Modeling shape distributions and inferences for assessing differences in shapes

AbstractThe general class of complex elliptical shape distributions on a complex sphere provides a natural framework for modeling shapes in two dimensions. Such class includes many distributions, e.g., complex Normal, Watson, Bingham, angular central Gaussian and several others. We employ this class of distributions to develop methods for asserting differences in populations of shapes in two dimensions. Maximum likelihood and Bayesian methods for estimation of modal difference are developed along with hypothesis testing and credible regions for average shape difference. The methodology is applied in an example from biometry, where we are interested in detecting shape differences between male and female gorilla skulls