As for initial request, Excel MUST HAVE a function to calculate p-value
of the test. The set of statistical functions built into Excel is
comprehensive enough for that; at least it was when I was teaching
statistics with Excel a year or two ago :).
Pretty old books like Kendall & Stuart say something like "all tests
with significance level up to 73% would not reject the hypothesis".
That is a good formulation of... of what you observe with your test
statistics. Another formulation is that p-value is the probability that
things are no better than observed in terms of your null hypothesis;
i.e., if you have F-test, that the true F-value is no less than the
observed F.
As for Jan de Leeuw's remark,
> Live is not simple. The term "observed significance level" or
> "achieved significance level" is fairly standard in statistics. It's
> basically the p-value. For a precise definition, see Efron and
> Tibshirani, Introduction to the Bootstrap, p. 203.
I do not have the book at hand, but I suppose it may refer to the simulated
significance level (at least, the context of bootstrap makes me think so). That
means, for instance, that if you want to check that your asymptotic procedure
works well enough with sample size 50, you generate a thousand samples under
your null hypothesis and check for how many samples your null was rejected by
the test. Now, if your 95% test rejects two hundred simulated samples (or just a
couple out of a thousand), you say it is not applicable, and if 49, you ought to
say that it seems to be reasonable, but further investigation is required :).
> The quotation of A. Lincoln merely gives one side in the age-old
> realism - nominalism debate. H. Dumpty would disagree.
Yeah, and it's really bad that people mix jargon with scientific terms. E.g. in
most papers in economic area, when people say "robustness check", it usually
means that they added another variable to their regression, or used another
proxy for the unobserved variable of interest, but not that they calculated
influence functions or something. And when they say "controlling for ownership",
it means they added dummies for private and state-owned firms, rather than ran
separate regressions for different groups of firms.
......Stanislav Kolenikov (7-095)-232-3613p -3739f
[EMAIL PROTECTED] http://giganda.komkon.org/~tacik/
..................Russian - European Center for Economic Policy
.................................................Moscow, Russia
===========================================================================
This list is open to everyone. Occasionally, people lacking respect
for other members of the list send messages that are inappropriate
or unrelated to the list's discussion topics. Please just delete the
offensive email.
For information concerning the list, please see the following web page:
http://jse.stat.ncsu.edu/
===========================================================================
