Maarten,
Thanks for the information you gave. I think you are right-- there is
no within-individual variation for a big proportion of the sample I
used. Can I ask though why the the note says that "multiple positive
outcomes within groups encountered"? If the problem is that there is no
within-group variation, why are there multiple positive outcomes within
groups? I checked the outcome variable, it is only valued as 0 or 1 so
I don't understand what this notes really means. Also, I have
encountered another notes saying "1,066 individuals dropped due to all
positive or all negative outcomes." when I was trying the same program
with another sample. Is this essentially the same problem happening
again? Thanks a lot!
Daisy
On Thu, 8 Oct 2009 05:56:10 +0000 (GMT)
Maarten buis <[email protected]> wrote:
>
>
> --- On Thu, 8/10/09, J. Li <[email protected]> wrote:
> > I am now running a fixed effect logit model using the
> > following lines.
> > (Note that the panel data I am using is with N=2,200 and
> > T=10.)
> >
> > iis id
> > xtlogit status treatment age group, i(id) fe
> > mfx, predict(pu0)
> >
> > STATA then gave the notes as following--
> >
> > note: multiple positive outcomes within groups
> > encountered.
> > note: 2054 groups (20506 obs) dropped due to all positive
> > or
> > all negative outcomes.
> >
> > After that STATA ran over more than 3000 iterations without
> > stopping nor giving any results. The log-likelihood values
> > didn't improve after iteration 40. For each iteration, it
> > showed "(not concave)" after every log likelihood value.
> > Do you know why is this happening?
>
> Fixed effects regression uses only information from changes within
> an individual (firm, country, or whatever your unit may be), it
> throws away all information that could be obtained from comparing
> individuals. So, a fixed effects regression can do nothing with an
> individual that doesn't change over time, and such an individual
> is removed. Apparently, your data mainly consists of individuals
> who do not change over time (as 2054 of your 2200 individuals where
> dropped). Which means there is only a very small sample left. I
> would check whether this is correct, e.g. make sure that your
> dependent variable is coded 0, 1 and not 1, 2. If it is true that
> there is very little variation in outcome within individuals then
> your problem is just not suitable for being analyzed with fixed
> effects regression, and you can move on to random effects regression.
>
> Hope this helps,
> Maarten
>
> --------------------------
> Maarten L. Buis
> Institut fuer Soziologie
> Universitaet Tuebingen
> Wilhelmstrasse 36
> 72074 Tuebingen
> Germany
>
> http://www.maartenbuis.nl
> --------------------------
>
>
>
>
>
> *
> * 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/