Generalised linear latent variable models, factor analytical extensions of generalised linear models, are a natural tool for modelling non-Gaussian data when there are many correlated responses.
While these models have been used in the social sciences for some time, we have only recently started using them in ecology to analyse multivariate abundance data, i.e. abundance (or presence-absence) data across species and sites, with correlation in abundance across species because of species interaction and other reasons.
This talk describes a range of applications of this model in ecology, and some innovations in model-fitting that we developed to accelerate model fitting and tailor models to ecological data.
Applications of the model include estimating patterns of residual correlation across species, ordination, multivariate inference about environmental effects and environment-by-trait interactions, accounting for missing predictors, and improving predictions in situations where one can leverage knowledge of some species to predict others.
Some of these are important applications for which there is currently no competing method in ecology. Innovations in model-fitting include the use of variational approximations and the TMB platform, to accelerate model-fitting by orders of magnitude and enabling the fitting of large metagenomics datasets in minutes rather than days.
Lunch will follow the seminar in the Glass Room, Level 4, CBE building.