Interpretation of Dummies in Pandeldata-Estimation

[Marc Almand]

Hi, I have the following question. How do I interpret time dummies in panel data?

I have the following model:

reg lncigs tax income i.year, r

The dataset contains data from 2000-2020.
Logically, the dummy variable from 2000 is removed in the estimation. The coefficient on the year.2001 variable is 0.012. Now how do I interpret this coefficient? He the number of smoked cigarettes in the year 2001 about 1.2% higher than in the year 2

[Carlo Lazzaro]

Marc:
welcome to this forum.
Your code has nothing to do with a fixed effect panel data regression, amd should rather be:

 
 reg lncigs tax income i.year i.panelid, vce(cluster panelid)

That said, I would start out with -xtreg,fe-.

[Marc Almand]

Marc:
welcome to this forum.
Your code has nothing to do with a fixed effect panel data regression, amd should rather be:

reg lncigs tax income i.year i.panelid, vce(cluster panelid)

That said, I would start out with -xtreg,fe-.

Hey Carlo, thank you. I will take a look into it. But how do i interpret the coefficients of the dummies?
And can i use the "vcec(cluster)" everytime?

[Carlo Lazzaro]

Marc:
1) within a -fe- setting, -i.year- gives the contribution of time to within-panel variation of the regressand (when adjusted for the other predictors);
2) if you go -fe- panel data regression with OLS -vce(cluster panelid)- is mandatory, as your observations are not independent and Stata does not know that you're carrying out a panel data regression, as -regress- is not an -xt- suite commands.
If you switch to -xtreg,fe- to go cluster-robust standard errors you can invoke either -robust- or -vce(cluster panelid)- as (unlike -regress- where -robust takes heteroskedastcity only into account) both options do the very same job (that is, taking both heteroskedasticity and autocorrelation into account).
About the smallest number of panels/groups that allows non-default standard error with no worries, there's no hard and fast rule, but some intersting papers:
a) http://cameron.econ.ucdavis.edu/rese..._February.pdf;
b) https://arxiv.org/abs/1710.02926.