Management advice provided by ICES has its roots in the work of stock assessment working groups. In addition to this expertise and the population models of the stock, another important factor is the knowledge of population dynamic parameters, their relationships derived from other stocks, and the historical experience in fisheries.
Traditional stock assessment methods can only use the data available for the stock of interest, which means that all other knowledge has to be left out from the quantitative analysis. The Bayesian approach to scientific reasoning provides a mechanism to incorporate this other knowledge and experience. The Bayesian approach is a mathematical logic for quantifying and processing expert knowledge, which enables direct integration of the prior information possessed by experts and their interpretation of the observed data.
Bayesian methods also provide a mechanism for the quantification and computation of uncertainty that is directly applicable to decision making. Traditional statistical methods only describe the sampling process while assuming known state of nature (stock size for instance); there is no measure of uncertainty about the state of nature itself. Thus, the scientists are not able to make probabilistic statements of uncertainty about the status of the stock or the population dynamic parameters. Bayesian analysis results in clear probability statements such as “there is a 90% probability the stock is between 1200 and 3000 tons”, and these probabilities can then be directly used in decision analysis to inform the management advice.
Objective
The objective of this course is to familiarize participants with the basic concepts of Bayesian inference and to provide skills for solving simple problems. The participants will have hands on experience using MS Excel and OpenBUGS software for Bayesian computation.
Course structure
• Principles of the Bayesian reasoning
• Differences and similarities between the Bayesian approach and conven-tional statistics
• Numerical integration methods: Markov chain Monte Carlo (MCMC) and Sampling importance resampling (SIR)
• Bayesian regression analysis (or estimation of a mean)
• Bayesian Mark-Recapture analysis