Robust estimation in the negative binomial model with applications to musculoskeletal problems

In intervention studies aiming at reducing the number of falls the standard method of analysis is negative binomial regression. Unfortunately more often than not a small group of multiple fallers unduly influences the fit and inflates the dispersion parameter estimate used in sample size calculations.

The objective of this project is to propose a robust estimation approach for this model and compare it to existing techniques. Part of William Aeberhard’s PhD, Geneva  University, co-supervised by S. Heritier and E. Cantoni, Geneva University