Some criteria for the estimation of evoked potentials

The estimation of an evoked potential from an ensemble of observations is considered from the statistical point of view. In comparing different estimators it appears that the common mean square loss function is inappropriate in evoked potential studies. A modified loss function is proposed which is insensitive to overall scalling factors in the estimate. Using that loss function the properties of different estimators are explored and lines are indicated along which the estimation can be improved. Simulation examples illustrate the theoretical concepts. It is concluded that it is usually possible to improve on averaging, although comparison of the results of more advanced techniques to those of averaging always remains essential to avoid possibly erroneous interpretations.