Analysing multivariate ecological data with Generalized Linear Latent Variable Models

Main organizer: Bert van der Veen, postdoctoral researcher, Department of Mathematical Sciences, NTNU, (bert.v.d.veen@ntnu.no)

Co-organizer: Audun Rugstad, PhD fellow, Department of Mathematical Sciences, NTNU

In recent years, new approaches for analysing co-occurrence patterns of species have gained in popularity. The Generalized Linear Latent Variable Modelling (GLLVM) framework unifies classical community and species distribution modelling. For binary data, the methods are better known under the popular term “Joint Species Distribution Models” (JSDM), but the framework generally represents model-based ordination methods.

Model-based ordination methods are akin to classical ordination methods, such as Principal Component Analysis (PCA), (Detrended) (Canonical) Correspondence Analysis, or even Non-Metric multidimensional scaling (NMDS). The model-based ordination framework is more flexible than classical methods, so that it is possible to retrieve better ordination, perform model diagnostics and selection, and incorporate random effects as in Generalized Linear Mixed effects Models.

In this workshop participants will learn to use state-of-the-art methods in the gllvm R-package for analysing typical community ecological data, from both JSDM and model-based ordination perspectives. Unlike other approaches (such as the HMSC package), the gllvm package fits models much more quickly, so that results even for relatively large datasets can be examined in real time. With short lectures and hands-on exercises, participants will fit models to real data. We encourage participants to bring their own data. Topics covered in the workshop will depend on interest from the participants, but can include 1) Fitting and interpreting JSDMs, 2) Fitting GLLVMs to different types of data (binary or abundance, ordinal or cover data, biomass data etc.), 3) Random effects in multi-species models, 4) Unconstrained, constrained and concurrent ordination, 5) Comparison with classical ordination methods, 6) Examination and interpretation of estimates of correlation between species (due to the environment and residual correlation), 7) Fast estimation of phylogenetic random effects models. By the end of the workshop, participants will be able to fit, interpret, and present GLLVM results for scientific publications.

Target audience:

Researcher with (ideally) their own multivariate data on an ecological community, with an interest in stepping up their multivariate analysis skills. Participants should be familiar with the R programming language, should possess some prior knowledge on ordination and/or GL(M)Ms, or a willingness to to do so on short notice.

Duration: full day, 08:30 -17:00

The material is available on github: https://github.com/BertvanderVeen/Nof2025GLLVMworkshop