Diagnostics for structured data and quality improvement

dms -0306304结构化数据和质量改进的诊断道格拉斯M.霍金斯，明尼苏达大学摘要研究者和他的合作者以及论文导师开发了各种问题的诊断方法，这些问题是由他们使用的数据所联系起来的，这些数据的结构超出了单变量随机样本的结构。例子包括多元回归、多元时间序列数据和大型矩形数据阵列中异常值和意外参数变化的识别。将这些方法应用于统计质量改进，解决了当前方法的许多不足，并提供了高性能的图表方法。数据集的规模和复杂性不断增长。蛋白质组学和微阵列实验常规提供了十年前难以想象的数据集。这就需要可靠的统计方法来提取信息，而无需手动检查，并且面对一小部分错误或缺失的读数。这项研究通过制订适当的统计方法和有效的执行方法来解决这些需要。在尺寸范围的另一端，及时生产和定制制造的能力创造了对统计质量控制工具的需求，这些工具能够快速反应，并且相对较小地偏离规格。变更点方法为这种检测和诊断提供了尖锐的工具。

DMS-0306304Diagnostics for Structured Data and Quality ImprovementDouglas M. Hawkins, University of MinnesotaAbstractThe investigator and his collaborators and thesis advisees develop diagnostic methods for a variety of problems linked by their use of data having structure beyond that of a univariate random sample. Examples include the identification of outliers and unexpected parameter changes in multiple regression, multivariate time series data, and large rectangular data arrays. Application of these methodologies to statistical quality improvement resolves many deficiencies of current approaches and provides high performance charting methods.Data sets keep growing in size and complexity. Proteomics and microarray experiments routine provide data sets of a size scarcely imaginable a decade ago. This creates a need for sturdy statistical methods to extract information without manual checking and in the face of a fraction of erroneous or missing readings. The research addresses these needs, by developing appropriate statistical methods and effective ways of implementing them. At the other end of the size spectrum, just-in-time and the ability to customize in manufacturing create a need for statistical quality control tools able to react quickly, and to relatively small departures from specification. Change-point methodologies provide sharp tools for this detection and diagnosis.