MerQuR+Statistics Glossary

Statistics Glossary → Inertia, singular values and profiles

Inertia, singular values and profilesconcept

In correspondence analysis: the measure of the total amount of association in a contingency table (inertia), the singular values that share this amount out across dimensions, and the percentage distributions of the rows or columns (profiles).

A row profile is the percentage distribution obtained by dividing the frequencies in a row by the row total; column profiles are defined in the same way. The farther the profiles are from the average profile, the stronger the association between the row and column variables. Total inertia is the weighted sum of these distances and equals χ² / n, that is, the chi-square statistic freed from the effect of sample size. Correspondence analysis splits the total inertia across dimensions through a singular value decomposition of the table's matrix of standardized residuals: each dimension's eigenvalue (its inertia) is the square of its singular value.

The ratio of each dimension's inertia to the total is the share of the association it explains; the first two dimensions are usually shown on the map, and more are examined if the cumulative share is not sufficient. The scale of the map coordinates depends on the normalization: in principal normalization the distances among points of the same kind (rows among themselves) can be interpreted, whereas distances between row and column points cannot be read directly; directional relationships (biplot) are assessed with asymmetric or symmetric normalization. In multiple correspondence analysis (MCA) the inertia is artificially inflated and adjusted inertia rates (e.g. Benzécri, Greenacre) are used.

An everyday example

In a table showing the sales shares of four kinds of coffee in five cities, if the cities' preference profiles are very similar, total inertia is small: there is little association between city and coffee choice. If one city drinks almost nothing but Turkish coffee and another almost nothing but espresso, the profiles move far from the average and inertia grows. If the first dimension explains 85% of this difference, the cities can be ordered along a single 'traditional-modern' axis.

In MerQur

In Correspondence Analysis (CA / MCA) the 'Profiles and singular values' output gives the row and column profiles and, for each dimension, the singular value, eigenvalue and χ² share; the 'Number of dimensions' and 'Normalization (Principal, Symmetrical, Row principal, Column principal)' parameters determine how the map is built.

Analyses where this concept appears

Correspondence Analysis (CA / MCA)

Related concepts

Supplementary (passive) pointsEigenvalue and eigenvector