2

Why can't I find nowhere, neither in books or help files on the internet, graphs of the "in between" estimators and the "within" estimator?

Imagine you have the logwages of 10 persons and also their work experience. Why does nobody show these ten lines of this panel data in a meaningful graph?

Is it correct that the within estimator would produce, in combination with the calculated 10 fixed effects, 10 lines that go as good as possible through each of the ten salary versus experience lines?

And is it correct that the in between estimator would produce a regression line going through the averages of experience and salaries? But what would happen then if all averages x's are equal?

In the graph below you can find 8 observed "experiences" versus "log wages", can you tell me my thoughts are correct?

First try

Kasper
  • 3,399
  • The plots obtained by estimating the within estimators would just be parallel lines since they differ only in the intercept. I think what you have in mind is plotting the predicted values as against plotting the estimators – Abhimanyu Arora Jan 25 '14 at 20:34
  • Is experience equivalent to time here? – Michael M Jan 25 '14 at 20:37
  • Experience is not equivalent to time, the logwages are taken at eight times, and every time the years of experience of the person is recorded to. – Kasper Jan 25 '14 at 20:39
  • I am happy to understand that the within estimators are parallel lines. But how does the "in between" estimator look? Is it just regressed on the dots (they are the averages of the persons experience and wage) – Kasper Jan 25 '14 at 20:40
  • If the individual points are panel averages, yes the between estimator would be plot through them – Abhimanyu Arora Jan 25 '14 at 20:56
  • And what happens than if all panel averages have the same average x (in this case, everybody has the same average experience)? Then it is impossible to draw a regression line? – Kasper Jan 25 '14 at 20:59
  • Yes, that's true (for both FE and BE) – Abhimanyu Arora Jan 25 '14 at 21:22

0 Answers0