15.2.2020 Reliable and computationally efficient methods for analysing multivariate abundance data (Niku)

The multivariate abundance data in ecology consist typically of multiple, correlated species encountered at a set of sites, together with records of additional covariates. Such data is often collected when interest is in studying species communities; their between-species interactions and interaction with the environment. In her dissertation at the Ä¢¹½Ö±²¥ M.Sc Jenni Niku showed how generalized linear latent variable models (GLLVMs) can be used in the analysis of multivariate abundance data.
Published
15.2.2020

Traditionally multivariate abundance data have been analyzed using classical algorithm-based methods, which often focus on visualizing multivariate abundance data in small dimensional form in order to make interpretations on the data structures (ordination). While these methods are easy to use, the results may depend on the used algorithm and the reliability of the results is hard to evaluate.

In this thesis we show how generalized linear latent variable models (GLLVMs) can be used in the analysis of multivariate abundance data. The method provides more reliable results and more versatile approach for the analysis as compared to the classical methods by modeling correlation structures in the data and by taking into account the typical properties of the data via statistical distributions. We study and illustrate the use of GLLVMs in ordination, studying between-species correlations, and hypothesis testing of the environmental and trait interactions by analyzing several typical multivariate abundance datasets.

In order to make the models more attractive among practitioners, new computationally efficient algorithms for the parameter estimation were developed. The accuracy and computational efficiency of the methods are investigated and compared to existing methods through extensive simulation studies. The developed algorithms and additional tools implemented for model diagnosis, visualization and statistical inference as well as several illustrative examples are collected in the R package gllvm.

The dissertation is published in JYU Dissertation series, number 192, Jyväskylä, 2020. 

Link to publication:  

M.Sc. Jenni Niku defends her doctoral dissertation in Statistics "On modeling multivariate abundance data with generalized linear latent variable models" on Saturday 15th of February starting at 12 at Mattilanniemi in hall Agora B222.1 (Gamma). Opponent Professor is Jouni Kuha from London School of Economics, UK and Custos is Docent Sara Taskinen from the Ä¢¹½Ö±²¥. The doctoral dissertation is held in Finnish.