Model selection for circular-circular models

Abstract

Circular-circular models, used for situations where a circular response is to be predicted using another circular variable, are frequent in the circular statistics literature as they appear in many practical situations. One of these is that of trying to relate the peak expressions of cell-cycle genes coming from two different species with possibly di erent cell phase lengths. Several different models have been considered for these situations. Some of them, such as those in Downs and Mardia (2002), Rueda et al. (2009), Di Marzio et al. (2013) or Rueda et al. (2016), will be described in the talk. We will see that it is not easy to compare and select among them as some involve non-standard features. To overcome this problem, we propose a selection criteria based on the work by Ye (1998) for the Euclidean setting. We will check the performance of this criterion for the aforementioned cell-cycle data.