Decision-making and learning speed predicted by executive functions (evidence from IGT)
Iowa Gambling Task (IGT) is one of the most widely used paradigm for assessing decision-making. IGT is a card game with material gain. Subject is provided with 4 decks of cards and 2000 “dollars”. Subject either wins or loses a certain amount choosing consistently one card from a deck. However some of decks are more profitable than others (in terms of probability) (Bechara et al.,1994). Absence of any explicit probability estimations in the instruction makes IGT more ecologically valid compared to other research methods.
We considered IGT as a learning task where participants have to establish and refine probabilistic representations of the game situation and keep the aim during the ambiguity environment and process of restructuring representation. The main purpose of our study was to reveal preconditions for individual differences in IGT performance. We hypothesized that due to the learning nature of the task executive functions (EF) like updating, inhibition and shifting (Miyake & Friedman, 2012) might be the source of differences in the success in IGT. The results of empirical studies investigating roles of intelligence and EF in IGT are contradictory (Toplak et al., 2010), thus we controlled IQ in the study.
44 people from 18 to 37 y.o. (27,9±4,5) participated in our study (31 were female).
“Symmetry Span” – task for updating (Foster et al.,2014), the final score was calculated as the sum of elements of sequences repeated correctly in the trials, where a distractor task was solved right.
“Go/no-go” for inhibition and shifting (Blakemore & Robbins,2012). Initially, two variables were included in the analysis: “NoGo” as percentage of actions correctly inhibited and “Go” as percentage of action correctly initiated – but the last one was excluded because of the extremely low standard distribution.
IQ was assessed using Matrix Reasoning and Three-Dimensional Rotation sub-tests from the ICAR (Condon & Revelle, 2014) and two verbal scales from the ROADS battery (Kornilov & Grigorenko, 2010). IQ score averaged sub-tests results.
IGT was administrated in Russian adaptation (Kornilov et al,2015) based on implementation by Grasman and Wagenmakers (2005). In data analysis we used total score and learning speed defined as the trial after which subject makes over 85% choices from “good” decks.
We used linear regression modeling to verify our hypothesis. We revealed a significant interrelation between learning speed and NoGo score (p=0.002, R2=0.41). For NoGo and IGT total score we also found significant interrelation, however, the coefficient of determination in this model was substantially lower (p=0.0315, R2=0.13). Thus, subjects with higher levels of inhibition and shifting were learning faster in IGT and were higher in the total score.
For updating we found a significant (t-test, p=0.03) and somewhat paradoxical result: people with higher levels of updating were learning slower. Possibly, it could be explained by the sort of “over-learning”: subjects with high-capacity working memory get used to relay on more details thus they need more time for collecting them.
Partial correlations IGT total score and EF with controlling level of intelligence didn’t reach the level of significance (p=0.063 and 0.064), which is most likely had been caused by the low power of the study.
We also found significant difference in the level of intelligence between people with higher and lower levels of updating (t-test, p=0.006), which corresponds previous studies (Thompson et al., 2013).
Supported by RGNF 150610404a