Being more precise: I have four regression models differing each other only for the dependent variable (always a continuous one), whereas independent variable are the same. Breusch-Pagan test and Hausmant test tell me to adopt different structure of the sample, i.e. once they suggest to adopt fixed-effects, then when I change the dependent variable these tests suggest to adopt random-effects, and when I change again the type of dependent variable tests suggest to adopt pooled OLS. So the question is: is this normal or usual? or there's something wrong in what I did? Thanks evereybody which is expert in panel vs. pooled regression model.
When you say these tests "suggest to adopt" some model, what do you mean?
A test shows that your assumptions are violated. If it shows p-value
Fabio -
I am not completely familiar with your situation, but I have had plenty of experience with linear regression for use in 'prediction' for establishment survey estimation for official statistics, for continuous data, and experience with hypothesis tests. Heteroscedasticity is a matter of degree, not really 'yes' you have it, or 'no' you do not. I found it a natural part of the error structure in my work:
https://www.researchgate.net/publication/263036348_Properties_of_Weighted_Least_Squares_Regression_for_Cutoff_Sampling_in_Establishment_Surveys
As for hypothesis tests in general, they can be very misleading, and are often misinterpreted:
https://www.researchgate.net/publication/262971440_Practical_Interpretation_of_Hypothesis_Tests_-_letter_to_the_editor_-_TAS
So you use the same independent variables, and look at different dependent variables? Well, as I said, my expertise is not exactly in your area, but I am not surprised that different y would relate differently to the same regressors. And, as noted above, it is all a matter of degree anyway.
Perhaps you should look at the estimated variance of the prediction errors to assess your results. Also, I think it best if you always consider heteroscedasticity.
Cheers - Jim
PS - Would any scatterplot graphics help you to assess your situation?
Article Practical Interpretation of Hypothesis Tests - letter to the...
Article Properties of Weighted Least Squares Regression for Cutoff S...
- Badges
- Science topic