Method at a glance
- Modality: behavioural (intuitive decision-making under ambiguity), paired with electrodermal and cardiac recording
- Measures: The degree to which a person learns, without reportable understanding of why, to prefer profitable over unprofitable options — scored as total profitable minus total unprofitable deck selections across 100 trials (range +100 to −100). Run with pre-decision EDA and HR recording, it also measures whether anticipatory bodily responses differentiate good from bad options.
- Procedure: Four decks of cards and a running money total. On each trial the participant contemplates the decks for 3 s, selects one, then guesses whether the card drawn will be the same or a different colour from an upturned central card; feedback follows for 3 s (money up plus a melodic flourish, or money down plus a buzzer), then a 3 s intertrial interval. Unknown to the participant, decks A and B score guesses correct on 6 of 10 trials (profitable, +200 pence per 10 trials) and decks C and D on 4 of 10 (unprofitable, −200 pence per 10). Magnitude is crossed with profitability: A and C pay 80–125 pence, B and D pay 105–180. Deck position and card backs are counterbalanced; participants may select from any deck as often as they like. A real cash bonus of up to £5 is paid on final winnings.
- Also called: IRT, Dunn’s card task
Strengths
- Built to repair the Iowa Gambling Task, point by point, by the people who catalogued its flaws. Nearly every change answers a specific objection in Dunn, Dalgleish & Lawrence (2006): the reversal confound is removed (no early advantage for the bad decks, so acquisition never requires un-learning), deck position and card backs are counterbalanced (no location bias), the deck-exhaustion limit is lifted (no penalty for early learners in the final block), and real money is at stake.
- Magnitude and profitability are crossed in a 2 × 2 — the design feature that matters most here. In the IGT the disadvantageous decks are also the high-variance decks, so ‘the anticipatory response marks badness’ and ‘the anticipatory response tracks reward magnitude/variance’ are confounded — the substance of Tomb et al.’s (2002) objection, and the sharpest live criticism on Skin conductance response (SCR). The IRT separates them. In Dunn et al. (2010) bodily responses tracked profitability and were unrelated to magnitude (Fs < 1, no interaction), which is the cleanest available answer to Tomb et al.
- The physiology is instrumented on the right window. Because the participant first picks a deck and only then guesses the card’s colour, the anticipation of making a decision is separable from the anticipation of receiving feedback. This addresses Amiez et al.’s (2003) objection that the IGT’s ‘anticipatory’ SCR may be post-decisional expectancy — the strongest structural criticism of the anticipatory window on Skin conductance response (SCR).
- Implicitness is validated rather than assumed. A separate 28-participant study probed hunch and conceptual insight after each block of 20 trials, using focused questions modelled on Maia & McClelland (2004) — the probe whose results dismantled the IGT’s impenetrability assumption. Behavioural learning appeared from block 3; conceptual insight reached significance only at block 5. Running the critics’ own probe on your own new task is the right instinct, and the IGT literature never did it prospectively.
Limitations
- Essentially a one-laboratory instrument. Introduced in Dunn et al. (2010) and, as far as this wiki’s sources show, not adopted since. Its test–retest reliability, its sensitivity to the VMPFC and amygdala lesions the Iowa Gambling Task was built around, and its correlation with IGT scores are all unestablished. Whatever the IGT’s faults, ~35 clinical studies had used it by July 2005; the IRT’s entire evidential base is two studies in one paper.
- Fixing the IGT’s confounds severs the link to the IGT’s literature. The IRT cannot inherit the Iowa results and cannot retrospectively decontaminate them. Its clean answer to Tomb et al. is an answer about the IRT — in the IGT the variance confound remains exactly as Tomb et al. described it. A reader should not treat IRT findings as IGT findings with better methods.
- The insight validation used a different sample from the main study (n = 28 vs. n = 92), so implicitness is established for the task, not for the participants whose data carry the result. The two groups also differ: mean NART IQ 111 vs. 117.
- Insight was not absent, only late and partial. Hunch insight for the chosen deck was above chance in blocks 1, 4 and 5, and calculated net profit reached significance by block 5. The honest description is that reportable knowledge lagged behaviour, not that the schedule was impenetrable — the same qualification Maia & McClelland forced on the IGT, arriving in gentler form.
