Statistical control
Also known as: Covariate adjustment, Controlling for
Holding other variables constant in analysis to isolate a relationship.
What it means
Statistical control is the practice of accounting for the influence of additional variables — through regression, stratification, or matching — so as to estimate the relationship between a predictor and an outcome as if those other variables were held fixed. It is the chief tool for reducing confounding in observational data when experimental control is unavailable. Its power is bounded by two hard limits: it can only adjust for variables that were measured, and adjusting for the wrong variables, such as mediators or colliders, can introduce rather than remove bias. Sound control therefore depends on a causal model of which variables belong in the analysis, not on throwing in every available covariate.
Examples
Estimating the effect of exercise on heart disease while statistically controlling for age, diet, and smoking.
A pay-gap study compares men and women in the same role, at the same level, with the same tenure — holding fixed the things that would otherwise do the explaining.
Control can backfire: adjusting the coffee–alertness link for caffeine intake removes the very pathway you meant to measure, and a real effect vanishes on paper.
First described in Regression and survey-analysis tradition.