Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.

RE: st: testing differences between coefficients

 From Ebru Ozturk To Subject RE: st: testing differences between coefficients Date Thu, 20 Dec 2012 12:56:35 +0200

```Dear Maarten,

The example below also does not compare two groups. They test two different dependent variables.

Example 3: A nonlinear Hausman-like test
. webuse income

. probit promo edu exp
. estimates store Promo

. regress inc edu exp
. estimates store Inc

. suest Promo Inc
. testnl [Promo]edu/[Promo]exp = [Inc_mean]edu/[Inc_mean]exp

----------------------------------------
> Date: Thu, 20 Dec 2012 11:26:40 +0100
> Subject: Re: st: testing differences between coefficients
> From: maartenlbuis@gmail.com
> To: statalist@hsphsun2.harvard.edu
>
> On Thu, Dec 20, 2012 at 10:49 AM, Ebru Ozturk wrote:
> > As far as I understand your e-mail, I want to do the comparison by looking at the difference. I do not do any group separation as I just focus on two different dependent variables. In terms of what is being compared (means, probabilities, odds etc.) I really do not know. How can I decide about that? or Does it depend on the model I use?
>
> A difference requires two or more groups. I cannot be diffrent from
> myself. It does not have to be two groups; you can think of a linear
> effect as a comparison of an infinite number of groups.
>
> A difference also requires an idea of what is being compared, and the
> decision depends on what interests you. You are probably very happy
> about the fact that I am not in a position to decide for you what you
> find interesting.
>
> The model is only there to estimate what you want to estimate, so the
> model is irrelevant at this stage. Your first step is to decide what
> you want to estimate, and than look for or derive a model that does
> what you want.
>
> I'll repeat this one final time: You need to do this on your own. It
> is not because I am mean(*) and don't want to help you, but this is a
> question that can only be answered by you.
>
> -- Maarten
>
> (*) I might or might not be, I'll leave that to others to decide.
>
> ---------------------------------
> 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/
*
*   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/
```