Individual-differences multidimensional scaling was applied to a set ofproximity dataforequiluminant lights (Paramei & Cavonius, 1999) In a recent study (Paramei & Cavonius, 1999), two different experimental paradigms were applied to a single set ofcolor stimuli (15 equiluminant lights) in order to collect matrices ofdissimilarities. First, response times (RTs) were measured as observers decided whether paired stimuli were the same or different; second, the observers rated the perceived dissimilarities between pairs of stimuli. Nosofsky (1992) has pointed out that these two tasks are not necessarily measuring the same thing. Each task mediates between the proximities perceived by the informants and the data collected; it may modulate the proximities in the process.In addition, the Paramei-Cavonius (1999) data inch.ide variations between observers. Three were normal trichromats; three had abnormal color vision. An immediate reaction is to analyze the matrices using individualdifferences multidimensional scaling (MDS), to take these variations into account. The result of individualdifferences MDS is typically a single group space, X o , plus a set of subject-specific parameters that describe a personal space, X m» for each data source, where X m is a compressed or an elongated version of X o ' Here X o and X m are N X P matrices, describing the locations of N points (where N is the number of stimuli) in a Pdimensional space.Individual-differences MDS will also detect differences between the rating and the same/different task, if these can be represented as a compression of color space. There are precedents for this: The distinctness ofthe border between areas of color can be used as an index of dissimilarity, but because border distinctness is tritanopic (Tansley& Boynton, 1976), color space is a line rather than a plane when mapped in this way. Another example is the elegant gestalt fusion method used by Stalmeier and de Weert (1988) to obtain triadic data. The stimuli were presented as contiguous blocks of color that could be seen in two possible configurations, depending on which pair of colors was more similar. Mapped with this method, color space is compressed along tritanopic confusion lines, in comparison with conventional triadic comparisons from the same subjects for the same stimuli.The reliability ofRT data is also an issue. A number of studies in which RT data were collected as indicators of dissimilarity or discriminability and analyzed with MDS found a tendency for the number ofdimensions in the MDS solutions to be inflated (Sergent & Takane, 1987, with human observers; Blough, 1988, with pigeons). For instance, in Blough's Experiments IB and IC, with stimuli varying on two known dimensions, three-and even fourdimensional solutions were required to recover the structure. These excess dimensions are possibly artifacts, generated by the relatively low reliability ofthe shorter RTs (corresponding to supraliminal dissimilarities). It is necessary to test this possibility as a preliminary step to analyz...