Correspondence Analysis Explained

What Is Correspondence Analysis?

Correspondence analysis (CA) is a multivariate statistical technique that visualizes the relationships between categories of two or more categorical variables on a single map. It takes a contingency table, like brands crossed with attributes, or products crossed with usage occasions, and reduces it to a two-dimensional plot where proximity indicates association. Brands that cluster near certain attributes are perceived as having those characteristics. Attributes that appear far from any brand represent white space in the market. The output looks like a scatter plot, but the positions aren't arbitrary, they're mathematically derived to capture the maximum amount of association structure in the fewest dimensions. It's one of the most powerful tools for making cross-tabulation data visually interpretable.

Why Correspondence Analysis Matters

Cross-tabulation tables are the backbone of survey analysis, but a 10-brand by 15-attribute table is nearly impossible to interpret by scanning numbers. Correspondence analysis transforms that table into a visual map where patterns jump out: which brands share similar perceptions, which attributes differentiate the category, and where positioning gaps exist. It answers questions that stakeholders care about, "How is our brand perceived relative to competitors?", in a format they can grasp in seconds.

How Correspondence Analysis Works

The Input Data

Correspondence analysis starts with a contingency table or cross-tabulation. The most common setup in market research is a brand-attribute association matrix:

Each cell contains a count or percentage, how many respondents associated that brand with that attribute.

The Mathematical Process

CA works through a process called singular value decomposition (SVD) applied to the standardized residuals of the contingency table:

1. Calculate expected values: what each cell would be if rows and columns were independent (no association).
2. Compute residuals: the difference between observed and expected values, standardized to account for row and column totals.
3. Decompose the residual matrix: extract dimensions (like principal components) that capture the most variance in the association patterns.
4. Project row and column categories onto these dimensions, creating coordinates for each brand and each attribute in the reduced-dimension space.

The first two dimensions typically capture 60-80% of the total variance (called inertia in CA terminology), making a 2D map a reasonable summary of the association structure.

Reading the Map

Interpreting a correspondence analysis map requires understanding a few rules:

Proximity indicates association. A brand positioned near an attribute is disproportionately associated with it. Brand B sitting close to "Innovative" and "Easy to Use" means respondents linked Brand B to those traits more than expected.

Distance from the origin indicates distinctiveness. Points near the center of the map have average profiles, they aren't strongly associated with anything in particular. Points far from the center have distinctive profiles.

Row-to-row and column-to-column distances are meaningful. Two brands close together have similar attribute profiles. Two attributes close together are associated with the same brands.

Row-to-column distances require caution. While proximity between a brand and an attribute suggests association, the geometric relationship between row and column points is technically angular rather than Euclidean. In practice, the directional interpretation (brands and attributes in the same region are associated) holds well for most research applications.

Supplementary Points

You can add supplementary (passive) points to the map without affecting the solution. This is useful for:

• Plotting a new brand or concept without recalculating the map.
• Adding demographic segments to see where different customer groups fall in the brand space.
• Projecting ideal brand positions to identify aspirational targets.

Multiple Correspondence Analysis

Standard CA handles two categorical variables. Multiple correspondence analysis (MCA) extends the technique to three or more variables simultaneously. It's commonly used in lifestyle and psychographic research where you want to map the relationships among many categorical responses at once.

When to Use Correspondence Analysis

• Brand perception mapping: visualizing how brands are positioned relative to each other on key attributes.
• Category structure analysis: understanding how products, occasions, and benefits relate to each other in consumers' minds.
• Competitive positioning: identifying white space (attribute combinations not owned by any brand) and overlaps (where you're competing head-to-head).
• Communication strategy: determining which attributes to emphasize based on where your brand currently sits and where you want it to move.
• Tracking studies: mapping brand positions over time to see whether repositioning efforts are working.

Common Mistakes to Avoid

• Over-interpreting small distances: points that are close together on the map may not be statistically distinguishable. Check the contribution of each point to the dimensions and the quality of representation (cos-squared values) before drawing conclusions.
• Interpreting dimensions as fixed constructs: the axes in correspondence analysis are mathematical abstractions, not named dimensions. Labeling them requires judgment based on which attributes load heavily on each axis. Avoid forcing interpretive labels that don't fit the data.
• Using correspondence analysis with continuous data: CA is designed for categorical data (counts, frequencies). If your data is continuous (rating scales), discretize it first or use principal component analysis instead.

Quali-Fi Support

Quali-Fi's cross-tabulation engine produces the brand-attribute association matrices that feed directly into correspondence analysis. The platform's data export to SPSS, R, and Tableau supports CA computation, and for teams using Quali-Fi's Intelligence product, pre-configured brand perception studies include perceptual mapping as a standard output.

Frequently Asked Questions

How is correspondence analysis different from factor analysis?

Factor analysis works with continuous data (like rating scales) and identifies latent dimensions. Correspondence analysis works with categorical data (like brand-attribute associations) and visualizes the association structure. They answer different questions: factor analysis asks "what underlying dimensions explain these ratings?" while CA asks "how are these categories related to each other?"

How many dimensions should I extract?

Two dimensions are standard for visualization and usually sufficient. If the first two dimensions explain less than 60% of inertia, consider examining a third dimension separately (plotted as dimension 1 vs. 3) rather than trying to interpret a 3D map.

Can I use correspondence analysis with small samples?

CA is sensitive to sample size because it's based on contingency tables. Low cell counts produce unstable maps. As a rule of thumb, you need at least 200-300 total respondents, and individual cell counts should ideally be 20+ for stable results.

Related Topics

• Perceptual Mapping
• Cluster Analysis (Segmentation)
• Segmentation Analysis
• Key Driver Analysis
• Data Visualization for Research
• Data Collection Methods

Build the brand-attribute data that powers correspondence analysis. Start your free 14-day Quali-Fi trial, no credit card required.