Separating Habit from Value in the Iowa Gambling Task
A new math model shows that separating our habit of repeating past choices from our calculations of future rewards better explains human decision-making.
Source
Decomposing the roles of perseveration and expected value representation in models of the Iowa gambling task
What they did
Researchers tested 35 undergraduate students on the Iowa Gambling Task, a decision-making test where players choose from decks to maximize points over 100 trials.
What they found
The new model, which separates the habit of repeating a choice from the calculation of expected rewards, fit the participants' choice patterns much better than previous models. Simulations from this model accurately predicted that participants would switch decks approximately 62 times during the task, whereas older models either over-predicted or under-predicted this switching behavior. Additionally, a strong link was found between a participant's calculated level of loss aversion and their overall performance, showing that a healthy fear of losing points drove better choices.
The limits
What it doesn't show
Because the study only tested 35 undergraduate students from a single university, the results might not represent the broader population or clinical groups. Additionally, the researchers did not use the generalization criterion method to test if the model's parameters could predict how the same participants would behave on an entirely different decision-making task. Finally, the study relies on correlations between model parameters and performance, meaning it cannot prove that loss aversion directly causes better decision-making.
Key terms
- Iowa Gambling Task (IGT)
- A psychological task used to evaluate decision-making by observing how participants balance short-term gains against long-term losses across multiple decks of cards.
- Perseveration
- The tendency to repeat a specific behavior or choice over and over, regardless of whether it continues to lead to a positive outcome.
- Loss Aversion
- A psychological phenomenon where the pain of losing something is felt more intensely than the pleasure of gaining an equivalent amount.
- Expectancy Valence (EV) Model
- A mathematical model of decision-making that assumes choices are driven primarily by a running average of the rewards and punishments previously received.
- Decay Learning Rule
- A mathematical rule assuming that the perceived value of unchosen options gradually decreases over time, making recently chosen options feel more attractive.
- Bayesian Hierarchical Estimation
- A statistical method used to estimate model parameters for individuals while pulling extreme values toward the group average to reduce measurement errors.
Flashcards
Want these cards to stick?
Save the deck to NoteFren and study it with spaced repetition.
Quiz yourself
What is the primary objective given to participants in the Iowa Gambling Task (IGT) experiment described in this study?
Common questions
Why is it important to separate perseveration from expected value?
If a model bundles these two concepts together, we cannot tell if a participant is repeatedly choosing a deck because they genuinely think it has high rewards or simply because they have fallen into a mindless habit of repeating their last move.
How does the new VPP model improve on previous models?
Unlike older models that either ignored habits entirely or mashed habits and reward calculations together, the VPP model uses separate mathematical terms for both processes, allowing it to predict human deck-switching rates with incredible accuracy.
What role does loss aversion play in the Iowa Gambling Task?
Participants who are more sensitive to losses tend to perform better because they quickly learn to avoid the 'bad' decks that offer high immediate wins but even higher eventual penalties.
What is the difference between maximum likelihood and Bayesian estimation in this context?
Maximum likelihood can push a participant's estimated traits to extreme, unrealistic boundaries, whereas Bayesian estimation balances individual data with group averages to produce more realistic and stable personality estimates.
More on Decision making