Fixed effect: trend of a group of countries in panel data

[Kasper Trier Jorgensen]

Dear all,

I am currently working on a thesis project on railway competition in the EU. I am using a fixed effects in a panel data setting and the following command: xtreg Y Var1..., fe robust

I suspect that there are differences between Western and Eastern European countries. Thus, I would like to control for this by establishing a Western and Eastern group and afterwards estimate the effect of being in one group on my dependent variable. I have tried with a dummy variable but it was omitted due to collinearity.

Could anyone help me with how to proceed?

Thank you very much!
Kind regards,
Kasper Trier

[Kasper Trier Jorgensen]

I have tried with the i.WestEast##c.Year command, where WestEast is a dummy variable, which is 1 when it is a Western country. Would this be a correct method?

xtreg ParlSHR OAcomp Gasprice GDPcap_log HighspeedrailSHR Raildensity Employmentrate Fincrisisdummy Verticalsep Horizontalsep Complevel DejureComp i.WestEast##c.Year, fe robust

[Clyde Schechter]

You cannot test for differences between Western and Eastern countries in a fixed effects model where the panel variable is the country. Every country is always Western or always Eastern, and the fixed-effects model is only capable of estimating effects of things that vary across countries. There is no way at all to estimate that difference in a fixed-effects model, so don't waste your time trying to figure out a way. You can't get around linear algebra. If this question of East vs West difference is important, you need a different model altogether. I would recommend either a random effects model -xtreg, re-, or, a hybrid model. The latter can be estimated using the -xthybrid- command, by Francisco Perales and Reinhard Schunck, available from SSC.

The code you show in #2 is legal. Whether it is correct depends on what you are trying to accomplish. It does not distinguish the level of ParlSHR between Western and Eastern countries. But it does tell you whether the rate of change in ParlSHR per year is the same in the Western countries as it is in the Eastern one. If that is of interest to you, and if it is reasonable to constrain the model to the assumption that ParlSHR actually grows (or shrinks) linearly with time, then you have estimated that correctly.

[Kasper Trier Jorgensen]

Thank you very much for your fast reply and help!
Ok.
My attempt was then not what I wanted to do I believe. Because it would be a too imple assumption for it to be growing linearly.
It's not of utmost importance, it was just to replicate a previous studies, where the trend of the two groups were controlled for.

[Carlo Lazzaro]

Kasper:
as an aside to Clyde's excellent advice, you may want to invstigate whether time, as a continuous predictor, has within panel non-linear relationship with the regressand:

xtreg ParlSHR OAcomp Gasprice GDPcap_log HighspeedrailSHR Raildensity Employmentrate Fincrisisdummy Verticalsep Horizontalsep Complevel DejureComp c.Year##c.Year, fe robust