Talk:Outlier

This article is rated B-class on Wikipedia's content assessment scale. It is of interest to the following WikiProjects:

Statistics Mid‑importance This article is within the scope of WikiProject Statistics, a collaborative effort to improve the coverage of statistics on Wikipedia. If you would like to participate, please visit the project page, where you can join the discussion and see a list of open tasks.StatisticsWikipedia:WikiProject StatisticsTemplate:WikiProject StatisticsStatistics Mid This article has been rated as Mid-importance on the importance scale.

What's a Renee Masse? (In the first sentence).

---

—Preceding unsigned comment added by 212.183.70.147 (talk) 15:04, 15 October 2007 (UTC)Reply

Adopted orphan redirects for Google: inner fence, outer fence, mild outlier, Extreme outlier

---

In sans-serif font, 1.5 IQR looks like a division. Patrick 11:15 Dec 23, 2002 (UTC)

---

I did it that way so I wouldn't have to use * or x or × for multiplication. What would you prefer? dcljr 13:53 Dec 23, 2002 (UTC)

---

When using multi-letter variables a multiplication sign avoids ambiguity, and in this case coincidentally there was even a little more ambiguity.

Any of the three is fine with me, × is neatest, but more cumbersome to write. - Patrick 14:26 Dec 23, 2002 (UTC)

more info but clearer answers i don't get a thing and my exam is tommorrow!!!!

Is there not a second defintion of outliers, as lying more than two standard deviations away from the mean? Or am I mixing other things up? I am a physicist, and it is a long time since I did "real" statistics... Batmanand | Talk 09:48, 28 September 2006 (UTC)Reply

This is a poor definition of outliers as it changes upon recursion, i.e. the standard deviation is highly dependent on the outliers. Check boxplot for a simple but easy to understand definition that is not distribution dependent.

Isn't an outlier (German: Ausleger, Swedish: utliggare) also a supporting 2nd keel for a canoe or sailing boat that makes it almost a catamaran? Hmm... apparently this is called a outrigger on an outrigger canoe in English. Other languages would use "rig" for things that have sails. --LA2 23:04, 1 August 2007 (UTC)Reply

-No, Ausleger is not outlier, that's a false friend. Outrigger, as you say, is the right term. —Preceding unsigned comment added by 212.183.70.147 (talk) 15:07, 15 October 2007 (UTC)Reply

I'm opposed to the "Mathematical definition" section in the article. Determining whether an observation is an outlier is ultimately a subjective decision, and any definition based on measures such as standard deviation or interquartile range is completely arbitrary.

If this section need be kept, perhaps it could be renamed? The term "mathematical" implies a logical certainty, which doesn't apply in this case. -3mta3 (talk) 11:49, 14 April 2008 (UTC)Reply

I want to know what is the source of the method of using the Interquantile range mentioned in the text?? Is it just a rule of thumb, or does it have a more objective explanation?--Forich (talk) 21:33, 9 June 2008 (UTC)Reply

It is a popular method, known as "Interquartile range" (IQR), where k = 3 is suposed to identify extreme outliers, and k = 1.5 mild outliers. This method has no scientific basis; it belongs to the category of Mumbo Jumbo methods in statistics.  --Lambiam 04:51, 10 June 2008 (UTC)Reply

A more thorough discussion of methods for identifying outliers should be added, for example Rosner's test. See e.g. Barnett, V., and T. Lewis. (1995): Outliers in Statistical Data. Agnerf (talk) 09:09, 24 February 2022 (UTC)Reply

The second citation

"2. ^ Grubbs, F. E.: 1969, Procedures for detecting outlying observations in samples. Technometrics 11, 1–21."

I googled it and found it on jstor at http://www.jstor.org/pss/1266761

Would it be a good idea to link directly to the article in the reference section? I didn't know how and I didn't know if that was appropriate or not. It seems appropriate though.

Your thoughts?

