Last reviewed: Next review due:
Why statistics matter for journalists
Most journalism errors involving numbers are not about complex statistics — they are about the basics: confusing mean and median, using percentage when you mean percentage point, or reporting a correlation as if it proves causation. This guide covers the concepts that matter most in everyday UK data journalism.
Core concepts
Averages: mean, median, and mode
Mean
Sum ÷ count
Use for: Symmetric distributions; test scores, temperatures
Note: Skewed by outliers
Median
Middle value when sorted
Use for: Skewed distributions; income, house prices, waiting times
Note: Most appropriate for journalism
Mode
Most common value
Use for: Categorical data; most common crime type
Note: Rarely the right choice for numeric data
# NHS waiting times example — why median matters
waits = [2, 3, 4, 5, 6, 8, 10, 52] # weeks (one outlier)
mean = sum(waits) / len(waits) # 11.25 weeks
median = 7.0 # middle of sorted list
# Reporting "average wait is 11.25 weeks" is technically true
# but misleading — most patients wait under 8 weeks
# Median is the fairer measure herePercentages vs percentage points
# Council tax support claimants: from 8% to 10% of working-age adults
absolute_change = 10 - 8 # = 2 percentage points (pp)
relative_change = (10-8) / 8 # = 0.25 = 25%
# Correct: "Claimants rose by 2 percentage points"
# "Claimants rose by 25% (relative to the previous level)"
# Wrong: "Claimants rose by 2 percent" when you mean 2ppPer-capita comparisons
# Comparing crime between Manchester (500k) and a village (5k)
manchester_crimes = 10000
village_crimes = 50
# Absolute numbers: Manchester has 200x more crime
# Per-capita rate: both have 20 crimes per 1,000 population
# Per-capita is almost always the correct comparison
# Rate = (count / population) * 1000Correlation vs causation
Key rule: Two things rising or falling together (correlation) does not mean one causes the other (causation).
Ice cream sales and drowning rates both peak in summer. They are correlated. Neither causes the other — hot weather causes both. Always ask: is there a third variable that explains the relationship? Could causation run in the other direction? Could the correlation be a coincidence?
When statistics are most likely to go wrong
- 1Reporting survey results without mentioning sample size or margin of error.
- 2Using the mean for income, house prices, or waiting times when the median is more appropriate.
- 3Writing "doubled" when you mean "increased by 100 percentage points" — these are different things if the base was not 100%.
- 4Presenting a relative change ("crime fell 50%") without the absolute context (from 2 to 1 incident).
- 5Treating correlation as causation in any data with two trends moving together.
Red flags in statistical claims
- A press release says "average" without specifying mean or median.
- A percentage change is impressive but the base number is tiny (e.g. 200% increase from 1 to 3 cases).
- Survey results are presented without a sample size or confidence interval.
- A government minister uses relative risk reduction; the absolute risk reduction is much smaller.
- "Studies show" without citing the study, its size, or whether it has been peer-reviewed.
Statistics accuracy checklist
- I have specified whether I am using mean or median, and chosen the appropriate one.
- I have distinguished between percentage change and percentage point change.
- I have compared rates or proportions, not absolute counts, when areas have different populations.
- I have not described a correlation as causation.
- I have included context: the base number for relative changes, the sample size for surveys.
- I have checked whether a claim of statistical significance also represents practical significance.
Go deeper
For survey data specifically, see our Survey Data Handling guide. For more advanced statistical concepts, see Regression and Correlation Basics.
Common mistakes
- Writing "percent" when you mean "percentage point" — these are different things.
- Quoting the mean income when the median income is more representative.
- Saying "X causes Y" based solely on a correlation in observational data.
- Not questioning whether a survey sample is representative of the population being described.
- Treating all official statistics as unimpeachably accurate — methodology notes exist for a reason.
Related guides
Regression & Correlation Basics
R-squared, coefficients, and scatter plots explained.
Survey Data Handling
Sample size, weighting, and reporting survey results.
UK Statistical Bodies
How to read ONS and other official statistical bulletins.
Data Storytelling
How to write about numbers without misleading readers.
Primary sources
Frequently asked questions
What is the difference between a percentage and a percentage point?
If the unemployment rate rises from 4% to 6%, it has risen by 2 percentage points (an absolute change). But it has also risen by 50% (a relative change — 2 is 50% of 4). Both statements are true but one sounds much more dramatic. Always state which you mean. Government communications often use whichever makes the change sound smaller; opposition politicians use whichever makes it sound larger. Your job is to use whichever is more informative for your reader.
What is a median and when should I use it instead of the mean?
The median is the middle value when all values are sorted in order. The mean is the sum divided by the count. When data is skewed by outliers — as income, house prices, and waiting times usually are — the median is a more representative 'typical' value. If ten people earn £20,000 each and one person earns £1,000,000, the mean salary is £109,000 but the median is £20,000. The median is almost always the better measure for reporting on income, wealth, and waiting times.
What does a margin of error mean in a survey?
A margin of error (typically quoted as ±X percentage points at 95% confidence) means that if you repeated the survey many times, 95% of the results would fall within that range of the true value. A poll showing Party A at 42% ±3% means the true figure is probably between 39% and 45%. When two candidates are within each other's margin of error, the race is statistically too close to call — even if one is numerically ahead.
What does statistical significance mean?
A result is statistically significant if it is unlikely to have occurred by chance given the null hypothesis. The conventional threshold is p < 0.05 — a less than 5% probability of observing the result if there were actually no effect. Statistical significance does not mean practically important. A tiny effect can be statistically significant in a very large dataset. Always report the size of the effect alongside its significance, and note that p < 0.05 is a convention, not a universal truth.
Related guides
Primary sources
- Code of Practice for Statistics — UK Statistics Authority— UKSA
- Full Fact — media guide to statistics— Full Fact
- ONS — statistical concepts and methodology— ONS
- Sense About Science — Making Sense of Statistics— Sense About Science
- NISRA statistics portal— NISRA
- StatsWales data catalogue— StatsWales