The effects of cash-out availability on horse-race betting: A common feature in many online betting platforms is the ability to “cash out” of a bet after it has been placed. This feature somewhat mitigates the risk of placing bets and may encourage people to gamble with more and riskier bets. To test the effects of this cash-out feature on betting, we ran an online experiment where people could place bets on horse-racing videos. In a first session, participants (N = 378) filled out some questionnaires and completed a real-effort task to earn a bonus of up to £3. They were then invited to a second session where they could place bets on a series of 3 horse races via a naturalistic interface. Participants were randomly assigned to either have cash-out available or not available during their betting session. People in the cash-out group were more likely to choose the betting task over a distractor (non-gambling) game and bet ~10% more of their available stake than people who did not have cash-out available. There was no effect on the average odds of bets placed. The effect of cash-out availability on bet size was more pronounced amongst those with low PGSI scores (<3), with those participants betting nearly 20% more of their stake. This differential impact makes low-risk bettors’ behaviour less distinct from that of high-risk bettors when cash-out is available. This similarity raises concerns about the ability of operators to identify those at risk of harm, as well as potential longer term effects of cash-out on low-risk bettors.

The dynamics of decision making with uncertain outcomes: In situations where an agent repeatedly makes decisions with uncertain outcomes there is no metric to definitively compare different strategies. A common approach is to use an expected value, this represents a scenario being repeated infinitely many times and outcomes averaged. However, this approach overlooks that a single agent observes an outcome only once, and there is no guarantee the expected value with be representative. Instead, the approach investigated in this talk is to consider the long-term rate of growth in wealth of a single agent - that is, rather than average over an ensemble we instead average over time. This talk will introduce the relevant theory and conclusions that a growth-based viewpoint of sequential decision making provides. We will also focus on novel extensions of growth-rate theory to a wider class of decision making problems that include phenomena, such as ruin, that are found in the real-world.