Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: RE: Panel data

From   Klaus Pforr <>
Subject   Re: st: RE: Panel data
Date   Wed, 01 Feb 2012 13:13:58 +0100


Dear James,

you can aggregate the duplicates by adding them up to one observation per year within firm. Alternatively, you can assume an order in time more fine grained than years (or find data supporting this assumption), to have e.g. firm X quarter data.

There is also statistical detail connected to this:
If you don't use lagged or lead variables and you don't specify a correlation in the error structure and you don't use first-difference-models (or similar models), you can ignore the time completely. For many nondynamic panel models you basically only compare coincidences of the dependent variable and the independent variables, regardless of the time order. You indicate with the id which observation belongs to the same panel/cluster with the same heterogeneity.
In practical terms this means that you only specify id with xtset:
xtset firmid

I guess if you stick to the utility functions provided by Stata, you will either be able to compute the model you want as you apparently do not need the time order or you will run in an error, as you will have to specify a time order with xtset (e.g. tsfill)



Am 01.02.2012 12:43, schrieb Nick Cox:
It seems that all you can reasonably do is -xtset firm_id- but you then still need to worry about what generating process assumptions match your situation and quite how your fixed and random effects arise. I have no advice on that.


Alexander James

I am working on a database that has firm year observations. Each observation regards the degree of familiarity that a firm has with a new acquired technology. Accordingly the dependent variable measures the number of new products released in year t2.

I am trying to run a FE or RE model capturing the effects of familiarity on the dependent variable (Probably a negative binomial). However, the problem is that I have multiple observations per year. In other words, the same firm may acquire more than one technology in the same year, that will have different degrees of familiarity.

When I try to run the xtset firm_id year, I get the message repeated time values within panel. I read some topics explaining how to correct in case it is a mistake, but in my setting that is how the data really is.

Would someone have any suggestion how to approach this issue?

*   For searches and help try:


Klaus Pforr
Universität Mannheim
D - 68131 Mannheim
Tel:  +49-621-181 2797
fax:  +49-621-181 2803

Besucheranschrift: A5, Raum A309

*   For searches and help try:

© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index