|
Abstract:
|
Self modeling regression, a method for functional data, is based on the simple assumption that the x and y axes can be separately shifted and scaled for each curve so that the data from all curves lie approximately on one typical curve. We propose to model the typical shape with a regression spline, and to treat the shift and scale parameters as random with a restricted covariance structure to ensure computational identifiability. The resulting nonlinear mixed effects model involves random effects that are correlated among experimental units, unlike random effects in usual mixed effects models, but the model can still be implemented using standard software. We describe our approach and apply it to pulse waveforms (where variability of the amplitude shift parameter represents short-term variability in blood pressure) from a large observational study of cardiovascular disease.
|
