GLLVMs: Advanced multivariate analysis of ecological communities in R

GLLVMs: Advanced multivariate analysis of ecological communities in R

Dates

7-10 July 2026

To foster international participation, this course will be held online

Course overview

Multivariate analysis of ecological communities is often performed with a set of loosely connected methods that rely on distance measures (e.g., C(C)A, NMDS, PCoA). In the last decade, Generalized Linear Latent Variable Models (GLLVMs) have been proposed as an alternative, unifying framework, for performing ordination, and for jointly analysing the occurrence patterns of multiple species as Joint Species Distribution Models do. GLLVMs can be seen as a multivariate extension of GL(M)Ms, so that they inherit many of the useful properties of both statistical models and ordination methods. The explicit statistical model formulation makes GLLVMs more flexible than classical ordination methods, and provide many useful tools for inference on species associations, and for model diagnostics.

Target audience and assumed background

The course is aimed at PhD candidates, postdoctoral researchers and researchers with basic statistical knowledge, that have multivariate data (which can be experimental or observational) and want to learn how to analyse it in a statistically appropriate manner. Familiarity with concepts in generalised linear models (e.g., distribution, link function, mean-variance relationship, uncertainty quantification) for different response types is required, whereas familiarity with generalised linear mixed models (e.g., variance parameters, quirks that come with mixed models) or with the basics of ordination (e.g., biplots, latent variables) is an advantage. Basic knowledge in the R programming language is required.

Learning outcomes

An understanding of the benefits of model-based ordination over classical ordination.
The ability to recognise the appropriate distribution for (multispecies) data.
Know how to fit multispecies Generalised Linear (Mixed effects) Models, model-based ordination, and JSDMs the gllvm R-package
Finding a "good" GLLVMs based on residual diagnostics and other criteria.
Understanding of the output of GLLVM to effectively visualize, interpret and communicate results.
Can utilize GLLVM outputs to answer ecological research questions

Session content

Tuesday – Classes from 2-8 PM Berlin time
On the first day, we will focus on properties of multispecies data, and the ability of (multispecies) generalised linear (mixed effects) models to accommodate these. This will include:

Important aspects of community data: sampling, data properties
(multispecies) Generalised linear models
(multispecies) Generalised linear mixed models.
Methods for finding a good model (diagnostics, convergence, assessing model fit)

Wednesday – Classes from 2-8 PM Berlin time
On the second day, we will dive into the different aspects of Joint Species Distribution Models for binary data, and explore how to summarize results for all species in the data:

Hierarchically modelling environmental responses (traits, phylogeny)
Incorporating species co-occurrence patterns: Joint Species Distribution Models
Predicting species richness and diversity from multispecies models

Thursday – Classes from 2-8 PM Berlin time
On the third day, we will focus on model-based ordination as a complex flavour of JSDM, suited to all possible data types.

Background concepts of ordination
Benefits of GLLVMs over classical ordination methods
Accommodating nested study designs with random effects
Fitting an ordination with covariates
Conditioning the ordination
Unimodal response models

Friday– Classes from 2-8 PM Berlin time
On the fourth day, we will consider:

Non-independence of sampling units in the latent variables (i.e., group-level ordination, spatial ordination or temporal ordination)
Multispecies models for mixed-response types (different types of data for different species)
A brief overview of other packages for model-based multivariate analysis
Participants will have the opportunity to fit models to their own data, with support.