Page 219 - Special Topic Session (STS) - Volume 2

STS486 R. Ayesha A. et al.

Recent advances in ecological networks:
Regularized grouped Dirichlet-multinomial
regression
1
2
R. Ayesha Ali , C. Crea
1 Department of Mathematics and Statistics, University of Guelph, Guelph, Canada.
2 Geosyntec Consultants, Guelph, Canada.

Abstract
Having a solid understanding of why species interact in ecological settings can
help policymakers in the development of conservation or restoration
management plans. Quantitative ecologists are now able to collect more
detailed species traits, either from the field or through published literature,
that can supplement observed counts of species interactions in ecological
networks. While there is great interest in using such data to further enhance
our understanding of the mechanisms driving the observed interactions, there
is a lack of statistical models that can accommodate the complexities
presented by such data. Here we discuss some of these modelling challenges
and present new advances in the context of plant-pollinator networks. In
particular, we present regularization for grouped Dirichlet-multinomial
regression.

Keywords
Grouped Dirichlet multinomial regression; interaction networks; ecological
modelling; adaptive lasso

1. Introduction
Mutualistic networks arise out of several ecological processes including
pollination, seed dispersal, and host-parasite relationships, to name a few.
Studies on these networks have revealed common structural patterns, such as
the nested organization of pairwise interactions and the skewed distribution
of links per species (Montoya et al. 2006; Bascompte and Jordano 2007;
Vázquez et al. 2009a). It is believed that these structural patterns are driven
by both evolutionary and ecological processes. Neutrality states that species’
interactions are totally random whereas linkage rule theory suggests that the
functional traits between species must match for species to interact with each
other.
For example, an insect’s tongue length must be long enough to reach a
plant’s reproductive parts for pollination to take place. The characteristics
pollinators seek in plant species may be better anticipated if species
interactions are modelled by the functional traits that drive them. Grouped
Dirichlet-multinomial (DM) regression provides a framework to quantify the
contributions of species traits and/or linkage rules to pollination.
208 | I S I W S C 2 0 1 9