Research On Cognitive Biases In Bayesian Task Reasoning

Bayesian reasoning has been used extensively in medicine, science, engineering and art, economic, politic and military strategy. If people can correctly reason, especially accurately identify the basic cause after a given result, or calculate the correlative risk of a prepared choice, the quality of their subsequent decision can be greatly improved. But the results of former research gives us cause for concern because findings indicate that people often make errors in the process of Bayesian reasoning, and their accuracy rate is very low. Therefore, exploring the participates' cognitive biases in an experimental context in the process of Bayesian reasoning and developmental trend of participates' Bayesian reasoning ability in the context of everyday regular, it is of great theoretical and practical importance that humans approach critical decision making with a healthy and active mind.In all studies on Bayesian reasoning, there are many researches about outcome-evaluate, but few on process-evaluate; research on frequency representation to improve participant's performance is abundant, while research into strategy training is rare. Moreover, most of the researches use classical reasoning tasks, but few on life contexts. Research on disciplinary property is also rare. To sum up, this study analyzed large quantities of related research achievements home and abroad. Along with the main line of discovering problems(oral reports)—analyzing problems(biases analyzing)—solving problems(relation transparent and meta-cognitive) and discussing problems(developmental study in life context), from four aspects: the analysis of cognitive biases in the process of Bayesian task reasoning, the transparence of nested-sets relations, the meta-cognitive, the task context and evidence information, this study systematically investigated Bayesian reasoning through six experiments. Some new phenomena, finding and conclusion have been obtained, but there still exist many problems that require attention.The main results of this research are the following:1. The relatively systemic biases studies on Bayesian reasoning have been completed. Based participates' oral reports and on-line reasoning process (narrating reasons), combined with the deep analysis of Bayesian theorem and the causes of leading to error in reasoning process, cognized and validated the biases' evidence, this research(experiment 1, 2, and 3) systematically put forward 'anchoring reference error' , 'intuitively adjust deviation' bias, 'base-rate neglected' bias, concretely illuminated that the 'conversing presentation' of 'hit rate proposition' and 'anchoring to the base rate' in the process of Bayesian reasoning led to 'anchoring reference error' bias; available heuristic was closely related with 'intuitively adjust deviation' bias in the process of probability's evaluation; 'base-rate neglected' biases existed in the process of solution on regular life program and Bayesian reasoning problem. This research analyzed the causes of people's commonly making mistakes in Bayesian reasoning from biases angle. All these findings offered psychological evidence to identify common mistakes in the process of people's Bayesian reasoning so as to learn the skill of probability's thinking and improve ability of Bayesian reasoning.2. The discussion on content effect of Bayesian reasoning has been completed. Former studies neglected whether or not the content of reasoning problem may affect the participants' evaluation of probability, with the exception of Girotto's (2001) and Xuyuan's (2003) studies. This research (experiment 2) adopted 'the study and explore department problem', 'the office problem' and 'the model workshop problem' as experimental materials, the results indicated that there was remarkable negative correlation between the evaluation of post-probability and the subjective probability of false report rate, this seems to indicate that the probability people experienced in the regular life can effectively illustrate the mind's internal mechanism in reasoning process.3. The strategy straining study on improvement of Bayesian reasoning has been carried out. Based the former studies, the research (experiment 4 and 5) used a classical mammography problem as experimental material, and discussed the facilitating effect of proposition and tree diagram represents on Bayesian reasoning. The results indicated that: (1)Significant effect of one-by-one-question had not been found; (2) Facilitating effect of completed tree diagram and uncompleted one on Bayesian reasoning was found by framing problem's nested-sets relations in terms, but the effect of a simple tree diagram was not significant. (3)Narrating reason inspired meta-cognition monitoring so that participants could effectively sense internal nested-sets relations by mobilizing their knowledge and schema and therefore improve their Bayesian reasoning performances. All these findings offered psychological evidence to deduce students' short-time memory burden so as to improve teaching efficiency and students' performance.4. Influence of disciplinary property of Bayesian reasoning has been carried out. It was commonly assumed that arts and science students differ in their knowledge system, and that the former ones do better in transferring learning techniques that they acquired in their study, such as hypothesis forming, testimony searching, verification of hypothesis into logical reasoning. But the results showed that the bias of participants majored in arts or science had no difference, and the difference remained the same even after strategies training.,5. The developmental study on Bayesian reasoning in regular life context has been completed. This research (experiment 6) selected life contexts and classical tasks to discuss the effect of different task and evidence information on the Bayesian inference of adolescents. The results indicated that: (1) Compared to classical mammography problem, ability of Bayesian reasoning had a apparent improvement from sixth grade in elementary school to second in college, it was reflected by the more accurate results, which were not found in the situation of classical task; (2) Bayesian inference affected not only task context but also different evidence information. When the evidences were consistent with fore-faith, people can take full advantage of clues to reason. This result offered psychological evidence to context teaching in the new course reform.