I have 10 sets. Each set corresponds to a user. Each set contains a different amount of floating point numbers. Each such number corresponds to a response time (of a user).
Now I would like to calculate a normalizing constant for the response time which I then will use to normalize the response time for a new user.
First, I will use median instead of average because it is more robust against outlier. Is this correct?
Second, I could either merge the 10 sets into one big set and then taking the median of the whole big set or I can take the median of each set and then calculate the median of the medians. Which one is the correct way to go?