MerQuR+Statistics Glossary

Statistics Glossary → Standardized and adjusted residuals (contingency tables)

Standardized and adjusted residuals (contingency tables)concept

Values showing, on a comparable scale, how far the observed count in each cell of a contingency table departs from the expected count.

The chi-square test says whether there is an association across the whole table, but not which cells drive it. The standardized (Pearson) residual is computed for each cell as (Observed - Expected) / √Expected; the sum of the squared residuals is the chi-square statistic. The adjusted (Haberman) residual divides this value by a denominator that also accounts for the row and column proportions, giving a quantity that is approximately standard normal under independence.

Cells whose adjusted residual exceeds about 1.96 (roughly 2) in absolute value contain significantly more (positive sign) or fewer (negative sign) observations than expected. Because Pearson residuals have a slightly smaller variance than a standard normal variable, the same threshold is somewhat conservative for them. In large tables many cells are inspected at once, so some large residuals can appear by chance; these values should be read as aids to understanding the pattern of association rather than as formal tests.

An everyday example

A school asks pupils which subject they like best, cross-tabulates the answers by sex, and the chi-square test is significant. The adjusted residuals show that only the 'girls and biology' cell is +3.1 and the 'boys and biology' cell -3.1, while every other cell stays within ±1. The association therefore comes mostly from a difference in liking for biology.

In MerQur

In Chi-Square Independence, the 'Standardized residuals' (Pearson, |r| > 2 flagged) and 'Adjusted residuals' (Haberman, |z| > 1.96 flagged) checkboxes; Pearson residuals in Chi-Square Goodness of Fit. In the Log-Linear Model, the 'Cell frequencies and residuals' output gives raw, standardized, adjusted and deviance residuals.

Analyses where this concept appears

Chi-Square IndependenceChi-Square Goodness of FitLog-Linear Model

Related concepts

Expected frequencyResidual plots and residual diagnostics