VIBASS3 - Invited Talks
Registration form

July 20, 2018

VIBASS3 - Invited Talks

Invited Talks

Bayesian modelling of environmental DNA data

Environmental DNA (eDNA) is a survey tool with rapidly expanding applications for assessing presence of a wildlife species at surveyed sites. eDNA methodology is known to be prone to false negative and positive errors at the data collection and laboratory analysis stage. Existing models for eDNA data require augmentation with additional sources of information to overcome identifiability issues of the likelihood function and do not account for environmental covariates that predict the probabilities of error or the probability of species presence.

In this talk, I will present a novel Bayesian model for analysing eDNA data, by proposing informative prior distributions for logistic regression coefficients that allow us to overcome parameter identifiability, while performing efficient Bayesian model-selection. The proposed methodology does not require the use of trans-dimensional algorithms and provides a general framework for performing Bayesian model selection under informative prior distributions in logistic regression models.


Eleni Matechou School of Mathematics, Statistics and Actuarial Science. University of Kent, UK.

Model Averaging techniques from a Bayesian perspective

A common situation in applied disciplines is that there is no a universally accepted theory leading to the construction of a single statistical model. More on the contrary, several different statistical models may be built based on the consideration of alternative sensible theories about the problem under study.

This context is normally handled choosing one of these competing models then producing inferences conditionally on this (now fixed) model. Nevertheless, this approach obviates the uncertainty regarding which is the true model leading to an underevaluation of the variability and potentially to incorrect inferences.

The statistical techniques that explicitly consider this source of extra variability through weighting inferences over the different models are called Model Averaging (MA). These procedures have received great interest in many disciplines and will be the subject of this talk.

We will review the basic aspects about the usage of MA with special focus on the Bayesian approach, which arguably is the natural way to approach the problem. The emphasis is placed on applicability and examples of applications with a review of the software that can be used to approach a MA problem.


Gonzalo García-Donato
Tweet @g_garciadonato

Department of Economics and Finance, Universidad de Castilla-La Mancha, Spain.