The management team is looking at slides showing consumer statements about eight brands and twelve brand image attributes, and the conclusion is that “everyone is somewhat modern and somewhat trusted.” At that point, correspondence analysis and a perceptual map stop being an academic curiosity and become a tool that reduces a cross-tabulation to a single chart showing which brand is genuinely associated with what. The text below describes the research approach: how to prepare the data, how to read the arrangement of points, and what to avoid in interpretation.<\/p>\n
Correspondence analysis (CA) and its extended version – multiple correspondence analysis (MCA) – are multivariate analysis techniques that transform contingency tables into spatial representations of relationships between categorical variables. In brand image research, the typical input is a “brand x attribute” matrix, with each cell containing the number of respondents who assign a given attribute to a given brand (pick-any associations). The result is a perceptual map – a two-dimensional chart in which the proximity of points reflects the similarity of profiles or the co-occurrence of categories, depending on how the map is constructed.<\/p>\n
The method works well when there are specific brand positioning questions that the data need to answer: which brands occupy the same association space, which attributes differentiate the category, and which are known as “table stakes” (assigned equally to all brands and therefore non-differentiating). A table of numbers does not show this, because the human eye cannot compare dozens of proportions simultaneously. A perceptual map makes this clear at a glance.<\/p>\n
Typical use cases in research projects include several recurring situations that should be distinguished before choosing a method:<\/p>\n
It is worth emphasizing that correspondence analysis and a perceptual map do not answer the question “why” – they answer the question “what is associated with what.” This is a diagnostic rather than an explanatory tool.<\/p>\n
The starting point is measurement. The most common approach uses pick-any questions (“Which of the following brands do you associate with…”) or pick-one questions for each attribute (“Which brand best fits the description…”). The first option produces a richer matrix, while the second provides sharper differentiation. The choice affects the geometry of the result, so this decision is made at the questionnaire design stage, not during analysis.<\/p>\n
From a mathematical perspective, CA decomposes the chi-square statistic of a contingency table into axes (dimensions) that explain as much inertia as possible. The first two axes form the chart plane. The percentage of explained inertia indicates how faithfully the two-dimensional image represents multidimensional relationships. If the first two axes explain only a small share of inertia, the map is a simplification that needs to be supplemented with a third dimension or additional tables.<\/p>\n
In practice, the analytical procedure consists of several steps that should be followed deliberately:<\/p>\n
The critical moment is naming the axes. Axes in CA do not have a predefined interpretation – they are emerging dimensions that organize differences. In FMCG categories, the first dimension may separate brands into “traditional vs. modern,” while the second may distinguish “premium vs. everyday,” but this is an outcome, not an assumption. In B2B categories, the axes may more often describe “specialist niche positioning vs. mass presence” and “relationship-based vs. transactional.” The meaning of an axis is determined by the attributes that contribute most strongly to the dimension.<\/p>\n
As Hume’s Institute experts point out, a well-prepared perceptual map can show in a few seconds what an hour of management discussion over a cross-tabulation cannot establish – because it reduces hundreds of pairs of comparisons to a single geometry that the brain reads intuitively. It does not replace numerical analysis, but it gives the team a shared language for discussing brand position.<\/p>\n
In Hume’s Institute projects, perceptual maps also provide the greatest value when they are compared with an “ideal brand” map – a point constructed from consumers’ preference statements. The distance between a brand’s position and the ideal point within the same space can then be compared over time, but its interpretation depends on the normalization adopted and on how the ideal point is constructed.<\/p>\n
A perceptual map is visually appealing, which is precisely why it can be dangerous in the hands of someone unfamiliar with its limitations. The first and most common error is interpreting distances between points of different types (brands and attributes) as Euclidean distances. In standard CA with symmetric normalization, brand-to-brand and attribute-to-attribute distances can be interpreted directly. Brand-to-attribute positioning, however, primarily indicates an above-average association and the direction of the relationship, rather than its exact numerical strength. This distinction disappears in many presentations seen on the market.<\/p>\n
The second error is ignoring the percentage of explained inertia. If the first two axes explain only a small share of variability, the map is like a projection of a three-dimensional solid onto a plane from an unfortunate angle – points that appear close may be far apart in the full space. A professional interpretation always reports the contribution of each dimension and the quality of point representation on the plane (e.g., cos2<\/sup>), because a point poorly represented on the plane should not be interpreted.<\/p>\n The third error concerns attribute selection. The list of associations included in the study determines the space in which the map will be created. If attributes are unbalanced (eight variants of “modernity” and one of “tradition”), the first axis will be artificially dominated. This is why a qualitative exploratory stage – IDIs or mini-groups – should be conducted before analysis to establish the language of the category and a list of attributes actually used by consumers, rather than invented at a desk.<\/p>\n The fourth pitfall is comparing maps from different research waves without an anchoring procedure. CA is sensitive to changes in the set of brands and attributes – adding a single brand can rotate the entire configuration. Brand image tracking based on CA requires retaining the same list of variables or using techniques such as supplementary points, which make it possible to place new elements in an existing reference space without changing the axes defined for the active set.<\/p>\n It is also worth remembering how CA differs from related techniques. Multidimensional scaling (MDS) works with similarity or distance matrices and does not require categorical data, but it requires a different type of measurement. Principal component analysis (PCA) works with quantitative variables, such as mean Likert scale ratings for brands – it produces a visually similar effect, but is based on variance and covariance rather than a contingency table. The choice between CA, MCA, PCA, and MDS is a methodological decision based on the type of data and the research question, not on chart aesthetics.<\/p>\n For a perceptual map to genuinely help a team understand brand positioning, the research project should meet several methodological requirements. The following list is a practical quality check used in analytical work:<\/p>\n A perceptual map is not an end in itself – it is an interface between data and team discussion. The value of correspondence analysis and a perceptual map comes from the fact that this interface is information-dense while remaining understandable to non-analytical audiences, provided it has been prepared with due regard for the technique’s limitations.<\/p>\n A perceptual map is a graphical representation of relationships between brands and brand image attributes in a space of two or three dimensions derived from the data. It is most often created using correspondence analysis, principal component analysis, or multidimensional scaling. The proximity of points may reflect the strength of association co-occurrence, the similarity of brand profiles, or perceptual similarity, depending on the method and visualization approach used.<\/p>\n Axes do not have predefined meanings – their content is determined by the attributes that make the greatest contribution to a given dimension. In practice, an axis is named by identifying the poles it runs between, for example, from traditional to modern. It is always necessary to check what percentage of inertia is explained by the pair of axes, because this determines the reliability of the flat projection.<\/p>\n It does not replace it – it complements it. CA shows the structure of associations quantitatively, but it does not explain why a given brand occupies its position or what language consumers use when speaking about it. The methodological standard is a sequence in which a qualitative exploratory phase establishes attributes and hypotheses, quantitative measurement builds the map, and an additional qualitative stage explains the mechanisms observed in the chart.<\/p>\nWhat distinguishes a project that delivers a useful map?<\/h2>\n
\n
Frequently asked questions<\/h2>\n
What is a perceptual map?<\/h3>\n
How should axes on a perceptual map be interpreted?<\/h3>\n
When does correspondence analysis replace qualitative briefing?<\/h3>\n
Ask about a brand positioning analysis for your category<\/h2>\n