Violin Plot Maker

A violin plot shows the distribution of a numeric variable for one or more groups. Each group gets a mirrored kernel density curve: the outline is wide where many values fall and narrow where few do, which gives the shape its name. Most violin plots also carry a small box plot or quartile lines inside, so the median and interquartile range stay readable.

The chart was introduced by Hintze and Nelson in The American Statistician (1998) as a box plot combined with a density trace. It is common in biology, neuroscience, psychology and machine learning papers, anywhere a group's spread or shape matters as much as its average. People also call it a violin chart or violin graph; they are the same thing.

Read each violin from the value axis outwards:

• Width at any height is the estimated density there. Wider means more observations near that value.
• Peaks are the most common values. One bulge means one cluster; two bulges (a bimodal violin) usually mean two subgroups mixed together.
• Ends of a trimmed violin are the smallest and largest values. Long, thin ends are tails or outliers.
• Inner box: the thick bar spans the first to third quartile (the middle 50% of values), the white dot is the median, and the thin line reaches the furthest values within 1.5 × IQR of the box. Quartiles are computed by linear interpolation, the default in R, ggplot2 and NumPy and the same values as Excel's QUARTILE.INC; SPSS and Minitab use a different default, so their quartiles can differ slightly for small groups.
• Mean (◇, when shown) sitting far from the median is another sign of skew.

Two cautions. The width is a smoothed estimate, so with a small n the outline reflects the smoothing more than the data; turn on the data points to check. And with the default equal area scaling every violin has the same area, so a narrow violin is not a smaller group. Switch Violin width to By count if group sizes should show.

Raw values, one observation per cell. Not means, not counts per bin. Two layouts work:

Details the tool handles for you:

• Separators: tabs (what Excel and Google Sheets copy), commas, semicolons or spaces. Semicolon files with decimal commas (3,14) are read as European CSV.
• Unequal groups: columns can have different lengths; blank cells are ignored.
• Extra columns: in long data with an ID column (sample_id, group, value), the text column with the fewest distinct values is used as the group, and the last numeric column as the value.
• Non-numbers such as NA or n/a are skipped and counted below the box, so you can see if something was dropped.

Groups appear in the order they first appear in your data. To reorder, move the columns or rows.

The outline is a kernel density estimate, and its shape depends on one setting: the bandwidth, i.e. how far each data point is smeared out. A small bandwidth gives a bumpy outline that follows every cluster; a large one gives a smooth bell that can hide real peaks.

So with default settings this tool matches ggplot2, and a seaborn violin of the same data will look slightly smoother and longer. To get closer to seaborn, choose Scott and turn off Trim tails. The exact bandwidth for each group is listed in the summary table under the plot: report it in your figure legend or methods if a reviewer may want to reproduce the figure, and do not tune the smoothing until a peak you were hoping for appears.

• Excel has a built-in box and whisker chart but no violin chart type. Building one by hand means computing the density in cells and drawing two mirrored area series per group. The quicker route is to copy your Excel columns, paste them here, and insert the PNG back into your workbook or report.
• R: ggplot(df, aes(group, value)) + geom_violin() + geom_boxplot(width = 0.1). Same default bandwidth as this tool.
• Python: seaborn.violinplot(data=df, x="group", y="value"), or plotly.express.violin for an interactive version.
• GraphPad Prism draws violin plots from a column table.

If you already work in R or Python, code is the reproducible choice. This page is for when you need the figure now: for a thesis chapter, a lab meeting slide or a quick check of your data's shape, without installing anything or writing code. For the colors, see the scientific color palette generator; for a bell-shaped reference curve, the bell curve maker.

Reviewers check the same handful of things:

1. n per group, on the plot or in the legend. The tool prints n = … under each group by default.
2. What the inner marks are: e.g. “white dot, median; thick bar, interquartile range; thin line, 1.5 × IQR”.
3. The bandwidth or method: “kernel density, Gaussian kernel, Silverman's bandwidth”.
4. Raw points for small groups: for fewer than about 20 values per group, overlay the data points or use a dot plot instead.
5. Units on the value axis: Plant height (cm), not Height.
6. Colors that survive grayscale and colorblindness: color here only identifies the group, and the group names are written on the axis, so the figure still reads in black and white.

A sample legend you can adapt: “Violin plots show kernel density estimates (Gaussian kernel, Silverman's bandwidth) of plant height per treatment; inner boxes show the median (white dot) and interquartile range; n = 30 per group.”