MerQuR+Statistics Glossary

Statistics Glossary → Homogeneity of regression slopes

Homogeneity of regression slopesconcept

The ANCOVA assumption that the relationship (slope) between the covariate and the dependent variable is the same in every group.

ANCOVA compares group means after removing the effect of a covariate. In doing so it assumes that the same slope linking the covariate and the dependent variable holds in every group, so that the regression lines are parallel. Only then is the difference between groups the same at every value of the covariate, so that a single 'adjusted mean difference' makes sense.

The assumption is checked by adding a factor x covariate interaction term to the model and testing it. If the interaction is significant, the slopes differ: the group difference depends on the value of the covariate and a single adjusted difference is misleading. The interaction model should then be reported, the difference examined at several meaningful covariate values, or the Johnson-Neyman technique used to find the covariate range in which the difference is significant.

An everyday example

Two diet programmes are compared, taking starting weight into account. In the first, both heavier and lighter starters lose about 5 kg; in the second, heavier starters lose 10 kg but lighter ones only 1 kg. Because the slope of weight loss on starting weight differs between the programmes, the answer to 'which programme is better?' depends on the person's starting weight.

In MerQur

In ANCOVA, the 'Assumption checks: Homogeneity of slopes' checkbox adds the factor x covariate terms to the model and runs a nested-model F test.

Analyses where this concept appears

ANCOVA

Related concepts

CovariateJohnson-Neyman technique (separate slopes)