Cognitively Plausible Models of Decision Making

描述（由申请人提供）：我获得了博士学位。2006年从加州大学洛杉矶分校获得社会学硕士学位，加州大学洛杉矶分校2008年的统计数据。作为加州人口研究中心的一名学生，我接受了社会人口学和社会分层方面的全面教育，以及分析大规模调查数据的研究方法和统计技术方面的严格培训。在此期间，我还开发了一种技术，用于将个人对社区偏好的经验估计与基于代理的模型联系起来，以探索个人居住流动决策所隐含的隔离动态。虽然这项工作是在社会研究的前沿，它依赖于一个原始的和高度不切实际的概念，个人的住宅选择行为。这并不是我的研究所独有的。定量社会科学和公共卫生研究中使用的统计模型很少（如果有的话）是引起所调查的社会现象的潜在行为或决策过程的合理模型。在过去的一年里，我了解到商学院市场营销系的研究人员拥有高度复杂的统计模型，可以用来衡量人们如何驾驭环境和做出决策，这些模型借鉴了认知科学和决策理论的见解。但这些方法从未应用于人口健康，例如，人们如何在社区，学校，工作或自助餐厅的主菜中选择。 基于我对这一领域的初步探索，我发现营销选择模型比我过去自学的技术更难掌握。没有标准的统计软件，程序通常是从头开始编写的;没有单一的模型或方法，而是一个松散的技术或战略“工具包”，为具体应用定制。此外，这些模型通常需要贝叶斯估计技术（这需要标准化软件包之外的大量专业知识）。K 01指导研究科学家奖将为我提供受保护的时间：（1）掌握估计这些模型所涉及的统计技能，并正式了解决策的基本理论;（2）使营销统计模型适应新的实质性应用-住宅流动性和配偶选择的研究（如在一个在线约会网站上观察到的）;（3）制定一个方法框架，将这些“在认知上合理的”个人决策模型与基于代理人的模型联系起来，以了解决策战略对总体人口动态的影响;及（4）探讨此统计架构如何应用于与健康研究有关的更广泛决策应用。我已经制定了一个课程计划，以提供更正式的培训，在统计技术和理论框架的基础上，在市场研究中使用的决策统计模型。此外，我打算对心理学中的决策文献进行全面的概述，以补充迄今为止的自学。我将与我的主要导师Fred Feinberg频繁互动，以补充统计建模课程。我的大部分培训将在密歇根大学进行，Feinberg和Diez Roux也住在那里。我将有权使用人口研究中心、社会学系和复杂系统研究中心的数据管理、计算和行政资源。社会学和人口研究都为我提供了办公空间的赠款期间。 我的短期目标是掌握统计技能， 估计这些模型，并获得决策的基本理论的正式理解;并适应市场营销统计模型，以新的实质性应用-住宅流动性和配偶选择的研究（如在线约会网站上观察到的）。从长远来看，我的目标是开发一个方法框架，将这些“认知上合理”的个人决策模型与基于代理的模型联系起来，以了解决策策略对社会整合或分离的总体模式的影响; 探讨如何将这一统计框架应用于与卫生研究有关的更广泛的决策应用。我提出的研究项目将选择建模框架应用于两个研究领域：配偶选择（在在线约会网站上观察到的）和邻里选择。这是一般选择问题中的两个具体例子，在这些选择问题中，人们根据不同程度的习惯或深思熟虑，从中等数量到大量的潜在选项中进行选择。当面对这类选择问题时，决策理论家们一再表明，人们倾向于援引“筛选规则”来简化选择问题。我将使用来自洛杉矶家庭和邻里调查的数据来估计认知上合理的选择模型，以在多个阶段捕获这些筛选规则。我的择偶项目利用了我最近从网上约会中获得的一个丰富的数据集。我将首先估计一个多阶段选择模型，旨在确定所使用的规则 在每一个阶段，并探讨如何寻找配偶和择偶策略不同的人口地区。稍后我将扩展我们的模型，以允许学习和适应性反应。在邻居和配偶选择的情况下，认知上合理的选择模型将与现实的基于代理的模型相结合，以探索个人行为和社会环境的共同进化。

DESCRIPTION (provided by applicant): I earned a Ph.D. in Sociology from UCLA in 2006, and an M.S. in statistics from UCLA in 2008. As a student affiliate of the California Center for Population Research, I received a comprehensive education in social demography and social stratification, and rigorous training in research methodology and statistical techniques for analyzing large-scale survey data. During this time, I also developed a technique for linking empirical estimates of individuals' preferences for neighborhoods with agent-based models to explore the segregation dynamics implied by individuals' residential mobility decisions. While this work is at the vanguard of social research, it relies on a primitive and highly unrealistic conception of individuals' residential choice behavior. This is not unique to my research. The statistical models used in quantitative social science and public health research are rarely (if ever) a plausible model of the underlying behavior or decision-making process that gave rise to the social phenomenon under investigation. Over the past year, I have learned that researchers in departments of Marketing at business schools have highly sophisticated statistical models of how people navigate their environment and make decisions, which draw on insights from cognitive science and decision theory. But these methods have never been applied in population health, for example, to how people choose among neighborhoods, schools, jobs, or entrees in a cafeteria. Based on my initial forays into this area, I have found marketing choice models are orders of magnitude more difficult to master than techniques I have taught myself in the past. There is no standard statistical software, and the programs are usually written from scratch; there is no single model or methodological approach, but rather a loose "toolkit" of techniques or strategies that are customized to a specific application. In addition, the models often require Bayesian estimation techniques (which require significant expertise outside of standardized software packages). The K01 