A RESEARCH ON ACCURACY AND UPDATING CORRECTED SUMS OF PRODUCTS COMPUTATION

博士 === 元智大學 === 工業工程與管理學系 === 104 === The computation of the corrected sum of products (CSP) is most common and important in statistical data analysis. Present algorithms are divided into two categories: One-pass requires passing through the data once, and two-pass requires passing through the data...

Full description

Bibliographic Details
Main Authors: Kuo-Hung Lo, 羅國弘
Other Authors: Tien-Lung Sun
Format: Others
Language:zh-TW
Published: 2016
Online Access:http://ndltd.ncl.edu.tw/handle/51153775134821096420
Description
Summary:博士 === 元智大學 === 工業工程與管理學系 === 104 === The computation of the corrected sum of products (CSP) is most common and important in statistical data analysis. Present algorithms are divided into two categories: One-pass requires passing through the data once, and two-pass requires passing through the data twice. All one-pass algorithms are updating algorithm. Two-pass algorithm is higher accuracy than all one-pass algorithms, but prohibits real-time data compilation and makes it infeasible for handling large datasets. This paper proposed and proved a new CSP updating formula that is based on successive differences. Performance evaluation using StRD shows that our algorithm is the same accurate as the corrected two-pass formula. Moreover, the performance of the new algorithm is better than the variance function implemented in popular data analysis software packages like SAS 9.1, SPSS 22.0 and Excel 2013.