We are working with panel data. But we want to study only the cross-section part of the panel data. So can anybody please tell me how to do any kind of data transformation, so that I can remove the time-series variance from the panel data?

Thank you.

  • 1
    $\begingroup$ Ari, what do you mean by "remove the time-series variance from the panel data"? You can certainly analyze the data as though they were cross sectional. And you can do so in ways that treats within-panel variance as a nuisance parameter (e.g. using a random intercept mixed effects model), but you cannot make the data not have been panel data in the first place, with the potential issues that may arise around, for example non-stationary/integrated processes in your data. $\endgroup$
    – Alexis
    Sep 3, 2014 at 19:22
  • $\begingroup$ Thanks Alexis for your comment. I wanted to "analyze the data as though they were cross sectional". But don't know how to do it. So I was thinking of data transformation. Can you suggest me any other method (other than using a random intercept mixed effects model)? Or refer me some literature. You can put that as an answer so that I can accept it. Thanks again. $\endgroup$
    – Beta
    Sep 3, 2014 at 19:45
  • $\begingroup$ OK, you simply analyze the data and ignore panel indicators. But that still does not address what you mean by "remove the time-series variance from the panel data". $\endgroup$
    – Alexis
    Sep 3, 2014 at 20:46
  • $\begingroup$ Ok. In panel data we have cross-section dimension & time-series dimension. Now, the total variation (or variance) of the dataset has both the cross-section & time-series variance. We can check variance of each of these component in sas by proc panel. So, I don't want to consider the time-series variance (or variation) when I'm analyzing my dataset. I check want to treat the panel data as cross-section data & proceed with my analysis. That's why wondering how to remove the time-series variation. $\endgroup$
    – Beta
    Sep 4, 2014 at 7:40

1 Answer 1


One idea would be to reduce the data to the cross-sectional dimension by running your analysis based on time-series averages. This is what statisticians refer to as the between estimator. For instance, if you want to find out what drives cross-sectional differences between your n individuals you could run a regression of the following type: $$ \bar{y_i}=\alpha+\beta \bar{x_{i}}+e_i $$ where $\bar{y_i}$ is the time series average of variable y of individual i and $\bar{x_i}$ is the respective time-series mean of variable x.

By taking time-series means you "remove" the time-series variation. I guess, to talk about variation rather than variance is to be preferred in this context.


Your Answer

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

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