Bar graph

Bar graphs compare discrete categories

Useful category questions include:

• Which of four settings had the most completed safety checks?
• How did three communication modes compare during one defined probe?
• What percentage of records in each clinic met one audit rule?
• Which response forms occurred during a fixed observation?

Bars can show counts, rates, percentages, means, medians, or other summaries. The axis and label must name the actual statistic.

Time-series questions need temporal order

A bar for “baseline average” beside a bar for “treatment average” hides session order, trend, variability, overlap, latency of change, and procedural shifts. Two phases can share the same mean while displaying very different patterns.

The BACB BCBA Test Content Outline, sixth edition includes selecting data displays, interpreting graphed data, and using data for decisions. It is exam content rather than a chart-design standard. The analyst still needs to choose a display suited to the question.

Keep the time-series graph available when repeated measurement and condition change matter. A bar graph can serve as a secondary summary for a different audience or decision.

Name the unit and denominator

“Success rate” has little meaning without a numerator and eligible denominator. State whether the bar represents sessions, opportunities, clients, records, minutes, or responses and which period it covers.

When group sizes differ, counts and percentages answer different questions. A bar of 8 completions can mean 8 of 10 or 8 of 40. Show raw counts near percentages or in an accessible table.

Avoid pooling incompatible cohorts. If rules, exposure windows, services, or measurement definitions differ, separate the bars or explain the limitation.

Category design changes the conclusion

Broad categories can hide important differences. Combining all communication into one bar may conceal whether speech, AAC, gesture, and writing had equal access. Combining every clinic can hide one site with a recurring failure.

Choose categories before reviewing the desired result when possible. Make them mutually exclusive when the total requires it, or say clearly when observations can appear in more than one bar. “Other” should have a defined rule and remain small enough to interpret.

When privacy or small cell size matters, aggregate through an approved method and state the resulting limitation. Never invent or suppress values silently.

Scale choices can distort magnitude

Bar length encodes magnitude from its baseline. Starting a quantitative bar axis above zero can exaggerate a small difference because the visible lengths no longer represent the full values. Use a zero baseline for ordinary magnitude comparison or provide a compelling, explicit reason and another complete view.

Use equal spacing, consistent width, readable labels, and a common scale. Avoid three-dimensional effects, decorative shapes, and color gradients that obscure length. Color should add meaning and remain distinguishable without relying on color alone.

Summary bars hide distributions

A mean bar can conceal outliers, skew, missing values, and within-group variation. Show the underlying observations, range, interval, or sample size when those features affect interpretation.

For clinical data, preserve the original observation record and time series. A polished summary cannot repair ambiguous definitions, unreliable measurement, missing opportunities, or selective exclusions.

A fictional training-cohort example

Mina reviews an access-training deadline across four staff cohorts. The due and completed counts are:

A bar graph of percentages makes rate comparison easy. Labels should retain 6/8, 8/10, 9/12, and 8/10, since equal percentages can contain different numbers of people.

The graph supports follow-up on four incomplete staff members. It cannot show when completion occurred, why items were late, whether training changed performance, or whether cohorts were comparable.

Predeclare the graphing decision

Before drawing bars, define:

• the question and audience
• category membership
• numerator, denominator, unit, and period
• treatment of missing and ineligible observations
• statistic and any aggregation
• scale, order, and label conventions
• underlying table or time-series source

Sort categories only when the order serves the question. Alphabetical, chronological, procedural, and value-based order communicate different structures.

Make the graph accessible

Pair the visual with a concise text conclusion and a data table. Use direct labels, sufficient contrast, readable type, and patterns or symbols when color distinguishes categories. Alt text should state the main comparison, units, and meaningful limitation.

A precision-teaching overview describes standardized time-series displays and timely data-based decisions. Its context differs from a bar chart, illustrating why the measurement system and decision should drive display choice.