B1357
Title: Robust parameter estimation in discrete data
Authors: Max Welz - Erasmus University Rotterdam (Netherlands) [presenting]
Abstract: Just like continuous data, discrete data can be contaminated by anomalous observations that, if unaccounted for, may cause large biases in parameter estimation. For instance, in rating-scale questionnaires, participants may not pay attention, or in grouped data, the frequency of some classes may be inflated. A unifying approach is proposed for robustly estimating statistical functionals in possibly multivariate discrete data, such as location, scale, and association. The estimator is root-$n$ consistent, asymptotically normally distributed, and depending on the choices of tuning parameters, can achieve asymptotic efficiency. In addition, various robustness properties of the proposed estimators are derived, such as bias curves and influence functions. The estimator's properties are verified by extensive simulation studies and demonstrate its practical usefulness by means of an empirical application.
