SEMIPARAMETRIC REGRESSION AND ENVIRONMENTAL HEALTH

本应用程序的目的是为设计和分析环境健康研究开发半参数回归方法。调查的主要领域是：(1)扩展广义加性建模方法，以协助若干环境健康研究，特别是那些涉及地理和缺失数据的研究；(2)扩展条件面板设计，允许纳入非参数关系；(3)将分布滞后模型与变系数模型相结合，辅助大气污染研究。第一个目标是最复杂的，是由一些涉及地理、集群、缺失和时序数据的研究推动的；但也受益于加法建模。这些研究包括疾病制图、心脏易感性和空气污染、花粉预测和糖尿病研究。提出了可加模型的扩展，以处理地质统计方法（“克里格”）、混合效应和自回归。有些数据集非常大，因此开发计算效率更高的模型是另一个关键组成部分。第二个目标涉及最近开发的一种低成本设计，用于评估空气过敏原和哮喘纵向研究中的时变协变量效应。重要的是能够灵活地控制协变量，如季节性和温度，因此提出了协变量平滑函数的扩展。第三个目标是由几项空气污染研究推动的，在这些研究中，刺激的影响是跨时间滞后分布的，并且受到辅助变量的修改。建议的研究将涉及实际数据的实证数据分析以及理论和方法。该基金的目的之一是为三位主要研究人员作为克雷斯基中心基金的主任和联合主任的工作中产生的想法提供方法研究支持。

The purpose of this application is to develop semiparametric regression methods for the design and analysis of environmental health studies. The primary areas of investigations are (1) extension of generalized additive modeling methodology for aiding several environmental health studies, particularly those involving geographical and missing data; (2) extension of conditional panel designs to allow for incorporation of nonparametric relationships; (3) combination of distributed lag and varying coefficient models to aid air pollution research. The first aim is the most involved and is driven by a number of studies that involve geographical, clustered, missing and time-ordered data; but also benefit from additive modeling. These studies include disease mapping, cardiac vulnerability and air pollution, pollen forecasting and diabetes research. Extensions of the additive model to handle geostatistical methods ('kriging'), mixed effects and autoregression are proposed. Some of the data sets are quite large so the development of more computationally efficient models is another key component. The second aim concerns a low-cost design that has been developed recently to assess time-varying covariate effects in a longitudinal study on aeroallergens and asthma. It is important to be able to control for covariates such as seasonality and temperature in a flexible fashion so the extension to smooth functions of covariates is proposed. The third aim is motivated by several air pollution studies where the effect of a stimulus is distributed across time lags, as well as being subjected to modification by an auxiliary variable. The proposed research will involve empirical data analysis with real data as well as theory and methodology. One purpose of the grant is to provide support for methodological research on ideas that arise out of the work of the three main investigators as director and co-directors of the Kresge Center Grant.