tappek
commented
2 years ago ad 1)
As you describe your data, they does not vary over time for the dependent variables. Computationally, this results in 0 values for the within transformation, hence model = "within" does not have anything to work with and stops rightfully with error message "empty model". The random effect estimator as per Swamy-Arora uses the variation of the associated within model and the associated between model. For a plm-based exposition see one of the package's vignettes https://cran.rstudio.com/web/packages/plm/vignettes/B_plmFunction.html, section "Unbalanced panels" (but this is not specific to unbalances panels). Any good text book about panel models will cover this, e.g., Wooldridge or Baltagi. Other random effect estimators like Wallace-Hussain use slightly other "base models" (but Amemiya's estimator uses the within model twice), see Baltagi's text book for an overview.
So, setting for the RE case, e.g., model = "random", random.method = "walhus" should work as this does not entail estimating a within model.
ad 2) It is unclear to me what more there could be to see if there is no time variation. Could you explain further what you expect to see? You could run the within transformation on the affected variables and see that they will be zero after transformation.
GokuuQ
Hi,
In my input data, I have 32 time periods and thousands of individual study areas. But the independent variables are 32-times repeated same data. So only the dependent variable varies over time. In my case, I have two questions when I apply the plm package:
Error in plm.fit(data, model = models[1L], effect = effect) : empty modelThanks!