Title: Bayesian modelling of underreported count data
Authors: Michaela Dvorzak - Joanneum Research Forschungsgesellschaft mbH (Austria) [presenting]
Helga Wagner - Johannes Kepler University (Austria)
Abstract: Regression models for count data subject to underreporting in a Bayesian framework are considered. We specify a joint model for the data-generating process of true counts and the fallible reporting process, where the outcomes in both processes are related to a set of potential covariates. Identification of this joint model is achieved through additional information on the reporting process provided by validation data and incorporation of variable selection in both parts of the model. For posterior inference, an MCMC sampling scheme is implemented, which is based on data augmentation and auxiliary mixture sampling techniques. The proposed method is illustrated using simulated data and applied to a real data set.