Two official surveys measured problem gambling in Britain at almost the same time. One put the rate at 2.4 percent. The other put it at 0.4 percent. Neither is wrong, and neither is a fake number. The gap is caused by how the questions were asked and who agreed to answer them — what statisticians call a mode effect. This page explains that effect in plain terms, because you cannot read any gambling statistic safely without it.
Understanding gambling survey methodology is not a technical sidebar. It is the difference between “one in forty adults has a gambling problem” and “one in two hundred and fifty does” — two claims that would justify very different policies. This article is the reference we link to from every statistics page on this site.
The two numbers, side by side
The Gambling Survey for Great Britain (GSGB) Annual Report 2025, published by the Gambling Commission on 16 July 2026, is the largest of its kind in Britain. The Adult Psychiatric Morbidity Survey (APMS) 2023/24, published by NHS England on 27 November 2025, is the long-running national mental health survey. Both use the same nine-question screen, the Problem Gambling Severity Index (PGSI).
| Feature | GSGB Annual Report 2025 | APMS 2023/24 |
|---|---|---|
| Published by | Gambling Commission | NHS England |
| Published on | 16 July 2026 | 27 November 2025 |
| Area covered | Great Britain | England |
| Fieldwork | January 2025 to January 2026 | March 2023 to July 2024 |
| Adults taking part | 20,775 | 6,912 (phase one) |
| How people were reached | Letter to a sampled address, “push-to-web” with a postal option | Interviewer visited the home; 96.5 percent took part face to face |
| How the PGSI was answered | Self-completed online or on paper | Self-completed, in the self-completion part of the interview |
| Screen used | PGSI (past 12 months) | PGSI (past 12 months) |
| PGSI 8 or more | 2.4 percent | 0.4 percent |
| PGSI 1 or more | 13.7 percent | 4.4 percent |
Note the row that surprises most people: in both surveys, the person answered the PGSI themselves. Nobody read the gambling questions aloud to APMS participants. The APMS methodology report confirms that around a third of the interview, including the gambling module, was completed by the participant on the interviewer’s laptop. So “an interviewer was listening” cannot be the whole explanation. The real story is more interesting.
What a mode effect actually is
“Mode” means the channel a survey travels down: an interviewer in your living room, a phone call, a web link, a paper form through the letterbox. Change the mode and the same question can get a different answer from the same person. Two separate things drive this.
1. Measurement: what people are willing to admit
People under-report embarrassing behaviour when another human being can hear the answer. Gambling losses, chasing, lying to family — these are hard things to say out loud. The Gambling Commission’s own technical report notes that “self-administered data collection methods are likely to mitigate social desirability in responses,” which is one reason the GSGB was designed the way it was.
2. Selection: who agrees to take part at all
This is the bigger and messier half. A survey invitation that says it is about gambling will appeal to some people and not others. The Office for Statistics Regulation (OSR) set out the problem from both directions in its 22 May 2025 review. On one hand, “the survey under-selects those with lived experience of gambling harms, as they may be unlikely to participate.” On the other, because the GSGB is openly gambling-focused, “it is possible that the survey disproportionately attracts those who gamble, so that this group may be over-represented.”
Those two biases push in opposite directions, and nobody can yet say by how much each one bites. That uncertainty is the honest state of the field.
Why weighting does not fix it
A fair objection: surveys are weighted. If too few young men answer, their replies are given extra weight so the final figures match the known age and sex profile of the population. Why does that not solve the problem?
Because weighting can only correct for things the statisticians can measure and match against population totals — age, sex, region, education and so on. It cannot correct for a difference it cannot see. If the people who open a letter about gambling and go online to answer it are more interested in gambling than their neighbours of the same age, sex and postcode, no amount of weighting will remove that. This is why the OSR asked for more research rather than more adjustment, and why the Commission ran a randomised experiment instead of reweighting its way out of the question.
What the gambling survey methodology experiments found
Rather than argue about it, the Gambling Commission paid for the experiment. Patrick Sturgis, Jouni Kuha, Shane Howe and Ioana Maxineanu ran three experiments on the causes of differences in estimates of gambling and gambling impacts in general population surveys (London School of Economics, 2025). The Commission summarised the results in a news release dated 14 August 2025.
| What was changed | Effect on PGSI 1 or more | Effect on gambling participation |
|---|---|---|
| Answering online instead of to a telephone interviewer | Up 4.4 percentage points — close to a 50 percent increase | Substantial increase |
| Saying the survey is about gambling in the invitation | Up 1.8 points, not statistically significant | Up about 4 percentage points |
| Using a longer, updated list of gambling activities | No significant effect | No significant effect |
Read that table carefully, because it is widely misreported. The mode of answering moved the harm estimate most. The wording of the invitation moved participation — how many people say they gambled at all — but did not move the harm estimate by a reliable amount. And the activity list, which critics had blamed, did almost nothing. A separate Commission analysis published on 9 October 2024 did find the GSGB list picked up more online gamblers than the older health survey list, but it flagged that the samples at PGSI 8 or more were too small to draw firm conclusions from.
Putting this against the APMS design matters. Because APMS respondents also answered the PGSI themselves, the gap between 2.4 percent and 0.4 percent cannot be explained by an interviewer reading the questions out. What is left is the overall context and channel of the survey, and above all differential recruitment: a gambling-branded postal invitation to a web survey reaches a measurably different slice of the population than an interviewer knocking on the door for a 90-minute interview about mental health. The Commission’s technical report estimates that around 5 to 6 percentage points of the difference in the PGSI 1 or more figure can be attributed to differences in survey invitations and modes.
