Bayesian Data Analyses Using
Linear Models with R and
WinBUGS
Fränzi Korner-Nievergelt & Regina Bispo
18 – 22 May 2015, Lisboa, Portugal
Course description
Bayesian data analysis becomes more and more standard in the analyses of biological
data. Bayesian methods are the only methods that provide exact estimates of the
standard errors in non-normal, hierarchical and more complex models (see e.g. Bolker et
al. 2008, TREE 24:127-135). They allow fitting models that are too complex to be fitted
easily using frequentist methods, e.g. hierarchical ecological models. Furthermore,
existing knowledge about a parameter can formally be used when analysing the data, and
the results have a natural interpretation such as the probability of a meaningful
hypothesis.
The course introduces the principles of Bayesian data analyses together with a sound
training in applying linear models. Today, life scientists, especially ecologists, are
expected to be familiar with normal linear models (LM), linear mixed models (LME),
generalised linear models (GLM), and generalised linear mixed models (GLMM).
These four types of models form the basis for a variety of more complicated models, such
hierarchical ecological models of which the occupancy, the point count model and the
Cormack-Jolly-Seber model will be introduced using the free software WinBUGS.
Participants will apply linear models including LM, LME, GLM, and GLMM using Bayesian
methods with the free statistical software R (www.r-project.org). The course follows the
book Korner-Nievergelt, Roth, von Felten, Guelat, Almasi and Korner-Nievergelt (2015).
Bayesian data analysis in ecology using linear models with R, BUGS and Stan, Elsevier
(http://store.elsevier.com/product.jsp?isbn=9780128013700&pagename=search).
Worked examples will include:
- graphical data exploration (various plotting functions)
- fit of the model to data (R-Functions lm, lmer, glm, glmer)
- assessment of model fit and model assumptions (diagnostic plots of residuals)
- visualization of the results and drawing conclusions (summary, anova, predict, sim)
- introduction to BUGS and/or Stan (depending on needs of course participants)
During the last day of the course, participants analyse their own data.
Contents
Day 1: Short introduction to R (optional)
Introduction to Bayesian statistics
Overview frequentist and Bayesian statistics
Normal linear model (LM): ANOVA, ANCOVA, regression
Day 2: Model selection
Linear mixed effects model (LME)
Generalised linear model (GLM): logistic regression
Day 3: Generalised linear model (GLM): binomial and Poisson-model
Generalised linear mixed model (GLMM)
Day 4: Extensions of GLMMs
Introduction to BUGS, Cormack-Jolly-Seber models and point count/occupancy
models
Day 5: Participants work on their own (or example) data and give short presentations
Teachers
Fränzi Korner-Nievergelt
FK has a first degree and PhD in Ecology and Organismal
Biology (University of Zurich, CH), a certificate in Applied
Statistics (ETH Zurich, CH) and a post-diploma in Applied
Statistics (University of Berne, CH). Since 2003, she is owner
of the statistical consulting company oikostat GmbH
(www.okoistat.ch) and works as a statistician at the Swiss
Ornithological Institute SOI. She further teaches Bayesian
statistics courses at the International Max Planck Research
School of Organismal Biology IMPRS. Author or the book
Bayesian data analysis in ecology using linear models with
R, BUGS and Stan, Elsevier, New York.
Regina Bispo
RB has a first degree and a PhD in Agronomic Engineering
(ISA, University of Lisbon, PT), a Master in Probability and
Statistics (FCUL, Portugal) and a PhD in Experimental
Statistics and Data Analysis (Faculty of Sciences, University
of Lisbon, PT). Assistant professor at ISPA – Instituto
Universitário (Lisbon, PT). Teaches statistics in a variety of
courses such as psychology, mathematics, statistics, biology
or pharmacy. Member of the Research Unit MARE – Marine
and Environmental Sciences and collaborator of the
Statistics and Applications Research Center, University of
Lisbon. Partner, senior consultant and operations manager
at Startfactor Statistical Consulting and Training Lda.
Details
Course form: Lectures and exercises
Participants: The number of participants is limited to 20. Participants are expected to bring their own laptop with
R installed.
Prerequisite: Make sure you are familiar with the following concepts: mean, standard deviation, standard error
and t-test (e.g. chapter 5 in Dalgaard 2008, Introductory Statistics with R, Springer). For basic knowledge in R
programming see, e.g., chapters 1, 2, and 4 in Dalgaard 2008, or chapters 1-5 in Crawley 2007, The R Book, Wiley.
Location: Taguspark – Parque de Ciência e Tecnologia, Oeiras, Portugal
Course hours: 9:00 to 17:30 including two coffee breaks and a one hour lunch break.
Costs: 500 Euros
Registration and information:
Startfactor, Lda
url: www.startfactor.pt
email: [email protected]