Performance of healthy participants on the Iowa Gambling Task.

Abstract

The Iowa Gambling Task (IGT; Bechara, Damasio, Damasio, & Anderson, 1994) is often used to assess decision-making deficits in clinical populations. The interpretation of the results hinges on 3 key assumptions: (a) healthy participants learn to prefer the good options over the bad options; (b) healthy participants show homogeneous choice behavior; and (c) healthy participants first explore the different options and then exploit the most profitable ones. Here we test these assumptions using 2 extensive literature reviews and analysis of 8 data sets. The results show that all 3 assumptions may be invalid; that is, (a) healthy participants often prefer decks with infrequent losses; (b) healthy participants show idiosyncratic choice behavior; and (c) healthy participants do not show a systematic decrease in the number of switches across trials. Our findings question the prevailing interpretation of IGT data and suggest that, in future applications of the IGT, key assumptions about performance of healthy participants warrant close scrutiny.

Approaches to Learning to Control Dynamic Uncertainty

“…As with the bandit task, people have to choose a card from each deck, and they receive feedback on their choices, which is used to help them decide on which decks to sample from to maximize their cumulative rewards [22]. In general the findings suggest that participants perseverate over an exploration strategy and rarely move from this to a stage of exploitation [23][24][25][26][27][28]. The reasons for this may be because decisions in the IOWA gambling task are predominately based on a variety of models of the environment and emotional factors (e.g., [22]), which we discuss in Section 1.3.…”

“…The reasons for this may be because decisions in the IOWA gambling task are predominately based on a variety of models of the environment and emotional factors (e.g., [22]), which we discuss in Section 1.3. For instance, one view is that card selection strategies are likely to change over time because people construe the IOWA gambling task as a non-stationary environment, despite the fact that it is a stationary one [27,29,30]. However, there has been considerable debate recently concerning the validity of the task and the reliability of the findings and whether much can actually be made of the learning profiles of participants in the task [27,29,30].…”

Data from 617 Healthy Participants Performing the Iowa Gambling Task: A “Many Labs” Collaboration

“…b Data collected by Annette Horstmann. These data were first published in Steingroever et al [10]. A subset of this dataset is published in Horstmann, Villringer, and Neumann [4].…”

“…The IGT is arguably the most popular neuropsychological paradigm to measure decision-making deficits in an experimental context. Part of the data was already reanalyzed elsewhere (i.e., [8], [10][11][12][13][14]) in order to assess basic assumptions underlying the performance of healthy participants on the IGT, and to compare reinforcement-learning models that try to disentangle psychological processes underlying performance on the IGT.…”

Understanding Optimal Decision-Making in Wargaming

“…At some point the subject gains enough confidence in their estimate of the payoff for each alternative that they begin to make use of this perceived information to maximize their long-term reward by choosing the alternatives with the greatest estimate -this phase is referred to as the exploitation phase. Typical decision performance measures are net score, frequency with which each deck was selected, an advantageous selection bias (proportion of good decks selected minus the proportion of bad decks selected), and intermediate deck selection frequencies (deck selection frequencies within each block of 20 trials) (Bechara et al, 1994;Steingroever et al, 2013). These summary statistics offer an understanding of the long-term value of each deck.…”