Meta-Analysis of the Effect of Natural Frequencies on Bayesian Reasoning
Meta-Analysis of the Effect of Natural Frequencies on Bayesian Reasoning
复制标题
DOI：
10.1037/bul0000126
复制
发表时间：
2017-12-01
影响因子：
22.4
通讯作者：
Jacobs，Perke
中科院分区：
文献类型：
--
作者：
McDowell，Michelle;Jacobs，Perke
自然频率促进效应（英语：Natural frequency facilitation effect）是指当人们被表示为通过自然采样获得的联合频率（称为自然频率）时，比被表示为条件概率时，能够更好地解决描述性贝叶斯推理任务。本荟萃分析回顾了20年的研究，旨在解决何时，为什么，以及对谁来说，自然频率格式是最有效的。我们回顾了与两个主要理论观点，生态理性框架和嵌套集理论相关的研究贡献，并测试了潜在的调节效应。对相关文献进行了系统性综述，获得了35篇文章，代表226个性能估计值。使用双变量混合效应模型对这些估计值进行统计学整合，该模型生成2种格式的平均性能的汇总估计值和不同研究特征对性能的影响估计值。这些研究特征的范围从代表个人特征（例如，计算能力，专业知识）的主持人，到方法差异（例如，使用激励措施，评分标准）和问题表征的特征（例如，短菜单格式，视觉辅助）。短菜单格式（计算复杂度较低的表示联合事件）和视觉辅助展示了一些最强的缓和效果，提高了条件概率和自然频率格式的性能。一些方法的因素（例如，暴露于两个问题格式）也被发现影响性能率，强调了系统的方法的重要性。我们建议如何贝叶斯推理的研究可以通过扩大成功的贝叶斯推理的定义，将选择和过程，并通过应用不同的研究方法来加强。（PsycINFO数据库记录（c）2017阿帕，保留所有权利）
The natural frequency facilitation effect describes the finding that people are better able to solve descriptive Bayesian inference tasks when represented as joint frequencies obtained through natural sampling, known as natural frequencies, than as conditional probabilities. The present meta-analysis reviews 20 years of research seeking to address when, why, and for whom natural frequency formats are most effective. We review contributions from research associated with the 2 dominant theoretical perspectives, the ecological rationality framework and nested-sets theory, and test potential moderators of the effect. A systematic review of relevant literature yielded 35 articles representing 226 performance estimates. These estimates were statistically integrated using a bivariate mixed-effects model that yields summary estimates of average performances across the 2 formats and estimates of the effects of different study characteristics on performance. These study characteristics range from moderators representing individual characteristics (eg, numeracy, expertise), to methodological differences (eg, use of incentives, scoring criteria) and features of problem representation (eg, short menu format, visual aid). Short menu formats (less computationally complex representations showing joint-events) and visual aids demonstrated some of the strongest moderation effects, improving performance for both conditional probability and natural frequency formats. A number of methodological factors (eg, exposure to both problem formats) were also found to affect performance rates, emphasizing the importance of a systematic approach. We suggest how research on Bayesian reasoning can be strengthened by broadening the definition of successful Bayesian reasoning to incorporate choice and process and by applying different research methodologies.(PsycINFO Database Record (c) 2017 APA, all rights reserved)