--Ted Wheeland (talk) 21:38, 3 January 2010 (UTC)Reply

How is it spoken ?

Like "Out Lier" or "ootlee-er".109.150.237.200 (talk) 09:46, 6 December 2012 (UTC)Reply

Former. Hear it at http://www.merriam-webster.com/dictionary/outlier Glrx (talk) 22:21, 7 December 2012 (UTC)Reply

Much of this section is directly plagiarized from A Survey of Outlier Detection Methodologies (2004) by Hodge & Austin. For example,

"Type 1 - Determine the outliers with no prior knowledge of the data. This is essentially a learning approach analogous to unsupervised clustering. The approach processes the data as a static distribution, pinpoints the most remote points, and flags them as potential outliers." is word-for-word identical, as are the definitions of the following 2 types. At the absolute minimum their work should be cited. — Preceding unsigned comment added by 208.105.82.93 (talk) 21:48, 7 March 2013 (UTC)Reply

I think the phrase "Deletion of outlier data is a controversial practice frowned on by many scientists and science instructors" needs some kind of citation or should be reformulated / removed. I mean - why is it controversial? What can happen? Examples of bad things that happend because of outliers exclusion? — Preceding unsigned comment added by 89.120.104.106 (talk) 10:26, 18 July 2013 (UTC)Reply

The whole article refers to outliers as possible errors (eg "An outlier may be due to variability in the measurement or it may indicate experimental error"). This is not right or actually incomplete. An outlier can also point at something real going on which is unusual. As a colleague pointed out, it could be your next Nobel prize. If I remember well, NASA had detected the ozone hole in data before Joe Farman published it, but NASA had excluded those. So, I would propose to change the tone of this article and emphasize that statistics helps to identify outliers, which are especially interesting points pointing at measurement errors, unusual distributions (the tails) and possibly new phenomena. — Preceding unsigned comment added by Pjtverheijen (talk • contribs) 05:58, 12 March 2014 (UTC)Reply

Modified Thompson Tau test is *exactly* the Grubbs' test for outliers, which makes things even more confusing. If it's just a different name convention, I think it should just link to Grubbs' test for outliers page. — Preceding unsigned comment added by Yannick Copin (talk • contribs) 16:24, 7 October 2016 (UTC)Reply

Grubb is already linked. Maybe simply remove the whole section on that modified Thomson Tau, if the only "citeable" version is Grubbs. HelpUsStopSpam (talk) 22:35, 7 October 2016 (UTC)Reply

This article seems to not include any mention of the corrections for parameter estimates when part of the sample is cut. Remember, some "good" points may be cut together with the outliers. For example, conditional probability can be used to correct the parameter estimates. — Preceding unsigned comment added by 84.245.245.196 (talk) 14:51, 23 March 2017 (UTC)Reply

Why isn't the control chart mentioned as method for outlier detection?

Seems like a blatant omission.

2601:14F:8002:CAD2:88A7:1D26:370A:AFD6 (talk) 11:40, 28 May 2018 (UTC)Reply

It better suits change detection (aka change point detection). Plus, the underlying model largely is that of a normal distribution, which is mentioned here. You can't put everything everywhere. Chire (talk) 21:18, 28 May 2018 (UTC)Reply

consenso: I am proposing the inclusion of a more modern and rigorous definition of outlier that is grounded in the context of applied statistical modeling. The definition comes from a peer-reviewed article published in Survey Review, a journal focused on measurement science. The proposed definition is: "An observation that has moved away from the most likely value to the point of not belonging to the mathematical model (functional and stochastic) stipulated." Citation:

Rofatto, V. F., Matsuoka, M. T., Klein, I., Bonimani, M. L. S., Rodrigues, B. P., de Campos, C. C., Veronez, M. R., & da Silveira, L. G. Jr. (2021). An artificial neural network-based critical values for multiple hypothesis testing: data-snooping case. Survey Review. https://doi.org/10.1080/00396265.2021.1968176

