Striving for perfection and falling short: The influence of goals on probability matching

Probability matching in sequential prediction tasks is argued to occur because participants implicitly adopt the unrealistic goal of perfect prediction of sequences. Biases in the understanding of randomness then lead them to generate mixed rather than pure sequences of predictions in attempting to achieve this goal. In Study 1, N = 350 participants predicted 100 trials of a binary-outcome event. Two factors were manipulated: probability bias (the outcomes were equiprobable or distributed with a 75%-25% bias), and goal type-namely, whether single-trial predictions or the perfect prediction of four-trial sequences was emphasized and rewarded. As we hypothesized, predicting sequences led to more probability-matching behavior than did predicting single trials, for both the bias and no-bias conditions. In Study 1B, we added a control condition to distinguish the effects of the grouped presentation of trials from the effects of sequence-level perfect-prediction rewards. The results supported goal type rather than presentation format as the cause of the Study 1 differences in matching between the sequence and single-trial conditions. In Study 2, all participants (N = 300) predicted the outcomes for five-trial sequences, but with different goal levels being rewarded: 60%, 80%, or 100% correct predictions. The 100% goal resulted in the most probability matching, as hypothesized. Paradoxically, using the inferior strategy of probability matching may be triggered by adopting an unrealistic perfect-prediction goal.