Estimating vital rates in the developing world: A Bayesian process modeling approach

 描述(由申请人提供)：我的最终目标是建立一个独立的研究议程，为发展中国家和其他数据受限的环境中的人口研究开发新的统计方法。我特别关注对重要指标的评估，这些指标对于了解人口动态、制定公共项目以及实施或评估公共卫生行动尤为关键。在发展中国家的大多数地区，即使是最基本的指标也存在巨大的不确定性。要实现这一目标，需要具备由三部分组成的跨学科技能：(1)统计建模方面的专门知识；(2)对人口统计核心主题的历史、社会/文化和经济基础的理解；(3)对收集发展中国家人口数据的复杂现实的经验。在完成统计学博士工作后，我已经准备好学习这三个组成部分中的第一个。我的培训和发展计划提出了一系列活动，以解决后两个组成部分。我的导师团队由斯图尔特·托尔奈(导师)、萨姆·克拉克(联合导师)、禤浩焯·雷富瑞(顾问委员会)和巴西亚·扎巴(顾问委员会)组成。首先，我将继续接受培训，以理解并最终为人口学和生态学中的实质性问题做出贡献。我将努力了解各种社会、文化和经济因素如何与个人的人口统计结果相关，以及这些结果如何与人口动态相关。虽然社会科学的问题激发了我学习统计学的动机，但我没有接受过人口统计学的正式培训，我唯一的社会科学正规培训是在本科水平。在我目前的培训中，我通过课程作业和高技能、有经验的指导团队指导阅读来弥补这一差距。其次，我的统计培训让我没有准备好应对发展中国家数据收集的复杂现实。我的统计培训强调对已经收集的数据的分析工具，通常是在限制性假设下进行的。然而，用于发展中国家人口研究的数据经常违反这些假设，非抽样误差非常严重。我通过课程学习和实地考察来解决这一差距。我提议在南非东北部的阿金库尔卫生和人口监测系统进行两次实质性的实地工作(每次大约6-8周)。阿金库尔网站在我的发展和培训计划中都占有重要地位，它包括年度人口普查和特别活动更新(系统地记录所有出生、死亡和迁移)，使阿金库尔成为极少数拥有高质量验证数据和基础设施来实施和评估新的数据收集方法的地方之一。在我的访问期间，我将在我的指导团队的监督下，观察访谈，会见主要的调查研究人员，并与Agin Court调查人员讨论我的研究建议的结果和想法。我在阿金库尔的经历是我提案中研究和培训部分之间的切实联系。该研究建议侧重于估计在这种情况下的生育率，并了解国家和区域生育率模式变化的主要驱动因素。生育率是人口规模和构成的重要决定因素。关于生育率的高质量信息对于制定国家和地区政策、制定公共项目以及实施和评估公共卫生行动至关重要。我提出了一种估计发展中国家生育率的技术，强调数据收集、模型和结果之间的关系。总体上的贝叶斯建模框架结合了非抽样误差，从相似的受访者那里汲取力量，并在不同的数据源之间自然地分享不确定性。拟议的方法将通过对通过非抽样误差引入的变异性进行调整来减少偏差，为国家和国家以下各级估计数提供具有统计原则性的不确定量度，并为有效的调查设计提出建议。使用相同的建模框架，我还将评估有关观察到的和预测的生育率趋势的具体假设。目标1开发了一个估算发展中国家国家和国家以下各级生育率的模型，并使用国土安全部和阿金库尔的数据对该模型进行了评估。Aim 2提出了一个微观模拟环境，以便于在个人或家庭层面上测试有关生育模式和动态的假设。这种环境还有助于测试有关测量误差的假设，将再次使用Agin Court和国土安全部的数据对其进行广泛评估。目标3开发了预测未来生育率的模型，该模型将不确定性纳入了潜在的个人水平协变量，这些协变量是 与国家和地区汇率的变化有关。我还将使用阿金库尔过去的数据(人口普查已经进行了大约20年)进行预测，并对阿金库尔未来的生育率做出实际预测，我将在项目结束时对其进行评估。

 DESCRIPTION (provided by applicant): My ultimate goal to establish an independent research agenda that develops novel statistical methods for population research in developing nations and other data-constrained environments. I focus specifically on developing estimates for vital indicators, which are especially critical to understanding population dynamics, developing public programs, and implementing or evaluating public health actions. In most parts of the developing world, there is massive uncertainty about even the most basic indicators. Achieving this objective requires an interdisciplinary skill-set that has three components: (i) expertise in statistical modeling, (ii) an understanding the historical, social/cultural and economc underpinnings of core themes in demography and (iii) experience with the complex realities of collecting demographic data in developing countries. After doctoral work in statistics, I am prepared for the first of these three components. My training and development plan proposes a series of activities to address the second two components. My mentoring team consists of Stewart Tolnay (mentor), Sam Clark (co- mentor), Adrian Raftery (advisory committee) and Basia Zaba (advisory committee). First, I will pursue training to understand, and eventually contribute to, substantive questions in demography and ecology. I will work to understand how various social, cultural, and economic factors relate to individuals' demographic outcomes and how these outcomes relate to population dynamics. Though social science questions motivate my study of statistics, I have no formal training in demography and my only formal training in the social sciences is at an undergraduate level. I address this gap in my current training through coursework and directed readings with a highly skilled and experience mentoring team. Second, my statistical training leaves me unprepared to address the complex realities of data collection in developing nations. My statistical training emphasizes analysis tools for data already collected, often under restrictive assumptions. Data used for demographic research in developing nations, however, often violates these assumptions and nonsampling error is rampant. I address this gap through coursework as well as fieldwork experiences. I propose two substantial (consisting of approximately 6-8 weeks each) fieldwork experiences at the Agincourt Health and Demographic Surveillance System in the northeast of South Africa. The Agincourt site, which features prominently in both my development and training plans, includes annual census and special events updates (systematic recording of all births, deaths and migrations), making Agincourt one of the very few places with both high-quality validation data and infrastructure to implement and evaluate new data collection methodologies. During my visits I will, under the supervision of my mentoring team, observe interviews, meet key survey research personnel, and discuss the findings and ideas of my research proposal with Agincourt investigators. My experiences in Agincourt are a tangible link between the research and training components of my proposal. The research proposal focuses on estimating fertility in such situations and understanding the key drivers of changes in national and regional fertility patterns. Fertility is an important determinant of population size and composition. Quality information about fertility is key for formulating national and regional policy, developing public programs, and implementing and evaluating public health actions. I propose a technique for estimating fertility in developing countries that emphasizes the relationship between data collection, model, and outcome. An overarching Bayesian modeling framework incorporates nonsampling error, draws strength from similar respondents, and naturally shares uncertainty between different data sources. The proposed methods would reduce bias by adjusting for variability introduced through nonsampling errors, provide statistically principled measures of uncertainty for national and subnational estimates and generate recommendations for efficient survey design. Using the same modeling framework, I will also evaluate specific hypotheses about observed and projected trends in fertility. Aim 1 develops a model to estimate national and subnational fertility rates in developing nations and evaluates that model using both DHS and Agincourt data. Aim 2 proposes a microsimulation environment that facilitates testing hypotheses about fertility patterns and dynamics at an individual or household level. This environment also facilitates testing hypotheses about measurement error, which will again be evaluated extensively using Agincourt and DHS data. Aim 3 develops models to project future fertility rates that incorporate uncertainty in the underlying individual-level covariates that are associated with changes in national and regional rates. I will also make projections using both past Agincourt data (a census has been in place for approximately 20 years) and make actual predictions of future fertility rates in Agincourt that I will evaluate at the end of the project priod.