# AW: st: Stability of gllamm results against linear transformation of independent variables

 From "Hanno Scholtz" <[email protected]> To <[email protected]> Subject AW: st: Stability of gllamm results against linear transformation of independent variables Date Thu, 20 Dec 2007 12:26:49 +0100

```Hello Maarten, thanks for your reasoning! Indeed, interaction effects would
explain the problem.

But (i) there are no interaction effects with time, and (ii) year was only
an example - if I linearly transform any other variable (e.g. logged GDP),
the complete regression results change as well. Are there any more hints
around?

Hanno

-----Urspr�ngliche Nachricht-----
Von: [email protected]
[mailto:[email protected]] Im Auftrag von Maarten buis
Gesendet: Donnerstag, 20. Dezember 2007 10:11
An: [email protected]
Betreff: Re: st: Stability of gllamm results against linear transformation
of independent variables

Does the model contain interaction effects with year? If so than in
your old model the main effects measure the effects in the year 0,
while in your new model they measure the effect in 1900 (and I would
expect the two to be different).

Also in fragile models that includes year I almost always find that the
model becomes much more stable if you let time start at some point
within the range of your data. This is especially true if the model
contains interaction effects with year. Often, I also devide year by 10
(or sometimes 100).

-- Maarten

--- Hanno Scholtz <[email protected]> wrote:
> I estimate democratization in the European Union's neighborhood as a
> function of EU incentives (e.g. whether the EU proposes the chance to
> get a member) and some controls. Since democracy (measured by Freedom
> House) is stepwise and censored, I use a multinomial probit; since I
> have panel data (36 countries and 13 years; for some missings
> altogether 385 observations),
> I need a random effects model. The model includes a time trend.
>
> The syntax applied is
>
>   gllamm [dependent] year [indepvars], link(oprobit) i([Count
> variable over countries])              (1)
>
> After estimating some models, I applied some linear transformations
> to the variables, for example
>
>   gen yeart = year - 1990
>   gllamm [dependent] yeart [indepvars], link(oprobit) i([Count
> variable over countries])             (2)
>
> Much to my surprise I found that the results of (2) deviate from
> those of (1) - not only in the coefficient of yeart vs year and the
> constant, as I expected, but in every coefficient and z stats.

-----------------------------------------
Maarten L. Buis
Department of Social Research Methodology
Vrije Universiteit Amsterdam
Boelelaan 1081
1081 HV Amsterdam
The Netherlands

Buitenveldertselaan 3 (Metropolitan), room Z434

+31 20 5986715

http://home.fsw.vu.nl/m.buis/
-----------------------------------------

___________________________________________________________
Yahoo! Answers - Got a question? Someone out there knows the answer. Try it
now.
*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/

*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/
```

• Follow-Ups:
• References: