4
$\begingroup$

I hate talking about "outliers," because I view that term as encompassing two entirely different concepts. The first is when it refers to data that was incorrectly recorded or measured. For instance, a 400000 bedroom house for 4 dollars or a person that is 180 inches tall. There are various techniques to handle these that are outside the scope of my question.

The second type of outlier refers merely to extreme points on the distribution. These might include things like Babe Ruth's home run totals or the net worth of Bill Gates. Although these can have adverse effects on certain types of analyses, they are nevertheless legitimate datapoints. They also have their own techniques that I'd rather not get into here.

Are there any terms that are used to distinguish between the two cases of outliers?

$\endgroup$
  • $\begingroup$ A term commonly used in signal processing and associated machine learning disciplines is 'artifact', i.e. samples of a time-series or signal that arise from an unwanted interference on the intended object of measurement. An error arising from the first scenario (human error) above could also be described as 'artifact' while outlier could refer to valid data that are statistically unusual/unlikely. $\endgroup$ – BGreene Feb 27 '13 at 22:19
  • $\begingroup$ I've seen the second type called influential observations. $\endgroup$ – Dimitriy V. Masterov Feb 27 '13 at 22:23
  • 1
    $\begingroup$ It ain't about the points per se. It is about whether you want to model them or not: observations that you want to model that are far from the centre we call extreme (or tail) observations. Data points that you do not want to model and are far from the centre we call outliers. $\endgroup$ – user603 Feb 27 '13 at 23:06
  • 1
    $\begingroup$ I've seen blunder for points that are impossible or implausible and are considered the result of errors in measurement or recording. I wouldn't agree with @user603 that outliers are points you do not want to model; to the contrary, one often wants a model on which they do not seem outliers. The simplest and most common example is that outliers can look acceptable on log scale. But you could always invert that by saying that if they look acceptable, they are not really outliers. $\endgroup$ – Nick Cox May 23 '13 at 7:53
1
$\begingroup$

I've come across three terms; outlier, leverage point, and influential point.

A reference for you to investigate is Chapter 11 of Applied Regression Analysis, Linear Models, and Related Methods By John Fox. This chapter contains a discussion about Unusual and Influential data. The figure on page 268 is rather instructive.

$\endgroup$

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.