Incidents are not victims
One person may experience several incidents, so an incident total and a victim total should not be substituted for each other.
Protection, evidence and public rules
Fraud statistics: what the numbers really mean
Fraud statistics can count incidents, victims, reports, losses, offences or survey estimates. A large number is not meaningful until you know what was measured, who was included, the time period and whether it can be compared with another figure.
Page T-014Guidance pageReviewed 29 August 2026
Key Guidance
Read the definition before the headline. Check whether the number describes incidents or people, an estimate or recorded reports, the population and period covered, and whether methods or reporting systems changed.
Start with Key Guidance for the short answer. Then read the sections that match what happened and what you need to do.
Understand the issue
Why this matters to you
Fraud is measured through several systems because no single source captures every experience. Survey estimates, police or reporting data, industry losses and regulatory incidents answer different questions.
Reports are not all fraud
Recorded reports depend on awareness, definitions, access to reporting and system changes. Many experiences are never reported.
Estimates have uncertainty
Survey results are estimates for a defined population and normally include confidence intervals or tests of statistical significance.
Methods can change
Question wording, classifications, processing and reporting platforms may alter what can responsibly be compared over time.
Consequences and movement
What may happen next
A useful statistic should help answer a defined question. It should not be stretched into a claim about a different population, harm or cause.
Start with the unit
Is the figure counting incidents, unique people, reports, transactions, organisations, records or money?
Find the denominator
A percentage needs a base: of all adults, all reports, all transactions, all surveyed respondents or another group.
Check the period and geography
A year ending in a stated month is not always a calendar year, and UK, Great Britain and England and Wales are not interchangeable.
Read the comparison note
A higher estimate may not be statistically significant, and a revised or redesigned series may not support a simple trend.
Practical response
What you should do
Use a short reading sequence before repeating or publishing a number. The same checks work for official releases, news reports, campaigns and social posts.
- Name the publisher and datasetIdentify the organisation and the specific survey, reporting system, industry return or regulatory collection.
- State what is countedUse the source’s unit and definition rather than replacing it with a broader word such as people, scams or cases.
- Add the period and coverageInclude the date range, geography, age group and important exclusions that shape the result.
- Separate estimates from recordsDo not add or compare a population estimate with administrative reports as though they measure the same set.
- Check revisions and significanceLook for correction notices, methodology updates, confidence intervals and whether a stated change is statistically significant.
- Link to the original releaseAllow readers to inspect the definition, table and limitations rather than relying on a cropped chart or secondary headline.
Look more closely
A large number can persuade without explaining
Statistics can create fear, certainty or urgency while hiding what was counted and how. Treat a number as evidence that needs context, not as proof that the claim, seller or solution attached to it is genuine.
A headline, chart or widely shared statistic attracts attention.
Scale, novelty or fear makes the issue feel immediate.
Official language, branding or precision gives a wider claim borrowed authority.
The reader is directed toward a product, donation, policy claim or unsafe response.
Current developments
Researched feedCrime in England and Wales: year ending March 2026
The ONS estimated 4.5 million fraud incidents and 3.8 million victims through the Crime Survey for England and Wales. The incident estimate was not a statistically significant change from the previous year, while the estimated number of victims increased.
Read the ONS bulletin and definitions
How to read the latest fraud estimates
The release illustrates why the unit, method and comparison matter. The survey estimates experiences in its covered household population. It is not a count of every report, business or public body.
The survey estimated 4.5 million fraud incidents in the year ending March 2026.
It estimated 3.8 million victims. One victim can experience more than one incident.
The incident estimate was not a statistically significant change from 4.2 million in the previous year, while the victim estimate rose from 3.4 million.
The household survey does not represent every business, public body, tourist, communal establishment or person under 16.
What to do with this information
When quoting the release, keep incidents and victims separate, retain the survey and period, and do not treat reporting-system totals as a direct substitute.
Selected history
Historical cases and lasting lessons
Quality and reporting-system changes
ONS published work on improving fraud and computer-misuse statistics, while changes to the national reporting service affected how some recorded data could be presented.
Lasting lesson: A gap or break in one administrative series does not automatically invalidate a separate household survey, but it changes which comparisons are possible.
Regulatory incident trends
Regulators and industries publish their own incident and loss datasets for the organisations and events within their remit.
Lasting lesson: These sources can show valuable patterns, but their definitions and reporting population must travel with the number.
Transferable understanding
Patterns over time
Precision can hide uncertainty
A detailed number may still be an estimate with sampling and methodological limits.
Different sources can both be correct
They may be measuring different populations, units, periods or stages of reporting.
Corrections are part of responsible statistics
A corrected release should replace the earlier value. The correction itself is not proof the entire dataset is unreliable.
Rankings require common definitions
Places, platforms or scam types should not be ranked using figures collected under incompatible rules.
Plain-language answers
Common questions
Why are there more incidents than victims?
A person can experience more than one fraud incident during the measured period.
Why do survey and police figures differ?
A survey estimates experiences in its defined population. Police or reporting data counts events that entered an administrative system.
Does a higher number always mean a real increase?
No. Check statistical significance, uncertainty, method changes and whether the comparison uses the same definition.
Can I add figures from different sources?
Only when their units, populations and periods are compatible and overlap has been addressed. Usually they should be presented separately.
Find the definition behind the figure
Use the original statistical release and methodology. For personal reporting or support, use the reporting and recovery routes rather than assuming a national statistic describes an individual case.
Scamology Insights
Use numbers as evidence, not borrowed authority
These apps, tools, books and courses can help you ask what a figure proves, what it leaves out and whether someone is using statistics to make a weak claim sound authoritative. Links are intentionally withheld in this production version.
Continue understanding
Explore related guidance
Evidence and currency