Comprehensive Software for Flexible Statistical Models

描述（由申请人提供）：本研究旨在进一步开发和 实现一类通用多元非参数统计模型 称为平滑样条方差分析（SSANOVA）模型。的 研究包括三个密切相关的方面：算法，方法， 和软件实现。对于算法，重点是开发 一种使用计算成本较低的近似值来精确计算 平滑样条：可以将操作从三次减少到稍微 同时保持相同的渐近收敛速度。 对于方法，最近提出的一些实证研究（有条件） 密度和危害模型将在可扩展的 算法和建模工具，如诊断， 将制定示范条款的意义。在将要研究的模型中， 是具有任意多个响应的非参数回归模型， 风险模型比比例风险模型更一般。还要 探索的是一种随机效应的方法， 具有相关误差的回归（高斯和非高斯）处理 以及具有脆弱性的危险模型。对于软件实现，友好的用户 接口将在R中实现，语法主要继承自 标准参数化建模工具。这项研究的最终成果将是 一个全面的R软件包，可免费下载，带有在线文档， 一个单独的手册充满了例子。SSANOVA模型的早期发展 在大型流行病学研究中发现了应用， 节律数据和激素数据的建模。用户友好，可扩展 软件将促进采用和日常使用这些多功能的 生物医学研究人员的建模工具， 地区

DESCRIPTION (provided by applicant): This research seeks to further develop and implement a class of versatile multivariate nonpararnetric statistical models known as the smoothing spline analysis of variance (SSANOVA) models. The research consists of three closely related aspects: algorithms, methodologies, and software implementation. For algorithms, the focus is on the development of a strategy that uses computationally less expensive approximations to the exact smoothing splines: one may reduce the operation from a cubic order to slightly over a linear order while maintaining the same asymptotic convergence rates. For methodologies, empirical studies of some recently proposed (conditional) density and hazard models will be conducted with the assistance of the scalable algorithms, and modeling tools such as diagnostics for the practical significance of model terms will be developed. Among the models to be studied are models for nonparametric regression with arbitrary multiple responses and hazard models more general than the proportional hazard models. Also to be explored is an approach to random effects that allows straightforward numerical treatments for regression (Gaussian and non Gaussian) with correlated errors and for hazard models with frailty. For software implementation, friendly user interface will be implemented in R with the syntax largely inherited from the standard parametric modeling tools. The final product of this research will be a comprehensive R package freely downloadable, with on-line documentations and a separate manual filled with examples. Earlier developments of SSANOVA models in regression have found applications in large epidemiological studies and the modeling of rhythms data and hormones data. The user friendly, scalable software will facilitate the adoption and routine use of these versatile modeling tools by biomedical researchers from a broader spectrum of application areas.