Actions and Detail Panel
Stats Central x58805 Multivariate Workshop
Tue., 22/11/2016, 9:00 am – Sat., 26/11/2016, 1:00 pm AEDT
Multivariate analysis in ecology has been changing rapidly in recent years, with a focus now on formulating a statistical model to capture key properties of the observed data, rather than transformation of data using a dissimilarity-based framework. In recent years, model-based techniques have been developed for hypothesis testing, identifying indicator species, ordination, clustering, predictive modelling, and use of species traits as predictors to explain interspecific variation in environmental response. These techniques are more interpretable than alternatives, have better statistical properties, and can be used to address new problems, such as the prediction of a species’ spatial distribution from its traits alone.
This course will provide an introduction to modern multivariate techniques, with a special focus on the analysis of abundance or presence/absence data, starting from a revision of fundamental tools in regression analysis, and extending these techniques to the case where there are multiple response variables.
Note: The course will run on Days 1 to 4 from 9.00 am to 5.00 pm and on Day 5 from 9.00 am to 1.00 pm.
Day 1: Revision of (univariate) regression analysis
Revision of key “Stat 101” messages, the linear model, generalised linear model and linear mixed model.
R packages: lme4
Day 2: Computer-intensive inference and multiple responses
The parametric bootstrap, permutation tests and the bootstrap, model selection, classical multivariate analysis, allometric line fitting.
R packages: lme4, mvabund, glmnet, smatr.
Day 3: Multivariate abundance data
Key properties, hypothesis testing, indicator species, compositional analysis, non-standard models.
R packages: mvabund
Day 4: Explaining cross-species patterns
Classifying species based on environmental response, species traits as predictors, studying species interactions.
R packages: SpeciesMix, mvabund, lme4.
Day 5: Model-based ordination and inference
Latent variable models for ordination, model-based inference for fourth corner models.
R packages: boral, mvabund.
Workshop prices: UNSW research student $400, UNSW researcher $700, External $1,000
For further details, email us at Stats.Central@unsw.edu.au
Date and Time
Pioneer International Theatre (Room G04)
Australian Graduate School of Management (AGSM - Building G27)
UNSW Kensington Campus, NSW 2052