Marginal Longitudinal Function-on-Function Regression

Music City Center 

We propose a novel inferential procedure for longitudinal function-on-function regression models. The method utilizes a marginal approach consisting of three steps: (1) fit pointwise longitudinal scalar-on-function regression models, (2) apply smoothers along the outcome functional domain, and (3) compute confidence bands for parameter estimates. A simulation study shows this approach provides accurate estimation and inference while being much more computationally efficient than existing approaches. Methods are motivated by a large physical activity study in older adults with data collected over multiple visits.

longitudinal functional data

physical activity

mixed models

smoothing