Panel data: repeated time values within panel

[Legov Sunil]

Dear Statalisters,

I have the following dataset:

company	industry	year	size_sboard	women
1	202	2017	16	0.25
1	202	2016	16	0.25
1	202	2015	12	0.33333333
1	202	2014	12	0.33333333
1	202	2013	12	0.16666667
1	202	2012	12	0.16666667
1	202	2011	12	0.16666667
1	202	2010	12	0.16666667
1	202	2009	12	0.16666667
1	202	2008	12	0.16666667
2	204	2017	6	0.33333333
2	204	2016	6	0.33333333
2	204	2015	6	0.16666667
2	204	2014	6	0.16666667
2	204	2013	6	0.16666667
2	204	2012	6	0.16666667
2	204	2011	6	0.16666667
2	204	2010	6	0
2	204	2009	6	0
2	204	2008	5	0
3	207	2017	12	0.16666667
3	207	2016	12	0.16666667
3	207	2015	12	0.16666667
3	207	2014	12	0.16666667
3	207	2013	12	0.16666667
3	207	2012	12	0.08333333

...and so on...

Trying to set up the following fixed effect regression

xtset industry year

I get the following error (which I understand)
repeated time values within panel
r(451);

Does anybody has a solution, how I can run a fixed effect regression controlling for industry and year?

Thanks in advance!

[Igor Paploski]

Hi Legov,

Most likely there are issues of repeated observations for year for industries not shown on the snippet of your dataset that you shared. Try running the following:

bysort year: distinct industry

The number of distinct observations for each year should be equal to the total number of observations of said year (in other words, you should not have repeated observations for year within industries).

Best;

[Legov Sunil]

Hi Igor,

thanks for your answer. Yes you are right, there are repeated observations for year within industries and I am totaly aware of this problem. I should have asked more precisely, sorry for that.

Give me a new try:

I want the following regression:

xtreg tobins_q size_sboard women_on_board.....

Originally I wanted to control for company and year. Since there many observations for companies, where for example the size of supervisory boards remains constant which reduces the expressiveness of the coefficient I came up with the idea to control for industries and years. But controlling for industries instead of companies leads to the problem of repeated time values.

Do you have a solution to solve this problem?

[Igor Paploski]

Hi Legov,

In Panel Data analysis, your rows are comprised of identifiers and time. Each combination of id and time must not occur more than once - a specific combination can not occur (those are called gaps), but two observations for the same id and time variables can not. This is what Stata is complaining to you. The regression you want to run (xtreg) requires you to set up your data as Panel Data, but you can't because you have repeated observations for the id x time combinations.

The way I would solve the issue is to find out why you have repeated id x time combinations. If they are truly repeated (all variables are the same for the rows that have the same values for id and time), than I would simply delete one the rows (because they are truly repeated). However, if the values for the other variables of the repeated observations are different, then either your time variable is not good (having two assessments of the companies, one in January and other in July and calling both "2016" is a simplification - you might want to refine this data or adjust how you record time in your dataset - semesters?) or your id variable is not good (you might have given different companies the same id, thus the repetition). In both scenarios, people will only be able to help you with the modeling per se (and I am not the best for that) after you fix these data related issues you have.

For more info look here.

Best;

[Phil Bromiley]

I'm not sure of your area, but many areas working on this kind of problem would expect panels to be at the firm level.

If you really want to work with industry panels, you can simply not set the year in the xtset. You won't be able to do lags, but will do a panel estimate.