In previous chapters, we have examined ways (specifically factor-analysis methods) by which the variation in a set of observed variables can be interpreted in terms of a dependence on a set of unobserved variables or factors. In most cases, obtaining the data in a form suitable to begin the analysis is reasonably straightforward. However, there is an important class of problem in which the initial data processing is more difficult. These problems are characterised by requiring a judgement of similarity* between variables; the difficulty here is that ‘similarity’ cannot be measured directly as the data resulting from such investigations is not in a metric or quantitative form. Multidimensional scaling is a technique which enables us to convert these non-metric measures into a form suitable for the application of those methods of factor analysis discussed previously.
Unable to display preview. Download preview PDF.