Re: st: AR(1) model with sign flip

[Clive Nicholas <clivelists@googlemail.com>]

Pedro Ferreira wrote:

> Hi All, I am trying to predict Internet penetration and thus running a
> model of the form I(i,t) = a +p I(i,t-1) + r G(i,t) + X(i,t) b +
> e(i,t). I run reg I L.I G X. p is positive and significant. X are
> controls and G is the level of engagement in a policy program, which I
> am interested in testing. Therefore, I am taking differences and
> trying reg D.I LD.I D.G D.X. Strangely, p flips to negative and
> significant! Any ideas? I must be dealing with a wrong specification,
> but I am clueless about what to do. I tried D.I L.I G X and get
> p=-0.66. The original specification I L.I G X gives me p=0.33. Makes
> sense, but the differenced one D.I LD.I D.G D.X yields p=-0.46! Help
> is much appreciated. Thank you, all the best, Pedro

I don't think that the -reg- command is suitable for the model you
want to fit, given that you have a lagged dependent variable in it.
Assuming serial dependence only, -prais- is better. Assuming serial
dependence _and_ heteroscedasticity, you'd be better off fitting it
with -ivreg2- from Baum, Schaffer and Stillman (SSC). Others may have
better suggestions.

-- 
Clive Nicholas

[Please DO NOT mail me personally here, but at
<clivenicholas@hotmail.com>. Please respond to contributions I make in
a list thread here. Thanks!]

"My colleagues in the social sciences talk a great deal about
methodology. I prefer to call it style." -- Freeman J. Dyson.
*
*   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/