Sorting backbone analysis: A network-based method of extracting key actionable information from free-sorting task results

Abstract

The free-sorting task is increasingly popular as a rapid sensory method to give a global picture of the similarities among samples. Sorting does not require training analysts, allows for the easy, simultaneous presentation of up to 20 samples, and provides stable results with 25–30 subjects. However, wide use of free-sorting is hindered by the current analyses for free sorting—for example DISTATIS and Correspondence Analysis—which require statistical expertise to conduct and interpret. In this paper a novel, alternative analysis is proposed, called “Sorting Backbone Analysis” (SBA), which is based on tools from network analysis. The similarity data produced from free sorting can represent a weighted network, and so a set of network-analysis tools can be used to identify groups of products which are significantly similar, and to visualize these results clearly and powerfully. SBA is simple and can be implemented with open-source software, provides interpretations that agree with current methods, and produces clear, powerful visualizations called “graphs,” which may offer new, interpretable insights to sensory scientists. This paper describes the mathematical and statistical background for SBA and applies SBA to four, previously published sorting datasets, with comparisons to DISTATIS. In each case SBA produces visual results that highlight all of the same features as the standard approach while being easier to interpret, and in many cases produces new insights. Therefore, SBA specifically and network analysis in general are suggested as new approaches for use in the analysis of sensory similarity data as produced through free sorting and related methods.