Recency and Primacy Biases Are Partially Dissociable: Evidence from the Florida-Arizona Gambling Task

What dimensions characterize memory-based decision making in everyday life? We test this question with a novel gambling task—the FLAG task. Reinforcement learning, factor, and neural network models reveal that the FLAG task is low dimensional and moderately predictable. Different cognitive components (e.g., recency and primacy biases) are partially dissociable.