Goals, usefulness and abstraction in value-based choice