On April 20th, the seminar series welcomed Philip Cooney, a PhD student in Trinity College. Philip spoke about novel methods for incorporating expert opinion in statistical analysis. Details of Philip’s talk are below. The manuscript on which the talk is based is available here and Philip has also prepared a Github repository with an implementation here.
Title
Incorporating Expert Opinion in Statistical Analysis
Abstract
Inclusion of Expert Opinion is particularly desirable in situations where data is sparse. Collection of expert opinion requires careful consideration of psychological biases, however, general consensus is that it is a valuable endeavour and elicitation should be in the observable space. There are a number of different approaches for encoding observed information onto a model parameter space, however, these are in general model specific methods which require specific steps or algorithms. We present a flexible method which can be motivated as either penalizing the likelihood or general Bayesian update using a loss function. This approach can be implemented with minimal effort within Frequentist and Bayesian methods and for a wide variety of observable outcomes for which probabilities may be elicited.