|
Abstract:
|
We model functional data from multiple subjects with a regression spline linear mixed model. In this model, the expected values for any subject (conditioned on the random effects) can be written as the sum of a population curve and a subject-specific deviate from this population curve. The population curve and subject-specific deviates are modeled as b-splines with k and k' knots located at t_k and t_k', respectively. We sample from the posterior p( k, t_k, k', t_k' | y), where y is the observed data, using reversible jump MCMC methods. Sampling from this posterior distribution is complicated by the fact that no analytical form for p(y | k, t_k, k', t_k') exists. We explore two approximations to this likelihood and study how each penalizes linear mixed models with too many random-effect knots.
|
