One-Way ANOVA

Test at once whether several groups - catalysts, conditions, operators - have the same mean. The ANOVA table, effect size, equal variance check, post hoc tests and residual diagnostics are all on one screen.

1Data Table

Paste a table copied from Excel into a cell (Ctrl+V). Click a column header (A, B…) before pasting to use the first line as column names. Double-click a header to rename it, right-click to insert/delete rows and columns, and formulas such as =A1*2 work too.

2One-Way ANOVA Options

How to run a one-way ANOVA

  1. Copy a group name column and a measurement column together with the header.
  2. Select the response column and the factor (group) column, and choose a post hoc method if needed (it tells you which pairs differ when the groups differ significantly).
  3. Press Analyze and check the p-value and effect size η² in the ANOVA table and the post hoc results.

Worked example with sample data

The sample data are 8 yield measurements (%) for each of 4 catalysts (A to D). The Catalyst column is the factor (group) column and the Yield column the response column.

Group means with 95% confidence intervals from the one-way ANOVA of catalyst A to D yields
Group means with 95% confidence intervals from the one-way ANOVA of catalyst A to D yields

The mean yields are A 82.39, B 84.9, C 84.35 and D 87.49%. The ANOVA table gives F(3, 28) = 7.962 and p-value 0.0005422, so the mean differs by catalyst, and the effect size η² = 0.4603 is large (the catalyst explains about 46% of the variation in yield).

The Levene check gives p=0.4947, so there is no problem with the equal variance assumption and the usual ANOVA is fine. Choosing Tukey HSD as the post hoc method shows A-D (mean difference 5.1, p=0.0002) and C-D (3.138, p=0.0286) as significant; the other pairs are not.

In the means with confidence intervals chart, D's interval is the highest. The Anderson-Darling p=0.7424 in the residual QQ plot shows the residuals can be regarded as normal, so the assumptions look fine.

How it is calculated

The total variation is split into variation between groups (SS factor) and within groups (SS residual), and F = MS factor / MS residual is compared with the F distribution with (k-1, N-k) degrees of freedom. The effect size is η² = SS factor / (SS factor + SS residual).

The equal variance check uses the Levene test (median-based) to see whether the group variances are similar. Groups with fewer than 2 valid values are excluded.

Post hoc tests

Tukey HSD compares all pairs with the studentized range distribution and assumes equal variances. Games-Howell does not assume equal variances, and Bonferroni/Holm correct the p-values of Welch t tests for multiple comparisons.

What the result shows

Descriptive statistics by group, the ANOVA table, the equal variance check, the post hoc table, a box plot, an individual value plot, means with confidence intervals, residuals vs fitted values, a residual QQ plot and a residual histogram.

How to read the one-way ANOVA result

  • If the ANOVA p-value is below the significance level, at least one group mean differs. To see which pairs differ, look at the 'Significant' rows of the post hoc table.
  • η² is read as a small, medium or large effect at 0.01, 0.06 and 0.14. Even with a small p-value, a small η² may mean little practical difference.
  • If the Levene check casts doubt on equal variances or the residual QQ plot departs strongly from the line, also check with Welch ANOVA.

Frequently asked questions

Is it the same as Excel's ANOVA: Single Factor?

Yes. You get the same SS, MS, F and p-value as 'Anova: Single Factor', plus post hoc tests and residual diagnostics.

Can I use it with 2 groups?

Yes. With 2 groups the ANOVA p-value equals the two-sided p-value of the equal variance 2-sample t test.

What if there are two factors?

For two factors such as temperature and pressure, use the two-way ANOVA.

Other chart tools