The programme is designed for U Bordeaux's Sciences and Environment Graduate School (with enrollment from students of other graduate schools). Requirements are a working knowledge of R, and some basic statistics (analysis of variance and regression). The course runs for 10 sessions of 3 hours each, roughly half-lectures half-practicals (TDs, travaux dirigés), although percentages may vary.
- Objectives and philosophy of Bayesian statistics. TD1 Bayesian estimation of a proportion
- Revisiting the ANOVA in a Bayesian framework. TD2 Getting acquainted with software (JAGS), coding the first models
- Markov Chain Monte Carlo (i.e., algorithms for Bayesian statistics). Practicals within the course: Monte Carlo integration, rejection sampling, Metropolis algorithm
- From fixed to random effects, introduction to mixed models. TD4 variance partitioning (with thorough convergence diagnostics).
- Mixed models. A hint of Poisson GLMs. TD5 mixed models following up on TD4 (done first)
- Generalized linear models for counts. TD6 GLM(Ms) Poisson LN (fitting diagnostics, posterior predictive checks).
- Binomial/Bernoulli GLM(Ms) (importance of priors in original and transformed scale). TD7 Binomial ANOVA.
- Nonlinear models (organism growth, population growth). TD8 Gompertz organism growth.
- Latent variable models. TD9 occupancy model (0/1 data with added observation process).
- Model selection in a Bayesian setting. TD10 Linear and nonlinear model comparison.