Good day,
I'm working with a panel dataset, I've used many models,
homogeneous (fixed effect, pooled ols and Driscoll and Kraay)
heterogeneous (swamy random coefficients) and would like to do a
post-estimation to select the model that best fit my regression.
Is there any method, command that may allow me to do this?
Any hint will be highly appreciated.
Ama
Tell me more
×
Cross Validated is a question and answer site for
statisticians, data analysts, data miners and data visualization experts. It's 100% free, no registration required.
|
|
|||||
|
|
According to my experience, model selection is not so much about seeking the model that best fits the regression. The first question to ask is : which model reflects my experimental design the best? The second question would then be : which model reflects the covariance structure in my data the best? Only then you can start worrying about which model fits the data the best. There are a number of approaches to evaluate and comppare non-nested models. A naive approach would be to do a cross-validation and compare those results. One could eg do a leave-one-out crossvalidation, get the SS values for each run, and treat these SS values as sample data for model comparison. There's a whole set of literature on comparing non-nested models, but that's always within the same framework. There's little possibilities for comparing models originating from different frameworks other than working with some general loss function. |
|||||||||||
|
