Semiparametric mixtures of nonparametric regressions (Q158425)

The authors propose a new class of semiparametric mixtures of regression models. Regression spline estimates are considered and the authors claim their simplicity. The proportions and variances remain constants and the component regression functions are considered as smooth functions of a certain covariate. Its identifiability is determined and the proposed estimation procedure seems to have an optimal convergence rate for both the global parameters and the nonparametric regression functions.