Lapses create troubles. You can take the median instead of the mean RT of each subject

Best to check the reaction time distribution, it's usually log-normal. In that case you can log transform the RTs and then use parametric stats. Another approach is to define a cut-off for lapses and then treat the lapse count as Poisson data (use a generalised linear model or mixed effects models with Poisson outcome data as implemented in the lme4 package for R).

Thanks for your responses, I will treat lapses separately and use these two approaches!

I take the median instead of the mean RT and it's indeed more appropriate for my results.

I checked the mean reaction time distribution and it's log-normal. So, I will log transform RTs. I have to do that for each subject havn't it?

Concerning the other approach, what is the goal? Observe distribution of mean RTs ?

Thanks!

For the log transform you can just take the log (natural or base 10) of all RTs, no difference between individuals. Make sure there are no zeros in the original RT values. Quick and easy to do, and the purpose is to have a dataset on which you can perform parametric stats, as well as to improve homoscedascity if you are planning to do ANOVAs.

Not sure what 'the other approach' refers to, but if you meant counting lapses then it would be to make explicit distinction between a slow response and a lapse. Since lapses are simply slower than whatever cut-off you choose, the actual RT on these becomes irrelevant. Hence you can only count the number of lapses, and that would be best treated as a Poisson distribution (for counts of discrete events in a given unit of time or space).

I hope this makes sense,

Stephan.

Yes it's makes sense. Really thanks you for your help!

Joy