None of this makes the GSGB the lesser survey. Its sample of 20,775 adults is three times the size of the APMS phase one sample, it runs continuously in quarterly waves rather than once every seven years, and it is large enough to break results down by activity and by group in ways a smaller survey cannot. The two surveys answer different questions well. The mistake is treating them as rival answers to the same one.
Why Britain changed survey method at all
Face-to-face surveys are expensive and fewer people answer the door each year. APMS 2023/24 achieved a 29.4 percent household response rate after interviewers visited addresses across England for sixteen months. That is a respectable figure for a modern in-home survey, and it is also a warning sign.
So the Gambling Commission moved to push-to-web. Before launch it commissioned an independent assessment from Professor Patrick Sturgis of the LSE. His report, Assessment of the Gambling Survey for Great Britain (February 2024), called the development work “exemplary in all respects” and endorsed the move away from in-person interviewing.
He also attached a serious caveat, which the OSR later quoted: policy-makers “must treat them with due caution, being mindful to the fact there is a non-negligible risk that they substantially over-state the true level of gambling and gambling harm in the population.” Both halves of that sentence are real. The method is sound; the level it produces may still be too high.
The 2025 experiments softened the warning somewhat. In its January 2026 response to the OSR, the Commission said the research showed “the risk of over estimation in the GSGB was not as high as initially thought,” and on that basis decided not to add an over-estimation warning to every release. Reasonable people disagree about that decision.
The OSR went further than asking for research. It recommended the Commission “include a clear statement, at the start of each statistical release, on Professor Sturgis’ conclusion that there is a risk that they over-state the true level of gambling and gambling harm in the population.” That recommendation was not adopted in the form it was written. Readers of the headline figures are therefore not warned about the risk on the page where they meet the number, which is a fair thing to know when you read a press story quoting it.
The rules for using these numbers
The Gambling Commission publishes guidance on using statistics from the GSGB, last updated 16 July 2026. It is short and worth following.
- Do not treat the PGSI as a measure of addiction. The guidance says plainly that it “should not be confused with a measure of gambling addiction.”
- Do not compare GSGB figures with older gambling or health surveys to show a trend. The methods differ too much.
- You may point out that two surveys differ — as this page does — provided you say the methods were different and do not present the gap as change over time.
- You may compare GSGB with GSGB. It now has three years of consistent data and is its own baseline.
The PGSI is a screen, not a diagnosis
This is the single most common error in gambling reporting. The PGSI is nine questions about the past twelve months. A score of 8 or more marks someone out for a closer look. It does not diagnose anyone, and it is not the same thing as gambling disorder in the clinical manuals. We explain the distinction in what is problem gambling and the clinical picture in gambling addiction explained.
So “2.4 percent of British adults are gambling addicts” is not a statement the GSGB supports, and the regulator that publishes the GSGB says so itself. The correct form is: 2.4 percent of adults who took part scored 8 or more on the PGSI.
How to read any gambling statistic
Five questions will catch almost every bad comparison you meet.
- Which survey, and which year? “UK problem gambling is 2.4 percent” means nothing without the source.
- Which instrument? PGSI, DSM-5 criteria, SOGS and the Lie/Bet screen all count different things and produce different rates. Germany’s national survey, for example, is built on DSM-5 criteria, so its figures are not interchangeable with PGSI figures.
- Which mode? Push-to-web, in-home interview, telephone, or paper. This is the one almost everybody skips.
- What time window? Past-year and lifetime rates are not comparable. Several Asian surveys report lifetime SOGS scores, which look alarmingly high next to past-year PGSI figures for no real reason.
- Who was excluded? Both surveys above sample private households only. People in prisons, hostels, care homes or sleeping rough are left out — and some of those groups have high rates of gambling harm.
Our guide to gambling prevalence applies these tests country by country, and global gambling harm in numbers sets out what can and cannot be said at world level.
What is still unresolved
The decisive test has not happened yet. The Commission has committed to benchmarking GSGB estimates directly against APMS and the 2024 Health Survey for England. As of its January 2026 update, both of those comparisons were still “awaiting raw data.” Until they are published, anyone claiming to know the “true” British rate is guessing.
What we can say now is narrower, and more useful: the direction of travel in the GSGB is measurable against itself, the PGSI band structure is stable across both surveys, and the headline gap between 2.4 percent and 0.4 percent is a methodological artefact rather than a disagreement about reality. Nobody should use it to argue that gambling harm is either invented or ten times worse than reported.
Why this matters if you are worried about your own gambling
Prevalence arguments can make people feel that their own problem is either too rare to mention or too common to take seriously. Neither follows. Population statistics describe groups; they say nothing about any individual. If your gambling is causing you harm, that is true regardless of whether the national figure is 0.4 percent or 2.4 percent.
Our step-by-step guide to getting help sets out what support exists and how to reach it, and what responsible gambling means covers the wider framework these surveys sit inside.
Sources and corrections
Every figure on this page is taken from the survey report or regulator page that produced it, and each is linked beside the claim it supports. We do not cite secondary summaries for statistics. If you find an error, our corrections policy explains how we fix it, and our editorial policy explains how these pages are written and reviewed.
Last reviewed: 7 October 2026. Sources checked on this date.