The performance of mixed truncated spline-local linear nonparametric regression model for longitudinal data

A mixed estimator nonparametric regression (MENR) model is an additive model that involves a combination of two estimators or more in multivariable nonparametric regression. The model is used when there are differences in data patterns among predictor variables. This study proposes the development of the MENR model on longitudinal data namely a mixed truncated spline-local linear nonparametric regression (MTSLLNR) model. A modified weighted least square (WLS) method through two-stage estimation is used to estimate the regression function in the proposed model. To illustrate the performance of the MTSLLNR model, a simulation study with a sample size variation of subjects and time points is provided. Additionally, the MTSLLNR model is also applied to model the poverty gap index data. Both simulated and real data results suggest that the proposed model has consistency findings and good performance in longitudinal data modeling. Some highlights of the proposed method are: • The method combines two estimators of local linear and truncated spline to accommodate the differences in data patterns in the nonparametric regression for longitudinal data. • Selection of optimal knots and bandwidth using the generalized cross-validation (GCV) method. • The consistency findings and general performance of the method is shown by simulation and real data application.