Cognitive Biases in Bet Selection: Pattern Recognition in Zero-Sum Games

We examine the role of cognitive bias in roulette bet selection through a combination of think-aloud protocol analysis and a controlled experiment with 148 participants. We identify five distinct cognitive patterns that systematically influence bet selection: the gambler's fallacy, the hot-hand fallacy, anchoring to recent outcomes, representativeness heuristic applied to sequences, and loss aversion expressed as bet escalation. The patterns are largely independent of statistical knowledge: participants who correctly identified the gambler's fallacy in abstract form were nearly as likely to exhibit it in practice as those who did not. We discuss implications for educational interventions.

The psychology of roulette bet selection is a domain where abstract statistical knowledge and in-the-moment behavior diverge sharply. Prior research has documented this divergence for the gambler's fallacy specifically; our study extends the analysis to a broader catalogue of biases and examines whether knowledge of the bias predicts behavior in practice.

We recruited 148 participants through the Roulette Community's member network and local university psychology departments. Participants completed a statistical knowledge assessment and then played 50 rounds of simulated roulette, providing verbal think-aloud reports of their betting reasoning. Sessions were recorded and transcribed; think-aloud protocols were coded by two independent raters.

The gambler's fallacy — betting against recent outcomes — was the most prevalent pattern, appearing in some form in 79% of participants. Its behavioral counterpart, the hot-hand fallacy — betting with recent outcomes — appeared in 61% of participants, sometimes within the same session. The two patterns are logically incompatible, yet 44% of participants exhibited evidence of both, switching between them depending on the salience of the recent streak.

Anchoring to recent outcomes was observed in 68% of participants: after a run of wins, participants reported feeling 'on a roll' and made larger bets; after a run of losses, they either reduced bets (consistent with loss aversion) or escalated them (consistent with the gambler's fallacy). Which response predominated appeared to be a trait-level individual difference rather than situationally determined.

The representativeness heuristic manifested as a strong preference for sequences that 'look random' — participants avoided repeating the same bet for more than three consecutive rounds even when the repetition was strategically neutral. This tendency produced a characteristic saw-tooth pattern in bet selection that had no mathematical basis.

Most relevantly for educational practice: participants who scored in the top quartile on our statistical knowledge assessment were not significantly less likely to exhibit any of the five cognitive patterns in behavior (all comparisons p > 0.10). The dissociation between knowing and doing is robust. This finding suggests that educational interventions focused purely on statistical knowledge are insufficient; behavioral interventions — such as structured pre-commitment and real-time decision prompts — may be necessary to bridge the gap.