A multichannel, model-free method for estimation of event-related potential amplitudes and its comparison with dipole...

We present a multichannel, model-free method for estimation of event-related potential (ERP) amplitude ratios and amplitudes using singular value decomposition (SVD), and compare with the Dipole Components Model (DCM). When the ERPs are generated by a single or multiple dipoles with equal amplitude ratios, the SVD method is superior to DCM in terms of reliable estimation of amplitude and is comparable with DCM for reliable and unbiased estimation of amplitude ratios. We show that dipole model misspecification leads to unbiased amplitude ratios and biased amplitudes when the ERP data sets are (1) generated and fit with a single dipole, or (2) generated by N dipoles with equal amplitude ratios and fit with M < or = N dipoles, because the effect of model misspecification 'cancels' for a ratio. Similarly proof that DCM estimates amplitude ratios more reliably than amplitudes for these cases is given.