4
$\begingroup$

Say I have a simple data set that describes height of males in a classroom.

I decide that whatever value falls 3.29 standard deviations away from the mean is actually an outlier. When removing these identified outlier values and reanalyzing my data set, should I recalculate the max 3.29 standard deviation mark based on the new dataset and keep removing outliers again and again until I have a data set where nothing falls over nor under 3.29 stdev away from the mean? Or is what I just said just silly?

$\endgroup$
  • 2
    $\begingroup$ An outlier would be someone who you don't believe is really part of the dataset you wish to analyze. Do you think that 3.29 sd from the mean makes them no longer male? You need another justification to remove a data point. $\endgroup$ – John Feb 13 '15 at 14:13
  • $\begingroup$ Many statistically-minded people would never remove outliers just because they are far out in the tails. The example of human height is one of many where you have ways of finding out whether values are credible and using judgment rather than some numerical rule. I'd read around widely on outlier threads here. There are many procedures easier to defend than the one you are contemplating. $\endgroup$ – Nick Cox Feb 13 '15 at 15:17
  • $\begingroup$ stats.stackexchange.com/questions/tagged/… $\endgroup$ – Nick Cox Feb 13 '15 at 15:18
  • $\begingroup$ OK I must state that the example I gave was a simple one just to make a point but I am actually trying to model the expected performance of machine after 1 year. But if you had a dataset for any type of analysis whether its machine performance, male height, player performance ..What is the proper way of getting rid of outlier values? Give me a way $\endgroup$ – alaboudi Feb 13 '15 at 15:31
  • $\begingroup$ Please explain why you think you need to "get rid" of such values. If you trust your data, then they reflect what really happened with the machine. Removing them could unrealistically alter your conclusions. $\endgroup$ – whuber Feb 13 '15 at 16:23

Your Answer

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

Browse other questions tagged or ask your own question.