The choice of approximative models in nonlinear regression