Does gambling spend predict harm? Open-banking evidence suggests the correlation with PGSI is nearly zero.
Session Title
Public Health: Harm Measurement & Indicators
Presentation Type
Paper Presentation
Start Date
26-5-2026 12:00 AM
Abstract
Regulatory frameworks rely on spending patterns to identify individuals at risk of gambling harm. Research shows higher gambling spend is associated with higher Problem Gambling Severity Index (PGSI) scores, but whether spend works as a diagnostic tool for individual harm remains unclear. We test this using longitudinal open banking data from 5,756 UK adults linked to PGSI scores. We analyse behavioral indicators spanning the year before PGSI assessment, including net deposits, frequency measures, and multi-operator use. Linear regression shows higher gambling activity is associated with higher PGSI, with net deposits as the strongest predictor: each £1,000 increase in monthly net deposits is associated with a ~7-point PGSI rise (95% CI: 5.95 to 6.96, p < .001). However, these associations translate poorly into individual-level prediction. Linear models explain just 14% of PGSI variance, leaving over 85% of individual differences unexplained by spending or behavioral indicators. Non-linear models improve this only to 24%. Critically, substantial heterogeneity exists within spending groups: both high (>£1,000/month) and low spenders (<£10/month) show PGSI scores from zero to severe. While behavioural indicators show significant population-level associations with harm, they fail to support reliable individual risk profiling, challenging the dose-response paradigm underlying current policies. Ongoing analyses examine whether psychosocial comorbidities improve PGSI prediction.
Does gambling spend predict harm? Open-banking evidence suggests the correlation with PGSI is nearly zero.
Regulatory frameworks rely on spending patterns to identify individuals at risk of gambling harm. Research shows higher gambling spend is associated with higher Problem Gambling Severity Index (PGSI) scores, but whether spend works as a diagnostic tool for individual harm remains unclear. We test this using longitudinal open banking data from 5,756 UK adults linked to PGSI scores. We analyse behavioral indicators spanning the year before PGSI assessment, including net deposits, frequency measures, and multi-operator use. Linear regression shows higher gambling activity is associated with higher PGSI, with net deposits as the strongest predictor: each £1,000 increase in monthly net deposits is associated with a ~7-point PGSI rise (95% CI: 5.95 to 6.96, p < .001). However, these associations translate poorly into individual-level prediction. Linear models explain just 14% of PGSI variance, leaving over 85% of individual differences unexplained by spending or behavioral indicators. Non-linear models improve this only to 24%. Critically, substantial heterogeneity exists within spending groups: both high (>£1,000/month) and low spenders (<£10/month) show PGSI scores from zero to severe. While behavioural indicators show significant population-level associations with harm, they fail to support reliable individual risk profiling, challenging the dose-response paradigm underlying current policies. Ongoing analyses examine whether psychosocial comorbidities improve PGSI prediction.