Analysis of Covariance (ANCOVA)
Analysis of Covariance · Also known as: analysis of covariance, covariance analysis, ANCOVA (Kovaryans Analizi)
ANCOVA is a parametric hypothesis test that compares the adjusted means of two or more independent groups while statistically controlling for one or more continuous covariates. By removing the portion of outcome variance explained by the covariate, ANCOVA increases statistical precision and produces fairer group comparisons. The method builds on the general linear model framework consolidated by Fisher in the early 1930s and is described comprehensively by Tabachnick and Fidell (2013).
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use ANCOVA when you have two or more independent groups, a single continuous outcome, and one or more continuous covariates that correlate with the outcome but are not the focus of the comparison. Four assumptions must hold: (1) residuals are approximately normally distributed — verify with Shapiro-Wilk applied to model residuals; (2) variances are homogeneous across groups — verify with Levene's test; (3) regression slopes are homogeneous — the covariate-by-group interaction must be non-significant; (4) the covariate is measured independently of the treatment — it should not itself be affected by group assignment. Minimum recommended sample size is 30 per group. When normality is violated or the sample is small, a permutation-based alternative is preferable.
Strengths & limitations
- Increases statistical power by reducing error variance attributable to the covariate.
- Produces adjusted group means that offer fairer comparisons when groups differ on a pre-existing characteristic.
- Partial eta-squared provides a directly interpretable effect size for the group factor.
- Well-supported by a century of general linear model theory and available in every major analysis package.
- Requires the homogeneity of regression slopes assumption, which must be verified before interpreting results.
- The covariate must be continuous and measured without substantial error; measurement error in the covariate can bias adjusted means.
- Covariate independence from treatment is essential; if the intervention itself changes the covariate, ANCOVA results are misleading.
- Minimum sample of 30 is needed to estimate the covariate effect reliably.
Frequently asked
How is ANCOVA different from one-way ANOVA?
One-way ANOVA compares raw group means ignoring any continuous covariates. ANCOVA adds a regression step that removes the variance the covariate shares with the outcome before computing the F ratio, producing adjusted means and reducing error variance. The result is typically a more powerful test when the covariate correlates substantially with the outcome.
What happens if the homogeneity of regression slopes assumption fails?
If the covariate-by-group interaction is statistically significant, the covariate-outcome relationship differs across groups, making a single pooled covariate adjustment inappropriate. Options include modelling the interaction explicitly, stratifying the analysis by group, or using Johnson-Neyman region-of-significance analysis to identify where the groups differ.
Can I use a pre-test score measured after randomisation as a covariate?
Only if the covariate was measured before the intervention began and could not have been influenced by group assignment. Using a measure collected after treatment starts introduces post-treatment bias and can produce seriously misleading adjusted means.
Which effect size should I report?
Partial eta-squared is the standard effect size for ANCOVA; it expresses the proportion of variance in the outcome accounted for by the group factor after the covariate is controlled. Values of 0.01, 0.06, and 0.14 are conventionally considered small, medium, and large. Always accompany partial eta-squared with adjusted marginal means and their confidence intervals.
Sources
- Tabachnick, B.G. & Fidell, L.S. (2013). Using Multivariate Statistics (6th ed.). Pearson. ISBN: 978-0205849574
How to cite this page
ScholarGate. (2026, June 1). Analysis of Covariance. ScholarGate. https://scholargate.app/en/statistics/ancova
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Kruskal-Wallis testStatistics↔ compare
- MANCOVAStatistics↔ compare
- One-way ANOVAStatistics↔ compare
- Welch t-testStatistics↔ compare