Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
From | Rodolphe Desbordes <rodolphe.desbordes@strath.ac.uk> |
To | "'statalist@hsphsun2.harvard.edu'" <statalist@hsphsun2.harvard.edu> |
Subject | st: RE: xtnbreg Initial values |
Date | Fri, 7 May 2010 10:03:51 +0100 |
Ivan, Are you referring to this strategy? http://cep.lse.ac.uk/pubs/download/dp0932.pdf If that is correct, see http://privatewww.essex.ac.uk/~jmcss/panelpoisson.do Rodolphe -----Original Message----- From: owner-statalist@hsphsun2.harvard.edu [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of Ivan Png Sent: vendredi 7 mai 2010 08:46 To: statalist@hsphsun2.harvard.edu Cc: maartenbuis@yahoo.co.uk Subject: st: xtnbreg Initial values Many thanks, Dr Maarten. There are quite enough observations left. Indeed, I ran xtreg (panel fixed effects linear regression) without a problem. As you say, random effects work. But, I need the interpretation of "within company", so I'd like to use fixed effects. Can you advise me how to specify the initial values in xtnbreg? That is my key problem. On Fri, May 7, 2010 at 3:33 PM, Maarten buis <maartenbuis@yahoo.co.uk> wrote: > --- On Fri, 7/5/10, Ivan Png wrote: >> I want to run a panel fixed effect negative binomial >> regression. >> * dep variable: patentcount >> * indep variables: xrd revt capx yr1976-yr2006 >> >> I ran: >> xtnbreg patentcount xrd revt capx yr1976-yr2006 , fe >> >> and got error message: "note: 395 groups (395 obs) dropped >> because of only one obs per group note: 2893 groups >> (20642 obs) dropped because of all zero outcomes >> >> initial values not feasible >> r(1400);" > > I would first focus on how many observations are left. > Remember that fixed effects regression only uses information > from changes within a individual. It looks like for a large > part of your data there are no such changes, so these are > dropped (that is what the two notes mean). If there are only > a few observations left then a fixed effects design just > doesn't make sense: You are trying to use information that > just isn't present in the data. In that case you could try a > random effects design, this way you can also use information > from variation between individuals, but the price is that > the individual specific error term is forced to be > uncorrelated with the observed variables. As the economists > say: there is no such thing as a free lunch. > > 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/ > -- Skype: ipng00 T: +65 6516 6807 * * 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/