A0803
Title: Generalized Laplace regression to model cylindrical responses with an application to physical activity in children
Authors: Marco Geraci - Sapienza University of Rome (Italy) [presenting]
Abstract: A multivariate regression model is proposed for cylindrical responses using the (symmetric) generalized Laplace (GL) distribution. This distribution has a shape parameter that captures the heaviness of the tails and includes the Gaussian and (classical) Laplace distributions as special cases. The likelihood is obtained by first projecting the scale-mixture representation of the GL onto the circle and then by numerically integrating it with respect to the latent random variance. In an application to accelerometer data collected from participants of the UK Millennium Cohort Study, children's physical activity behaviors are studied in free-living conditions. The outcome consists of two components, one related to the intensity of the physical activity (linear and univariate or multivariate) and one related to its timing over the day (circular and univariate). In summary, the proposed model represents a useful extension of models based on the normal and the projected normal distribution.
