There are various types of sorting techniques described in the sensory literature. The challenge we face in sensory testing is knowing which method is the best one to use.
In a recent cat food study we focused on three techniques: free sorting, hierarchical free sorting and Napping. The objective was to determine which of these methods would be the most suitable for screening a large number of samples (26) down to a more manageable number for descriptive sensory profiling (about 8). Although Napping is not a sorting technique, it was included as it gives a similar output and has been widely used in sensory studies for sample pre-screening and selection.
In free sorting, the assessor is presented with all of the samples and asked to place them into groups; putting samples that are similar into the same group. The number of groups is not pre-specified nor is the criteria on which they are to be sorted. This means that the assessor will be able to choose the parameters that are most dominant or most important to them.
Hierarchical free sorting starts with the same premise but once the assessor has formed his initial groups, he is then tasked with combining two groups that are the most similar, and to continue to do so until only two groups remain.
From this we see that sorting organises samples on the basis of similarity. Napping works differently in that it does not task assessors to group samples on the basis of how similar they are. Rather, the instruction is to create a (dissimilarity) sensory map in two dimensions, where the distances between the samples on the map reflect the sensory differences detected between them. You can read more information about Napping in our White Paper: What is Napping?
The analysis of free sorting data starts by working out how often samples are grouped together and from this calculates a co-occurrence matrix. In the case of hierarchical free-sorting a set of distance matrices is derived; one per assessor. Napping generates a set of sample co-ordinates and hence a dissimilarity matrix for the samples from each assessor.
The output from each technique is a sensory map of our samples reflecting the differences between them. How we get this map varies with the technique. Factor based methods (usually Multiple Factor Analysis for Napping and Multi-Dimensional Scaling or Multiple Correspondence analysis for Sorting techniques) are used to generate the sensory map. Clustering may also be applied and in fact in our cat food study, we used both types of analysis.
For both sorting and Napping, adaptations are reported whereby assessors label the groups of samples or areas of their plot with words explaining their criteria for grouping or positioning the samples.
In theory, sorting and napping techniques can work really well for reducing large sample sets to smaller numbers and, with word labelling, for revealing the sensory or other attributes relevant to consumers or sensory assessors. What they lack is detail, so if you need to understand precisely what the differences are between products then sensory profiling of all samples may be the most cost-effective route.
In the cat food study, we found that all three methods would have led us to select slightly different sub-sets of samples for our sensory profiling study. This is not too surprising when we consider that cat food is a highly heterogeneous product and so 26 samples would have exhibited many, many sensory attributes. Assessors would undoubtedly have focused on different attributes as the criteria for their sorting and Napping, due to differences in their individual sensitivities, in their judgement of what was important and partly due to memory and fatigue.
Sorting is claimed to be simple for assessors to carry out as it mimics what our brains do naturally. However, the techniques require all samples to be presented and assessed at one time and so sensory fatigue and memory have a major influence, particularly if flavour and aroma assessment is required.
The issue of attending to different attributes may be reduced by using more assessors… we only had 12 and the question we were left with was ‘would we have found better consensus across the methods with more assessors?’ Only further research can answer that question, but pragmatism dictates that any cost benefits accrued from sample pre-selection by sorting will soon be eroded if we have to use large numbers of respondents to achieve a reliable result.
We also need to think about what we are expecting to achieve? Like many researchers, we measured the effectiveness of the techniques by their ability to recover the sensory space derived by descriptive sensory profiling. However, is this fair when sensory profiling is an analytical technique that sets out to find differences, whereas sorting works on the basis of sensory similarity?