Statistically robust measurement of evoked response onset latencies

Abstract

Onset latencies of evoked responses are useful for determining delays in sensory pathways and for indicating spread of activity between brain areas, therefore inferring causality. Previous studies have applied several different methods and parameters for detecting onsets, mainly utilizing thresholds based on the mean and standard deviation (SD) of the pre-stimulus “baseline” time window, or using t-tests of group data to determine when the response first differs significantly from the baseline. However, these methods are not statistically robust, have low power when the baseline data are not normally distributed, and are heavily influenced by outliers in the baseline. Here, we examine using a modified boxplot method known as the “median rule” for determining onset latencies. This rule makes no assumptions about the baseline distribution, is resistant to outliers, and can be applied to individual level data therefore allowing intersubject and interregional comparisons. We first show with simulations that the median rule is significantly less sensitive to outliers in the baseline than the SD method. We then use simulations to demonstrate the effect of skewness on onset latencies. Finally, we use magnetoencephalography (MEG) to show that the median rule can be easily applied to real data and gives reasonable results. In most situations the different methods give similar results, which enhances comparability across studies, but in data sets with a high noise level there is a clear advantage to using a statistically robust method. In conclusion, the median rule is an excellent method for estimating onset latencies in evoked responses.