More potential in statistical analyses of event-related potentials: a mixed regression approach

Despite many developments in the methods of event-related potentials (ERPs), little attention has gone out to the statistical handling of ERP data. Trials are often averaged, and univariate or repeated measures of analysis of variance (ANOVA) are used to test hypotheses. The aim of this study was to introduce mixed regression to ERP research and to demonstrate advantages associated with this method. Eighty-five healthy subjects received electrical pain stimuli with simultaneous electroencephalography (EEG) registration. Analyses first showed that results obtained with mixed regression analyses are highly comparable to those using repeated measures of ANOVA. Second, important advantages of the mixed regression technique were demonstrated by allowing the inclusion of persons with missing data, single trial analysis, non-linear time effects, time×person effects (random slope effects) and a within-subject covariate. Among others, the results showed a strong trial (habituation) effect, which contraindicates the common procedure of averaging of trials. Furthermore, the regression coefficients for intensity and trial varied significantly between persons, indicating individual differences in the effect of intensity and trial on the ERP amplitude. In conclusion, using mixed regression analysis as a statistical technique in ERP research will advance the science of unravelling mechanisms underlying ERP data.