Comparison of spline estimator at various levels of autocorrelation in smoothing spline non parametric regression for longitudinal data. Issue 21 (2nd November 2018)

• Record Type: Journal Article
• Authors: Achmad Rinaldo Fernandes, Adji; Janssen, Paul; Sa'adah, Umu; Solimun, ; Effendi, Achmad; Nurjannah, ; Amaliana, Luthfatul
• Is Part Of: Communications in statistics. Volume 47:Issue 21(2018)
• Page Start: 5265
• Page End: 5285
• Publication Date: 2018-11-02
• Subjects: Smoothing spline -- Longitudinal -- Autocorrelation -- Non parametric regression; 62G08; Mathematical statistics -- Periodicals; Mathematics; Statistics; 519.2
• DOI: 10.1080/03610926.2017.1388404
• Languages: English
• ISSNs: 0361-0926

Abstract

ABSTRACT: The purpose of this research are: (1) to obtain spline function estimation in non parametric regression for longitudinal data with and without considering the autocorrelation between data of observation within subject, (2) to develop the algorithm that generates simulation data with certain autocorrelation level based on size of sample ( N ) and error variance ( EV ), and (3) to establish shape of spline estimator in non parametric regression for longitudinal data to simulation with various level of autocorrelation, as well as compare DM and TM approaches in predicting spline estimator in the data simulation with different of autocorrelation observational data on within subject. The results of the application are as follows: (a) implementation of smoothing spline with penalized weighted least square (PWLS) approach with or without consideration of autocorrelation in general (in all sizes and all error variances levels) provides significantly different spline estimator when the autocorrelation level >0.8; (b) based on size comparison, spline estimator in non parametric regression smoothing spline with PLS approach with (DM), or without (DM) consideration of autocorrelation showed significantly different result in level of autocorrelation > 0.8 (in overall size, moderate and large sample size), and > 0.7 (in small sample size); (c) based on level of variance, spline estimator in non parametric regression smoothing spline with PLS approach with (DM), or without (DM) consideration of autocorrelation showed significantly different result in level of autocorrelation > 0.8 (in overall level of variance, moderate and large variance), and > 0.7 (in small variance).