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 on April 23, and its replacement, statalist.org is already up and running.


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

Re: st: Test coefficients


From   Nick Cox <njcoxstata@gmail.com>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: Test coefficients
Date   Wed, 18 Apr 2012 17:33:00 +0100

No other ideas from me, other than smoothing your graph with e.g.
conditional means. As said, using indicator or dummy variables is
surely the standard approach here.

On Wed, Apr 18, 2012 at 5:27 PM, Miguel Angel Valencia Garza
<miguel.valencia@sat.gob.mx> wrote:
> I originally tried to plot the predictions for both models, but old panel has 432,104 observations and new panel has 994,931 observations. Since no_observations = no_predicted_values, the plot looks pretty awful and one cannot suggest anything from it. So, I compared the MSEs and they are pretty close to each other.
>
> Finally, I tried to use a Chow test to prove the H0 I wrote in the original mail. The problem was that old panel is included in new panel. Hence, I gave different id to taxpayers in old from the new, ignoring that they might be the same taxpayer. In this point, I felt that I was forcing it. And here I am, stuck with this problem.
>
> Do you have any other ideas, please?
>
> Thank you very much!!
>
> -----Mensaje original-----
> De: owner-statalist@hsphsun2.harvard.edu [mailto:owner-statalist@hsphsun2.harvard.edu] En nombre de Nick Cox
> Enviado el: miércoles, 18 de abril de 2012 10:01 a.m.
> Para: statalist@hsphsun2.harvard.edu
> Asunto: Re: st: Test coefficients
>
> The problem with the test you imagine is that the set (old + new)
> includes the subset new.  It is easier to define one or more indicator
> variables (1 for new data and 0 for old) and phrase a test in terms of
> those.
>
> As you're an economist you unsurprisingly tend to think in terms of
> such tests. I prefer to think of the whole model and how it works in
> predicting the data. Even if the new data are very similar some
> coefficients will shift one way and others another way as a matter of
> chance, but it's usually how the model works as a whole that interests
> me most. So, I would might start by plotting predictions for old data
> from old model versus predictions for old data from new model. My
> guess is that would seem a strange thing to do in economists' circles.
>
> Nick
>
> On Wed, Apr 18, 2012 at 3:36 PM, Miguel Angel Valencia Garza
> <miguel.valencia@sat.gob.mx> wrote:
>> Dear all,
>>
>> I had estimated the regression coefficients for a panel (using xtgee). This was 8 months ago. I use this model to analyze the behavior of taxpayers and asses certain features. Now, I have updated the panel; i.e. I have fitted a new xtgee model with this panel. However, the coefficients look similar to the ones I'd estimated before.
>>
>> What test (or tests) should I perform in order to proving this hypothesis (H0: coeff_betas_8months_ago = coeff_betas_updated)? Can this test be done in Stata? Thank you all.

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


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