Remarks on the relaxation-time approximation