Decision Making and Modeling: Computational Modeling in Different Iowa Gambling Task

The Iowa Gambling Task (IGT) is a laboratory-based gambling paradigm which involves a conflict between immediate gain and delayed loss. In original IGT, the bad decks with lower expected values are associated with higher magnitude of reward and punishment, and the good decks with higher expected value are associated with lower magnitude of reward and punishment. Normal participants usually chose the bad decks in the beginning but switched to good decks through gradual learning. It is usually interpreted that normal participant’s performance is guided by the expected value. However, there are many confounding variables in IGT. One of such factors is risk level. In the present study, the original IGT and three modified IGT were studied. In the modified IGT, the expected values and risk levels were manipulated. However, there are structure differences among the four IGT, for example, whether there is a conflict between immediate and delayed results and whether decks only contain immediate gain and delayed loss. One way to clarify the possible underlying processes in IGT is to apply cognitive modeling to identify the specific processes. Therefore, we further applied Expectancy-valence learning model (EV model, Busemeyer & Stout, 2002) and prospect-decay-independent model (PDI model, Ahn, Busemeyer, Wagenmakers, & Stout, 2008) for four IGT behavioral results. By applying these two cognitive decision models, it is found that PDI model outperformed than the EV model, not only on the overall goodness of fit, but also the ability to reflect the experimental manipulations.