That's just the problem, neither the time dummies nor the time variable (indicating 2004, 2005 etc..) are significant..

> Date: Thu, 29 Jan 2009 17:54:27 +0100
> From: svetosch@gmx.net
> To: gretl-users@lists.wfu.edu
> Subject: Re: [Gretl-users] help
>
> Am 29.01.2009 17:32, Dennis van Amelsfort schrieb:
> >
> > I hope someone can help me..
> >
> > When I run a fixed and random effects model for my data, I obtain very
> > differing results my variable of interest when using time dummies or a
> > continuous year variable.. When I use a year variable my variable of
> > interest (AS5) is only just insignificant (p=0,15) nut when I use time
> > dummies it is totally insignificant (p=0,73). The hausman test is also
> > inconclusive when I use the year variable, but points towards fixed
> > effects when using time dummies.
>
> Time dummies of course are more flexible than a single time trend (if
> that's what you mean by 'continuous year variable'). So provided the
> time dummies themselves are significant, you seem to be looking at an
> omitted variable bias when you don't include the dummies.
>
> >
> > When I insert an interaction term between AS5 and MW (another variable
> > already in the model), both the interaction term and AS5 become
> > significant when using the year variable, but only the interaction term
> > becomes significant when using time dummies.
>
> Same thing here, the time dummies soak up a lot of variation that
> otherwise are (probably spuriously?) attributed to other variables.
>
> without having looked at your results (sorry, no time), IMHO keep the
> time dummies and throw out everything else that isn't significant.
>
> (that's of course a quick-and-dirty advice over the net, you probably
> should go see a local doctor -- ahem, econometrician)
>
> good luck,
> sven
> _______________________________________________
> Gretl-users mailing list
> Gretl-users@lists.wfu.edu
> http://lists.wfu.edu/mailman/listinfo/gretl-users

---

Express yourself instantly with MSN Messenger! MSN Messenger