Flexibility of reinforcement biases and reaction times in competitive zero-sum games

In competitive zero-sum games with mixed equilibria, two rational players should make each of their game choices randomly, with no contingencies between their choices. However, people often deviate from this equilibrium by following a reinforcement heuristic of repeating moves that won on the previous round (win-stay) and avoiding the repetition of moves that did not win (lose-shift). In this thesis, I examine the flexibility of these reinforcement biases, and the speed of decision-making: under what circumstances do people make biased choices, and under what circumstances do people choose quickly or stop to deliberate? In Chapter 1, I review the current state of knowledge on how well people can produce or detect randomness, how reinforcement biases influence decision-making, and how processing speeds might differ between different game situations. In Chapters 2 and 3 I present four experiments where I examined performance in the games Rock, Paper & Scissors (RPS; Chapter 2, Experiments 1 and 2) and Matching Pennies (MP; Chapter 3, Experiments 3 and 4). Surprisingly, I found no reinforcement biases in RPS, but consistent reinforcement biases in MP. Additionally, participants made slower decisions after losses when they succeeded in the game due to finding an appropriate strategy to exploit an opponent’s pattern (Chapter 2), but not when they succeeded no matter what they did (Chapter 3). In Chapter 4, I present two experiments (Experiments 5 and 6) directly comparing performance in RPS and MP, designed to replicate the findings from Chapters 2 and 3, and examine why the previous studies only found reinforcement biases in MP. The results of these two last experiments suggest that reinforcement biases differ between RPS and MP due to different cognitive demands, and that there is considerable variability in reinforcement biases both between individuals and between the two types of bias. In Chapter 5, I discuss the contributions of the findings on the wider literature on bias and randomness detection, the generality of the reinforcement biases, and present some suggestions for future studies.