A Comparison of Different Methods for Estimating Single-trial P300 Latencies

Inferences from comparative analyses of reaction time and P300 latency are stronger when the various aspects of the distribution across trials are treated in the same way for both variables. To this end, a number of studies have resorted to estimation of P300 latency at the single-trial level. This report presents a comparative evaluation of two common methods for such single-trial analysis, i.e., peak-picking and template-matching. Both methods were applied to a representative set of real data, comprising different task conditions and two age groups. Relevant scoring parameters were varied: low-pass filter settings (down to 0.94 Hz) for peak-picking, template duration (250-970 msec) and use of covariance vs. correlation for template-matching, and use of a noise-range criterion for both methods. It is concluded that peak-picking with a 3.4 Hz filter, and template-matching using covariance and template duration between 600 and 800 msec, are best in terms of sensitivity and reliability, with peak-picking surpassing template-matching. Also, the marked increase in the number of rejected trials when the noise-range criterion was applied resulted in unwanted modulation of behavioral effects of task conditions and age groups.