Bookmark and Share

Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down at the end of May, and its replacement, statalist.org is already up and running.


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: Ologit question


From   Maarten Buis <maartenlbuis@gmail.com>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: Ologit question
Date   Wed, 31 Oct 2012 14:46:06 +0100

On Wed, Oct 31, 2012 at 2:23 PM, Anat (Manes) Tchetchik wrote:
> I ran an Ologit and received the following cutoffs:
>
>                 Coef.         Std. Err.           [ 95% Conf. Interval]
>  /cut1 |  -1.798009   2.900799             -7.483471    3.887453
>  /cut2 |   .7973023   2.793147             -4.677166    6.271771
>  /cut3 |   2.902347   2.794626              -2.57502    8.379713
> which means that the cutoffs are not significantly different.

You seem to perform this test by looking at whether the confidence
intervals overlap. This is not a correct way to perform such a test,
as that way you ignore the covariances in the sampling distribution of
these estimates. Instead you should use -test- if you want to perform
such a test.

> Does it mean that the entire model is worthless?

No, statistical tests do not represent a relevant tradeoff you need to
make when choosing a model. The tradeoff implicit in a statistical
test is all about the probability of rejecting a true null hypothesis
and the probability of not rejecting a false hypothesis. However, it
is by definition impossible for a model to be "true". A model is by
definition a simplification of reality, and a simplification is just
another word for "wrong in some useful way". So you want your model to
be wrong (simplify) in such a way that the results become manageable
for a human brain to process, but at the same time you don't want your
model to be too wrong so that it deviates too much from your raw
observations. So selecting a model and doing a statistical test are
fundamentally different beasts working with very different logics,
which means that a statistical test just cannot inform you about which
model is "right" or "wrong".

-- Maarten

---------------------------------
Maarten L. Buis
WZB
Reichpietschufer 50
10785 Berlin
Germany

http://www.maartenbuis.nl
---------------------------------
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/


© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   Site index