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 at the end of May, and its replacement, statalist.org is already up and running.


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

Re: st: OLS pooled regression


From   Maarten Buis <maartenlbuis@gmail.com>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: OLS pooled regression
Date   Thu, 12 Jul 2012 09:56:34 +0200

On Wed, Jul 11, 2012 at 10:30 PM, Erhan Kilincarslan wrote:
> I set my basic model as follows; Y x l.y ..N= 875 and T=11 years. It s unbalanced data set. I run OLS pooled and have some normality and linearity problems. Then, I run FE and RE. FE overruled both OLS and RE. But as I included lagged dependent variable, and it is the case of large N small T, it has got some endogeneity related issues, right? Then, I tired dynamic model, namely xtabond. Then, it gave more or less the same result with FE and I think it got some autocorrelation problems etc.. So, should I  go for RE? Also, if OLS has normality and linearity problems, then running FE or RE sort out that problems or not?

You seem to be looking for a model without problems, i.e. a true
model. Such a model by definition cannot exist. A model is a
simplification of reality, and simplification is just another word for
"wrong in some useful way". So you need to make a trade-off, and
unfortunately you cannot rely on any automatic techniques to make such
decisions for you. You certainly should not rely on statistical tests:
a statistical test has no idea of what "useful" is, so it cannot
represent the necessary trade-off. This is the job of you, the
researcher.

In linear regression(*) the normality of the errors is hardly ever an
issue. So, that is something you do not need to worry about. If you
still think you do worry about it, than you are in all likelihood
wrong but you can still ease your conscience by specifying the
-vce(robust)- option.

Sometimes people make the mistake of thinking that the dependent
variable in a linear regression needs to be normal. This is never
true. An example showing this can be seen at the bottom of this page
<http://www.maartenbuis.nl/software/hangroot.html>, also see:
<http://www.maartenbuis.nl/software/margdistfit.html>.

Linearity of effects can be a bigger issue in linear regression. It
can be a problem, but it does not have to be. Again, a model is just a
simplification of reality. So it is your decision whether or not the
relationship can be summarized with a linear line. Anyhow, if you
think that that is not the case, than random effects models or fixed
effects models will not help you with that. I personally like linear
splines for such problems, see -mkspline-, as they often represent a
nice compromise between simple, and thus interpretable, results and
allowing for some non-linearity.

-- Maarten

(*) Linear regression is the model, Ordinary Least Squares (OLS) is
only the algorithm used for estimating that model.

--------------------------
Maarten L. Buis
Institut fuer Soziologie
Universitaet Tuebingen
Wilhelmstrasse 36
72074 Tuebingen
Germany


http://www.maartenbuis.nl
--------------------------

*
*   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