-xtoverid- after robust Standard Errors

I have to choose between Fixed and Random effects to estimate a panel data model (T=5 N=182). I read that Hausman test doesn't work well in presence of heteroskedasticity and/or autocorrelation in the error terms. After I detected these problems in some models, I reestimated them using robust/clustered SE. I read that I should use the command -xtoverid- instead of -hausman-, given that the first is robust to these problems.

My doubt is, should I use -xtoverid- when I detect heteroskedasticity and/or autocorrelation in one of these estimators (FE or RE) or just when I detect it in both (FE and RE)?

Thanks in advance!