Hi

[Ina Ode]

I am trying to run a fixed effect regression estimate. I dont know how to incorporate both sector and time dummies into one equation.

xtreg tobinq csr econ environ social size lvrg saleg i.year i.sector, fe. The time dummy works but the sector doesnt. I appreciate any help

Many thanks
Ina.

[Joao Santos Silva]

Dear Ina,

What do you mean when you say the sector dummy does not work? Is it collinear with the fixed effects?

Joao

[Ina Ode]

Dear Joao,
Thanks for the response, yes when I enter xtreg DepVar Var1 Var2 Var3 Var4 Var5 Var6 Var7 i.country i.industry, robust in stata it ommits all my sector due to collinearity

Many thanks
Ina

[Clyde Schechter]

There are two different possibilities here, depending on how you had -xtset- your data.

If sector is your panel variable (-xtset sector year-) then everything is fine. When you run any of the -xt, fe- models, Stata automatically identifies the panel variable fixed effects from the preceding -xtset- command, and incorporates it automatically. That is, you don't need to mention i.sector in your -xtreg, fe- command in this case. When you do mention it in the -xtreg, fe- command, it is redundant, and Stata lets you know that and omits it. This does no harm: you get the exact same results you would have gotten had you not, unnecessarily, included i.sector in your -xtreg- command. So you can just use the results you got and not worry about it. (Or, if you want "clean" output to look at, just re-run the same command without the i.sector term and you will get identical, but clean, results.)

If, on the other hand, sector was not the panel variable in your -xtset- command, then you have a problem. For example, if your panel variable in the -xtset- command was, say, firm, and if each firm is always in just one sector, then you are attempting to do the impossible. Once you include a fixed-effect at, say, the firm level, then you have automatically adjusted for the effects of any time-invariant attributes of a firm (such as the sector it is in) and it is, in principle, impossible to separately estimate the effects of any such attribute. This is not a peculiarity of Stata: it is an inherent property of fixed-effects regression. Generally speaking, one is often not interested in actually estimating such effects anyway, but only interested in adjusting for them. If adjusting for them is all you want, then it is already accomplished without mentioning i.sector in your command. (And as above, you can either go with what you have, or re-run without i.sector to get cleaner tables of identical results.)

But, if you need to actually estimate effects of such attributes (not just adjust for them) to accomplish your research goals, you must use a different model such as -xtreg, be- or -xtreg, re-, or a multi-level model such as -mixed-.

[Marcos Almeida]

Hello Ina,

Welcome to the Stata Forum.

You've got excellent advice so far, However, I believe you can get even more, provided you follow the FAQ recommendations. For example, you could present a short display of the data, plus the commands and the output, You can do this by using the CODE delimiters or with the SSC - dataex - as also informed in the FAQ.

To end, since the title is sort of a key term, very important to entice the attention of fellows with a particular expertise as well as those who share a similar problem, I kindly recommend to choose an informing title for the thread in the forthcoming messages. Thank you in advance.

Best,

Marcos

[Ina Ode]

Hi Clyde
Thank you very much for the excellent clarifications. Marcos sorry I am quite new to this forum. I will make sure to always put in the appropriate key term for the title.

Many thanks,
Ina.