A note on assessing the stability of eigenvectors (principal components) derived form ERP data.

Although principal component analysis (PCA) may be applied to psychophysiological data as a data reduction and modeling technique, prior to interpreting parameter estimates thus obtained, the authors posit that some insight into the stability of the results is needed. A suggested method to acquire such information is based on a perturbation technique suggested by W. J. Krzanowski (1984). This method uses the standard output of statistical programs for carrying out PCA and requires very little computation. No assumptions are necessary about the distribution of the event-related potential (ERP) data of the type of dispersion matrix. An example is given using ERP data