On gamblers' use of language in the context of their financial transaction behaviour, gambling harms and financial harms
Session Title
Gambling Behavior: Language, Finance & Digital
Presentation Type
Paper Presentation
Start Date
28-5-2026 12:00 AM
Abstract
How people talk about gambling may reveal harms that are not fully captured by screening instruments or transaction data alone. We examine whether free‑text responses to openly worded survey questions provide incremental insight into gambling harm and financial distress beyond established screens and behavioural markers. We analyse banking data from a random sample of 4,000 UK adults, linked to two surveys containing PGSI, open‑ended questions on gambling and financial wellbeing. Using natural language processing (embedding‑based and lexicon/topic approaches), we extract linguistic themes and markers from participants’ written responses. We then apply supervised learning to (i) predict PGSI scores and (ii) detect self‑reported financial distress, benchmarking models that use free‑text features, transactional markers, and their combination. We present results showing the association between free‑text signals, e.g., references to advertising; emotional responses; addictiveness and vulnerability, and PGSI scores and show how these can add incremental predictive value when combined with behavioural transaction-based indicators. By integrating behavioural and linguistic evidence, this work offers a richer view of how gambling‑related harms manifest and are articulated. Our findings highlight the complementary strengths of language and transactional behaviour for identifying gambling‑related harms and inform more sensitive screening and support pathways.
On gamblers' use of language in the context of their financial transaction behaviour, gambling harms and financial harms
How people talk about gambling may reveal harms that are not fully captured by screening instruments or transaction data alone. We examine whether free‑text responses to openly worded survey questions provide incremental insight into gambling harm and financial distress beyond established screens and behavioural markers. We analyse banking data from a random sample of 4,000 UK adults, linked to two surveys containing PGSI, open‑ended questions on gambling and financial wellbeing. Using natural language processing (embedding‑based and lexicon/topic approaches), we extract linguistic themes and markers from participants’ written responses. We then apply supervised learning to (i) predict PGSI scores and (ii) detect self‑reported financial distress, benchmarking models that use free‑text features, transactional markers, and their combination. We present results showing the association between free‑text signals, e.g., references to advertising; emotional responses; addictiveness and vulnerability, and PGSI scores and show how these can add incremental predictive value when combined with behavioural transaction-based indicators. By integrating behavioural and linguistic evidence, this work offers a richer view of how gambling‑related harms manifest and are articulated. Our findings highlight the complementary strengths of language and transactional behaviour for identifying gambling‑related harms and inform more sensitive screening and support pathways.