Workshop Presenter: Ben Bolker.

About Ben Bolker: “I’m a professor in the departments of Mathematics & Statistics and of Biology at McMaster University. My interests range widely in spatial, theoretical, mathematical, computational and statistical ecology, evolution and epidemiology; plant community, ecosystem, and epidemic dynamics; and whatever else takes my fancy. I run the mac-theobio research group jointly with Jonathan Dushoff and David Earn.

I am currently Director of the School of Computational Science and Engineering and Acting Associate Chair (Graduate) for Mathematics.”

Workshop Title: Intermediate and advanced linear and generalized linear mixed models

Workshop Date: Monday, May 13, 2024

Focus: Troubleshooting and advanced modeling techniques with existing R packages

Topics:

• Review of concepts and definitions
• Troubleshooting convergence and singular-fit issues
• Random effects with structured covariance matrices and alternative bases (e.g. multivariate models, multimembership models, penalized spline bases)
• Inference and prediction: likelihood profiling, non-parametric and parametric bootstrapping, bias correction for GLMMs, confidence intervals on predictions
• Frequentist (empirical Bayesian) and fully Bayesian methods

Expected background: Equivalent of a (completed or in-progress) master’s in applied statistics, basic R proficiency

Dr. Bolker is also giving the keynote talk on Tuesday morning.

Title: Progress and challenges in open-source multilevel modeling

Abstract: In parallel with the revolution in machine learning and AI, more traditional multilevel models are gradually becoming more flexible, powerful, and readily available in convenient open-source software. I will give a brief overview of the current landscape of mixed modeling tools, exciting new directions, and open challenges. I will focus on methods available in R but touch on methods in Julia and Python, as well as comparing to some commercial software. Other topics will include model selection, troubleshooting (convergence and singular-fit problems), and the use of latent variables to fit nonlinear patterns in data.