Dear all!

For my thesis project I investigated the influence of age, sex (male/female) and education (average/high) on five tasks that tested reaction time and executive functions (N = 43). The reaction time parameters were RT mean (s), intraindividual coefficient of variation (RTsd/RTm) and total time in seconds to finish the task. I wanted to do a repeated measures ANCOVA (SPSS 26).

After checking the assumption it became clear that almost all scales were positively/negatively skewed and outliers were present on all scales. I therefore did a log 10 transformation (LG10(x) for the positive skew and a reflextion LG10(maximum value + 1- x) for the negative skew) on the dependent RT values of the five subtasks. This helped with the skewness. Values that were due to incorrect use of the subtask were removed and replaced with the mean (before doing the log). I didn't take out the other 'outliers', though.

The literature on reaction time gives me somehow mixed messages when it comes to removing outliers. Some argue that one should keep them in a small sample as they can reflect effects of the actual population groups.

But I do wonder whether or not it is sensible to keep them?

Thanks a lot in advance for your responses!

Kind regards,

Nina Wiesner