Report
Understanding and measuring consumer trust in the gambling industry - Full technical report
The understanding and measuring consumer trust in the gambling industry full technical report from The Gambling Commission.
Pilot and validation exercise
Objectives
Following the development of the new ‘Trust question’ above it was piloted on Wave 4 of Year 2 (2024), of the online Gambling Survey for Great Britain (GSGB). After analysis of the first set of data from the GSGB, the Commission came to Yonder with the need to understand why there was a large proportion of ‘don’t know’ responses than expected across the statements being tested, as well as a need to understand whether the same trends appeared when surveying Yonder’s Y-Live panel.
Sample
To understand this in more detail, Yonder fielded a quantitative survey to a sample of 1,500 people who had gambled in the last 12 months (excluding National Lottery only players). This sample size was chosen to ensure readable base sizes for follow-up questions concerning each of the statements, based on estimated fallout of those responding ‘Don’t know’ calculated using data from the Commission’s GSGB; the last 12 months' timeframe was also used to reflect that used in the GSGB.
Methodology
Firstly, respondents answered a series of questions about their demographic characteristics, as well as their gambling activity, for routing purposes. Respondents would then answer the ‘Trust question’, to sense check the nationally representative results from the GSGB against those from Yonder’s proprietary panel. Across the sample of 1,500 people who had gambled in the last 12 months (excluding National Lottery only), 43 percent selected ‘Don’t know’ for at least one statement.
Those who had selected a “Don’t know” response were then shown follow-up questions to delve deeper to understand why they had selected a ‘don’t know’ response. The approach here was 2-fold, with an initial mini-qualitative question intended to allow respondents to go into more depth about their response, and a prompted multi-code question with potential reasons, which included confusion or misunderstanding concerning each ‘Trust question’ statement. For the latter, the multi-code lists were designed in collaboration with the Commission.
To limit respondent burden and repetitiveness, a least fill approach was employed to ensure a maximum of 4 follow-up questions (1 mini-qualitative, and 3 multi-code prompted lists) were seen per participant. Where a participant had said ‘Don’t know’ in response to four or more ‘Trust question’ statements, the least fill ensured that they were not asked about the same statement in the mini-qualitative exercise, and the multi-code prompted list.
The mini-qualitative exercise asked participants to imagine they were explaining why they were unsure about the statement in question as if they were speaking in a thread and/or forum online or having a conversation with a friend who gambles. After they provided an open-text response detailing what they would say, a smart AI probe tool was used to follow up for further detail. The AI probe was instructed to ‘understand how the statement can be refined so it made sense to as many people as possible, as well as digging deeper into why people answered ‘Don’t know’. It was instructed to ‘probe for extra detail on how the statement might be improved – if they didn’t understand a certain word and/or phrase, could this be switched out?’ If the API did not pick up sufficient detail to include a tailored follow-up, the default probe ‘Which part of the sentence, if any, do you think is unclear?’ was used.
Strengths and limitations
Strengths
The mini-qualitative approach elicited useful qualitative feedback within the survey environment, allowing the Commission to get in-depth feedback on statements without needing to speak with participants via interviews or groups.
AI probing can sometimes garner more honest insights as respondents do not qualify their responses as much as they perhaps would with a human interviewer
Yonder’s ringfenced panel of people who gamble can provide insights from a more involved perspective, which can then be applied to the broader audience on the GSGB.
Limitations and mitigations
In keeping with the foundational samples for this programme of research, findings should not be treated as representative of all those who gamble, and caution should be taken in assessing the significance of sub-group differences such as gambling frequency and PGSI score, as these groups may also not be representative, and may represent a higher than natural proportion within the sample than they would in a survey of the general population.
Our sample size for this survey was 1,500. When filtering down to those that responded, ‘don’t know’, and were willing to give a verbatim response, this left base sizes in the 75 to 90 region for the mini-qualitative verbatims. Although larger base sizes are unlikely to have led to different thematic analysis, they would have given more weight to findings.