This definition complements the existing content by reflecting how outliers are treated in fields such as geodesy, engineering, and measurement science, where functional and stochastic models are explicitly defined. It is not a repetition of existing definitions but rather adds an important perspective grounded in real-world modeling applications. Additionally, it is relevant to note that the first three authors (Rofatto, Matsuoka, and Klein) are established researchers who have published multiple scientific papers specifically on outlier detection. I would appreciate community input regarding the inclusion of this definition in the article. Thank you. Tomiomatsuoka (talk) 02:03, 23 July 2025 (UTC)Reply

Wikipedia is not a place to self promote or to try to force in citations to yourself, kindly stop treating it as one. This is not belong in the article, let alone in the lead paragraph - it is redundant with what's already present, and also more confusing and less well written. It is not an improvement to the article. MrOllie (talk) 02:41, 23 July 2025 (UTC)Reply

With all due respect, we believe there has been a misunderstanding. Our intention is not self-promotion or to force citations of our own work, but rather to contribute to a topic we have studied extensively over the years, based on serious, peer-reviewed academic research.

The definition we suggested is grounded in published scientific work and aims to address a conceptual gap that still exists in many articles on the subject. We are not outsiders trying to alter Wikipedia irresponsibly. On the contrary — our goal is to collaborate, bringing a well-founded technical perspective supported by publications.

We fully acknowledge the importance of Wikipedia’s guidelines on neutrality and conflict of interest, and we are willing to suggest the inclusion through the article’s talk page, if that is the more appropriate route.

If there are specific parts that could be better written or more clearly contextualized, we are open to constructive suggestions. We believe we all share the same goal: building a free, reliable, and well-structured encyclopedia. Vfrofatto (talk) 03:04, 23 July 2025 (UTC)Reply

You've been repeatedly adding mention of yourself to Wikipedia, over obvious objections. You absolutely are self promoting, and you absolutely are not 'collaborating', you are coming into a community and violating its policies and guidelines. If you really want to respect the Wikipedia community's practices on this, my 'constructive suggestion' is that you remove what you've added from the article. MrOllie (talk) 03:14, 23 July 2025 (UTC)Reply

The citation in question was not added arbitrarily, but in good faith, to address a specific conceptual gap with a published, verifiable source. If there's a better path forward, we're happy to take it. Vfrofatto (talk) 03:30, 23 July 2025 (UTC)Reply

I've just told you what that better path forward is: show some good faith and respect for Wikipedia's guidelines by removing it yourself. MrOllie (talk) 03:32, 23 July 2025 (UTC)Reply

I honestly didn’t know I was violating any guidelines — I’m new here and still learning how things work on Wikipedia. If that’s the case, then I sincerely appreciate it being pointed out.

That said, I believe a more constructive path would have been to explain the appropriate process in a more didactic way. I’m here to collaborate in good faith, not to create conflict or push personal content.

As you can see, I’m now requesting the change properly through the Talk page, and I’m open to feedback regarding relevance, neutrality, or placement. My goal is simply to improve the article by contributing reliable, peer-reviewed sources on a topic I’ve studied extensively.

Thank you for your understanding. Vfrofatto (talk) 03:49, 23 July 2025 (UTC)Reply

@MrOllieMrOllie,

Thank you for your attention and vigilance regarding the Outlier article.

After approximately 15 years researching and publishing on this topic in scientific journals, we had the idea to contribute to Wikipedia within our field of expertise. We initially attempted to do so by including a definition of “outlier” that we carefully developed over time and that has already been peer-reviewed and published.

We were also planning to contribute further by describing different aspects of the topic, such as common outlier detection methods.

However, for now, we have decided to step back from contributing and will instead continue focusing on publishing in scientific journals.

If you're interested, we encourage you to search academic journal databases for additional articles we’ve authored, which we believe demonstrate our good intentions. As a suggestion, perhaps consider adding a broader and more modern definition to complement the one currently cited in the first paragraph of the Outlier article.

