Practical and Innovative Uses of Correspondence Analysis

Abstract

Brand or perceptual maps have been used extensively in the presentation of survey data, in order to demonstrate relationships that exist in otherwise user-unfriendly data sets. Correspondence analysis has become widely used in South Africa and is gaining currency elsewhere, it having the advantage of only requiring nominally scaled data. This paper demonstrates 'traditional' uses of correspondence analysis in brand mapping and a variety of simple, innovative applications allowing the inference of motivators, the determination of segmentation variables, the elicitation of trends and the establishment of scale interpoint distances. An account is presented of its use in setting up an urban-rural gradient to allow a person's level of urbanisation to be determined as a numerical score. This has become important in developing countries, as the pace of urbanisation increases. Knowledge of this gradient provides a means for people marketing different goods to divide consumers into whatever groups are appropriate for their purpose along a continuum. 1 Background One of the principal needs of a survey statistician is to be able to communicate the generally multivariate relationships that exist in typical marketing research data sets in a fashion which summarises the data while allowing insights into dimensions of perception and hence inferences as to attribute saliency, positionings, market gaps and strategic direction. Brand or perceptual mapping has been extensively used to achieve some of these objectives. Principal components and factor analysis, discriminant analysis, biplots and multidimensional scaling are some of the more 'traditional' approaches which might be adopted. Such approaches may suffer from problems such as the need for measurement at a better than nominal or ordinal level-often interval level is required. This has often led to the inappropriate application of various mapping techniques. Some techniques are more appropriate for small samples and a more intensive examination of perceptions at an individual level. Finally, many techniques call for a subjective (and often very difficult to agree upon) interpretation of 'factors'-it is often not possible to plot, say, brands and attributes, together on a chart without an intermediate 'translation'. Correspondence analysis originated in France in the early 1960s and was introduced to the English-speaking world in 1984 (Greenacre, 1984). It is a graphical display technique requiring only a nominally scaled task from respondents and can be applied to almost any rectangular matrix. This versatility becomes one of its strongest advantages, to the extent that it can provide a single methodology to solve a variety of mapping problems (Shahim & Greenacre, 1988). In market research, it is most commonly applied to aggregated matrices and allows the simultaneous consideration of multiple categorical variables (either nominal or ordinal), although on occasions its application to disaggregated data is indicated. The result is a graphical representation which plots the row and column profiles in a chi-squared metric such that profiles close to the 'average' are plotted near the map's origin, and those with differing profiles are plotted further away. Hence, it is primarily used to display the degree of differentiation of row and column profiles.