Omitted Regional Dummies in FE Panel Data

[James Horton]

Hi,

Hoping someone can help. I am currently running a FE regression, assessing the impact of gender inequality in China on economic growth at the provincial level, so I have 30 provinces with data from 2000-2019.

I am looking to control for regions, as a proxy for technical efficiency, as the data is available for technical efficiency but other literature argues well that regions vary in technical efficiency and thus can use them as dummies to control for this.

I have manually created my dummies, following another threads advice. My regression is below:

"
global id p_code
global t year
global ylist ln_grp_pc
global xlist NC NEC EC CC SC WC ln_ngd ln_elec_con1 ln_fyschool ln_myschool ln_n_medical_beds1

xtreg $ylist $xlist, fe

"

where NC NEC EC CC SC WC are my dummies, for North, North East, East, Central China etc respectively.

They are dropped however due to collinearity, which I understand is because they do not vary over time. Is there a way I can control for the regions, to proxy for a measure of technical efficiency?

Any help would be greatly appreciated. Thanks!

[Clyde Schechter]

Well, you don't show how you -xtset- your data. But from your description, I'm guessing that you specified province as the panel variable. If I have that right, the issue is not that the regions don't vary over time: in fact I imagine each region occurs in combination with each year. But what will be true is that region is the same over time for any given province. Once you have specified province as your fixed effect it is not possible to estimate the effects of region: region is a time-invariant attribute of the province. This is not some quirk of Stata: this is linear algebra, and there is no getting around it.

As for "controlling" for region effects, there is no need to do that. In a fixed effects model with province as the panel variable, the effects of everything, including region, that is a time-invariant attribute of the province are automatically adjusted for--no explicit mention of those variables is needed.

So just forget about region and move on with the rest of your project.

[James Horton]

Will do, thanks for your help Clyde! James