Dear Mandy,
there is a command -felsdvreg- (try -search felsdvreg, all-) by Thomas
Cornelißen (see Cornelißen 2006, 2008 for descriptions). My experience
is that it works better (in terms of memory) than e.g. the
spell-fixed-effects-approach described in Andrews, Schank and Upward
(2006). To give an example: I recently had a problem estimting a model
with person, county and time effects with a rather large dataset.
Spell-fixed-effects did not work with roughly 14 GB of memory
allocated, -felsdvreg- worked with about 4 GB.
Best regards,
Nils
Refs.
-------
Andrews, Martyn, Thorsten Schank and Richard Upward, 2006: "Practical
fixed effects estimation methods for the three-way error-components
model", The Stata Journal 6(4), pp. 461-481.
Cornelißen, Thomas, 2006: "Using Stata for a memory saving fixed
effects estimation of the three-way error component model", FDZ
Methodenreport 03/2006, Nuremberg. (see
http://doku.iab.de/fdz/reporte/2006/MR_03-06.pdf)
Cornelißen, Thomas, 2008: "The Stata command felsdvreg to fit a linear
model with two high-dimensional fixed effects", The Stata Journal
8(2), pp. 170-189.
On Mon, Jan 19, 2009 at 10:27 PM, Mandy fu <[email protected]> wrote:
> Hi everyone,
>
> I ran into some problem when trying to estimate a panel data from
> three ways (worker, firm, year). I find in statalist previously this
> question has been posted here:
>
> http://www.stata.com/statalist/archive/2004-07/msg00229.html
>
> And there's one reply:
>
> http://www.stata.com/statalist/archive/2004-07/msg00230.html
>
> I find the four methods given in the reply are all two-way fixed effects.
>
> What I find is that a paper seems to talk about my question "Practical
> fixed–effects estimation methods for the three–way error–components
> model" in Stata Journal Volume 6 Number 4: pp. 461-481. But I have to
> wait for several days to get this paper since library at my university
> does not have this journal.
>
> So, I was wondering if anyone here could give me some hints about my questions:
>
>
> (1) If the equation to estimate is: Yijt = ai + bj + ct + BXijt + eijt
>
> Can I demean the Y and X in year dimension, add dummy variables for
> firm1 till firm j , and then use xtreg,fe? That is:
>
> ---xtreg demeaned_Y demeaned_X firm1-j ,fe--
>
> (2) If the above way is fine, then, how to deal with if we have many
> workers, many firms and many years, let's say, 5000 workers, 1000
> firms, and 50 years? It'll be inconvenient to add any dummy varaibles.
>
> (3) Besides what I mention above, is there any other discussion or
> resource about this three-way fixed effects using Stata ?
>
> Thanks your your help! Really appreciate it.
>
> Mandy
>
> *
> * 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/
