Gain-Loss Separability and Coalescing in Risky Decision Making

T experiment tested two behavioral properties of risky decision making—gain-loss separability (GLS) and coalescing. Cumulative prospect theory (CPT) implies both properties, but the transfer of attention exchange model (TAX) violates both. Original prospect theory satisfies GLS but may or may not satisfy coalescing, depending on whether editing rules are assumed. A configural form of CPT proposed by Wu and Markle (2004) violates GLS, but satisfies coalescing. New tests were designed and conducted to test these theories against specific predictions of a TAX model. This model used parameters estimated from previous data, together with simple new assumptions to extend TAX to gambles with negative and mixed consequences. Contrary to all three forms of prospect theory, systematic violations of both coalescing and of GLS were observed. Violations of GLS were confirmed by analyses of individual data patterns by means of an error model in which each choice can have a different rate of error. Without estimating any parameters from the new data, the TAX model predicted the majority choices in the new data fairly well, correctly predicting when modal choices would violate GLS, when they would satisfy it, and when indifference would be observed.