In article <[EMAIL PROTECTED]>, [EMAIL PROTECTED]
says...
> Excpet that in the case of contingency tables, one test does not
> necessarily dominate another. If, for example, you were to
> choose the smaller P value from the Pearson chi-square and the
> likelihood ratio tests, your true level would be greater
> than the nominal 0.05 (unless there's been some recent research
> I'm unaware of, which is a possibility; last time I checked many
> years ago the two tests were "unsurpassed".)
>
The Pearson chi-square was useful before computers became common because
it was easy to calculate. I'd say it's a good example of a statistic that
could be scrapped. In other cases it's statistical theory that's at
fault. Pseudo R^2 measures for logistic regression come to mind. There's
a large number of them, with no clear winner. Lisrel has a whole pack of
fit measures, also no clear winner there (although Lisrel is an example
of overkill, I'd say).
A nice package in this regard is Stata. Program results are stored in
macro variables so you can write your own macro to present what's
important to you and scrap the rest. I wrote a nice little macro program
that presents estimates, standard errors, and symbols for significance at
5% and 1% for estimation procedures. The rest of the information is
redundant: z- or t-values, confidence intervals are derived from
estimates and s.e.-s. Significance levels are only important for a few
critical values, "*", "**", or " " are all you're interested in that
regard. Statistics is basically about extracting relevant information
from a chaos of data. Why shouldn't statistical software reflect this?
An earlier version of this macro in SAS could do the same thing, but only
worked in PROC GENMOD, using the then experimental Output Delivery System
(ODS). Recent releases of SAS have expanded ODS, that could be another
option for trimming redundant output information. Scripting in SPSS is
another option.
Back to work,
John Hendrickx
=================================================================
Instructions for joining and leaving this list and remarks about
the problem of INAPPROPRIATE MESSAGES are available at
http://jse.stat.ncsu.edu/
=================================================================
