Skip to main content

Table 2 Summary of classification solution evaluators in this paper

From: The assignment of relevés to pre-existing vegetation units: a comparison of approaches using species fidelity

Evaluator

Optimality criteria*

Non-geometric evaluator

Morisita’s index of niche overlap (Horn 1966)

This index represents a measure of group overlap for a particular clustering solution. High proportional occurrence of species within a single group causes to niche overlap is decreased. It means minimal niche overlap indicates optimal solutions, so we used 1- Morisita instead of Morisita in the analysis.

  ISAMIC-indicator species analysis to minimize intermediate constancy (Robert 2010)

This index is a measure of consistent presence or absence of species in groups and it is bounded 0-1. The higher values the better.

  ISA (number of significant indicators) (Aho et al. 2008)

High ISA values indicate high fidelity and abundance of species within clusters. P-values for ISA-values calculated with Monte-Carlo procedures.

Geometric evaluators

  C-index (Hubert and Levin 1976)

This index shows the ratio of within to between group distances.

This index is confined to interval 0–1 and minimum C-index scores is its optimality. So 1- (C-index) was considered instead of C-index in the analysis.

  PARTANA ratio (Robert 2005)

High PARTANA value implies Low within group dissimilarity and high dissimilarity of relevés within groups to relevés outside of groups. So higher PARTANA values were considered as optimal clustering solutions.

  Point biserial correlation (PBC) (Brogden, 1949)

The higher PBC values were considered as optimal clustering solutions

  ASW-average silhouette width (Rousseeuw 1987)

High ASW indicates samples within clusters are compositionally similar, and dissimilar to nearest neighbor samples outside clusters.

This index is confined to interval 0–1 and a high value indicates optimal solutions.

  ANOSIM-analysis of similarities(Clarke 1993)

This index uses the rank of dissimilarity values in between and within groups. This index is confined to interval − 1, 0, + 1 and a high value indicates optimal solutions. Value 0 indicating completely random grouping.

  1. *Equations and descriptions for all evaluators included in Appendix 2