Indicator Species Analysis for bioassessment: part one Dufrene-Legendre ISA

I have long been a fan of indicator species analysis for bioassessment and identifying trends in experimental treatment groups. There are many applications of indicator species analysis (ISA) in both applied and basic research, and there is myriad free software to allow everyone in government or the non-profit sector to learn about both classical Dufrene-Legendre ISA and alternative methods for identifying species thresholds and multiple group membership.

The initial concept of indicator species analysis was to develop a statistical method to develop "indicator species" that occur under certain environmental conditions or in certain communities - both goals of ecology and conservation. The math behind this method is also extremely simple - for a given species, calculate the product of the relative abundance and relative frequency for a given group. A higher value is more indicative of a better "indicator species" for a given group. Some software multiply this value by 100, scaling the values from zero (non-indicator) to 100 (perfect indicator), while others do not creating a range between zero and one. The strength of this association is then tested with Monte Carlo simulations to calculate a probability value for that indicator species in a given group relative to that indicator species in other group combinations.

The original 1997 Ecological Monographs paper by Dufrene and Legendre is where one should start investigating ISA. It starts by discussing the statistical firepower of that era that helped ecologists decide where to prune a dendrogram from cluster analyses. They quickly develop a case that using ordination to enumerate clusters is a subjective slope in need of some rules. McGeoch and Chown (1998) quickly ran with the idea that this could be applied to other ecological contexts, like a priori treatment groups.

Jon Bakker authored a 2008 paper in the Journal of Applied Ecology (freely available) on improving the utility of indicator species analysis that suggests using meta-analytical techniques, using exact permutations and potentially simplifying ISA to presence-absence can make ISA much more flexible.

These papers provide just enough background to make someone dangerous. They provide full context on the rationale and methods used to assign indicator values to treatment groups or algorithm-derived clusters. Next time I'll break out two newer methods, the controversy that one of them has recently generated and some R links and code from one of my recent papers. Since I'll be moving fast, I can't help but recommend Legendre and Legendre's Numerical Ecology and Borcard et al.'s, Numerical Ecology with R as great references for many things multivariate and ecological.

Greatest Hits