Different panel data can have very different "shapes", e.g. regarding the length and resolution of the time series component (yearly, daily, hourly intervals) or the number of individuals and cross-sectional variables. When I read about panel data methods the examples mainly refer to economic data of different countries (N) over a number of years (T). Both N and T are usually <100.

The panel data set I am working with has a comparably high temporal resolution (days) and thus the time-series component is predominating. The number of individuals and cross-sectional variables is comparably small.

My questions are:

1) Is there any existing categorization of panel data with respect to its shape? I have heard the expression "long panels" once but do not know exactly what it means.

2) There are references explaining which regression method (FE,RE,pooled) to use, but is there any intuitive preference for a dataset like mine?

Thanks in advance!


1 Answer 1


this is not going to be a huge help for you, but actually the number of N's and T's does not speak for or or against FE or RE in your case. In FE you should avoid to have too few T's, but that refers to datasets with T<5 for example.

In general you should take distance from pooled-modells when working with longitudinal data, because they are going to be distorted in nearly every case.

If you use FE or RE should rather be a decision based on what question you want to answer, and in special: which independet variables you would like to use. Because FE can not handle time-invariant variables which seems like a disadvantage but is actually a plus for real longitudinal questions.

Best wishes, Marcel


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.