How to Read a Statistic Without Being Fooled
Last updated June 2026
A number feels like proof. It looks objective, settled, hard to argue with — and that is exactly why statistics are such a popular tool for misleading you. Learning to read one carefully is one of the most useful habits a sceptical news reader can build.
You don't need to be a mathematician to spot a dodgy statistic. Most misleading numbers in the news aren't wrong in a technical sense — they're true figures stripped of the context that would change how you feel about them. Once you know the handful of tricks that get used again and again, you'll start seeing them everywhere. Here is what to watch for.
Relative risk versus absolute risk
This is the single most common way a headline inflates a finding. "Eating this doubles your risk of a rare disease" sounds alarming. But if the risk started out tiny, doubling it leaves it almost as tiny.
Imagine a hypothetical condition that affects 1 person in 100,000 each year. A study finds that some habit doubles that risk. The relative change is huge: a 100% increase. The absolute change is from 1 in 100,000 to 2 in 100,000 — still vanishingly rare. Both numbers describe the same finding, but they leave very different impressions.
Whenever you see "doubles", "triples", "50% more likely" or "cuts your risk in half", ask the obvious follow-up: more likely than what, starting from what baseline? A big relative change on top of a tiny absolute risk is rarely worth losing sleep over. A reputable report gives you both figures; a misleading one quietly drops the absolute number.
Percentages without the base number
A percentage is a ratio, and a ratio is meaningless until you know what it is a ratio of. "Complaints rose 200%" could mean complaints went from 1 to 3, or from 10,000 to 30,000. The percentage is identical; the story is not.
The same trick works in reverse. "Only 0.5% of cases were affected" sounds reassuring until you learn the total was several million, at which point 0.5% is a large number of real people. Small percentages of huge populations, and huge percentages of tiny populations, are both classic ways to make a figure feel bigger or smaller than it really is.
The defence is simple: when you see a percentage, look for the raw count behind it. If the article never gives you the base number, treat the percentage as decoration rather than evidence.
Averages that hide the spread
"The average" is one word doing at least two different jobs, and the gap between them is where a lot of misleading claims live.
Mean versus median
The mean adds everything up and divides by the number of items. The median is the middle value when you line them all up. When data is lopsided — a few very large values pulling the figure up — the mean can sit far above what a typical person actually experiences, while the median stays closer to reality.
Pay rises are the textbook example. If a handful of top earners get enormous increases and everyone else gets very little, the average rise can look healthy even though the typical worker barely moved. The mean went up; the median tells the more honest story. When a report quotes an average, ask whether the median would say the same thing — and be suspicious if only one of the two is mentioned.
Correlation dressed up as causation
Two things moving together is not proof that one causes the other. Ice cream sales and drowning rates rise at the same time of year, but ice cream does not cause drowning — warm weather drives both. That hidden third factor is called a confounder, and it lurks behind a great many "linked to" headlines.
Watch the verbs. Careful writing says a habit is associated with or linked to an outcome. Misleading writing slides into causes, triggers or leads to without the evidence to back it up. Observational studies — which simply watch what happens — can show associations but rarely prove cause. For that you usually need a controlled experiment. Our companion guide on whether a scientific study is trustworthy goes into how to tell the two apart.
Cherry-picked timeframes and baselines
Where you start the clock decides what the trend looks like. Pick a starting point right after an unusually low dip and almost anything will look like a dramatic recovery. Pick one right after a record high and the same data can be made to look like collapse.
This is why you should be wary of charts and claims that begin on an oddly specific date. If a figure is compared to "this time last year" but last year happened to be extreme, the comparison is doing the misleading work, not the underlying reality. Always ask: why this baseline, and would the story survive a longer view? A trend that holds up over several years is far more trustworthy than one that depends on a carefully chosen month.
Sample size and margin of error, in plain terms
A finding drawn from a handful of people is fragile; the same finding across thousands is far sturdier. Tiny samples produce noisy results that can swing wildly by chance, which is why a striking number from a small study should be treated as a hint, not a verdict.
Polls and surveys come with a margin of error — the wiggle room around the headline figure. If a poll puts support at 48% with a margin of error of plus or minus 3 points, the true figure could reasonably sit anywhere from 45% to 51%. So a "two-point lead" inside that range is not really a lead at all; it is a tie dressed up as a result. When coverage treats movements smaller than the margin of error as meaningful, the coverage is overreaching.
"Up to", "as many as" and other weasel framings
Some phrases are engineered to sound precise while promising almost nothing. "Up to 70% off" is satisfied by a single item at 70% and everything else at 5%. "As many as a million people" is true even if the real figure is far lower. "Could", "may" and "linked to" all hedge a claim so heavily that it can never quite be wrong — and never quite be informative either.
When you meet one of these phrases, mentally replace it with its weakest honest version. "Up to 70%" becomes "at most 70%, possibly far less". If the claim still impresses you after that translation, it may have substance. If it deflates, the framing was the point.
Misleading graphs and truncated axes
A chart can mislead even when every number on it is accurate. The classic move is the truncated axis: a bar chart whose vertical scale starts at, say, 90 rather than 0, so a difference between 92 and 95 looks like a landslide instead of the small gap it is.
Other tricks include stretching the time axis to exaggerate a trend, or using two different scales on one chart to imply a link. Before you trust a graph, glance at the axes: where do they start, and is anything left unlabelled? A picture is persuasive precisely because we tend not to interrogate it the way we would a sentence.
Questions to ask of any statistic
You can pressure-test almost any number in the news with a short mental checklist:
- Compared to what? A change only means something against a clear baseline.
- What's the raw number? A percentage needs the count behind it.
- Who is in the sample, and how many? Small or unrepresentative samples are fragile.
- Is this relative or absolute? A scary relative change can sit on a tiny absolute risk.
- Mean or median? Averages can hide an uneven spread.
- Association or cause? Watch for a confounder doing the real work.
- Where did the figure come from? A statistic with no traceable source is just an assertion.
That last question matters most. A number is only as good as the source it came from, and following it back is the heart of how careful readers work — a skill we cover in primary versus secondary sources.
Not sure whether a figure in an article is properly sourced and shown in context? Paste the article into Fact or Fiction News and it will flag claims that lean on numbers, check whether each one is backed by a credible source, and tell you when the surrounding context is missing.