AN S-PLUS FUNCTIONAL DATA ANALYSIS MODULE

功能数据出现在许多医学研究领域，例如对生长模式、步态、黑色素瘤发病率、CD4计数和许多其他方面的研究。事实上，任何一组随时间（或空间）收集的测量，包括生存分析中的时间相关协变量，都可以被认为是功能数据。将这些数据视为函数而不是不相连的点有许多优点，也许最重要的是能够将派生信息常规地包含到分析中。从历史上看，功能数据一直使用多变量或时间序列方法进行分析，但这些方法对于不规则间隔的数据或不同受试者在不同时间测量的数据来说是薄弱的。最近的进展使分析函数等数据成为可能。在这里，我们建议实现一个S-Plus模块，用于功能数据分析。该模块将是Ramsay和Silverman（1997）开发的探索性方法的商业实现，扩展到自适应回归样条和基于局部多项式核的平滑。我们还计划扩展方法论，包括广义线性模型和生存分析模型。新模块将无缝集成功能数据分析方法到S-Plus。建议的商业应用：随着计算机融入日常生活，研究人员收集功能数据的能力变得越来越普遍。目前还没有可用于处理功能数据的商业产品。与现有技术相比，所提出的方法具有显著的优势。一个设计良好和全面的模块来实现这些模型将会找到一个现成的市场。

Functional data arise in many fields of medical research, with examples from studies of growth patterns, gait, melanoma incidence rates, CD4 counts and many others. Indeed, any set of measurements gathered over time (or space), including time dependent covariates in survival analysis, may be thought of as functional data. There are many advantages to viewing such data as functions rather than disconnected points, perhaps the most important being the ability to routinely include derivative information into the analysis. Historically, functional data has been analyzed using multivariate or time series methods, but these methods are weak for irregularly spaced data or data measured at different times for different subjects. Recent advances make it possible to analyze such data as functions. Here we propose to implement an S-Plus module for functional data analysis. The module will be a commercial implementation of the exploratory methods developed by Ramsay and Silverman (1997), with extensions to adaptive regression splines and local polynomial kernel based smoothers. We also plan to extend the methodology to include generalized linear models and survival analysis models. The new module will seamlessly integrate functional data analysis methods into the S-Plus. PROPOSED COMMERCIAL APPLICATIONS: As computers become integrated into daily life, the ability of researchers to collect functional data is becoming more common. There are currently no commercial products available for handling functional data. The proposed methods have significant advantages over existing techniques. A well designed and comprehensive module for implementing these models will find a ready market.