The group deviations are constrained by the overall mean, leaving k - 1 independent degrees of freedom in the basic one-way comparison.

Can MS between be smaller than MS within?

Yes. Sampling variation can produce that result even when both estimate the same variance under the null.

Does MS between identify the best group?

No. It summarizes overall variation among means rather than ranking practical desirability.

AI Glossary / Statistics and probability / Intermediate

MS between groups (MS group )

Definition

MS between groups, also written MS group or between-groups mean square, is the between-groups sum of squares divided by its degrees of freedom in an analysis of variance. In a basic one-way ANOVA with k groups, its denominator is k - 1. The numerator measures how far group sample means lie from the overall mean, weighted by group sizes. It is compared with the within-groups mean square to assess whether observed differences among group means are large relative to variation within groups. It is not itself a probability or a standalone measure of practical importance.

What it means

Group averages differ partly because populations may differ and partly because samples fluctuate. The between-groups mean square reflects both possibilities. Under the classical equal-means null hypothesis and common-variance assumptions, it estimates the same variance quantity as the within-groups mean square. When population means differ, it tends to include an additional contribution from those differences. Calling it between groups does not mean it contains only a pure treatment effect. Its interpretation depends on the ANOVA design, and more complex designs require the appropriate model-specific sums of squares and comparison terms.

How it works

For group i, calculate its sample size ni and mean yi-bar. Calculate the overall mean using all observations, so larger groups receive more weight. Form SS between = sum of ni times (yi-bar minus overall mean) squared. Divide by k - 1 to obtain MS between. In the basic one-way model, divide this by MS within to obtain the F statistic. Interpret that ratio using its numerator and denominator degrees of freedom and the model assumptions. Inspect the group data and residuals, and use planned contrasts or suitable follow-up comparisons when the scientific question requires more than an omnibus test.

Three practical examples

Process comparison: compare average output measurements across several machines under an appropriate sampling design.
Training evaluation: summarize differences among average assessment scores for independently assigned training conditions.
Service analysis: examine whether average completion times differ across defined operating groups. These applications need attention to independence, variance assumptions and possible confounding before differences are attributed to the grouping factor.

Think of it this way

Think of comparing the centers of several clusters of points. The between-groups calculation measures how spread out those centers are, while accounting for how many observations support each center. To judge whether that spread is surprising, it must be compared with the spread of points around their own centers. Widely separated averages can be more or less convincing depending on that background variability.

Worked example

Consider three artificial groups: A contains 1 and 3, B contains 4 and 6, and C contains 7 and 9. Their means are 2, 5 and 8, with overall mean 5. SS between is 2(2 - 5)^2 + 2(5 - 5)^2 + 2(8 - 5)^2 = 36. With k - 1 = 2 degrees of freedom, MS between is 36/2 = 18. Within each group, deviations from its mean are -1 and 1, so SS within is 6. Its degrees of freedom are 6 - 3 = 3, giving MS within = 2 and F = 18/2 = 9. The ordinary sample variance of the three group means is 9, not 18; the between-groups mean square includes the group-size weighting. This original arithmetic example does not establish a causal effect or report a significance level without the relevant inferential assumptions.

Illustrative case study

In a fictional case, Willow Manufacturing compares measurements from three machines. The report initially highlights a large MS between value without stating the measurement units or within-machine variability. Analysts add the full ANOVA table, group summaries and residual checks. They also discover that machines were operated on different shifts, making machine and shift effects difficult to separate. The revised interpretation distinguishes the observed group differences from a causal claim about the machines themselves.

Common misconceptions and mistakes

MS between is not generally the unweighted variance of group means. Unequal group sizes require the specified weighting, and the grand mean is based on all observations. A large numerical value can simply reflect the scale of measurement. Multiplying every observation by ten multiplies mean squares by one hundred, while leaving the F ratio unchanged. A significant omnibus result does not identify which pairs differ. The basic F comparison also depends on independence and appropriate variance and distributional assumptions; clustering or unequal variances can require a different analysis.

Business use

Businesses should report MS between as part of a complete analysis rather than as a performance score. Include sample sizes, group means, uncertainty and the within-group comparison. Explain whether the groups arise from random assignment, ordinary observation or a hierarchical process. Evaluate effect sizes and operational consequences alongside statistical evidence. When the design is unbalanced or includes several factors, document how the sums of squares were defined. This helps prevent a familiar ANOVA label from obscuring a different statistical question or a confounded comparison.

Check your understanding

If SS between is 36 for three groups, what is MS between?

Answer: 36 divided by 3 - 1, which is 18. Dividing by the total number of observations would be incorrect.

Related terms

Degrees of freedom
Contrast
ANOVA
between-groups sum of squares
within-groups mean square
F statistic
grand mean

Learn next

Work through the sum-of-squares decomposition for a small one-way ANOVA. Then examine how unequal group sizes and repeated measurements change the design and the appropriate analysis.

One-way ANOVA

Sources

NIST/SEMATECH e-Handbook, One-Way ANOVA: https://itl.nist.gov/div898/handbook/ppc/section2/ppc231.htm NIST/SEMATECH e-Handbook, One-Factor ANOVA: https://itl.nist.gov/div898/handbook/eda/section3/eda354.htm

Previous: MRF optimization via dual decomposition Next: MS-SSIM