SEMI-VARYING COEFFICIENT MODELS FOR INTENSIVE DATA

密集数据的半变系数模型

• 批准号: 7311360
• 负责人: RUNZE LI
• 金额: $ 17.41万
• 依托单位: PENNSYLVANIA STATE UNIVERSITY, THE
• 依托单位国家: 美国
• 资助国家: 美国
• 起止时间: 至
• 项目状态: 未结题
• 来源: https://reporter.nih.gov/project-details/7311360
• 关键词: computer simulation; computer system design /evaluation; drug abuse prevention; drug abuse therapy; health science research; human data; longitudinal human study; mathematical model; method development; model design /development; statistics /biometry

With the advent of modern data collection devices, intensive longitudinal data are being collected more and more in drug use studies. Intensive longitudinal data have many closely spaced measured occasions, and usually, many variables. In theory, they can provide answers to important questions in drug abuse research. However, in practice, it is not immediately clear what statsitical procedures can be applied to such kind of data to address questions, such as: How does the subjective sensation of withdrawal vary over a day or a week? What is the relationship between mood and drug use? In this project, we propose new semi-varying coefficent mixed models for intensive longitudinal data and time-to-event data. These models possess many valuable features which make them the most appropriate to use for addressing important drug abuse questions and testing key hypotheses in drug use research using intensive longitudinal data. The proposed new models allow effects varying over time and changing across individual subjects, and keep the structure of random error process very flexible. We will propose effective estimation procedures and variable selection procedures for the proposed new models. We will develop software to implement the proposed procedure and assess the performance of the proposed procedures by extensive Monte Carlo simulations. We will also apply the new procedures we have developed to address important drug use questions using empirical data on tobacco, alcohol and marijuana. These are the most widely used substances within the US and have been linked to a myriad of both short and long-term consequences. Findings in this project will help public health researchers to undertand the etiology of drug use and dependence. This understanding can be used to inform drug abuse and treatment efforts.

随着现代数据收集设备的出现，在药物使用研究中越来越多地收集密集的纵向数据。密集的纵向数据有许多紧密间隔的测量场合，并且通常有许多变量。理论上，它们可以为药物滥用研究中的重要问题提供答案。然而，在实践中，目前尚不清楚可以对此类数据应用哪些统计程序来解决问题，例如：戒断的主观感觉在一天或一周内有何变化？情绪与药物使用有何关系？在这个项目中，我们提出了新的半变化 用于密集纵向数据和事件时间数据的系数混合模型。这些模型具有许多有价值的特征，这使得它们最适合用于解决重要的药物滥用问题和使用密集的纵向数据测试药物使用研究中的关键假设。所提出的新模型允许效果随时间变化以及个体受试者的变化，并保持随机误差过程的结构非常灵活。我们将为所提出的新模型提出有效的估计程序和变量选择程序。我们将开发软件来实施建议的程序 并通过广泛的蒙特卡罗模拟评估所提出程序的性能。我们还将应用我们开发的新程序，利用烟草、酒精和大麻的经验数据来解决重要的药物使用问题。这些是美国境内使用最广泛的物质，并与许多短期和长期后果有关。该项目的研究结果将帮助公共卫生研究人员了解药物使用和依赖性的病因。这种理解可用于指导药物滥用和治疗工作。