Scale break

A break removes numerical distance

Imagine a vertical axis that jumps from 20 to 80. The physical space between those labels no longer represents the full sixty-unit difference. A zigzag, gap, or paired slash alerts the reader that ordinary equal-interval interpretation stops at that location.

The plotted values remain real; the spatial relationship has been compressed. Readers should be able to identify the omitted range without consulting hidden settings.

Use a necessity test

A scale break may be reasonable when:

• one or two legitimate outliers flatten the clinically important range
• a cumulative total begins far above zero after a documented continuation
• two meaningful ranges are separated by a large unused interval
• page or screen constraints make an unbroken display unreadable

Ask whether a full-scale graph, separate panel, inset, table, or alternate chart would communicate the data more directly. Convenience alone is a weak reason.

A fictional outlier example

Noor’s team graphs latency in seconds across 12 observations. Eleven values range from 6 to 18 seconds. One equipment failure produces a valid 112-second observation.

A 0-to-120 axis makes the eleven ordinary values cluster near the bottom. The team publishes a full-scale panel and a second detailed panel covering 0 to 25 seconds. The outlier remains listed in the accompanying table.

If the team instead uses a break from 25 to 100, it labels both axis segments, uses a visible discontinuity, and explains that slopes crossing the break cannot be read normally. The 112-second point stays in the graph and analysis.

The team also asks whether the equipment failure belongs in the clinical series, a systems-reliability series, or both. The answer follows the measure’s inclusion rule. Excluding a valid observation after seeing its effect on the graph would bias the display. If the point is ineligible under a rule written before review, record the exclusion and show the operational incident elsewhere.

Interactive views can let readers toggle between full and detailed scales. The static export still needs a complete explanation because screenshots, printed reports, and copied images may lose controls and hover text.

Keep that rationale with the graph’s version history.

Breaks can exaggerate small changes

Removing most of the axis may make a two-unit change occupy half the graph. This risk is especially high with bars, whose lengths are often read from a zero baseline. For bar graphs, a truncated or broken value axis can create a strong visual claim that the numbers do not support.

Line graphs can also mislead. A segment that crosses the omitted range may appear shorter than its numerical change. A steep-looking within-segment trend cannot be compared casually with a segment on a differently scaled panel.

Preserve consistent comparisons

When comparing people, conditions, sites, or time periods, use the same scale and break wherever possible. Different breaks can make similar data look different or different data look similar.

If one panel genuinely needs another scale, label it directly and avoid a shared visual claim. Provide the raw table so readers can inspect exact values.

Keep missingness separate

A scale break concerns the value axis. Missing observations concern the data path or time axis. A calendar gap concerns elapsed time. A phase-change line concerns a planned condition boundary. Use distinct marks for each.

A broken horizontal axis can compress a long passage of time, but it must show the omitted dates or sessions. Otherwise the graph may imply continuous observation or immediate change.

Audit every broken axis

Review:

• why the break is necessary
• exact omitted minimum and maximum
• visible break symbol and labels
• raw-value agreement
• effects on slope, bar length, and comparison
• consistency across panels and exports
• availability of an unbroken view or table
• accessible text describing the discontinuity

Break transparency can be reported as broken-axis graphs meeting every required disclosure divided by broken-axis graphs due for review. That measure assesses display compliance, not clinical quality.

Choose scale from the question

An equal-interval graph emphasizes additive differences. A semilogarithmic chart emphasizes proportional change. Adding an axis break keeps the original scale type while removing a visible span.

The BACB BCBA Test Content Outline, sixth edition covers graphing and visual analysis. A comparison of linear and ratio graphs illustrates how display format changes visual interpretation. These sources support deliberate graph selection while leaving the case-specific choice to qualified reviewers.