Box plot

Updated
View as text
Open in Confluence
Open in ChatGPT Open in Claude

A box plot compresses a distribution into five numbers and shows the outliers separately. It is the right chart when comparing distributions across groups rather than a single average.

Anybody comparing performance across services or teams should be looking at one of these rather than at a bar chart of means.

#When this is the right chart

  • Comparing distributions across groups.

  • Performance data, where the tail matters more than the mean.

  • Any comparison currently being made with averages.

#What the data needs to look like

Field

Type

Group

Nominal

Value

Quantitative: the raw measurements

#A worked example

Latency by service, showing the spread:

Box plot

The source, which you can paste into a new diagram and edit:

{
  "$schema": "https://vega.github.io/schema/vega-lite/v5.json",
  "description": "Response time distribution by service",
  "width": 380,
  "height": 240,
  "data": {"values": [
    {"service": "api", "ms": 42}, {"service": "api", "ms": 61}, {"service": "api", "ms": 70}, {"service": "api", "ms": 78}, {"service": "api", "ms": 96}, {"service": "api", "ms": 310},
    {"service": "search", "ms": 120}, {"service": "search", "ms": 160}, {"service": "search", "ms": 180}, {"service": "search", "ms": 240}, {"service": "search", "ms": 610}, {"service": "search", "ms": 980},
    {"service": "reporting", "ms": 900}, {"service": "reporting", "ms": 1400}, {"service": "reporting", "ms": 1800}, {"service": "reporting", "ms": 2400}, {"service": "reporting", "ms": 5200}
  ]},
  "mark": {"type": "boxplot", "extent": "min-max"},
  "encoding": {
    "x": {"field": "service", "type": "nominal", "title": null},
    "y": {"field": "ms", "type": "quantitative", "title": "Response time (ms)", "scale": {"type": "log"}}
  }
}

#A few things that catch people out

  • Say what the whiskers mean. Min-max and 1.5 times the interquartile range are both common and different.

  • Box plots hide bimodal distributions. If you suspect one, use a histogram too.

  • With few data points per group, plot the points instead. A box plot of five values is theatre.

Pick the chart for the question, not for the look.