On 25 Jul 2001 11:44:34 -0700, [EMAIL PROTECTED] (Rich Strauss)
wrote:
> Thanks to Rich Ulrich for the suggestion below -- that was the direction I
> was heading, but there seem to be difficulties. The general problem is
[ snip; looking at rows and columns ... ]
> But the problem is: what if the row totals and column totals are not
> independent? I've done a few 2-way chi-square contingency tests on these
> matrices (using randomized null distributions, of course, since the
> matrices are binary), and some of the results are statistically
> significant. Doesn't this mean that I can't simply accumulate the row and
> column totals for a goodness-of-fit test, since they're not always
> independent? And even if I did the goodness-of-fit tests for rows and
> columns independently, how do I combine the p-values to get a single level
> of singificance for the entire matrix, if the tests are not independent?
Was the sample 50% male? ... as expected beforehand.
Did the ethnic claims in the sample match the most recent census?
- Those are the hypotheses of the margins. I don't see much
excitement in either one. Or in combining the two. And I
especially don't see it happen, often, where someone combines
those two hypotheses with their Interaction: males and females
equally distributed, PLUS, equal-by-ethnicity.
In its loglinear modeling,
BMDP used to provide two sets of tests at a given level (no
variable, or one-variable, or two-way, or three-way...).
Likelihood chisquares are additive, the Pearson chisquare are not.
BMDP also provided a sum of the ML chi-squares, cumulatively;
that *was* a test, if you could justify using it. That seems to me
to be exactly what your were asking for; but I doubt if it is a good
idea.
Statistical consultation works best if you look at your problem
and your hypothesis, and from there, figure what data
you MIGHT collect.
If you have already gathered your data, well, you still
might stop to figure out what the *best* model would
have been, to see how to approach it.
But you should look at your data with an eye on the questions.
What have *other* people done? What numbers would
make sense of *these* relations that you guess at?
Move *from* questions, *to* a computer program or algorithm.
It is certainly helpful to learn new stat-procedures by exploring
your own data. But there are dozens of ways to tests for Fit,
with different alternatives as their strengths;
so the random test is unlikely to be the best choice for
whatever data the random person happens to have on hand.
Where *do* they come from?
Who cares? why does some randomness matter?
--
Rich Ulrich, [EMAIL PROTECTED]
http://www.pitt.edu/~wpilib/index.html
=================================================================
Instructions for joining and leaving this list and remarks about
the problem of INAPPROPRIATE MESSAGES are available at
http://jse.stat.ncsu.edu/
=================================================================
