A chart can be numerically accurate and still mislead through scale, encoding, omission, or decoration. These seven recurring failure patterns are more useful than a gallery of “worst charts” because each comes with a diagnostic question and a repair.

1. A truncated baseline exaggerates bars

Bar length encodes magnitude from a baseline. If the baseline is truncated, a small difference can look many times larger than it is. For example, $10.10 is about 39% greater than $7.25, not several hundred percent greater.

Fix: Start ordinary bar charts at zero. If a nonzero baseline is analytically necessary, prefer a dot or line chart, label the scale conspicuously, and show exact values. Line charts do not universally require zero because position and slope—not bar length—carry the comparison.

2. An inverted axis reverses intuition

The widely discussed 2014 Reuters Florida gun-deaths graphic used a vertical scale with larger values lower on the page. Its labels made the values recoverable, but the orientation conflicted with the conventional “up means more” reading. It is safer to call the chart confusing than to assert intent.

Fix: Make values increase upward unless the domain has a familiar alternative, such as depth. If inversion is necessary, label it directly and test the chart with readers.

3. A pie chart represents overlapping answers

Pie slices encode mutually exclusive parts of one whole. Survey answers that allow “select all that apply” can legitimately total above 100%, so forcing them into a pie invents a part-to-whole relationship.

Fix: Use a sorted bar chart and state that respondents could choose multiple answers. Before charting, run the validation checks described in data cleaning with Python.

4. Dual axes manufacture a relationship

Two vertical scales can be independently stretched until unrelated series appear to rise together, cross at a dramatic point, or diverge. The visual correlation depends partly on arbitrary axis domains.

Fix: Use aligned small multiples, direct indexing to a common baseline when appropriate, or a scatterplot for association. Show uncertainty and avoid implying causation; review spurious-correlation examples.

5. Three-dimensional effects corrupt area and angle

Perspective makes foreground marks look larger and can hide categories. Exploded 3D pies add depth, angle, area, and occlusion even though only one variable needs encoding.

Fix: Remove 3D perspective from quantitative charts. Use position on a shared scale—usually bars or dots—for accurate comparisons.

6. Too much color and clutter hides the message

Excess gridlines, icons, labels, gradients, and saturated colors compete with the data. A rainbow scale also introduces false boundaries and may be inaccessible to readers with color-vision deficiencies.

Fix: Establish a visual hierarchy. Use a restrained palette, direct labels, light reference lines, and redundant cues such as shape or text. Check contrast and do not rely on color alone. See data visualization best practices.

7. Missing context turns accurate marks into a bad claim

A chart can omit denominators, time windows, inflation adjustment, uncertainty, missing values, source, or a policy change. Counts without exposure can confuse growth in activity with growth in risk. Cherry-picked start dates can reverse the apparent trend.

Fix: Name the measure and unit, define the population and period, disclose exclusions, show uncertainty where relevant, cite the source, and annotate structural breaks. Separate descriptive association from causal attribution.

A pre-publication chart audit

  1. Data: Are totals, units, denominators, missingness, and transformations correct?
  2. Encoding: Does area, length, position, angle, or color represent the intended quantity faithfully?
  3. Scale: Are domains, baselines, intervals, and ordering defensible and visible?
  4. Context: Are comparison group, time window, uncertainty, and source stated?
  5. Accessibility: Is the chart understandable without color, and does it have useful alt text or a data table?
  6. Claim: Does the caption say only what the design supports?

Good visualization supports data-driven decisions only when the underlying measure and decision context are sound.

Frequently asked questions

Must every chart start at zero?

No. Bar charts normally should because length represents magnitude from a baseline. Line and dot charts may use a focused range when it is clearly labeled and does not distort the intended comparison.

Are pie charts always bad?

No, but they work only for a small number of mutually exclusive parts of a meaningful whole. Position on a common scale is usually easier to compare.

Is a misleading chart necessarily deceptive?

No. Diagnose the visual effect and evidence; do not infer intent without support.