Re: st: testing differences between coefficients

[Maarten Buis <maartenlbuis@gmail.com>]

On Thu, Dec 20, 2012 at 9:58 AM, Ebru Ozturk wrote:
> Thank you Maarten. I looked at the examples given in Stata after suest. They also run testnl in Example 3 A nonlinear Hausman-like test. Do I need that?

That depends on the exact hypothesis you are testing "the effects are
the same" is typically not precise enough, as you have to define what
you mean with "effect". Remember that an effect is a comparison of
(hypothetical or counterfactual) groups. So you need to define the
groups  (probably the most important and most difficult decision),
what is being compared (means, probabilities, odds, latent scores,
...), and how you want to do the comparison (difference or ratio).
Than you can look at the different kinds of statistical models to
figure out what "effect" they are estimating. Than you can decide
which one gets close enough to what you want to do (and its
limitations). Once you have done that, it will become obvious whether
you need a linear or non-linear test. So, the answer to this question
is that you really need to figure this one out for yourself, as you
are the only one who can make these decisions.

> Another question is after suest test we get the results table and how do we interpret it? Do we just compare the coefficients for different models? For example, if one coefficient (variable) in X model is higher than the coefficient in Y model, the effect of this variable is greater for X model relative to Y model. Is this right interpretation?

Again, this depends on what it is exactly you want to do, and what you
mean exactly with "effect". Once you have gone through the steps
specified above, the answer will be obvious. So, again you need to
figure this one out on your own, because it critically depends on
decisions only you can make.

-- 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/