Basic NeuroscienceControlling for experimental effects in event-related potentials by means of principal component rotation

- A method for the minimization of the misallocation of variance effect is proposed.
- The method allows for an integration of treatment-effects into component rotation.
- The elimination of misallocation of variance is shown in a simulation study.
- An empirical example shows how to improve the parietal topography of the P3.

BackgroundMisallocation of variance means that experimental effects do not only occur on the original event related potential (ERP) component, which should normally represent the respective experimental effect. Simulation studies have shown that misallocation of variance occurs when principal component analysis with subsequent Varimax-rotation is performed. Previous research has also shown that the amount of misallocation of variance substantially depends on component rotation.New methodAccordingly, in this study component rotation has been conducted in a way that allows for a minimization of misallocation of variance. It is proposed to generate dummy variables representing the relevant effects and to generate a loading matrix from the covariances of the dummy variables and of the ERP waves with the components. In the next step, a target matrix for partial Procrustes-rotation is generated by means of specifying the target loadings of the dummy variables on the intended components according to theoretical expectations.ResultsThe elimination of the misallocation of variance effect is shown in a simulation study.Comparison with Existing Method(s):Substantial misallocation of variance occurred for PCA with subsequent Varimax-rotation in the simulation study, but no misallocation of variance occurred with the new method. An empirical example showed how to improve the parietal topography of the late P3 by means of the new method.ConclusionsThe study shows clearly that misallocation of variance is not due to PCA, but to component rotation. The new method of component rotation can be used in order to minimize misallocation of variance.