Burden on the respondent in this survey had potential to be high, with multiple mini-qualitative questions. It is also inherently difficult to describe why you don’t know something. To mitigate this burden, a least fill approach was used so respondents did not need to answer more than four follow-up questions.
Analysis process
To analyse the verbatim data from the mini-qualitative exercise, Yonder’s proprietary AI summary dashboard was used. This dashboard takes the initial verbatim responses for each statement and provides the user with the top 10 themes that are emerging, alongside a short overall summary. The relevant verbatim responses are also shown on screen to enable human validation of the AI thematic ranking and summaries.
Yonder’s research team used this dashboard alongside supplementary analysis of verbatims and follow-up responses to present the ‘top three’ emergent themes for each ‘Trust question’ statement, and assign a statement clarity score out of 5 to each, ranging from 1 for ‘Room for improvement’ to 5 for ‘Strong’.
Scores were assigned as follows:
| Statement | Score | Comments |
|---|---|---|
| Gambling companies are held accountable by a regulator if their conduct falls short of standards | 4 | Statement should be left as is – there is a potential to add examples of enforcement, but this would come with risk of priming |
| Gambling companies are free from corruption | 5 | No specific recommended change |
| Gambling games and machines are fair and free of errors | 2 | ‘Fairness’ was thought to be a particularly subjective phrase, which could be further refined to focus on a specific aspect of the gambling experience |
| Activities, offers and odds are clear, easy to understand and not misleading | 4 | No specific recommended change |
| Effective measures are in place to protect young, and other vulnerable people when gambling | 4 | Statement should be left as is – there is a potential to add examples of measures in place, but this would come at risk of priming |
| Gambling companies have effective checks in place before they allow gambling (ID, financial) | 5 | No specific recommended change |
| Gambling promotions and incentives do not encourage excessive gambling | 4 | Prompting respondents to consider what ‘excessive’ looks like to them may elicit less don’t know responses, but this statement is broadly strong |
| It is quick and easy to withdraw and collect my winnings | 4 | Prompting respondents to consider what ‘quick’ looks like to them may elicit less don’t know responses, but this statement is broadly strong |
Responses to the multi-code question lists were used to reinforce these findings, by picking out issues with a clear proportional lead versus others (for example, for those that said they ‘don’t know’ whether ‘gambling companies are free from corruption’, they were overwhelmingly more likely to say it was because they ‘didn’t know if gambling companies are corrupt or not’ (84 percent) verses not knowing ‘how gambling companies can be corrupt (8 percent), or not knowing ‘what is meant by corruption’ (4 percent)).
The multi-code findings validated Yonder’s understanding that the notable proportions answering ‘Don’t know’ was due to a lack of awareness or engagement with a specific element of the gambling experience, rather than a lack of understanding of the statement itself.
Outcomes
To that end, Yonder recommended adding a ‘Not applicable’ code across all statements, to afford respondents without direct personal experience a more accurate response option.
Additionally, Yonder recommended adding qualifying text to the ‘Trust question’ text itself, along the lines of “please answer your personal impression, even if you can’t be certain whether it is truly the case”, to ensure respondents answered based on interpretation, rather than solely relying on absolute knowledge.
The outcome of the validation exercise demonstrated similar trends as the GSGB survey and assured the Commission that the trust statements themselves are fit for purpose. The revised question was also shared with NatCen, who administer the GSGB for their review, and further suggestions were made to ensure the question would work effectively when placed within the survey.
NatCen recommended reframing the opening to the ‘Trust question’ to ensure that respondents were answering the question based on their own views and experiences rather than perceptions or opinions drawn from elsewhere. They also recommended streamlining the statement on activities, odds and offers to make it easier for respondents to answer. Finally, the question was split into two shorter questions due to space restrictions on the paper copies of the questionnaire. Yonder were consulted on these final changes and after full consideration, the Commission accepted all recommendations, which were implemented ready for the 2026 fieldwork of the GSGB.
Previous sectionStatement selection process - Trust technical report Next section
Revised Trust Question - Trust technical report
Last updated: 23 July 2026
Show updates to this content
No changes to show.