We introduce a new class of distributions to model directional data, based on hyperspherical log-splines. The class is very flexible and can be used to model data that exhibits features that cannot be accommodated by typical parametric distributions, such as asymmetries and multimodality. The...