MerQuR+Statistics Glossary

Statistics Glossary → Covariate

Covariateconcept

A variable, usually continuous, that is not of primary interest but is added to the model because it affects the dependent variable, so that its influence is statistically controlled.

When a covariate enters the model, part of the variability in the dependent variable is explained by it and the remaining error variance shrinks. The effect of the factor of interest is therefore estimated more precisely and the test gains power. In ANCOVA the group means are adjusted as if every group had the same covariate value (usually the overall mean); these adjusted means are called estimated marginal means.

Covariate adjustment rests on assumptions: the relationship between covariate and outcome should be linear and its slope the same in every group (homogeneity of regression slopes). The covariate should not itself be affected by the factor; using a variable measured after treatment and influenced by it as a covariate distorts the estimated effect. In observational studies, adding covariates partly controls for confounders, but it cannot remove unmeasured confounding and does not by itself establish causation.

An everyday example

You compare how two textbooks affect maths achievement, but one class was already stronger at the start of term. Adding the start-of-term test score as a covariate lets you compare the two classes 'as if they started from the same point'. You can then separate how much of the end-of-term gap comes from the textbook and how much from the initial difference.

In MerQur

It is entered as 'Covariates' in Two-Way ANOVA and Difference-in-Differences, and as a covariate added to the REML model in the Nested and Crossed Mixed Models; in ANCOVA the 'Covariate checks' extra output examines its assumptions, and in Latent Class Analysis covariates predict class membership. It is also used in repeated-measures and mixed-design ANOVA, MANOVA, multinomial and ordinal regression, Linear Mixed Models, Propensity Score Analysis and GWR.

Analyses where this concept appears

Two-Way ANOVARepeated Measures ANOVAMixed-Design ANOVA, Split-PlotMANOVAANCOVAMultinomial Logistic RegressionOrdinal Logistic RegressionLinear Mixed Models (LMM)Nested Mixed Model (Nested LMM, Multi-Level Hierarchical Mixed Model)Crossed Mixed Model (Crossed LMM, Crossed Random Effects Mixed Model)Propensity Score AnalysisDifference-in-Differences (DiD)Latent Class Analysis (LCA)Variance Components Analysis (VARCOMP)Bayesian ANOVAGWR (Geographically Weighted Regression)Mediation AnalysisCox Regression (Proportional Hazards Model)Parametric Survival (AFT)Competing RisksTime-dependent Cox RegressionInterval-Censored SurvivalCluster-Robust Cox (Marginal and Frailty Models)Response Surface Regression (RSREG)Variogram + Kriging (Spatial Interpolation)

Related concepts

Homogeneity of regression slopesEstimated marginal means (EMM)Types of sums of squares (Type I, II, III)