Use of mixed revealed-preference and stated-preference models with nonlinear effects in forecasting