So I need a way of ruling out outliers and "the ice skater method" has been suggested. The person who suggested it has a good deal of experience of doing the task I am doing, so I am certainly compelled to use it! However, I can't find anything about anyone else using/testing it...

The figure skating method: in figure skating competitions, there are five judges and after a routine you get 5 scores ("five samples from a 1D distribution"). To calculate your final score they discard the highest and lowest individual score(the outliers) then average the three scores in the middle.

To me it is a vaguely intuitively appealing method of getting an average that is not biased by outliers. To scale it to an average coming from more than 5 values you could either discard the top and bottom 20% or just the top and bottom datapoints. At this point I guess I could describe the data we're measuring and argue that it's a good fit (it feels like it is!), but that's not exactly my question...

...what worries me is that from googling is that I haven't found any mention of it. Is it at all a formalized idea, perhaps by another name? Is it widely used? Have its properties been studied/is there literature on when it might/might not be a good idea?

  • 3
    $\begingroup$ Looks similar to a trimmed mean. I would not call this a "good" option, you are throwing away possibly good information. $\endgroup$ Sep 10, 2019 at 11:15
  • 2
    $\begingroup$ Discarding data requires a good reason. Just because a point is different from the rest does NOT make it invalid. $\endgroup$
    – mkt
    Sep 10, 2019 at 11:55
  • 4
    $\begingroup$ This sounds more like a resistant estimator of a mean than "ruling out outliers." Could you elaborate on how you intend to use or interpret the result? $\endgroup$
    – whuber
    Sep 10, 2019 at 12:52
  • $\begingroup$ Why do you need a method of excluding outliers? Perhaps you really do, but probably not. And this method does not exclude outliers. It is the trimmed mean. That automatically excludes a certain proportion of values, regardless of whether they are outliers or not. In figure skating, they are almost never outliers. $\endgroup$
    – Peter Flom
    Sep 11, 2019 at 11:44

1 Answer 1


"Trimmed mean" appears to be what we're talking about here! https://en.wikipedia.org/wiki/Truncated_mean apparently this is sometimes called the "Olympic average".

Agree that discarding loads of data is not ideal. In this context data is cheap... but I am thinking of using the mode instead of a trimmed mean now. Thank you!

  • 1
    $\begingroup$ The mode discards even more information. $\endgroup$
    – Peter Flom
    Sep 11, 2019 at 11:44

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.