How to Visualise Sports Data Well

Spatial charts. Shot charts, heatmaps of field or court position, and pitch location plots. The natural choice for location data. They are also the easiest to over-read, for reasons covered below.

Step charts for prices. Odds do not glide from one value to another; they jump. A step chart that holds each price flat until it changes is an accurate representation. Suspensions and missing snapshots should appear as visible gaps.

Win probability and game flow charts. Probability over game time, which conveys how a game unfolded far better than a box score. They show one model's view, and the model deserves a label.

Calibration plots. Predicted probability on one axis, observed frequency on the other, grouped into bins, with the diagonal as reference. The most informative single chart for judging a probability model. Show the number of predictions in each bin, because the extreme bins are usually sparse.

Distributions. Histograms or density plots of margins, totals, and player outputs. They reveal skew, clustering on common scoring values, and fat tails, all of which an average conceals.

Small multiples. The same chart repeated per team, season, or player on shared axes. Far easier to compare than a single chart crowded with colours.

Heatmaps on small samples. A shot chart that divides the court into many zones gives each zone few attempts. A zone with three attempts and three makes shows as the hottest spot on the chart, and it means almost nothing. Use larger bins or smoothing, set a minimum count below which a zone is greyed out, and display counts.

Interpolated price lines. Connecting hourly snapshots with sloped lines implies prices were gradually moving between observations. They may have been flat and then jumped, or moved and reverted. The chart invents a path that was never observed.

Forward filled gaps. Carrying the last value across a collection failure or a suspension draws a flat line that looks like stability. Gaps should look like gaps.

Truncated axes on percentages. A probability axis running from forty five to fifty five percent makes small differences look dramatic. Sometimes that zoom is right; if so, label it clearly.

Cherry-picked windows. A trend line starting at a convenient date can show almost anything. Show the full available period, or state why a window was chosen.

Averages without spread. Two players with the same average can have completely different distributions. An average plotted without any indication of spread invites exactly the wrong comparison.

Uncertainty is part of the data

Label what model produced any derived value. Win probability, expected goals, and expected points all come from particular models. Name the source in the caption.

Use consistent colour for the same entity across every chart in a piece of work. Readers learn the mapping once.

Prefer direct labels to legends when there are only a few series. It reduces the back and forth between chart and key.

Show the reference point. The diagonal on a calibration plot, zero on a margin chart, the league average on a player comparison. Charts without a reference invite readers to supply their own.

Choose accessible colour palettes. Red and green alone are indistinguishable to a meaningful share of readers, and sports charts use them constantly for good and bad.

Write the takeaway in the title. A title that states what the chart shows, rather than what it plots, makes the chart useful to someone skimming, and forces the author to check whether the chart actually supports that statement.

This page describes visualisation practice and is not betting advice.

One chart, one claim