Lets compare three panel most basic models: pooled, random effects, fixed effects.
I observe common belief, that the fixed effect model estimation is "less biased", than in the other models. The only cost of this model is said to be less effectiveness of this estimator. I can not give source of this, since it is commonly repeated by applied scientist I know.
I wonder if there are described situations, when fixed effect estimator is most biased? In other words, can we unconditionally trust Hausman test which model should we choose?