Panel regression error: year is not regularly spaced

[Daniel Huynh]

Dear all,

I have a data set of 2763 firms ỉn 3 years 2011, 2013, 2015. I considered it as a panel data: id is firm id, time space is 2 years. I want to run Discroll - Kray estimation by xtscc command but the error given back is as follow:

year is not regularly spaced: there are contemporaneous gap(s) across all subjects in id
r(101);

Maybe the problem is that my time space is 2 years instead of 1 year. Since I'm new to Stata, I could not find a proper solution to this problem.

I look forward to be supported by you all to complete my graduation thesis.

Thanks and best regards,

[Nick Cox]

Studying

help tsset

or

help xtset

points up that you should specify delta(2) as an option on whichever of those commands you used.

If you did do that, then it's a misfeature of xtscc not to cope.

But which xtscc are you using? You're asked to explain where community-contributed commands come from (FAQ Advice #12). In this case, the version on SSC is more up-to-date than that in the Stata Journal. At worst, create your own time variable that runs 1, 2, 3

egen time = group(year), label

and xtset or tsset in terms of that (now the delta() option is not needed). That might help before you use xtscc.


Whether panels with 3 observations in each are good for Driscoll-Kraay I really don't know.

I look forward to be supported by you all to complete my graduation thesis.

You're welcome here. But on Statalist, or any other similar site, the key for support is having a question that someone wants to answer, and nothing else.

[Daniel Huynh]

Studying

help tsset

or

help xtset

points up that you should specify delta(2) as an option on whichever of those commands you used.

If you did do that, then it's a misfeature of xtscc not to cope.

But which xtscc are you using? You're asked to explain where community-contributed commands come from (FAQ Advice #12). In this case, the version on SSC is more up-to-date than that in the Stata Journal. At worst, create your own time variable that runs 1, 2, 3

egen time = group(year), label

and xtset or tsset in terms of that (now the delta() option is not needed). That might help before you use xtscc.


Whether panels with 3 observations in each are good for Driscoll-Kraay I really don't know.



You're welcome here. But on Statalist, or any other similar site, the key for support is having a question that someone wants to answer, and nothing else.

Dear Mr. Nick Cox,

Thanks for prompt support. May I ask one more question? My data has small T and large N. When I use Hausman test, it recommends to use FEM rather than REM. However, the regression of FEM reveals that 4/6 independent variables and 1/5 control variables are significant. On the other hand, when using Pooled OLS with i.year i.industry dummies, it reveals 10/111 signficant variables. Therefore, could I consider regression model with year and industry dummies as a fixed effect model?

Thanks and best regards,
Daniel

[Nick Cox]

That's a new question -- the test is that the thread title no longer applies -- and should be asked in a new thread. I have no idea what to advise myself.