What it means
Comparing two averages is easy; comparing several at once invites false conclusions, because running many separate two way comparisons raises the chance of finding a difference that is not there. ANOVA solves that by asking one question across all the groups at the same time.
The answer is a single test statistic called F, together with a probability value. In business settings the groups are usually things you can control: three price points, four store layouts, five supplier batches, or several marketing channels.
The outcome is something measurable such as basket value, conversion rate, defect count or days to pay. Used well, the test keeps a team from rebuilding a process on the strength of a difference that is pure noise.
The arithmetic splits the total variation in the data into two parts, the variation between group averages and the variation within groups. Each part is divided by its degrees of freedom to give a mean square, and F is the ratio of the between group mean square to the within group mean square.
A large F, checked against a critical value or a probability value, means the group averages are unlikely to be the same. Two points of nuance matter in practice.
ANOVA tells you that at least one group differs, not which one, so a follow up comparison is needed to identify it. It also assumes the groups have broadly similar spread and that observations are independent, so results from clustered or heavily skewed data should be treated with care.
In practice
Real-world examples.
Example
A coffee chain tests three menu board designs across 60 matched stores and measures average basket value. ANOVA shows the design effect is unlikely to be chance, and a follow up comparison identifies the second design as the one driving the gain.
Example
A manufacturer receives castings from four suppliers and records defect counts per thousand units. The test shows the suppliers are not equivalent, which gives procurement the evidence to renegotiate with the weakest of them.
Example
A finance team compares days to pay across five customer segments to decide where to focus collections effort. ANOVA confirms a real difference between segments, so the credit controller reassigns two staff to the slowest paying group.
Formula
Calculation
F = mean square between groups / mean square within groups. Take average deal values from three sales regions, four deals each, in thousands of dollars. Region A records 10, 12, 14, 12 with a mean of 12; Region B records 16, 18, 20, 18 with a mean of 18; Region C records 22, 24, 26, 24 with a mean of 24. The grand mean is 18. Between group sum of squares is 4 x [(12 - 18) squared + (18 - 18) squared + (24 - 18) squared] = 4 x [36 + 0 + 36] = 288, with 2 degrees of freedom, so the mean square between groups is 144. Within each region the deviations are -2, 0, 2 and 0, giving 8 per region and 24 in total, with 9 degrees of freedom, so the mean square within groups is 24 / 9 = 2.667. F is therefore 144 / 2.667 = 54, far above the critical value of roughly 4.3 from standard tables at the 5% level, so the regional averages are not plausibly equal.Case study
Seen in the real world.
Pellard Home Fittings is an illustrative, fictional retailer that trialled three delivery promises: next day, two day and a named hour slot. Each option ran in 20 stores for eight weeks, and the measure was average order value.
The marketing team saw the named hour slot produce the highest average and wanted to roll it out everywhere at an annual cost of $2,400,000. The analytics lead ran an analysis of variance and found the differences were well within the normal week to week variation of the stores, so the apparent winner could not be separated from noise.
In this illustrative case the rollout was paused and a longer trial run instead. The test did not tell the company what to do, but it stopped a seven figure commitment being made on the strength of a chart.
Watch out
Common mistakes.
- Running many separate two group comparisons instead of one ANOVA, which inflates the chance of finding a difference that is not real.
- Concluding which specific group is best from the ANOVA alone, when the test only says that at least one group differs.
- Ignoring the assumptions, especially unequal spread between groups and observations that are not independent of each other.
Questions
People also ask.
What does a high F value actually mean?
It means the differences between group averages are large compared with the variation inside the groups, so chance is an unlikely explanation.
How many observations do I need per group?
Enough for the variation within groups to be measured sensibly, and in business testing that usually means tens rather than a handful per group.
Is ANOVA the right test for two groups?
It will work and gives the same conclusion as a two sample t test, but the t test is the simpler and more usual choice for two groups.
From the founder's library
Take it further with the book.
Build your financial confidence beyond this definition. Shihan's full-length guide, Accounting Fundamentals, takes the same plain-English approach and turns it into a complete, practical playbook for non-finance managers, business owners and students - with chapter-end quiz answers and presentation slides included.
25% off with code MMHQ25, applied at checkout. Priced in USD - checkout may show the equivalent in your local currency.
View the book and save 25%