Sunday, 5 July 2020 | 9:00 - 17:00 | Location: TBDInstructors:
- Lars Rönnegård
- Youngjo Lee
Summary:
Complex problems do not always require complex tools, and searching for these simple solutions is what makes science really exciting. Hypothesis testing and prediction in various modern applications can be rather complex because they often require modelling of correlated responses. Data can for instance be correlated either because we have repeated measurements on individual subjects, there are genetically related individuals or data have been sampled spatially. Furthermore, non-Gaussian outcomes need to be handled to be able to perform adequate inference in real life problems. Here hierarchical generalized linear models can be applied, which is a broad class of models that can be fitted using a deterministic likelihood-based optimization algorithm.
Lee, Rönnegård & Noh published the book “Data analysis using hierarchical generalized linear models with R” in 2017 that has been used on several courses already. The course is well suited for IBC2020 and includes a combination of lectures and hands-on exercises in R. The course covers topics such as generalized linear models with random effects, heteroscedasticity and dispersion modelling, modelling of multivariate outcomes and further likelihood-based methods. The core part of the course includes exercises using the hglm and dhglm packages in R developed by the course instructors.
Prerequisites:
The course level is suitable for first-year graduate students and young researchers in applied statistics. The participants should be well acquainted with R. It is an advantage if the participants have experience of working with: maximum likelihood estimation, generalized linear models, and linear mixed models or other multilevel models.
Outline:
This is a full-day course introducing the participants to the book by Lee, Rönnegård & Noh (2017).
Morning session lecture:
Motivating examples and overview of R packages to be used.
Algorithm to fit hierarchical generalized linear models (HGLMs).
Morning session exercises:
Analyses of the epileptic seizure count using the hglm package in R.
Working with model checking plots for generalized linear models and HGLMs.
Afternoon session lecture:
Extended likelihood principle and inference for random effects using the h-likelihood.
Dispersion modelling using double HGLMs.
Further possibilities using the h-likelihood.
Afternoon session exercises:
Analysis of crack growth data using the hglm and dhglm packages.
Comparison to other packages to fit mixed models in R.
The afternoon session exercise includes analysis of the crack growth data using the hglm and dhglm packages. A summary of this analysis is available here for interested participants to have a look at.
Learning Outcomes:
After completing the course the participants will gained knowledge on: an algorithm for fitting hierarchical generalized linear models, likelihood inference for random effects and variance components, and available model checking tools for hierarchical generalized linear models.
After completing the course the participants will be able to analyze non-Gaussian data using dispersion modelling with the hglm and dhglm packages in R.
After completing the course the participants will gain insights into the extended likelihood principle and ongoing developments within the field of hierarchical likelihood inference.
About the Instructors:
|
Lars Rönnegård is a professor of statistics at Dalarna University, Sweden, and guest researcher the Swedish University of Agricultural Sciences. His research interests covers applications in genetics and ecology including spatial modelling. He has a twenty-year teaching experience both at undergraduate and graduate level with a majority of the courses given in English.
Youngjo Lee is a professor of statistics at Seoul National University, Korea. His research interests are extension, application and software development for GLM class of models, including joint GLMs, hierarchical GLMs (HGLMs), double HGLMs etc. He has developed R packages, dhglm, mdhglm, jointdhglm and frailty.HL to fit various models in HGLMs. He is an experienced teacher used to giving courses in English. Professor Lee collaborated with John Nelder for several years in the UK.
|
Recommendations for the Course:
Lee, Rönnegård & Noh (2017) Data analysis using hierarchical generalized linear models with R. CRC Press. ISBN 9781138627826.
The course requires all participants bring their own laptops.