Significance letters (compact letter display), explained

The small letters above bars in papers — a, ab, b — are the compact letter display (CLD): a summary of every pairwise comparison in one glyph per group — it reproduces the significance pattern exactly, though not the p-values or effect sizes behind it (reviewers sometimes want those too, which the results table provides). The rule is simple: groups sharing a letter are not significantly different; groups sharing none are.

How the letters are made

First, a multiple-comparison procedure tests every pair of groups while controlling the overall error rate — classically Tukey’s HSD, with the Tukey–Kramer extension handling unequal group sizes. The resulting yes/no matrix of differences is then compressed into the fewest letters that reproduce it exactly. Done by hand this is genuinely fiddly — four groups already have six pairs — which is why letters are so often copied wrongly between drafts; MinutePlot computes and places them on the figure automatically.

Reading edge cases

“ab” means the group is statistically indistinguishable from both the a groups and the b groups — typically a middle group the experiment lacked power to separate. Letters can also be non-transitive: A may differ from C while B differs from neither; the display “a, ab, b” states exactly that, and it is a fact about evidence, not an error.

What to write in the caption

“Different letters indicate significant differences between groups (Tukey HSD, p < 0.05); groups sharing a letter do not differ significantly.”

MinutePlot generates this caption sentence with every lettered figure, matched to the test actually used.

When not to use Tukey

Tukey’s HSD controls the family-wise error rate — right for confirmatory all-pairs comparisons. In screening settings with very many comparisons, some fields prefer to control the false discovery rate instead (Benjamini–Hochberg), which preserves power; MinutePlot’s post-hoc options are Tukey HSD, Games–Howell and Dunn, so this is background rather than a setting.

Try it yourself — in a minute. Open MinutePlot in your browser and try it on your own data — free 14-day trial, no card needed.
References
  1. Tukey, J. W. (1949). Comparing individual means in the analysis of variance. Biometrics, 5(2), 99–114.
  2. Kramer, C. Y. (1956). Extension of multiple range tests to group means with unequal numbers of replications. Biometrics, 12, 307–310.
  3. Benjamini, Y., & Hochberg, Y. (1995). Controlling the false discovery rate: a practical and powerful approach to multiple testing. Journal of the Royal Statistical Society: Series B, 57(1), 289–300.