"then a "mid value" is chosen (not "average", as a wildly erroneous value would affect the data too much"
Well, if the stats have been properly configured the sampled range should be examined, remove the highest and the lowest values to remove extremes, and the total remainder divided by the remaining number of values... sounds a bit geeky, admittedly, but potentially significantly differences can be gained from "mean" values as opposed to more representative properly configured actual "averages".
All quiet on the eastern front.