County, time-fixed effects

[Max Hurd]

Hello,

I am currently using Stata for my senior thesis and had some questions about fixed-effects. Currently I have county-level data on 6 states over the years 2007-2012, with monthly observations for each state. Total, I have 41,760 observation.

To give a better idea of my dataset, this is how I have it:

Buyer_location	month	year
xyz, ga	1	2007
xyz, ga	...	...
xyz, ga	12	2012
123, ga	1	2007
123, ga	...	...
123, ga	12	2012
abc, nc	1	2007

I want to make sure I am setting up my fixed effects correctly.

egen id = group(Buyer_location)

reg y x1 x2 i.id i.year#i.month

Is this the correct way to go about this? Any insight would be greatly appreciated.

[Carlo Lazzaro]

Max:
welcome to this forum.
If you're dealing with a panel dataset and you have a continuous regressand, your first choice should be -xtreg-.
An aside, I fail to get why you interact -year- with -months- (and do not include the conditional main effect of those two predictors in the right-hand side of your regression equation).

[Max Hurd]

Hi Carlo,

Thanks for the welcoming.

I was under the impression that my id variable would be my "County Fixed Effect" and the grouping of month/year would be my "Time Fixed Effect". Is that not the best way to go about this?

[Carlo Lazzaro]

Max:
I have a different take about that.
Your -panelid-="Buyer". I would consider -location- as a predictor. This way I will capture -Buyer- fixed effect.
Your -timevar"=

egen new_timevar=group(year months)

I would start with -xtset-ting your data:

xtset Buyer new_timevar

Then, provided that a panel-wise effect does exists (see -xtreg postestimation-, you should test which specification fits your data better than the other oone (ie, -fe- or -re-)?
An aside, I still fail to get why you interacted -year- with -months-.