Narasimhan Sowmyanarayanan wrote:
> I just noticed something in my regressions. Not sure if it is a
> problem with what I am trying to do or if this is expectes. I ran an
> OLS of the mean centered variables by each case in a panel data and
> then a fixed effect regression (xtreg), fe. I find that the estimates
> are a bit differet (though not by a big margin). How can such a thing
> happen. I checed that the number of observations in both regressions
> is exactly the same.
It would have been helpful for you to have included some output for us to
look at: Mark Schaffer very politely asked you to do this once already.
Nevertheless, this is almost certainly because the two regressions make
different degress-of-freedom (df) adjustments, which help to change the
parameter estimates, standard errors and t-ratios. Centering the variables
makes no difference to these, apart from in the constant term. Run
. webuse grunfeld, clear
. reg invest mvalue kstock
. center invest mvalue kstock
. reg c_invest c_mvalue c_kstock
. xtreg c_invest c_mvalue c_kstock, i(company) fe
to see what I mean (NB: -center- is a user-written program devised by Ben
Jann and downloadable from SSC). Notice how in the fixed-effect model, the
number (10) of groups is subtracted from the df, which it isn't in the OLS
model. For more sophisticated answers to a more finessed problem posed by
Garrett Glasgow back in July 2004, check out the replies made by Mark and
by Kit Baum at
http://www.stata.com/statalist/archive/2004-07/msg00616.html (Kit)
http://www.stata.com/statalist/archive/2004-07/msg00620.html (Mark)
The latter link contains the original post.
> Another detail issue is concerning mean centering if we want to run a
> regression with a lag variable. Should one mean center and then lag to
> create the mean centered lag variable, rather than create a lag
> variable first and then mean center. how is this implemented in stata
> ?
My preference would be to take the lag, then -center- it. Doing this is easy:
. g lfoobar=l.foobar
. center lfoobar
Then you simply drop this and your other -center-ed variables into your
model. Again, this doesn't change any of the statistics in the centered
model from the uncentered one, apart from those belonging to the constant
term (which may or may not matter to you). I've yet to read anything
telling me that centering a lagged variable (_if it's continuously
observed!_) invalidates its usage.
However, I have read quite a bit of stuff (at least, quite a bit for this
non-econometrician) telling me that the use of a lagged variable in
certain panel-regression models (in particular, the least-squares dummy
variable FE model) invalidates the use of these models as the parameter
estimates become inconsistent. In sum, it depends upon what the variable
is being used for.
Hope that helps.
CLIVE NICHOLAS |t: 0(044)7903 397793
Politics |e: clive.nicholas@ncl.ac.uk
Newcastle University |http://www.ncl.ac.uk/geps
Whereever you go and whatever you do, just remember this. No matter how
many like you, admire you, love you or adore you, the number of people
turning up to your funeral will be largely determined by local weather
conditions.
*
* 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/