Correspondence analysis¶

A dimension-reduction and visualization method for contingency tables using chi-square geometry to jointly map row and column profiles.

Version: v2 · 2026-09-19 · History
Domain-specific #: 3925
Domain group: Formal Sciences
Origin domain: Experimental Design & Statistics
Subdomain: Multivariate Statistics → Experimental Design & Statistics

Core Idea¶

Simple, multiple and canonical correspondence analyses are distinct; masses, expected counts, rare categories, supplementary points and scaling convention control interpretation. Observed counts are normalized into profiles, deviations from independence are weighted by row and column masses and singular-value decomposition yields low-dimensional principal coordinates. The abstraction is therefore identified by a declared carrier, a transformation or constraint over that carrier, and an invariant that tells an analyst whether the named structure is genuinely present.

The load-bearing residual is not the broad topic of multivariate statistics. It is the domain-specific identity determined by the nonnegative table and population, row and column categories, total and masses, independence baseline and chi-square metric, standardized residual matrix, singular values and inertia, retained dimensions, row and column scaling and supplementary and uncertainty treatment are explicit.

How would you explain it like I'm…

Picture From a Count Chart

Imagine a chart counting how many kids in each class picked each ice cream flavor. Correspondence analysis turns that big chart of numbers into a simple picture that shows where the choices are different from what you'd expect if every class picked flavors in the same way.

Turning Count Tables into Maps

Correspondence analysis is a way to turn a big table of counts into a picture. Say the rows are towns, the columns are ice cream flavors, and each box counts how many kids in that town chose that flavor. First, each town's counts are turned into shares (percentages) so big and small towns can be compared. Then the method looks at how each town and flavor differs from what you'd expect if every town liked flavors in the same proportions. It squeezes those differences into a simple map with just two or three directions, where similar towns sit near each other and similar flavors sit near each other. How to read the map depends on choices about scaling, so those choices need to be stated.

Chi-Square Profile Mapping

Correspondence analysis (CA) is a method for visualizing a table of counts, such as a contingency table crossing two categorical variables. It converts each row and column into a profile — its counts divided by its total — and weights each by its mass, the share of the grand total it holds. The baseline is independence, where every row would have the same profile; CA measures deviations from that baseline using the chi-square distance. A singular value decomposition of the matrix of standardized residuals then gives a few principal dimensions that capture most of the 'inertia' (total chi-square deviation, scaled). Rows and columns get coordinates on these dimensions, producing a low-dimensional map. Simple, multiple, and canonical correspondence analysis are different methods, and interpretation depends on masses, rare categories, supplementary points added after the fact, and the scaling convention used for the plot.

Simple correspondence analysis takes a nonnegative contingency table with total n, forms the correspondence matrix P = N/n with row masses r and column masses c, and measures departure from independence against the baseline rcᵀ using the chi-square metric. The standardized residual matrix S = D_r^{−1/2}(P − rcᵀ)D_c^{−1/2} is decomposed by singular value decomposition; the squared singular values partition the total inertia, which equals the table's chi-square statistic divided by n. Principal and standard coordinates for rows and columns are obtained from the singular vectors scaled by masses and singular values, and the retained dimensions give a low-dimensional map. The scaling convention — principal, standard, symmetric, or asymmetric — determines which distances are interpretable: within-set distances approximate chi-square distances between profiles, but row-to-column proximity is not a distance in the symmetric map. Rare categories with small mass can be unstable or dominate outlying positions, supplementary points can be projected without influencing the solution, and uncertainty should be addressed. Multiple CA (for several categorical variables) and canonical CA (constrained by explanatory variables) are distinct methods.

Scope of Application¶

Correspondence analysis belongs to multivariate statistics and is useful where the analyst can specify the typed multivariate statistics carrier, defining objects and relations, parameters, conventions, evidence, boundary cases, and comparison targets, then evaluate the nonnegative table and population, row and column categories, total and masses, independence baseline and chi-square metric, standardized residual matrix, singular values and inertia, retained dimensions, row and column scaling and supplementary and uncertainty treatment are explicit. The scope is broad within that domain but bounded by the need for the nonnegative table and population, row and column categories, total and masses, independence baseline and chi-square metric, standardized residual matrix, singular values and inertia, retained dimensions, row and column scaling and supplementary and uncertainty treatment are explicit.

Clarity¶

The abstraction clarifies a crowded vocabulary by making the nonnegative table and population, row and column categories, total and masses, independence baseline and chi-square metric, standardized residual matrix, singular values and inertia, retained dimensions, row and column scaling and supplementary and uncertainty treatment are explicit the center of the account. A claim should name the carrier, the governing operation or relation, the applicable assumptions, and the recognition test.

Manages Complexity¶

Without the abstraction, an analyst must reason directly over many local details: the carrier roles, admissibility assumptions, competing conventions, derived invariants, boundary cases, and proof or validation obligations specific to Correspondence analysis. Correspondence analysis compresses them into the roles in the structural signature. That compression permits comparison across instances without erasing the variables that determine validity. It also exposes which details may be varied safely and which are constitutive.

Abstract Reasoning¶

Identify the carrier. State what the elements, states, objects, or observations are: the typed multivariate statistics carrier, defining objects and relations, parameters, conventions, evidence, boundary cases, and comparison targets. Reject examples whose alleged carrier belongs to a different problem.
Lock the constitutive rule. Express the nonnegative table and population, row and column categories, total and masses, independence baseline and chi-square metric, standardized residual matrix, singular values and inertia, retained dimensions, row and column scaling and supplementary and uncertainty treatment are explicit independently of one notation or implementation.

Knowledge Transfer¶

Knowledge transfers strongly among subfields of multivariate statistics because they reuse the typed multivariate statistics carrier, defining objects and relations, parameters, conventions, evidence, boundary cases, and comparison targets, Observed counts are normalized into profiles, deviations from independence are weighted by row and column masses and singular-value decomposition yields low-dimensional principal coordinates., and type the carrier, state every parameter and convention in the definition, test that the nonnegative table and population, row and column categories, total and masses, independence baseline and chi-square metric, standardized residual matrix, singular values and inertia, retained dimensions, row and column scaling and supplementary and uncertainty treatment are explicit, compare the nearest accepted identity, and report counterexamples, uncertainty, and limiting cases.

Relationships to Other Abstractions¶

Current abstraction Correspondence analysis Domain-specific

Parents (1) — outgoing typed relations

Correspondence analysis is a kind of Dimensionality Reduction Prime

It remains its own entry because its identity is fixed by the nonnegative table and population, row and column categories, total and masses, independence baseline and chi-square metric, standardized residual matrix, singular values and inertia, retained dimensions, row and column scaling and supplementary and uncertainty treatment.

Hierarchy paths (4) — routes to 3 parentless roots

Correspondence analysis → is a kind of Dimensionality Reduction → is a kind of Approximation → is a decomposition of Representation → presupposes Abstraction

Show alternative paths (3)

Neighborhood in Abstraction Space¶

Correspondence analysis sits in a crowded region of the domain-specific corpus (5^th percentile for distinctiveness): several abstractions share nearly its structure, so a description that fits it tends to fit its neighbors too.

Family — Descriptive Statistics & Correlation Measures (40 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-10-25