Is MCA the same as correspondence analysis?

It extends related ideas to several categorical variables.

Does a two-dimensional map show everything?

No. Additional dimensions may contain substantial structure, and individual points can be poorly represented.

AI Glossary / AI and machine learning / Intermediate

Multiple correspondence analysis

Definition

Multiple correspondence analysis is an exploratory method for summarizing associations among several categorical variables. It represents individuals and category patterns in a lower-dimensional space, helping analysts visualize how combinations of responses vary across a dataset.

What it means

A survey may record preferred channel, product category and reason for purchase. These are categories rather than measurements on a continuous numerical scale. Multiple correspondence analysis, or MCA, provides a way to explore their joint structure.

The method is related to correspondence analysis and is often introduced as a counterpart to principal component analysis for categorical data. That analogy is helpful, but MCA uses a particular indicator representation and weighting, so it is not simply ordinary PCA applied to arbitrary category numbers.

The resulting axes are constructed summaries of associations. They do not prove that respondents belong to natural types, that nearby categories cause one another or that a two-dimensional plot preserves every important relationship.

How it works

Each categorical variable is expanded into indicator columns, with a one marking the category selected by an individual and zeros for the alternatives. MCA analyses the structure of this table using correspondence-analysis geometry.

The method produces coordinates, dimensions and diagnostic quantities such as contributions and quality of representation. Analysts inspect which categories help define an axis and whether the displayed dimensions adequately represent the points being discussed.

Rare categories can strongly influence the geometry, so their counts require attention. Missing values and supplementary variables need an explicit treatment. Supplementary information can help interpretation without determining the main fitted axes.

Three practical examples

A survey team explores associations among preferred communication channels, purchase reasons and product categories.
A service analyst examines recurring combinations of categorical issue codes before designing a more focused study.
A researcher uses MCA coordinates as exploratory inputs to clustering, while validating the resulting groups separately.

Think of it this way

Think of arranging survey response patterns on a map so that important differences can be inspected at a glance. The map offers a compact view of a larger table.

The analogy has limits: distances and relationships depend on the chosen scaling, and flattening the information onto two axes can distort it. A point's visible position should be interpreted with its representation diagnostics, not with map-like intuition alone.

Worked example

From categories to indicators

Consider two variables: channel has categories email and phone, while purpose has categories purchase, repair and advice. Each respondent becomes a row with five indicator columns.

A respondent choosing email and repair is coded (1, 0, 0, 1, 0). Someone choosing phone and advice is coded (0, 1, 0, 0, 1). Each complete row contains exactly two ones because there are two active variables.

MCA uses the full table and its category frequencies to construct a lower-dimensional representation. It does not interpret the categories as numerical scores such as email = 1 and phone = 2.

Suppose repair appears near phone on a plot. Before claiming an association, inspect the relevant coordinates, contributions, representation quality and original cross-tabulation. Apparent nearness on a partial projection can be misleading, especially for rare or poorly represented categories. This example illustrates preparation and interpretation, not a hand-calculated MCA solution.

Illustrative case study

Fictional case: exploring customer research

Birch & Bay, an invented homeware business, collects a voluntary categorical survey. An analyst uses MCA to explore response patterns and notices an axis associated with several purchase motivations.

A striking category belongs to only a few respondents. The team checks those records and reports the small count rather than building a major customer segment around the visual result. It uses the exploration to formulate questions for follow-up research.

The case illustrates hypothesis generation and careful interpretation. It does not claim that the analysis established stable market segments or caused a real improvement in sales.

Common misconceptions and mistakes

MCA is not a classifier by itself and does not require a target outcome. Its axes do not automatically carry simple business labels. Assigning numerical codes to categories and running ordinary PCA is not equivalent to the intended analysis.

Distances between different kinds of points on a combined plot require care. Rare categories and low-quality projections can create visually dramatic patterns that are not reliable grounds for a decision.

Business use

MCA can help businesses explore categorical surveys, coded feedback and other records where several nominal variables are important. Its strongest role is often to reveal questions and patterns that deserve closer examination.

Review category definitions, missingness and sample composition before fitting the analysis. Show counts and diagnostics alongside visualizations. Keep supplementary variables separate when they are intended only to aid interpretation. If MCA features support a later predictive model or clustering exercise, evaluate that later step independently. Avoid presenting exploratory axes as causal explanations or fixed identities of the people represented.

Check your understanding

A category with three respondents sits far from the center of an MCA plot. Does that alone establish an important market segment?

Answer: No. Check its frequency, contribution, representation quality and stability. A striking position for a rare category can be an exploratory clue rather than a dependable segment.

Related terms

Correspondence analysis
Categorical data
Exploratory data analysis
indicator matrix
contribution
supplementary variable

Learn next

Learn next how contributions differ from quality of representation. Then compare an MCA plot with the underlying frequency tables so that a visually appealing pattern can be checked against the actual response counts.

Sources

Primary documentation and author teaching material: FactoMineR MCA, https://search.r-project.org/CRAN/refmans/FactoMineR/html/MCA.html ; Magalie Houée-Bigot and François Husson, https://husson.github.io/MOOC_GB/RMarkdown_MCA_Hobbies.pdf . The MCA reference on indicator and Burt formulations, supplementary variables and representation diagnostics was consulted, together with the author tutorial opening and analysis setup. Coding example and case are original.

Previous: Multiple correlation coefficient Next: Multiple discriminant analysis