Written by: Bert van der Veen, Inland Norway University of Applied Sciences
For a while now I’ve wondered what this “Bayesian statistics” is that researchers have been talking about. It’s been frustrating to have to sit through talks on “priors” and “credible intervals”, without knowing what they meant and without being able to contribute to discussions.
Fortunately, I had the privilege of attending the International Summer School on Bayesian Modelling: An Introduction for Ecologists and Environmental Scientists taught by Dr. Joe Chipperfield, Prof. Florian Hartig and Dr. Jörn Pagel, during the last week of September 2017. The course was taught at the Heathland Centre on the incredibly beautiful island of Lygra, and accommodation and food (the food was truly amazing) were covered by the course fee. The tutors were excellent, they adapted the course and their explanations to the circumstances (transport problems, some attendees missed the first day) and their attendees. Where necessary they provided applicants with the opportunity to ask questions related to their own projects. My thanks goes out to IRSAE for making this possible for me, by providing me with a mobility grant.
Even though a reasonable level of statistical knowledge was assumed (i.e. knowledge of various probability distributions and the parameters that define them), the tutors took plenty of time to work various ecological examples. My previous knowledge of R turned out to be useful, but not completely necessary, as most of the code was provided (both R and JAGS).
We covered basic concepts of Bayes and the philosophical differences between “classical” statistics (frequentist) and Bayesian, as well as various modelling approaches: linear and generalized linear models, mixed-effect models including ones with zero-inflation and overdispersion parameters, followed up by more complicated hierarchical models (e.g. occupancy and state-space). During the last day we also received an insight into simulation models.
Even though applying more complex hierarchical models might still be a step too far for me, I am confident I am now able to perform analysis in the Bayesian framework with the more general modelling approaches (LM, GLM and mixed-effect models). If you are, like me, looking for a deeper understanding of (Bayesian) statistical inference in natural sciences, you will not be disappointed by this annually repeated course (see www.bayessummerschool.com).