Re: st: fixed effects glm - fractional dependent variable

[Steve Samuels <sjsamuels@gmail.com>]

Please note:

"You are asked to post on Statalist using your full real name. This is a long-standing practice on Statalist. Giving full names is one of the ways in which we show respect for others."





On Jul 30, 2012, at 4:09 AM, joe j wrote:

Thank you very much Jeff,

Somehow I missed your message, but nevertheless I ended up using the
solution you suggested (thanks to the paper, Papke, L.E., J.M.
Wooldridge (2008) Panel data methods for fractional response variables
with an application to test pass rates Journal of Econometrics 145:
121–133). I used time averages, as well as time dummies. As you
indicate, the results from both glm and xtgee were quite similar.

However, my data is unbalanced. Are your solutions for unbalanced data
available yet?

Best,
Joe.

On Sun, May 13, 2012 at 11:10 PM, Jeffrey Wooldridge
<jmwooldridge60@gmail.com> wrote:
> proposes a correlated random effects approach. As you suspect, putting
> in cross-sectional dummies introduces an incidental parameters problem
> (not to mention the computational problem). If you have a balanced
> panel you put in the time averages. With unbalanced panels it is
> trickier but I have recently worked on some solutions.
> 
> We used xtgee and glm in our panel data work and found that, even
> though glm does not exploit the panel structure, it was almost as
> efficient. The important decision was including the time averages to
> control for the heterogeneity being correlated with the time-varying
> covariates.
> 
> Jeff
> 
> On Fri, Mar 30, 2012 at 4:26 AM, joe j <joe.stata@gmail.com> wrote:
>> Thanks for the link. There is indeed some discussion on this topic.
>> Joe.
>> 
>> On Thu, Mar 29, 2012 at 8:47 PM, Anders Alexandersson
>> <andersalex@gmail.com> wrote:
>>> I do not have an answer to your question but -glm- ignores that you
>>> have panel data.
>>> For example, see http://www.stata.com/statalist/archive/2012-01/msg00595.html
>>> 
>>> Anders Alexandersson
>>> andersalex@gmail.com
>>> 
>>> 
>>> 
>>> 
>>> On Mar 29,, joe j <joe.stata@gmail.com> wrote:
>>>> I have a panel data with the dependent variable being a faction,
>>>> including some zeros (about 1%) and ones (about 10%). These 0s and 1s
>>>> are real outcomes indeed (that is, not the results of censoring).
>>>> 
>>>> So I am going in favor of a glm model as proposed in the literature
>>>> (e.g. Papke, Leslie E. and Jeffrey M. Wooldridge. 1996.  Econometric
>>>> Methods for Fractional Response Variables with an Application to
>>>> 401(k) Plan Participation Rates. Journal of Applied Econometrics
>>>> 11(6):619-632.):
>>>> 
>>>> "glm dependent_variable independent_variable, family(binomial)
>>>> link(logit) robust"
>>>> 
>>>> What I would like to do is run a fixed effect model. However, there
>>>> are too many dummy variables to create (over 16,000 in a sample of
>>>> over 40,000 observations); moreover, I am not sure dummy variable
>>>> approach is appropriate given the non-linear nature of the model.
>>>> 
>>>> My first thought was to use:
>>>> 
>>>> vce(cluster panel_variable)
>>>> 
>>>> Is that the closest I could get to a fixed effect model?
>>> 
>>> *
>>> *   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/
> 
> *
> *   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/


*
*   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/