data at the three periods, and the 
degree of signification given by the averaged univariate F test is specified 
in a footnote.
Errors bars [means +- sem] describes the two linear combinations of 
differences, showing their situation in relation to the zero of scale. The 
degrees of signification given by the univariate F tests are specfied by 
annotations in the graphs....  »
Secondly, here is the critic of the reviewer: the p value is for the ANOVA. 
Usually, vhen this is significant, secondary testing is done to identify 
which of the specifics points are different from each other.
When assessing this many outcome variables, some methods needs to be used to 
reduce the level of the p values for the multiple tests being done.
Thirdly here is in part my answer:
The observed significance levels specified by annotations in the graphs help 
us to identify which individual contrasts contribute to overall differences. 
However, these observed significance levels for the individual parameters 
[univariate F tests] are not adjusted for the fact that several comparisons 
are being made. By reason of this, a single graph represents the contrasts 
and their 95% Bonferroni intervals in relation to the zero of scale 
[simultaneous intervals for parameters within each variable]; then, where the 
95% intervals don’t include the zero we can state positively that the linear 
combination of the differences we tested is different from zero.
Finally, in spite of my researchs on this suject (the control of the p value 
in repeated measures designs) I must recognize that I have not really 
answered to the question and that I don’t really know how to do the 
correction of p value.
Many thanks in anticipation.
Merci infiniment par avance.

harryl (harryl@value.net) wrote:
: I was recently hired on a contractor basis to tabulate and analyze
: results of a customer satisfaction/demograghics email survey.  The
: company anticipated a response of about 500, but so far has received
: over 3,000 surveys.  As a cost saving measure (I'm basically being paid
: per questionnaire), they want me to take a sample of 200-300 responses
: and project the results over the 3,000-4,000 responses they will
: actually receive.  
: 
: My experience in survey analysis has been limited to 4 small, simple
: surveys.  So I'm not sure if their sample "suggestion" is mathematically
: or ethically sound.  That is, is it statistically appropriate to, in
: effect, take a sample of a sample, and would you have any confidence in
: the accuracy of the results.
Yes, it is statistically appropriate.  Depending on the accuracy you are
seeking, 200-300 responses may or may not be enough.  But there is nothing
wrong with subsampling (randomly).
In reading your description, I would have more concern about those who did
NOT respond.  What was the response rate?
-- 
Michael P. Cohen                       home phone   202-232-4651
1615 Q Street NW #T-1                  office phone 202-219-1917
Washington, DC 20009-6331              office fax   202-219-2061
mcohen@cpcug.org

In article <2.2.16.19961213051004.21d753da@email.psu.edu>, Dennis Roberts
 wrote:
> someone else suggested that we would be better off changing NULL to ...
> statistical?  I tend to agree ...
I tend to disagree. There is nothing statistical about a hypothesis.
Statistics is used to evaluate the hypotheses. Perhaps "working
hypothesis" or "default hypothesis" are viable.
OTOH, perhaps we should eliminate the word "hypothesis".
No-one in his right mind would hypothesise that a regression
coefficient is zero. We might be prepared to believe it is small,
but never exactly zero. Maybe "proposition" would be better.
So we now have the "working proposition" and the "alternative
proposition". That gets closer, but I'm really looking for a
word meaning "flight of fancy".
Terry Moore, Statistics Department, Massey University, New Zealand.
Imagine a person with a gift of ridicule [He might say] First that a
negative quantity has no logarithm; secondly that a negative quantity has
no square root; thirdly that the first non-existent is to the second as the
circumference of a circle is to the diameter. Augustus de Morgan

In article , T.Moore@massey.ac.nz
(Terry Moore) wrote:
> Perhaps "working
> hypothesis" or "default hypothesis" are viable.
> OTOH, perhaps we should eliminate the word "hypothesis".
> No-one in his right mind would hypothesise that a regression
> coefficient is zero. We might be prepared to believe it is small,
> but never exactly zero. Maybe "proposition" would be better.
> So we now have the "working proposition" and the "alternative
> proposition". That gets closer, but I'm really looking for a
> word meaning "flight of fancy".
how about "pretense"?
One could begin the discussion of testing with
"P0: pretend that beta = 0..."
That would make it far easier for students to understand how to proceed
thereafter; they are to pretend that the null pretense is correct and find
a p-value. ("p" standing now for both probability and pretension).
-- Paul Velleman