Bias and sensitivity in two-interval forced choice procedures: Tests of the difference model

We assess four common claims concerning the two-interval forced choice (2-IFC) procedure and the standard Difference Model of 2-IFC performance. The first two are (1) that it is unbiased and (2) that the structure of the 2-IFC task does not in itself alter sensitivity. The remaining two concern a claimed sqrt(2) enhancement in sensitivity in 2-IFC relative to that measured in a Yes-No task. We review relevant past research and re-analyze seventeen experiments from previous studies across three laboratories. We then report an experiment comparing 2-IFC performance with performance in a second task designed to elucidate observers' decision processes. This second task is simply two successive Yes-No signal detection tasks with the same timing as in the 2-IFC experiment. We find little evidence supporting the claims that 2-IFC is unbiased and that it does not alter sensitivity and we also reject the two claims associated with the Difference Model as a model of performance in our own experiment.