중복분석의 점수조정에 따른 행렬도 비교

Abstract

This paper examines redundancy analysis, a highly useful technique for analyzing species-abundance-environment data in ecology, often used in conjunction with canonical correspondence analysis. Redundancy analysis evaluates the influence of environmental variables on species by performing principal component analysis on the results of regression analysis for species-abundance data. However, the scaling of site scores and species scores can lead to different results. Therefore, in this study, we introduce various scaling methods for redundancy analysis and investigate how the bi-plot changes according to the scaling method. Specifically, we introduce several adjustment methods of Oksanen's scaling methods. Furthermore, we demonstrate that the adjustment results of Oksanen (2018) are mathematically equivalent to the projection of the data used in principal component analysis. In summary, site scaling minimizes row scores and maximizes column scores, species scaling increases row scores and decreases column scores, and symmetric scaling represents a compromise between these two approaches. In contrast, principal component scaling is a score adjustment method that maximizes row scores and minimizes column scores.