With respect, and thank you again for your attention. Tomiomatsuoka (talk) 04:25, 23 July 2025 (UTC)Reply

Hello, I’d like to request an edit to the article Outlier.

I am one of the authors of a peer-reviewed paper that proposes a definition of "mathematical outlier" specifically in the context of stochastic and functional modeling. While I understand that citing one's own work directly is discouraged, I believe the proposed definition is both relevant and clearly sourced.

Proposed addition (under the introduction section):

"An outlier can also be defined as an observation that has moved away from the most likely value to the point of not belonging to the mathematical model (functional and stochastic) stipulated."<ref>Rofatto, V. et al. (2021). "An artificial neural network-based critical values for multiple hypothesis testing: data-snooping case". Available at: https://doi.org/10.1080/00396265.2021.1968176

This definition has been cited in the context of geodetic and statistical quality control and offers a modern interpretation aligned with model-based diagnostics. If the placement or wording needs to be adjusted for neutrality or clarity, I’m open to suggestions.

Thank you for considering. — Preceding unsigned comment added by Vfrofatto (talk • contribs)

I don't think this is a particularly good definition, because observations do not "move", and a sample does not have a "most likely value" (but distributions have a mode). The "deviation from the expected value" is just one possible approach. Overall, this is IMHO a poor rephrase of Hawkins 1980 "an observation which deviates so much from other observations as to arouse suspicions that it was generated by a different mechanism", hence redundant. Also, many common techniques do not stipulate a particular model, but are nonparametric in order to not make too strong assumptions. I don't think this can encompass common popular techniques sich as LOF or Isolation Forests well. 2A02:3100:4713:AA00:79EA:B2E1:E861:AC91 (talk) 21:07, 23 July 2025 (UTC)Reply

We greatly appreciate the careful and detailed evaluation of our definition of outlier.

It is very interesting to see the perspective of experts from another field.

Our definition works very well in the context of our field — Geodesy — specifically within the framework of least squares adjustment, assuming normally distributed measurements and a well-defined functional and stochastic model.

In retrospect, we should have placed our definition elsewhere in the text and emphasized that it refers to the concept of an outlier within the context of Geodesy.

Taking this opportunity, we should also consider applying the same level of scrutiny to the definition of outlier currently presented on Wikipedia, which states: “an outlier is a data point that differs significantly from other observations.”

By the same rigorous standard, this is arguably not a good definition either. The claim that a data point “differs significantly from other observations” is vague and often misleading.

In fact, in many cases within our field of Geodesy, this would not hold at all.

For example, in GPS navigation, the estimation of a receiver’s 3D coordinates is based on observations of distances to at least four satellites. These measurements naturally differ significantly from each other — but they are not outliers. They are all valid inputs to a mathematical model that links them to the unknown coordinates.

In that context, “being different from other observations” tells us nothing about whether the data point is valid or not.

This kind of definition seems overly simplistic and only applicable in trivial 1D situations — where the parameter being estimated is essentially the observed quantity itself, without reliance on a functional model.

In more complex, multidimensional systems like ours, such a definition can be misleading or even counterproductive.

What do you think? Should we keep the current definition on the Wikipedia page as it is, or consider revising it to reflect broader and more robust applications? Tomiomatsuoka (talk) 20:05, 24 July 2025 (UTC)Reply

The first sentence of the second paragraph; "Outliers can occur by chance in any distribution," is hard to understand, and false in many cases (e.g. the degenerate distribution can only take on one value, and so by no definition can a sample thereof contain an outlier). I imagine there exists some concise statistical notion for a distribution which can take on any real number, however I am also sure that that is not the widest qualifier to make this sentence true (e.g. a bounded uniform distribution, which may have a sampled outlier given wide enough bounds). I think the notion of this sentence is correct and useful, which is why I have not removed it, but it could be more clear and more correct while not taking a specific definition of an outlier.

Some-sort-of (talk) 12:24, 18 May 2026 (UTC)Reply