Dear all
I'm using Stata 17.1. My dataset is panel unbalanced, with N=340. Observations correspond to political parties' positions in 25 countries across different elections. My model's dependent variable is shifts (from t-1 to t) in variable Y (or dY), and I have 9 independent variables . After running xtserial, I got that my data have autocorrelation (p=0.016), so I decided to run Prais-Winsten regression, with standard errors clustered by country and election. The model contains the 9 independent variables plus 24 country dummies. The model runs smoothly (prob > F=0.000, R squared =0.119 and rho= -0.552). However, in the results for the Durbin-Watson test, I get that the value of rho(original)=1.62 is higher than the value of rho(transformed)=1.15. In all the examples I've seen of Prais-Winsten regression, the transformed value is always higher. Does this imply that the model is wrong? Do you suggest any other alternative?
Thanking you in advance
Pepe
I'm using Stata 17.1. My dataset is panel unbalanced, with N=340. Observations correspond to political parties' positions in 25 countries across different elections. My model's dependent variable is shifts (from t-1 to t) in variable Y (or dY), and I have 9 independent variables . After running xtserial, I got that my data have autocorrelation (p=0.016), so I decided to run Prais-Winsten regression, with standard errors clustered by country and election. The model contains the 9 independent variables plus 24 country dummies. The model runs smoothly (prob > F=0.000, R squared =0.119 and rho= -0.552). However, in the results for the Durbin-Watson test, I get that the value of rho(original)=1.62 is higher than the value of rho(transformed)=1.15. In all the examples I've seen of Prais-Winsten regression, the transformed value is always higher. Does this imply that the model is wrong? Do you suggest any other alternative?
Thanking you in advance
Pepe
