Hello,

I am struggling with my data panel analysis. I am conducting a research with information of 20 cities in a 6-year period. Before deciding if using fixed effects (fe) o randome effects (re) for the regressions, I run the Hausman test to see what the p-value is. The issue that I found is that, when running the FE regression, the output of the software indicates collinearity among CITIES (not among variables). Even if a try to orthogonalize variables, substract their mean, apply the logarithm or combine variables differently, I do not get to eliminate collinearity.

Well, my question then is: can I use RE for my regressions even if I find collinearity running FE ? if yes, could you recommend some literature to justify its use ?

Thank you very much in advance.