It is shown empirically, analytically and in simulations that common and average referenced recordings differentially affect the accuracy of equivalent source estimates. This effect is mediated by the influence of the reference on noise correlations. The general conclusion of this analysis is that, if software only allows for ordinary least squares estimation (OLS), then average referencing should be preferred, although these estimates will still be sub-optimal. Optimal estimates are derived by generalized least squares (GLS) which accounts for correlated noise. With GLS, average and common referenced recordings give rise to comparable accuracy.
|Number of pages
|Electroencephalography and Clinical Neurophysiology
|Published - Dec 1996
All Science Journal Classification (ASJC) codes
- General Neuroscience
- Clinical Neurology