A0391
Title: To the limit and beyond for the frequency modulated Mobius periodic regression model
Authors: John Kent - University of Leeds (United Kingdom) [presenting]
Charles C Taylor - University of Leeds (United Kingdom)
Norah Almasoud - University of Leeds (United Kingdom)
Abstract: The frequency-modulated Mobius periodic regression model is an elegant and tractable model to describe how a real-valued response depends on a periodic explanatory variable, e.g. time with a daily cycle. The underlying idea is to warp time using a Mobius transformation and then to model the expected response as a first-order Fourier function of the warped time. Although the model is simple to describe and straightforward to fit for "nice" data, there are various subtle features worthy of deeper study. First, two summary measures are introduced to help interpret the effects of the parameters on the shape of the regression function. Second, limiting (and beyond the limit) versions of the model are developed to clarify and extend the range of possible behaviors covered by the model. Finally, several issues related to estimation are discussed, including multimodality, singularity and reparameterization.
