Perceptual color spacing derived from maximum likelihood multidimensional scaling

Abstract

The canonical application of multidimensional scaling (MDS) methods has been to color dissimilarities, visualizing these as distances in a low-dimensional space. Some questions that remain are how well the locations of stimuli in color space can be recovered when data are sparse, and how well can systematic individual variations in perceptual scaling be distinguished from stochastic noise? We collected triadic comparisons for saturated and desaturated sets of Natural Colour System (NCS) samples, each set forming an approximate hue circle. Maximum-Likelihood MDS was used to reconstruct the configuration of stimuli more accurately than the standard 'vote-count' approach. Individual departures from the consensus response pattern were minor, but repeated across stimulus sets, and identifiable as variations in the salience of color-space axes. No gender differences could be discerned, contrary to earlier results.

Gender difference in color preference across cultures: An archetypal pattern modulated by a female cultural stereotype

“…A putative gender‐specific pattern of association between color and personality traits or gender schemata failed to replicate with Indian participants, ruling out these factors as a universal explanation for the color gender difference. Although the evolution of a gender‐specific behavioral use of trichromacy that would shape the opponent‐color mechanisms would be consistent with a pan‐cultural expression of gender difference in color preference, recent results obtained from color similarity judgments with participants stemming from the same Indian population and tested in the same laboratory conditions, call for prudence as the gender difference observed in British was not replicated in Indian participants …”

“…Although the evolution of a genderspecific behavioral use of trichromacy that would shape the opponent-color mechanisms would be consistent with a pancultural expression of gender difference in color preference, recent results obtained from color similarity judgments with participants stemming from the same Indian population and tested in the same laboratory conditions, call for prudence as the gender difference observed in British was not replicated in Indian participants. 29 Indian subjects were also questioned about their education, interests and hobbies, as other potential correlates or predictors of color preference. Gender differences would have been informative, but the results were negative.…”

Estimation of perceptual scales using ordinal embedding

“…As an example, consider (Ramsay, 1977;Takane, 1978) a situation in which the authors make a number of explicit model assumptions and then apply a maximum likelihood approach. This approach has been adapted for the analysis of data gathered by the method of triads (Bonnardel et al, 2016), where the triad responses are used to estimate the similarity matrix between items directly. The main differences between these more statistical approaches and machine learning might, up to some point, be of a philosophical nature: Rather than making many explicit model assumptions (for example in Ramsay (1977), a lognormal noise model, explicit weights on coordinates, powers of Euclidean distances to deal with non-Euclidean data), machine learning algorithms try to operate with minimalist assumptions -because they tend to be applied to data that rarely satisfy statistical model assumptions.…”

Variation of Saturation Across Hue Affects Unique and Typical Hue Choices

“…Colour categories and category prototypes are defined along all three dimensions of colour perception, including saturation (e.g., Figure 8 in Olkkonen, Witzel, Hansen, & Gegenfurtner, 2010). Observers tend to choose saturated rather than desaturated colours as prototypes of chromatic colour categories (see also Bonnardel et al, 2016). The tendency to choose saturated colours as prototypes could explain why observers from fundamentally different languages choose Munsell chips as prototypes that coincide with English prototypes (Witzel, 2018b).…”