Investigating the failure to best respond in experimental games

In experimental games, a substantial minority of players often fail to best respond. Using two-person 3x3 one-shot games, we investigated whether 'structuring' the pre-decision deliberation process produces greater consistency between individuals' stated values and beliefs on the one hand and their choice of action on the other. Despite this intervention, only just over half of strategy choices constituted best responses. Allowing for risk aversion made little systematic difference. Distinguishing between players according to their other-regarding preferences made a statistically significant difference, but best response rates increased only marginally. It may be that some irreducible minimum level of noise/imprecision generates some proportion of sub-optimal choices. If so, more research might usefully be directed towards competing models of stochastic strategic choice.