A cautionary note on the use of singular perturbation methods for “small noise” modelst †