The predictive accuracy of intertemporal-choice models.

Area of Science:

• Behavioral Economics
• Decision Science
• Computational Neuroscience

Background:

• Intertemporal choice, the selection between smaller sooner rewards and larger later rewards, is a fundamental aspect of decision-making.
• Existing models of intertemporal choice are often evaluated by goodness of fit or their ability to explain anomalies.
• Predictive accuracy offers an alternative, crucial metric for assessing model quality.

Purpose of the Study:

• To compare the predictive accuracy of 10 different models of intertemporal choice.
• To evaluate model performance using cross-validation on a binary-decision task.
• To identify robust models for predicting choices involving delayed rewards.

Main Methods:

• A 100-trial binary-decision task was employed to collect behavioral data.
• Cross-validation techniques were used to assess predictive accuracy.
• Ten distinct models of intertemporal choice were tested, including a novel logistic-regression model termed the difference model.

Main Results:

• Many models achieved high accuracy (around 85%), even with limited training data.
• The difference model demonstrated superior predictive performance, particularly when noise was introduced into the training data.
• Differences in predictive accuracy between models were often small, challenging existing debates.

Conclusions:

• The difference model exhibits strong simplicity and robustness, making it suitable for future research on intertemporal choice.
• Predictive accuracy is a valuable criterion for evaluating intertemporal choice models.
• The findings suggest that simpler models can be highly effective in predicting complex decision-making behaviors.