covariate

In statistics, a covariate is a variable that is possibly predictive of the outcome and is not the primary variable of interest. Covariates are collected to adjust for differences among study units or to reduce residual variance, thereby increasing the precision of estimated effects.

In experimental and observational studies, covariates are included in statistical models (such as linear regression or analysis of covariance) to account for variability associated with these variables. This adjustment helps control for confounding and can improve the accuracy of estimated associations between the main exposure and outcome.

Examples include age, sex, baseline disease severity, body mass index, or year of diagnosis. In a randomized trial, covariates may be balanced by randomization, but adjustment remains common to increase statistical power or to account for imperfect balance between groups.

Covariates differ from confounders, mediators, or moderators, though a covariate may act as a confounder if it influences both exposure and outcome. A mediator lies on the causal pathway between exposure and outcome, while a moderator changes the strength or direction of that relationship. All are considerations in model specification.

In other contexts, covariates equal features in predictive modeling; some covariates may be fixed (measured once) while others are time-varying in longitudinal analyses. Related concepts include covariate shift, which refers to differences in the distribution of covariates between training data and target populations.