Stratified Data in Factor Analysis

• Main Authors: Hui-Chu Lin, 林惠珠
• Other Authors: Chong-Yau Fu
• Format: Others
• Language: zh-TW
• Published: 2004
• Online Access: http://ndltd.ncl.edu.tw/handle/39878436022507385287

碩士 === 國立陽明大學 === 公共衛生研究所 === 92 === Abstract Factor Analysis is one method of exploring the data structure. The main point is to classify correlated variables to reduce the dimension of data. Based on covariance matrix, factor analysis extracts common factors and remains most of variance through matrix algebra. In practice, stratified data can reduce the additional variance due to the extra variable. In modeling, Conditional logistic regression model, and Stratified PH model are available for stratified data. Factor analysis for stratified data is somehow complicated and not available in software. In this thesis, two methods, pooled-amount-variation and group-mean-corrected methods, are proposed for data correction before factor analysis used. In this fetal data, Pulsality Index (PI) values are measured from several parts of vessels. The effect caused from gestational age effect is reduced by corrected methods. Two factors remaining most of variation are generated from factor analysis. Combining the binary fetal outcome (normal or abnormal), logistic regression analysis and CART analysis are used for binary discrimination. The discrimination result has lower misclassification rate from corrected data when data highly correlate with gestational age. Keywords：Stratified Data、Factor Analysis、 Fetal Data、Pulsatility Index﹒