Gamble evaluation and evoked reference sets: Why adding a small loss to a gamble increases its attractiveness

When presented with a gamble involving a chance of winning $9, participants rate it as only moderately attractive. However, when other participants are presented with a gamble that adds a chance of losing 5 cents - resulting in gamble that is strictly worse - they rate it as much more attractive. This surprising effect has previously been explained in terms of the small loss increasing the affective evaluability of $9. This paper argues for an alternative model, in which the baseline and small-loss gambles evoke different reference sets for comparison. In inferring a relevant reference set, people are sensitive to both the objective content and the framing of a gamble. The model distinguishes between two effects of evoked reference sets on behavior - an obligatory (and rational) effect on scale interpretation, and an optional (but not rational) effect on the internal representation of value. Five experiments provide strong evidence for the evoked reference set model. Data from attractiveness ratings suggest large and consistent reference set effects on scale interpretation, while data from willingness-to-pay and choice tasks suggest that effects on the internal representation of value are less robust.