The relationship between gambling behaviour and gambling-related harm: A data fusion approach using open banking data.

Abstract

UK-based gambling policymakers have proposed affordability checks starting at monthly losses of £125. The present study combines open banking data with self-reports of the Problem Gambling Severity Index (PGSI) and other relevant information to explore the harm profiles of people who gamble at different levels of electronic gambling behaviour. This was a data fusion study in which participants consented to share their bank data via an open banking application programming interface (API) and who also completed relevant self-report items. Hierarchical hurdle models were used to predict being an at-risk gambler (PGSI > 0) and being a 'higher-risk' gambler (higher PGSI scores among those with non-zero scores) using four specifications of electronic gambling behaviour (net-spend, outgoing expenditure, incoming withdrawals, interaction model combining expenditure and withdrawals), and by adding self-reported data across two additional steps. The study took place in the United Kingdom. Participants were past-year people who gamble (n = 424), recruited via Prolific. Self-report measures were used of gambling-related harm (PGSI), depression [Patient Health Questionnaire 9 (PHQ-9)], age and gender; bank-recorded measures of income and electronic gambling behaviour. Unharmed gamblers had an average monthly gambling net-spend of £16.41, compared with £208.91 among highest-risk gamblers (PGSI ≥ 5). Being an at-risk gambler (PGSI > 0) was predicted significantly by all four types of gambling behaviour throughout all three steps [1.08 ≤ odds ratios (ORs) ≤ 2.92; Ps < 0.001), with only outgoing expenditure being significant in the interaction model (2.26 ≤ ORs ≤ 2.81; Ps < 0.001). Higher PHQ-9 scores also predicted at-risk gambling in steps 2-3 (1.09 ≤ ORs ≤ 1.10; Ps < 0.001), as did lower age (0.95 ≤ ORs ≤ 0.96; Ps < 0.001) and male gender identity in step 3 (2.51 ≤ ORs ≤ 2.95; Ps < 0.001). Being a higher-risk gambler was predicted significantly by gambling behaviour only in the expenditure-only (1.16 ≤ ORs ≤ 1.17; Ps ≤ 0.048) and withdrawal-only (1.08 ≤ ORs ≤ 1.09; Ps ≤ 0.004) models, and was not predicted by income (0.98 ≤ ORs ≤ 1.14; Ps ≥ 0.601), age (0.98 ≤ ORs ≤ 0.99; Ps ≥ 0.143) or male gender identity (1.07 ≤ ORs ≤ 1.15; Ps ≥ 0.472). The UK government's proposed affordability checks for gamblers should rarely affect people who are not experiencing gambling-related harm. At-risk gambling is predicted well by different types of gambling behaviour. Novel insights about gambling can be generated by fusing self-reported and objective data.