Statistics in the News: Why Relative Risk, Poll Margins of Error, and Chart Scales Often Mislead

"9 out of 10 doctors recommend" means something very different from what most readers assume

Percentages appear constantly in news reporting, health studies, and political discourse — and they're routinely misrepresented or misunderstood in ways that lead to poor decisions. The two most important distinctions — relative risk vs absolute risk, and the margin of error in polls — are the ones most often obscured by how statistics are presented.

Relative risk vs absolute risk in health statistics

Relative risk reduction tells you the percentage change in risk between two groups. Absolute risk reduction tells you the actual change in probability of an outcome.

Example: Suppose a drug reduces the risk of a heart attack from 2% to 1% over 5 years.

• Relative risk reduction: (2% − 1%) ÷ 2% = 50% (a 50% reduction)
• Absolute risk reduction: 2% − 1% = 1 percentage point

Headlines often report the relative risk: "New drug cuts heart attack risk by 50%!" This is technically correct but potentially misleading — the 50% figure sounds dramatic; the 1 percentage point figure sounds modest. Both describe exactly the same clinical result.

Number Needed to Treat (NNT): the reciprocal of absolute risk reduction. If absolute risk reduction is 1%, the NNT = 1 ÷ 0.01 = 100. That means 100 people must take this drug for 5 years for one additional heart attack to be prevented. At what cost and side effect risk? That context is usually missing from headlines.

Margin of error in polls

Opinion polls report that "Party A is at 42%, Party B is at 40%, with a margin of error of ±3%."

Many readers interpret this as: "Party A leads Party B by 2 points."

The correct interpretation: given the ±3% margin of error, Party A could be anywhere from 39% to 45%, and Party B could be anywhere from 37% to 43%. The ranges overlap substantially. The poll is not showing a statistically significant lead for Party A.

Statistical significance and poll methodology:

• The margin of error in a poll is typically the 95% confidence interval — meaning if the poll were repeated 100 times, the true value would fall within that range 95 times
• Two values are statistically distinguishable at 95% confidence when their ranges don't overlap
• A 2-point lead with a ±3% margin of error is not a lead — it's statistical noise

The 1000-sample problem: most national polls use 1,000 respondents. The margin of error for a proportion near 50% with n=1,000 is approximately ±3%. Increasing to 2,000 respondents halves the error to ±2.1%, but doubling the sample size only reduces error by √2 ≈ 1.4×. To get a margin of error below ±1%, you'd need roughly 10,000 respondents.

Base rates and conditional probability

The same Bayes' theorem problem that appears in medical testing appears in everyday risk communication:

"People who eat breakfast every day are 20% less likely to develop obesity."

The question: 20% less likely than what? If the baseline obesity rate in the non-breakfast-eating population is 40%, then breakfast eaters have a 32% rate — a meaningful difference. If the baseline is 10%, then breakfast eaters have an 8% rate — still a relative 20% reduction, but a much smaller absolute change.

Always ask: what is the baseline rate, and what is the absolute difference?

Percentage changes over time: the compounding distortion

Headlines often compare current figures to last year, or to a reference year, in ways that distort perception:

"Crime down 10% this year, after being up 15% last year."

Starting value: 100 After 15% up: 115 After 10% down: 115 × 0.90 = 103.5

The net result is still 3.5% above the starting point, despite the headline making it sound like crime has fallen back to base. Percentage changes are not additive — they compound on a changing base.

Misleading chart scales

Percentage data is often displayed in charts with truncated y-axes, which exaggerates apparent change:

• A bar chart showing values ranging from 45% to 55% with a y-axis starting at 44% makes a 10-percentage-point difference look like a massive change
• The same chart with a y-axis starting at 0% shows a much less dramatic picture

Look at the y-axis scale before interpreting any bar or line chart. The choice of scale can make a trivial change look dramatic or a significant change look flat.

How to use the Percentage Calculator on sadiqbd.com

1. Calculate absolute and relative differences — enter the baseline and the new value, calculate both the percentage change and the absolute change
2. Cross-check headline statistics — if a news story says "X% risk reduction," use the calculator to find the absolute risk if you know the baseline
3. Verify compound percentage claims — test the "down 10% after being up 15%" type claims

Frequently Asked Questions

What is "statistical significance" and does it mean a result is important? Statistical significance means the result is unlikely to have occurred by chance (typically p < 0.05, or less than 5% probability of a false positive). It does NOT mean the result is practically important or large in magnitude. A study with millions of participants might find a statistically significant but clinically irrelevant 0.01% difference.

How do I spot a misleading percentage in the news? Three questions: (1) Is this a relative or absolute change? (2) What is the baseline? (3) What is the confidence interval or margin of error? Answers to these three questions reveal almost all common percentage misrepresentations.

Is the Percentage Calculator free? Yes — completely free, no sign-up required.

Try the Percentage Calculator free at sadiqbd.com — calculate relative and absolute changes, percentage differences, and percentage-of-percentage problems instantly.