Description
Places items in a low-dimensional map so that distances match judged dissimilarities, revealing the perceptual dimensions people use and where products or designs sit on them.
Multidimensional scaling takes a matrix of pairwise dissimilarities, from direct ratings, card-sort co-occurrence or confusions, and positions items in two or three dimensions to minimise stress, the mismatch between map distances and the data. Dimensions are then named from the properties of items at their extremes; timbre studies, for example, consistently recover attack time and brightness. Marketing positioning adds an ideal product point and reads real-to-ideal distances, on the assumption that shorter distance means stronger preference, to find open positions. MDS needs no predefined attributes, unlike factor analysis of feature ratings, so it can surface dimensions the team did not anticipate. The rigour lies in collecting the full pairwise matrix from users, reporting stress, choosing dimensionality on evidence, and checking whether the data are better represented by a tree. In UX the 'perceptual map' often survives only as an opinion-based 2x2.
Attributes
- Discipline
- Statistics & Modelling
- Loss type
- Vestigial
- Data
- Quantitative
- Effort
- Days
Steps
- Choose 8-25 items (products, sounds, icons, features) equated on irrelevant properties such as loudness or size.
- Collect dissimilarities: rate every pair on a scale (for example 1-10), or derive them from card-sort co-occurrence or confusion counts.
- Average across participants into a dissimilarity matrix, or keep individual matrices for an individual-differences (INDSCAL) solution.
- Run MDS for one to four dimensions and choose the solution from the stress elbow and interpretability.
- Name each dimension from the measurable properties of items at its ends, ideally by regressing item properties onto the coordinates.
- For positioning, elicit the segment's ideal product and place it on the map, then read gaps and real-to-ideal distances.
- Before trusting the map, check nearest-neighbour centrality and reciprocity, and use a tree representation if hubs dominate.
Sources
- Handbook of Data Mining (Ye, 2003)
- The Science and Psychology of Music (Thompson & Olsen, 2021)
- Innovation and Product Management (Gaubinger et al., 2015)
- Statistics for Marketing and Consumer Research (Mazzocchi, 2008)
- The SAGE Handbook of Quantitative Methods in Psychology (Millsap & Maydeu-Olivares, 2009)
- A User's Guide to Principal Components (Jackson, 1991)
- Quantitative Social Research Methods (Singh, 2007)
- Methods of Multivariate Analysis (Rencher, 2002)
- Foundations of Measurement, Vol. 1: Additive and Polynomial Representations (Krantz et al., 1971)
- Foundations of Measurement, Vol. 2: Geometrical, Threshold, and Probabilistic Representations (Suppes et al., 1989)
- R for Marketing Research and Analytics (Chapman & Feit, 2019)
- Pattern Classification (Duda et al., 2001)
Also known as
- Multidimensional scaling
- MDS
- Classical scaling
- Stress minimisation
- Multidimensional scaling of timbre dissimilarity ratings
- Timbre space
- Perceptual sound map
- Perceptual mapping for product positioning
- Positioning model
- Positioning analysis
- Ideal point map
- PREFMAP
- Internal and external preference mapping
- Multidimensional unfolding
Categories
- Analytics & Behaviour
- Statistics & Modelling