Mentored Research Scientist Award would provide protected time for me to: (1) master the statistical skills involved in estimating these models, an gain a formal understanding of the underlying theories of decision-making; (2) adapt the marketing statistical models to new substantive applications-the study of residential mobility and mate choice (as observed on an online dating site); (3) develop a methodological framework for linking these "cognitively plausible" models of individual decision-making with agent-based models to understand the implications of decision strategies for aggregate population dynamics; and (4) explore how this statistical framework may be applied to a broader range of decision-making applications relevant to health research. I have developed a course plan to provide more formal training in the statistical techniques and theoretical frameworks that underpin the statistical models of decision-making used in marketing research. In addition, I intend to get a comprehensive overview of the decision literature in psychology to supplement what has up to now been self-study. I will supplement the coursework in statistical modeling with frequent interactions with my primary mentor, Fred Feinberg. The bulk of my training will take place at the University of Michigan, where Feinberg and Diez Roux are also in residence. I will have access to the data management, computing, and administrative resources of the Population Studies Center, the Department of Sociology, and the Center for the Study of Complex Systems. Both Sociology and Population Studies have provided me with office space for the duration of the grant period. My short-term goals are to master the statistical skills involved in estimating these models and gain a formal understanding of the underlying theories of decision-making; and also to adapt the marketing statistical models to new substantive applications-the study of residential mobility and mate choice (as observed on an online dating site). In the longer term, my goals are to develop a methodological framework for linking these "cognitively plausible" models of individual decision-making with agent-based models to understand the implications of decision strategies for aggregate patterns of social integration or separation; and to explore how this statistical framework may be applied to a broader range of decision-making applications relevant to health research. My proposed research projects apply the choice-modeling framework in two areas of research: mate choice (as observed on an online dating website) and neighborhood choice. These are two specific instances of a general class of choice problems where people choose-with varying degrees of habit or deliberation- from a moderate to large number of potential options. When confronted with this sort of choice problem, decision theorists have repeatedly shown that people tend to invoke "screening rules" to simplify the choice problem. I will use data from the Los Angeles Family and Neighborhood Survey to estimate cognitively plausible choice models to capture these screening rules at multiple stages. My mate choice project makes use of a rich dataset I recently acquired from an online dating. I will first estimate a multi-stage choice model aimed at identifying the rules used at each stage, and explore how strategies for mate search and mate choice differ across demographic areas. I will later extend our models to allow for learning and adaptive responses. In both the neighborhood and mate choice case, the cognitively plausible choice models will be coupled with realistic agent-based models to explore the co-evolution of individual behavior and the social environment.