Premium
DATA TRANSFORMATION AND STANDARDIZATION IN THE MULTIVARIATE ANALYSIS OF RIVER WATER QUALITY
Author(s) -
Cao Yong,
Williams D. Dudley,
Williams Nancy E.
Publication year - 1999
Publication title -
ecological applications
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.864
H-Index - 213
eISSN - 1939-5582
pISSN - 1051-0761
DOI - 10.1890/1051-0761(1999)009[0669:dtasit]2.0.co;2
Subject(s) - standardization , multivariate statistics , water quality , principal component analysis , variance (accounting) , multivariate analysis , relevance (law) , quality (philosophy) , transformation (genetics) , ecology , computer science , environmental science , data mining , biology , machine learning , artificial intelligence , business , philosophy , biochemistry , accounting , epistemology , political science , gene , law , operating system
Multivariate approaches are being increasingly applied to aquatic studies; however, their ecological relevance has not been adequately addressed. This paper examines the effects of data transformation and standardization on principal‐components analysis (PCA) of river water quality. Two currently popular methods, log( x + 1) and standardizing to 0 mean and 1 variance, were evaluated. The choice of the data‐handling methods substantially affected the analytic output. However, both methods tested were found to be flawed, with the major problem being disregard for the biological/ecotoxicological significance of water‐quality variables and their ranges. We developed a new standardization model, which incorporates water quality standards into data standardization. This provides a mechanism to improve the biological relevance of multivariate analysis. The new model‐based PCA yields an assessment of water quality that is more ecologically meaningful than any others tested. We discuss the ecological implications of the new model and of the established methods tested.