Hello,
if I may add another question under this thread:
what if you have a very normal panel data set, with the DV being
measured in all panels at the same time (as opposed to Ken's dataset),
but the time periods are unevenly spaced? In my specific case, 20
years apart at the beginning of the dataset, 10 years after that, and
5 years towards the end. I am very reluctant to drop observations in
order to achieve evenly spaced intervals. What if one now wants to
estimate a dynamic panel in this case?
The problem I am thinking of is that if the autocorrelation factor is
equal to, say, "a" for two periods that are 5 years apart, it should
be "a^2" for two periods that are ten years apart, and "a^4" in the
earliest part of the dataset.
Does anyone of you know whether this problem has been dealt with in
the literature in the past, (and if there are any implementations for
that in stata)?
Any suggestions are much appreciated!
Augusto
On Sat, Aug 2, 2008 at 9:27 AM, Clive Nicholas
<[email protected]> wrote:
> Kenneth F Greene wrote:
>
>> 1. The observations are country-election-years
>> 2. The DV is the vote percent gap between the winning party and the first loser
>> 3. I use a LDV in some specifications and not in others, but the main model does not contain a LDV
>> 4. I have an average of 7.9 time points (elections) per country and 55 total observations.
>>
>> For reasons that are aren't helpful for present purposes, the data set used to be larger and I used xtgee, i(id) robust with success. Now that the data set is smaller, xtgee does not converge. I can make it converge by changing the tolerance, although I'm not yet sure what that actually does and how best to play with it, if at all.
>
> Thanks for the extra details.
>
> Hmmm, not sure why you would use -xtgee- on this data given the way
> your DV is measured, but I'll make two suggestions:
>
> (1) if you want your model to obey a panel structure, then using
> -xtpcse- with or without the -corr(ar1)- option seems reasonable here;
>
> or
>
> (2) if you're indifferent, then you could normalize your DV and fit a
> fractional logit model: -glm y x1 x2 ... xk, fam(bin) link(logit)
> robust-.
>
> Both (1) and (2) can accept LDVs, and I've not yet read a rationale as
> to why you shouldn't include them in (2).
>
> Hope that helps.
>
> --
> Clive Nicholas
>
> [Please DO NOT mail me personally here, but at
> <[email protected]>. 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/
>
*
* 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/
