Detail Analysis of designed experiments using generalized linear models Bibliografi Author: Montgomery, Douglas C. (Advisor); Lewis, Sharon Louise Topik: ENGINEERING; INDUSTRIAL|STATISTICS Bahasa: (EN )    ISBN: 0-591-80699-1     Penerbit: Arizona State University     Tahun Terbit: 1998     Jenis: Theses - Dissertation Fulltext: 9828191.pdf (0.0B; 18 download) Abstract In recent years, attention has focused on designed experiments for situations with a variable of interest such as a count of defects, the proportion defective, or the time to failure. Classical linear least squares is the most widely used technique for modeling the relationship between this response variable of interest and one or more independent variables. The optimality properties of least squares, however, depend on the assumptions of normality and constancy of variance. Clearly, when dealing with this type of data, there is not constancy of variance. Traditionally, to correct for non-constant variance, a variance-stabilizing transformation is applied to the response variable to bend the data into shape. This allows application of classical least squares to the transformed data. However, the literature suggests that there are significant problems with data transformations and the resulting inverse transformation to return the variables to their original units. This research finds that the generalized linear model (GLM) provides a powerful alternative to more traditional approaches, such as the data transformation. GLM extends the traditional linear model to allow for responses that are not normally distributed. This research makes major contributions in three areas. First, this research investigates and illustrates the analysis of designed experiments using generalized linear models. The findings of this examination are clearly important. Many statistical tools and techniques from normal linear theory least squares transfer efficiently and appropriately to GLM. Second, this research presents a comprehensive evaluation of the coverage and precision of confidence intervals on the mean response for designed experiments analyzed with GLM. Confidence interval theory in GLM relies on asymptotic properties. This pointedly raises the issue of the behavior of these properties for small samples. This research shows that confidence interval coverage and precision perform closely to theoretical claims even for very small samples. This is a significant finding in that it allows the use of confidence intervals for comparing the predictive performance of models. This research also investigates how this coverage and precision are affected in situations of model or link misspecification and for models with ill-conditioned design matrices. Third, this research demonstrates that GLM will likely produce a better model in terms of predictive performance than that obtained by more traditional methods. A comprehensive study comparing data analyzed with GLM to the same data analyzed with more traditional methods is given. Opini Anda Klik untuk menuliskan opini Anda tentang koleksi ini! Lihat Sejarah Pengadaan  Konversi Metadata   Kembali

Copyright © 2006, 2007 Unika Atma Jaya, all rights reserved