I have a question regarding a reaction time task. For starters, I’m using a reaction time task with multiple conditions. I will analyse both cue-locked and movement-locked time-frequency activity. Regarding the cue-locked activity, I would like to remove outlier trials from the data. However, in numerous previous studies with similar designs, this step was not mentioned. Furthermore, if I were to remove the outliers, I find it difficult to choose wheter to use mean ± 3 standard deviation or median ± 3 median absolute deviation (reaction time is typically skewed). There does not seem to be consensus on this matter, although it seems to be quite important.
Any input is appreciated
With kind regards
Yeah, that’s a tricky one. Mean and median are basically the same when the data are roughly normally distributed, but the mean is larger for skewed distributions.
I’ve also struggled with this question in the past, and as you note, both are commonly used. I usually use the median. You can also look at the RT distributions and see whether the mean-based threshold is preserving outliers.
Thanks for the answer! And when using median, would you use a cut-off value to delete outlier trials from your time-frequency analyses? In various references, this was not mentioned although I can imagine that outlier trials would negatively impact your time-frequency plots when locking them to cue-onset.
I remove outlier trials during the data preprocessing, so they’re gone from the dataset before even getting to the time-frequency (or any other) analysis.