The pervasive nature of online gambling has pushed it to the forefront of social concerns in Great Britain (GB). Understanding how and where gambling-related behaviours manifest is essential for informing targeted interventions and evidence-based public policy.
The GB2C dataset provides the first national areal classification of gambling behaviours in GB. Uniquely, it is based on observed online transactional behaviours drawn from industry data. The classification is built using circa 1.2 million anonymised online gambling accounts recorded throughout 2022, provided by one of the ‘Big 5’ British gambling operators. This work was conducted independently by GeoDS researchers, with data access facilitated through collaboration with the gambling service provider (which had no influence over the research or reporting of it). Using this unique data resource, customers were segmented into 11 Active Subgroups, with additional estimates for non-Active account holders and the remaining adult population. Local Authority District (LAD) estimates of the incidence of each Subgroup are available through the GeoDS for bona fide research purposes. More granular Lower layer Super Output Area (LSOA) data are also available.
This classification extends what is possible using conventional survey instruments alone. Linkage of georeferenced, anonymised individual customer records to neighbourhood attributes from the GeoDS UK Output Area Classification (UK-OAC) and the GeoDS Harmonised Index of Multiple Deprivation (IMD) enables GB-wide profiling of the geographic context in which actual patterns of gambling behaviour occur – rather than relying on coarser regional scale reports of stated behaviour which are subject to recall errors.
This independent, ethically approved GeoDS Research Ready Data product pushes the frontiers of social science methodology. It empowers researchers at all career stages to develop deeper insights into the complexities of gambling behaviour in GB, at spatial scales previously unavailable – all while maintaining the highest standards of data protection and ethical research practice.
Content
The data are provided in CSV format. Additional resources, including a detailed glossary of terms, descriptive statistics and pen portraits are also available for download.
This dataset applies a small-area estimation approach to model the geographic distribution of online gambling behaviours across GB. Regional-level customer counts drawn from circa 1.2 million accounts in 2022 were used to create LAD level estimates using decile-ranked estimates of gambling penetration profiles and local population data from the 2021/2022 Census. Market share adjustments, benchmarked to national prevalence rates derived from the Gambling Survey for Great Britain (GSGB) are used to ensure consistency of estimates with known patterns of gambling participation online. The resulting estimates provide neighbourhood-level counts of adults segmented by 13 classifications of online gambling behaviours.
Full methodological details, including descriptions of input features, will be provided in a forthcoming academic paper.
Quality, Representation and Bias
The GB2C dataset is based on anonymised behavioural records from a single major British gambling operator, covering online gambling activity throughout the 2022 calendar year. While this operator is among the largest in the market with broad national reach, the dataset captures only a partial view of total gambling engagement across GB. Cross-operator activity, land-based gambling and online lottery participation are not observed, potentially leading to underestimation of some individuals’ total gambling behaviour. Our implicit assumption is that these effects are uniform between different gambling behaviours.
To enhance representativeness, estimates were triangulated with national survey benchmarks from the GSGB, helping to align prevalence and ensure even coverage across GB regions. However, survey estimates are subject to response biases (e.g., recall and interviewer/interviewee interaction effects), which may propagate into small-area estimates.
These limitations should be considered when interpreting the data in applications where they are relevant.