- The pre-decision window is of variable length (‘of variable length but at least 3 s long’), since it runs until the participant selects a deck. Response time therefore varies with the physiological window it defines, and quick and slow deciders contribute differently-sized windows to the same average. Not addressed in the source.
- EDA’s status as a somatic marker is asserted, and the authors concede the gap. Note 3 of the source: ‘The source of feedback from EDA to the brain is currently unclear.’ The task’s bodily-differentiation composite averages EDA with HR; whether either is perceptible — and they are not the same channel as the cardiac Heartbeat detection task used to measure the perceiver — is left open. See Skin conductance response (SCR).
- Like the IGT, it registers that a somatic state occurred, not which one or of what valence — EDA is a single sympathetic channel and HR is dually innervated but scalar here. See Do basic emotions have distinct autonomic signatures?.
Connections
- Key studies: Dunn et al. (2010), Dunn et al. (2006)
- Related methods: Iowa Gambling Task, Skin conductance response (SCR), Heartbeat detection task
The Iowa Gambling Task’s purpose-built successor, introduced in Dunn et al. (2010). Its interest to this wiki is less what it measures than who built it and why: Barnaby D. Dunn co-authored the field’s most thorough catalogue of the IGT’s design flaws (Dunn et al. (2006)) and then constructed a task that fixes them, item by item.
The repairs, mapped to the objections
| IGT problem (Dunn et al. 2006) | IRT fix |
|---|---|
| bad decks pay more early, so learning requires reversal — and removing the reversal abolishes the VMPFC deficit (Fellows & Farah 2005a) | reinforcement schedule rebuilt so acquisition never requires un-learning |
| magnitude confounded with profitability, so anticipatory SCR may index variance not value (Tomb et al. 2002) | magnitude × profitability crossed in a 2 × 2 |
| deck position not counterbalanced, so choices may be location bias | position and card backs systematically counterbalanced |
| 80-card deck limit penalizes early learners in the final block | participants may select from any deck as often as they wish |
| the ‘anticipatory’ window may be post-decisional (Amiez et al. 2003) | deck choice and colour guess separated, isolating pre-decision from pre-feedback physiology |
| the schedule is cognitively penetrable (Maia & McClelland 2004) | insight probed with Maia & McClelland’s own focused questions in a validation study |
Read as a list, it is the most direct case in this wiki of criticism converted into instrumentation. Whether it worked is a separate question, and the answer is partly: the reversal, position, magnitude and window problems are genuinely addressed; penetrability is mitigated rather than solved, since hunch insight still appears above chance in block 1.
Why “intuitive,” and what the task actually indexes
Dunn et al. define intuitive ability as the degree to which a participant acquires the optimal strategy without being able to say why — operationalized simply as profitable minus unprofitable selections across all 100 trials. The framing borrows Kahneman (2003): intuition as automatic emotional judgement about whether an option is good, deployed where deliberation is unavailable.
The scoring is where a caveat belongs. “Intuitive ability” is a net choice score, exactly as in the IGT, and it carries the same interpretive burden: the label attributes the behaviour to intuition, but the measure only records the behaviour. What licenses the attribution is the insight validation — and that validation was run in a different sample, and found insight late rather than absent.
What it found
See Dunn et al. (2010) for the full result. In brief: anticipatory bodily responses differentiated profitable from unprofitable decks and correlated with performance (r = .41) — the Somatic marker hypothesis supported in a task purged of two of its critics’ central confounds. Interoceptive accuracy, meanwhile, was unrelated to performance (r = .08) and mattered only in interaction: better perceivers did better when their bodies were right and worse when their bodies were wrong.
Individual variation was wide — M = 25.74, SD = 40.71, with 27% of a healthy sample preferring the unprofitable decks. That figure is worth holding beside the Iowa Gambling Task page’s note that 20–37% of IGT controls perform disadvantageously. Rebuilding the task did not shrink the fraction of healthy adults who fail it, which is either a fact about decision-making or a fact about four-deck card games, and no source in this wiki distinguishes them.