A chart misleads when its visual encoding, metric definition, or selected data creates an impression unsupported by the underlying values. Preserve proportional scales, show denominators and relevant counts, disclose filters and missingness, compare like periods and populations, and match visual prominence to decision importance. Review the data claim before polishing the graphic.
Who this is for: Anyone creating or reviewing charts, scorecards, or dashboards that influence operational and financial decisions.
- Check population, denominator, aggregation, period, and exclusions before judging the chart type.
- Use visual encodings whose geometry accurately represents the magnitude and uncertainty of the data.
- Expose filters, freshness, missing values, and comparison context so viewers can reconstruct the claim.
Audit the metric beneath the picture
A clean chart can faithfully display a flawed measure. Confirm the unit counted, eligibility, numerator, denominator, time basis, and source. A defect rate may fall because low-risk products entered the mix. Average order value may rise because small orders disappeared. Show counts beside rates when volume changes alter the meaning.
Check whether aggregation hides variation. An overall improvement can coexist with declines in every major segment when segment proportions shift. Means can be dominated by extreme values, while medians can hide tail risk. Choose summaries based on the decision and include distributions or segment views when the single statistic is insufficient.
Use honest visual geometry
Bar lengths encode magnitude and generally need a zero baseline. Truncating the axis can make a small difference appear enormous. Lines emphasize change and may use a narrowed scale when clearly labeled, but readers still need context about practical magnitude. Avoid three-dimensional effects, area symbols, and decorative perspective that distort comparison.
Use equal intervals on time and numeric axes. Sort categories to support the intended comparison rather than a convenient source order. Keep dual axes rare and explicit because unrelated scales can manufacture apparent correlation. If two measures need comparison, indexed trends or aligned small multiples often reveal the relationship more honestly.
Reveal selection and missingness
Display active filters, date range, timezone, data freshness, and excluded records. A dashboard screenshot without filter state is weak evidence. Distinguish zero from missing and not yet available. Connecting a line across absent observations implies continuity that the data did not establish. Mark breaks and incomplete current periods.
Use comparable windows. Month-to-date should be compared with the same elapsed portion of a prior month, not its completed total. Account for different numbers of business days and seasonal events. Explain revision policies when late-arriving records change history. A current result that remains provisional should look provisional.
Match emphasis to evidence
Color and size draw attention, so reserve strong emphasis for meaningful exceptions. Red and green should correspond to governed thresholds and include text or symbols for accessibility. Do not color any negative change red when lower may be better. Use neutral styling for descriptive variation that has no agreed judgment.
Ask a reviewer to state the chart's claim without hearing the author's explanation, then compare that statement with the data. Test alternative scales and segment views. Include uncertainty when sampling or forecasting matters. The goal is not to make every visual cautious and dull; it is to make the fastest reasonable reading accurate.
Repair an exaggerated conversion chart
A dashboard bar chart shows conversion rising from 4.8 to 5.1 percent, with the axis beginning at 4.7 and a headline claiming dramatic growth.
- Confirm both periods use the same eligible sessions, attribution window, bot exclusion, and complete observation period.
- Show converted counts and eligible counts, then calculate absolute percentage-point and relative changes.
- Redraw bars from zero or use a clearly labeled line with enough history to show normal variation.
- Add an uncertainty interval or experiment result if the values come from sampled or randomized data.
- Replace the headline with a statement proportional to evidence and note any material channel-mix change.
Chart integrity review
Run this review before publishing a decision-facing visualization.
- Metric: entity, formula, denominator, units, exclusions, source, owner, and reconciliation.
- Comparison: equal population, complete periods, seasonality, mix, absolute movement, and practical significance.
- Geometry: chart type, axis origin, intervals, ordering, area or length encoding, and dual-axis risk.
- Disclosure: filters, timezone, freshness, missingness, provisional values, revisions, and uncertainty.
- Reading test: independent viewer's claim, intended claim, accessible emphasis, and corrected ambiguity.
Common mistakes
- Starting a bar axis near the smallest value to turn a minor difference into a towering visual gap.
- Reporting a percentage change without the underlying counts or the change in percentage points.
- Treating missing observations as zero or silently connecting a trend line across a collection outage.
Try one
A map uses circles whose radius is proportional to city sales, making a city with twice the sales appear four times as large by area. Explain and fix the encoding.
The visual area, not radius, is what viewers perceive, so scaling radius directly exaggerates magnitude. A correct fix scales circle area to sales, which means radius follows the square root of value, or uses another accurate encoding such as a sorted bar chart. The answer should also check population, geographic coverage, missing cities, filters, units, and whether precise comparison is actually required.
Sources
- NIST exploratory data analysis handbookNIST methods for exploring data, checking assumptions, and revealing structure.
- Tableau dashboard documentationOfficial Tableau guidance for building dashboards from worksheets and interactive objects.