Problems with bins: A critical reassessment of Gotelli and Ulrich's Bayes approach using bird data

• Published in: Acta oecologica (Montrouge) 2015-11, Vol.69, p.137-145
• Main Authors: von Gagern, Melanie, von Gagern, Martin, Schmitz Ornés, Angela
• Format: Article
• Language: English
• Subjects: Aves; C-Score; Co-occurrence; Null model; Presence-absence matrix; Segregation
• ISSN: 1146-609X; 1873-6238
• DOI: 10.1016/j.actao.2015.10.003

Null model analyses are a common technique used to detect co-occurrence patterns in presence-absence matrices of species. One method which aims to identify interesting pairs of species has been introduced by Gotelli and Ulrich (2010). Based on the “fixed–fixed” null model constraint, it uses a pair-wise C-Score measure and partitions pairs into a number of bins in an attempt to reduce the number of false positives. Applying this technique to ornithological observations from Mecklenburg-Western Pomerania, Germany, provided some insights into the suitability of the method for this kind of data in particular, and raised some more fundamental questions about the method in general. Specifically, the number of bins, which can be chosen arbitrarily, is shown to have significant impact on the number and identity of the returned species pairs. Other parameters, like the number of null model iterations, or the algorithm used to generate these null model matrices, have less impact as long as certain minimal requirements are met. The computations have been performed using different implementations of the method mentioned, “Pairs” by Ulrich and our own code “RePairs”. The latter also introduces a randomization algorithm based on a network flow model. Comparing these implementations exposed an error in “Pairs”, which might invalidate results obtained using it. •We introduce a new and improved fixed–fixed algorithm for null model generation.•This algorithm guarantees termination and perfect independence from the input matrix.•We re-implement and extend the Bayes approach from Gotelli and Ulrich (2010).•Testing uncovers bin-dependency of species pair selection in co-occurrence analyses.•Arbitrary bin choices lead to unstable results, invalidating previous studies.