No open-access location is known for this record yet.
No open-access location is known for this record yet.
This dissertation examines the effects of a decision-maker's goals on what is learned from experience in a complex environment. The overarching hypothesis is that goals affect learning by changing the allocation of attention, and that less attended stimuli contribute less to learning. This over-arching hypothesis was tested in a series of three experiments by manipulating goals and attention across environments. In each experiment, subjects learned to predict prices based on three product attributes. Goals were manipulated by defining the learning task to be either price prediction or an action-based categorization task that depended on the continuous price value. The three experiments demonstrate that the information extracted from literally identical stimuli and informationally identical feedback is strongly affected by the decision-maker's goal at the time of learning. Performance is only weakly affected by the decision-maker's goal at the time of testing and the match between goals during learning and test. The results are consistent with a model in which problems within an “attentional band” of prices received more extensive processing than other problems. When the goal at learning is price prediction, the attentional band is wide; when the goal is action-based categorization, the band is narrow and focused on the category boundary. Also, when engaging in action-based tasks, subjects are unlikely to use the maximum information available on all problems (i.e., price in addition to category membership). Rather, they are likely to “think less,” allocating their attention to all available information only for critical problems (i.e., those within the attention band). This results in selective learning. It was expected that selective learning would be detrimental to knowledge transfer from specific tasks (categorization) to general tasks (prediction). Instead, subjects in the categorization conditions equaled or exceeded the performance of those in the prediction conditions. Moreover, the “cost of learning” (i.e., time spent making decisions and processing feedback at the time of learning) was approximately twice as high for the prediction group than for the action-based group, even though their performance was no better.
Loading…
Imported from OpenAlex (CC0) — nothing here is generated.