Studies of experience based decision making

Decision Making;Experience;Learning

Broomell, Stephen B.

关键词: Decision Making;    Experience;    Learning;   

Others  :  https://www.ideals.illinois.edu/bitstream/handle/2142/16719/1_Broomell_Stephen.pdf?sequence=2&isAllowed=y

美国|英语

来源: The Illinois Digital Environment for Access to Learning and Scholarship

【 摘 要 】

A series of papers aimed at characterizing how decision makers (DMs) make choices based on past experiences have revealed patterns that differ from those reported from description based decision making studies. Choices based on past experiences typically involve providing a DM direct experience with gambles. Experience takes the form of feedback provided from independent random plays (or realizations) from a gamble. The focus of this research is on experience based decision making where experience does not have real consequences, but is thought of as a learning process. The DM is free to sample from the source generating the gambles, without consequence, as many times as needed. During sampling, outcomes from the gamble have no bearing on the final payment. Once the DM feels ready, he/she chooses a gamble for a one-shot play for real money. Two experiments are presented to test the impact of three manipulations: experience, initial information states, and motivating incentives. The dependent variables include the DMs' choices, experienced samples, and estimated functional forms of Cumulative Prospect Theory.Results indicated that the manipulation of experience in the presence of a full description produced almost no effects. The manipulation of initial information states significantly influenced choices, the perceptions of risk, and the experienced samples. Choices show significant differences due to initial information that support under-weighting of small probabilities for groups exposed to experience without a full description. Ratings of risk perception are less discriminating between gambles with rare wins and rare losses for groups without a full description when compared to groups with a full description. Initial information states also influence experienced samples showing different search patterns between the two gambles that can be simultaneously sampled. The third manipulation of motivating incentives significantly influenced the experienced samples and perception of risk.Experienced samples are longer and produce less estimation error when motivated by probability estimation.The perception of risk also shows significantly less discrimination between gambles with rare losses and rare wins when compared to groups with full descriptions. Finally, estimates of weighting functions from Cumulative Prospect Theory are more linear for groups with incentive to estimate probability than for groups with incentive of a one-shot play.