how to set up panel data xtset

[Daniel Hu]

Hi, I am new to STATA and would need some help. I have a large set of data that looks like the following, with multiple firms per year. How can I xtset? When I tried this " xtset ID Year, yearly", there is an error "repeated time values within panel". So I decided to go ahead with "xtset ID" instead. But I am not sure whether am I doing it right, I guess not because when I run a regression"xreg, fe" I have 6000 observations missing, furthermore, there are two variabes that are omitted due to collinearity. Thanks in advance for your help.

ID	Year	Name
1	2000	ABC
1	2001	ABC
1	2002	ABC
1	2003	ABC
2	2002	CDE
2	2003	CDE
2	2004	CDE

[Jorrit Gosens]

You have multiple observations for some combinations of Year and company ID, and Stata wants one at most. See further explanation and remedy here, note in particular the suggestions on 'duplicates'. Tag these, and figure ou what you want to do with these.

[Daniel Hu]

Hi Jorrit,

The problem is those are not duplicates, but they are observations that are required.

[Jorrit Gosens]

What Stata is telling you is that for some ID's, there are multiple rows with the same year, and Stata cannot decide which to use.
These observations are duplicates, at least in the sense that ID and Year are the same in multiple rows. Perhaps you mean to say that they are 'not duplicates', because values in other variables are not the same? If that is the case, you are going to have to think of a way to resolve this, because Stata cannot decide which of the two rows to use.

Did you try to use below code, and did it report observations with 2 or more copies?

duplicates report ID Year

[Phil Bromiley]

It depends a lot on what you want to do with the within-panel analysis. If you don't care about the order of the observations (i.e., the years), then you can xtset ID [without the time indicator] and then use xtreg. This gives you the panel effects but ignores time completely. While if you xtset ID year, xtreg will not accept duplicate years, I don't see that this actually modifies the estimates. I think the actual estimate from xtreg will be the same with and without the time variable, assuming you end up using the same observations and don't want to mess with more sophisticated issues (like serial correlation).

For fixed effects, if you just want different intercepts for each panel, then you can also use regress with i.ID or areg.

This does not preclude you putting in dummy variables for years in any of the analyses.

Alternatively, you could generate an artificial time variable by enumerating the observations within panels and using that enumeration as the "time" variable in xtreg. It would like something like:

sort ID Year
by ID: g seq=_n
xtset ID seq
xreg ...

Given more than one observation per year, I'm not sure how you should handle time. The discussion above assumes you ignore time or just have dummy variables for year.

With more than one observation per year, I don't think the built-in procedures would properly estimate serial correlation. You may have some very curious error structures (since some observations differ by years and others are in the same year).