Lies and statistics

Core idea

Most statistical deception falls into three buckets. Graphs can be made to lie by mismatching chart type to data type, by truncating axes, by stretching aspect ratios, or by ordering categories illogically. Percentages can lie by omitting the base rate, by confusing rate-of-change with absolute change, or by switching the denominator mid-argument. Averages can lie by hiding spread — a single number cannot tell you how widely the values vary, but listeners often assume it does.

Why it matters

These tricks aren't rare. Politicians describing crime rates, advertisers describing freshness, employers describing salaries, and tabloid editors describing virtually anything all reach for the same playbook. Each move is technically defensible while being morally misleading. Learning to spot the move is the difference between consuming statistics and being consumed by them.

Mental model

The three families of distortion

Every statistical lie you'll meet maps onto one of three classes — and each class has a small kit of countermeasures.

Copy sourceZoom

The truncated-axis trick

A bar chart where the vertical axis starts above zero can amplify a small absolute difference into a visually huge gap. The countermeasure is to always look at where the axis begins.

Copy sourceZoom

Practical application

When confronted with a statistical claim that supports a conclusion, run this audit.

Example

A regional newspaper runs a story headlined: "Crime in our town is up 20% — vote out the council!" Walk it through the audit.

• Axes. The accompanying bar chart shows last year (90 incidents) and this year (108 incidents). The vertical axis starts at 80, so the bar for this year is roughly 3 times the height of last year's bar visually — even though the numerical increase is only 20%.

• Denominators. "20% up" — out of what base? 18 extra incidents out of 90. Compare to the per-capita rate: if the town's population grew by 10% (say a new housing estate), the per-capita rate barely moved.

• Rate vs level. Maybe the regional average was up 40% over the same period. The town's growth in crime, while real, was below the regional baseline. Stripped of context, the headline is alarming; in context, it's news of relative improvement.

• Spread question. Was the increase driven by one bad month (a single event) or by a steady trend across all twelve months? A boxplot of monthly incidents would settle this in a heartbeat. A single year-on-year mean buries the dynamic.

• Missing comparison. What kinds of crime? Petty theft, violent assault, parking violations? A 20% jump in parking enforcement and a 20% jump in burglary mean very different things for the community.

The headline isn't false — there are 18 more incidents. But every move in the audit above either softens or reframes the story. The first move of statistical literacy is to refuse to react until you've completed the audit.

Related lessons