Dear all,
I'm running a fixed effect estimation (each observation is a country-year combination) with cluster-robust standard errors at the country level. The most basic model includes the income shares of each quintile of the population as regressors. All of the quintiles are significant, which is what theory would predict. But the joint test for model significance varies widely for the different quintiles, ranging from .0053 for one of the quintiles to .436 for another. In all other estimations ( including other regressors -just one or two more) the joint significance is well below the 1% level. I'm puzzled by the results of the lack of joint significance of some quintiles in the basic model, can someone explain why the significance level would vary so much. is there another test I can run? Any insights would be greatly appreciated.
I'm running a fixed effect estimation (each observation is a country-year combination) with cluster-robust standard errors at the country level. The most basic model includes the income shares of each quintile of the population as regressors. All of the quintiles are significant, which is what theory would predict. But the joint test for model significance varies widely for the different quintiles, ranging from .0053 for one of the quintiles to .436 for another. In all other estimations ( including other regressors -just one or two more) the joint significance is well below the 1% level. I'm puzzled by the results of the lack of joint significance of some quintiles in the basic model, can someone explain why the significance level would vary so much. is there another test I can run? Any insights would be greatly appreciated.
Comment