Estimating Branching Curves in the Presence of Subject-Specific Random Effects

• Activity Number: 495
• Type: Topic Contributed
• Date/Time: Wednesday, August 6, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Section on Nonparametric Statistics
• Author(s): Sarah J. Ratcliffe*+ and Angelo Elmi and Wensheng Guo
• Companies: University of Pennsylvania and George Washington University and University of Pennsylvania
• Keywords: Branching curves ; B-splines ; nonparametric ; mixed effects

Abstract

Branching curves are used in nonparametric regression to estimate curves with time-varying treatment. For cross-sectional data, Silverman & Wood (1987) introduced a smoothing spline solution with a roughness penalty controlling the local variability at each branching point. However, in the longitudinal setting, this approach is difficult to implement in the presence of subject specific random effects. Instead, we propose a B-spline solution with finite support knots controlling the smoothness at the branching points. Estimates are found using a B-spline based semiparametric nonlinear mixed effects model with adaptive Gaussian quadrature (Elmi et. al. 2011). This solution results in a more straightforward and intuitive estimation of the average curve set. We illustrate the techniques using data from a labor and delivery study where the administration of the labor stimulant oxytocin results in a divergence in the estimated treatment curves.