Statalist The Stata Listserver


[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

Re: st: xtivreg2 - little questions


From   "Clive Nicholas" <Clive.Nicholas@newcastle.ac.uk>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: xtivreg2 - little questions
Date   Sun, 4 Jun 2006 20:18:11 +0100 (BST)

Debbie Enders wrote:

(Never having used -xtivreg2-, my comments must be heavily qualified and
Rodrigo's more expert comments should be carefully studied. It is worth
pointing out here that -xtivreg2- is a user-written program downloadable
from SSC, as is -ivreg2-.)

> 1. Is it OK to specified lagged value(s) of the
> dependent vaiable as regressors, as in
> xtivreg2 profit l.profit l4.profit ...(other
> regressors etc)?

Yes, it is - but you do need to worry about the possibility of lots of
lagged values 'sucking out' the (potential) explanatory power of your
other independent variables in your model, where there is also a non-zero
risk of their signs changing (and sometimes implausibly).

> 2. xtivreg2 does not allow one to put in time dummies
> in order to carry out 2-way fixed effects by
> specifying i.date, as one could with OLS. With
> xtivreg2 one gets the error message:
> i:  operator invalid
> r(198);
> But if I generate individual time dummies (tab date,
> g(time)) and then put these time dummies into the
> specification as (time1-time48), the regression seems
> to run OK. However, I wondered whether, since the
> programme does not allow the i.date option, this
> approach was wrong anyway? I did include time dummies
> (i.date) in my OLS regression as this seemed to make
> sense and some broad comparability between the
> specifications would be nice, but perhaps it is
> nonsensical for xtivreg2.

In my view, no: there is nothing wrong with this approach. We know that it
is perfectly acceptable to use a full set of unit dummies in order to make
adjustments to the intercept to individual units in one's dataset (be they
countries, regions, etc), so that differences around the mean of your
dependent variable don't bias the parameter estimates in your named
independent variables. The same principle surely extends to time dummies.
The only problem is that it can make for a rather bloated regression with
too many variables (49 time dummies is quite a lot): the simple solution
is just to report those variables of _direct_ interest.

> 3. What criteria should one use for the # of the bw(3)
> option, the bandwidth of the HAC (in the case of
> robust) covariance estimator, in order to get
> estimates robust to autocorrelation? The examples I
> saw used 3 or 4, but I was not sure how to choose
> this.

Well, I was using another user-written pacakge, -ivreg2-, with -bw(3)- as
the only option, just the other day, and I'm pretty sure that gave me OLS
estimates robust to AC, which was what I wanted. However, somebody on here
will be sure to correct the error of my ways, if need be (again).

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/



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