Iowa Gambling Task (IGT) is frequently used to index individual differences in decision-making under uncertainty, particularly in atypical (clinical) populations. However, it is rarely analyzed as a learning task, and research on the predictors of performance on the IGT in normative populations is scarce. Here, we focused on tolerance and intolerance for uncertainty as two traits that could potentially influence subjects' IGT performance. Using mixed modeling analysis of longitudinal experimental data (n=60, 5 blocks) we showed that tolerance for uncertainty predicted the initial level of risk in IGT as manifested in the proportion of “bad decks” chosen; at the same time, intolerance for uncertainty predicted explorative learning in IGT as manifested in the number of deck switches after a loss and its decline over the course of the experiment. The results are discussed in the context of viewing IGT as capturing a set of dynamic decision making processes that rely on learning, risk taking, and exploration.
Despite a considerable body of research generated in the
field of decision making in the recent several decades,
sources of individual differences in decision making remain
largely understudied, in part due to the absence of
wellestablished individual differences-focused research
paradigms
The IGT requires the participant to choose cards from
four decks that have a systematically varied intermittent
gain and loss structure that the participants uncover by trial
and error during the experiment. The two disadvantageous
IGT decks (A and B) are associated with high immediate
rewards but long-term net losses. The two advantageous
decks (C and D), on the other hand, are associated with
lower immediate rewards but also significantly smaller
long-term losses. Initially used to test the somatic marker
hypothesis in patients with lesions to the ventromedial
prefrontal cortex
Perhaps surprisingly, a recent review of the associations
between participants’ performance on the IGT task and
cognitive traits found that IGT performance was largely
unrelated to general cognitive ability, working memory,
executive functions, and set shifting, although no aggregate
effect sizes were computed
Note that IGT performance is most frequently analyzed in terms of the resulting proportion of disadvantageous choices to all choices [(A+B)/(C+D)] in the second half of the experiment (that typically consists of 5 blocks of 20 trials, for a total of 100 trials) or the overall game money net gain by the end of experiment. Yet, the IGT task can also be considered to be a learning under uncertainty task where participants are faced with the neccessity to establish and continuously refinine probabilistic representations of the reward and punishment structure of the environment (i.e., the experimental deck setup). Correspondingly, decision making in IGT unfolds over time and within-participant learning trajectories can be established and related to individual differences in participants’ basic cognitive and personality characteristics.
Uncovering these trajectories and explaining them from
the standpoint of individual differences was the main aim of
the reported study. Based on our previous findings of the
importance of tolerance and intolerance for uncertainty for
understanding the nature and mechanisms of decision
making, we hypothesized that longitudinal indices of IGT
performance should be related to traits of tolerance and
intolerance for uncertainty as capturing the fundamental
regulatory elements of decision making
The participants were undergraduate students from Moscow State University and military instructors. A total of 60 adult participants took part in the study (age ranged from 18 to 52, M = 30.58, SD = 10.61; 41 were males, and 19 were females).
All participants were first administered the Iowa Gambling
Task, followed by personality assessments. For the purpose
of the study, we translated and adapted the standard
computerized IGT protocol developed by Grasman and
Wagenmakers
We analyzed the following indices of performance on the IGT: 1) cumulative Net Gain, 2) proportion of disadvantageous deck choices (Bad Decks) to total deck choices, and 3) proportion of deck switches after experiencing a loss (Loss Switches). Participants’ performance was averaged across 20 trials within each of the five blocks, and the resulting data were subjected to mixed modeling (or growth curve) analysis (see below).
The previously validated New Questionnaire of Tolerance
for Uncertainty (NQTU, or NTN in Russian) was used to
measure variables associated with acceptance of uncertainty
The data were analyzed using a set of mixed linear models
First, we found that over the course of the experiment, participants exhibited significant learning that could be described by a quadratic function (see Table 1). There was a trend for the association between participant’s net gains for the first IGT block (i.e., the intercept parameter) and TU (B = 5.68, SE = 2.93, t = 1.94), suggesting that tolerance for uncertainty modulates the baseline IGT performance level.
We also found that the proportion of Bad Decks decreased over the course of the experiment linearly. Importantly, TU was a significant predictor of the baseline level for this dependent variable (B = .006, SE = .003, t = 2.13), corroborating results reported in the previous paragraph.
Finally, we found that Loss Switches were relatively constant over the course of the experiment for our “average” participant. Yet, ITU predicted the baseline level (B = -.05, SE = .02, t = -2.32), with higher ITU associated with lower number of deck switches after losing experimental money. ITU also predicted the linear growth parameter (B = .05, SE = .02, t = 2.14) and showed a trend for a significant association with the quadratic growth parameter as well (B = -.01, SE = .005, t = -1.86). This result suggests that individuals with higher ITU are less likely to explore other decks after losing money in the beginning of the experiment, and potentially, display a relatively constant (or slightly increasing, compared to constant or slightly negative average growth rate, see Table 1) level of deck switching throughout the course of the experiment.
The reported study investigated individual differences in the dynamic indices of decision making as captured by the IGT. We found that tolerance and intolerance for uncertainty predicted several of these indices, suggesting that these traits modulate decision making on-line.
Tolerance for uncertainty predicted the participant’s
baseline performance on the IGT – i.e., individuals with
higher tolerance for uncertainty were more likely to choose
disadvantageous decks in the beginning of the experiment,
and yet showed higher net gains than individuals with lower
tolerance for uncertainty. These results suggests that
tolerance for uncertainty regulates baseline risk taking
propensity during online decision making under uncertainty,
consistent with previous reports of TU being linked to risk
taking
On the other hand, we found that higher intolerance for
uncertainty was associated with lower baseline exploratory
activity after failure (i.e., number of deck switches after
experiencing a monetary loss in the IGT). This finding is
consistent with viewing intolerance for uncertainty as
indexing risk aversion and uncertainty rejection
Overall, the results of our study indicate that decisionmaking under uncertainty is partially modulated and regulated by tolerance and intolerance for uncertainty. Interestingly, TU appears to regulate baseline risk propensity that underlies exploratory learning at the initial stages of decision making, while ITU regulates risk propensity after failure/loss, potentially constraining learning under uncertainty through risk aversion and outcome sensitivity.
Our study was also instrumental in showing the added value of investigating dynamic (as opposed to static) indices of decision making in the IGT task. Future studies should investigate the incremental predictive value of TU/ITU with respect to IGT performance over and above cognitive traits (i.e., nonverbal and verbal intelligence, working memory, and executive functions) in larger samples and attempt to clarify the mechanistic role of TU and ITU in constraining the online development of probabilistic representations and risky exploratory behavior under uncertainty.
This research and the preparation of this paper were supported by the Russian Foundation for Humanities (RGNF), grant №15-06-10404a (PI: Smirnov).