Bean plot

Each bean is a mirrored density curve of one group; each short line inside is one observation and the thick line is the group mean.

Placed side by side, two groups can be compared for location, spread and shape while every data point stays visible; this suits moderate sample sizes (tens to hundreds).

Brackets show stars for the adjusted P value: * < 0.05, ** < 0.01, *** < 0.001, ns = not significant. Two groups use the Welch t test or Mann–Whitney U test; more groups use ANOVA with Tukey or Kruskal–Wallis with Dunn (BH-adjusted).

Choose another value column to compare other measures; with very many values the lines merge and a violin plot is clearer.

Method

Boxes span the quartiles (Q1–Q3, R type 7) and whiskers reach the furthest value within 1.5 × IQR; densities use a Gaussian kernel (Silverman bandwidth). Tests: Welch t test, Mann–Whitney U (as scipy), one-way ANOVA with Tukey HSD, Kruskal–Wallis with Dunn’s test (Dunn 1964, Technometrics 6:241–252); multiple comparisons are adjusted by Benjamini–Hochberg.

Data size

Up to 40 groups; at most 10 comparison brackets are drawn (smallest P first); all comparisons are in the downloadable table.