I am conducting a study with three energy sectors over 15 years of timeline. I intend to perform a panel data regressions.
I wanted to confirm if there is any rule which says that I cannot take more than three regressors or interdependent variables in the panel model equation because my cross-section observation number is only three (three energy sectors).
Since you have 45 observations in total, if they are yearly, I see that using three regressors or a few more is fine. However, the number of observations should be sufficiently greater than the number of regressors to ensure the estimators' desirable properties.
If you want to perform panel data regressions with conventional estimation methods (fe, re, ols), then you may be not be able to use more than 3 regressors, as the degree of freedom will become a negative number.
However, now there are some new techniques which could help you solve this problem, such as LASSO (least absolute shrinkage and selection operator), though you need time to learn how to perform that in the statistical software.
I do not think so, but you must have a sufficient number of observations for reliability and accuracy of the data model.
No, it is not mandatory that the cross-section observation numbers should always be more than the number of regressors. The number of observations required depends on the statistical power of the test and the effect size.
- Badges
- Science method