Jump to content

Talk:Accuracy and precision

Page contents not supported in other languages.
From Wikipedia, the free encyclopedia

Clarification needed

[edit]

In science there is a clear distinction between accuracy and precision.

  • Accuracy is a measure of the magnitude of systematic error in the value of a measurement.
  • Precision is a measure of the magnitude of random error in the value of a measurement.

In the real world, measurements are affected by both types of error. Measuring instruments are calibrated for accuracy and graduated for precision. Both accuracy and precision for a quantity derived from measurements can be obtained by using error propagation methods. Petergans (talk) 14:12, 4 August 2019 (UTC)[reply]


What do you mean "in science"? I do science and most often we just talk about accuracy. Clearly, we are interested in both systematic and random error. Exactly because in the "real world" the errors are both systematic and random...and science describes the real world — Preceding unsigned comment added by 82.130.95.205 (talk) 08:40, 18 November 2019 (UTC)[reply]

Urgent to edit: there are NOT two different definitions of accuracy.

[edit]

Wikipedia is great because it spreads knowledge. With this article it is just spreading the confusion! I will edit it as soon as I can, but first I'd like to keep up the discussion. Excerpt from the Wikipedia Accuracy and precision "Common technical definition": "In the fields of science and engineering, the accuracy of a measurement system is the degree of closeness of measurements of a quantity to that quantity's true value.[1] The precision of a measurement system, related to reproducibility and repeatability, is the degree to which repeated measurements under unchanged conditions show the same results.[1][2] Although the two words precision and accuracy can be synonymous in colloquial use, they are deliberately contrasted in the context of the scientific method. "

There is no contrast. When you say "the accuracy is the degree of closeness to the true value" this includes BOTH the average value (describing the trueness) AND the random errors (associated with the precision). Let's say my repeated measurements have an average value that is close to the true value. We can use the example of the darts. Let's say I shoot 100 darts and they spread homogeneously on the entire target circle. Let's say the average value is zero. Can I say that the shoots are accurate? NO! Because none of the darts are "close to the true value" (which is the center in this simplified example). The definition of precision is correct, though. It only refers to how close all my shoots are to each other, not to the true value. So If my 100 shoots all end up in the exact same point I have a high precision. But if this point is not the center, my accuracy is still low. This is exactly according to the definition of BOTH the ISO 5725-1 and the JCGM2008. An article worth reading: https://www.researchgate.net/publication/226202473_Understanding_the_meaning_of_accuracy_trueness_and_precision (disclaimer: I'm not on of the authors nor I'm anyhow affiliated with the authors)

It is sometimes easy to confuse the two terms precision and accuracy: if the bias (linked to the trueness) of the measurement is zero the only contribution to accuracy is given by the precision. In other terms accuracy = bias + precision. If bias is zero then accuracy = precision. 82.130.95.205 (talk) 10:26, 18 November 2019 (UTC)[reply]

I've included a citation to the source suggested, thanks. fgnievinski (talk) 05:46, 21 September 2024 (UTC)[reply]

Clarification for the content of the first line

[edit]

I think the first words of the article should be ″In a set of measurements″ instead of ″In measurement of a set″. I must be mistaken. I have not studied advanced topics. — Preceding unsigned comment added by 2402:4000:2280:B5C1:A8BD:B307:5034:8F4B (talk) 09:10, 15 November 2020 (UTC)[reply]

Too technical

[edit]

I added Template:Technical to the article. An example of overly technical language occurs in the first sentence of the article: "In measurement of a set ...." I would guesstimate that over 95% of readers do not know what that means. See Wikipedia:Make technical articles understandable for some helpful guidelines and suggestions. I'll try to work on making the article less technical too. Mark D Worthen PsyD (talk) [he/his/him] 22:55, 8 February 2021 (UTC)[reply]

Missing citations

[edit]

This page is prescriptive and lacks citations. For example, the section titled "Common technical definition" has 6-7 paragraphs with no citations, reflecting someone's opinion as opposed to a well sourced scientific position. This is misleading and needs to be clarified, removed, or tagged. I propose tagging this section as "lacking citations" until the issue is resolved.

For example, there's a line saying "The terminology is also applied to indirect measurements—that is, values obtained by a computational procedure from observed data." What does "indirect" or "computation procedure" mean here? Are we talking about a state estimation problem, where the state cannot be directly observed? What is the role of observability in such cases? 2600:1700:3EC2:B000:ADCB:15F7:FDAF:89D (talk) 16:40, 19 June 2023 (UTC)[reply]

Binary classification section should be removed?

[edit]

The section on binary classification seems like 100 per cent red herring. It is exclusively about two indicators in the field that happen to be called accuracy and precision. It does not tell us what indicators in binary classification, if any, embody the ISO concepts. So it would be better I think to delete it. — Preceding unsigned comment added by Willbown (talkcontribs) 18:25, 8 August 2024 (UTC)[reply]

I have reverted the removal for now, because I think that some treatment of the concepts wrt binary classification is needed here. From personal experience, when first encountering the need to interpret a confusion matrix, the reader will have a definite need to read up on these terms, and may sensibly expect them to be covered here. They are not dealt with at any length at either Binary classification or Confusion matrix. An argument could be made that the in-depth description should be moved to either of these, but then we'd still need a summary and a link in this article. --Elmidae (talk · contribs) 11:53, 12 August 2024 (UTC)[reply]
Thanks. Instead of deleting the section, I've now replaced it with a section that aims to allow the reader to see that the concepts are relevant in the classificatory context. I haven't tried to explain them there, but I have of course removed the material that appeared to explain them but in fact explained different concepts.
If you know of a good reference for applying these ideas to classification, I'll look at it and enrich this section.
I also added a section that provides a simple reference for the very confusing overlaps in terminology that dog this topic, and this page.
With the benefit of that section, I also edited the Intro with the aim of making it clearer, in part by indicating the point at which the explanation becomes limited to non-classificatory scenarios. Willbown (talk) 13:08, 12 August 2024 (UTC)[reply]
More radically, why does this page even exist independently of Observational error, which explicitly states that such error is composed of accuracy and precision. Willbown (talk) 13:55, 12 August 2024 (UTC)[reply]
Accuracy, precision (and trueness) are summary statistics about multitude of error values in a dataset or sample compared with the true value. fgnievinski (talk) 05:59, 21 September 2024 (UTC)[reply]

Double meaning

[edit]

Willbown has "simplified summary [in the lead] by relegating second meaning of accuracy". The assumption that ISO's definitions are unimportant needs proper sourcing; without it, both meanings should be summarized in the lead. Also, WP:SELF forbids text such as "The three core concepts on this page... This Wikipedia page". fgnievinski (talk) 05:10, 21 September 2024 (UTC)[reply]

Yeah, not too happy about the self-referential setup either. Feel free to adapt :/ --Elmidae (talk · contribs) 10:08, 21 September 2024 (UTC)[reply]