A Semiparametric Model via Local Polynomial Smoothing for Unevenly Sampled Longitudinal Data

• Activity Number: 415
• Type: Contributed
• Date/Time: Tuesday, August 5, 2014 : 2:00 PM to 3:50 PM
• Sponsor: Section on Nonparametric Statistics
• Author(s): Lei Ye*+ and Ada Youk and Susan Sereika and Stewart Anderson and Lora Burke
• Companies: University of Pittsburgh and University of Pittsburgh and University of Pittsburgh and University of Pittsburgh and University of Pittsburgh
• Keywords: semi-parametric ; local polynomial ; longitudinal data

Abstract

Semi-parametric mixed models can be used for exploring longitudinal data when the sampling frequency of response and covariate are the same. A new semi-parametric model using local polynomial smoothing is proposed when the response is less frequently sampled than the covariates within the same time frame. We insert pseudo data points for the response at time points where only covariates are measured. We assume the unmeasured response values between two adjacent measured responses are on the straight line between the two measured responses. To apply the semi-parametric model, the parametric components model covariates that affect the response parametrically, and the nonparametric components model covariates that affect the response non-parametrically. For the nonparametric components, rather than a kernel distribution to assign weights, analytic weights that indicate the importance of the response data point were used. The weight used was the time distance between the inserted pseudo data points for the response and observed measurement points. The resulting estimated coefficients will have straightforward interpretation and individual response profiles can also be estimated.