Abstract:
|
Principal component analysis (PCA) is commonly used to investigate the relationships within and among groups of persistent pollutants that have similar fate and transport pathways or toxicity mechanisms. Measured concentrations of these chemicals in environmental samples require some form of transformation before analysis by PCA in order to meet the underlying assumptions of the method. This poster investigates how transformation of the input data can influence the results, interpretations, and conclusions from PCA. Four commonly used data transformations, including log-transform, normal variate (or z-score), normal variate of log-transform, and percent contribution (or row-sum), are compared and contrasted for a simulated dataset. Simulated data allows control of starting conditions, and a priori knowledge of each sample to allow comparisons on a common basis. We address how well each transformation meets the underlying assumptions of PCA, affects on interpretation of results, and the potential for true underlying patterns to be misinterpreted or missed entirely. Recommendations are provided regarding what types of use would be applicable for each transformation type.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.