Abstract
In previous work, we have shown that violating certain assumptions underlying multidimensional decision models has little effect on the performance ratio in an uncertainty paradigm when modeled as a yes/no task. In certainty conditions, the observer knows in advance which of two components or dimensions will carry a signal to be discriminated. In uncertainty conditions, the relevant dimension or component is unknown. The performance ratio is the ratio of d' values in uncertainty and certainty conditions. However, many absolute d' values could lead to the same performance ratio. Here, we extend the analysis to a rating task and directly examine d' in the uncertainty condition. We varied the relative saliency of each potential cue by varying the relative mean and variance values defining each underlying distribution, violated assumptions of equal variance, etc. Despite large changes in these values, d' values based on 10000 simulated trials varied over a range less than 0.1 d' until. We conclude from these and our performance ratio results that the paradigm is very robust with respect to underlying assumptions, and that violations of them do little to skew results.
Supported by NIH grant EY13953 to LAO.