Am 17.07.2009 21:12, Iveta Paterova schrieb: > Hello everybody, > it may sound silly, but I simply cannot understand why when calculating > fixed effects panel estimation, some variables are "omitted due to the > collinearity" and the constant remains. Should not also constant disappear > from the estimation - as it bears the fixed effects??? > I am calculating log-gravity models and the important variables are > omitted. What do I do wrong? > How can I get know the coefficient of omitted variables? IT IS IMPROTANT > FOR MY ESTIMATION? > > Thank you for your help. > Iveta, it sounds as if the omitted variables only vary in the cross-section dimension, not over time. That you collinearity then is a general problem of a fixed-effects specification, not of gretl's implementation. You might try a random-effects model, but if that happens to be rejected as biased, then you're out of luck. You would then need to think of a different approach to get your coefficient estimates, probably by checking out what textbooks like Wooldridge suggest. good luck anyway, sven _______________________________________________ Gretl-users mailing list Gretl-users(a)lists.wfu.edu http://lists.wfu.edu/mailman/listinfo/gretl-users

On Fri, 17 Jul 2009, Iveta Paterova wrote: > thank you for your answer. The problem is that every literature > recommeds fixed effect model as a correct estimator. It is a > gravity equalition, therefore some county-specific variables > remain and do not chage over the time. You cannot include regressors that are time-invariant in a fixed effects specification.  Mathematical fact of nature, doesn't matter what software you are using.  Please see the discussion in Chapter 15, "Panel Data", in the Gretl User's Guide. The fixed effects model sweeps out between-group variation in either of two ways: by including dummy variables specific to each of the groups, or by subtracting the group means from the regressors. On the dummy variables approach, a time-invariant variable (for example, something basically geographic) will (inherently) be perfectly collinear with the group dummies, it's just another group-specific constant.  On the approach of subtracting the group means, such variables will collapse to nothing but zeros, since every observation has value equal to the relevant group mean, If anyone told you the fixed effects specification would work with such variables, they were misinformed. Allin Cottrell _______________________________________________ Gretl-users mailing list Gretl-users(a)lists.wfu.edu http://lists.wfu.edu/mailman/listinfo/gretl-users

On Sat, 18 Jul 2009, Iveta Paterova wrote: > thank you Allin for your reply.But I still do not know which estimator > to use for gravity if OLS seems to be unappropiate (low R squared) for > gravity equation? A low R-squared value is not in itself a signal that OLS is inappropriate.  But an alternative is the random effects model. Allin Cottrell _______________________________________________ Gretl-users mailing list Gretl-users(a)lists.wfu.edu http://lists.wfu.edu/mailman/listinfo/gretl-users