Wikidata:Requests for comment/How to capture negative results in Wikidata?

The following discussion is closed. Please do not modify it. Subsequent comments should be made in a new section. A summary of the conclusions reached follows.

RFC closed. Inconclusive result.--Micru (talk) 18:08, 26 December 2018 (UTC)

---

An editor has requested the community to provide input on "How to capture negative results in Wikidata?" via the Requests for comment (RFC) process. This is the discussion page regarding the issue. If you have an opinion regarding this issue, feel free to comment below. Thank you!

In our efforts to capture the content of the CIViC database in Wikidata, we are dealing with negation. That is, in CIViC there is the notion of "Does Not Support", the question is how to model that into Wikidata. Currently, I am using the qualifier "Statement disputed by". Since, all the related references have all been added to Wikicite, it is straightforward to capture those supporting and not supporting references. So, currently all supporting and not supporting references are captured in the reference section of the relevant Wikidata statement (e.g. Q21851559 see statement "positive therapeutic predictor" for Panitumumab). The not supporting references are then also added as qualifiers to the same statement. This creates some level of redundancy, which is causing some confusion. However, if we would capture the disagreeing references as qualifiers alone, Wikicite related queries - which focus on reference alone - would miss the valuable negative results. However, since AFIK it is not possible to qualify a reference if it supports a statement or not, I added those negative references as qualifiers.

The question is if this is the ideal way of modelling this.

There are two questions I would like to raise:

1. Should supporting and not-supporting references be jointly captured in a single statement, or do negating reference vouch for a separate statement? This would create two statements, one which captures the supporting references and the second to capture the not-supporting ones, where the negative result is added as qualifier.
2. How should we treat negation in Wikidata. Is the qualifier "Statement disputed by", appropriate or should we distinguish more levels of negations. The phrase "disputed by", might not always be appropriate, since "does not support" could also indicate a lack of evidence. Should we introduce various qualifiers to capture negative results (e.g. valid, lacks evidence according to, etc)

I would really appreciate some thoughts on how to capture negative results in Wikidata.

--Andrawaag (talk) 11:08, 11 July 2017 (UTC)

Discussion

To add another example (in the context of Civic):

BRAF is a NOT a positive therapeutic predictor for Panitumumab (link). You can see this from the primary source here where the "Evidence Direction" is "Does Not Support". And the question is, how should this be modeled? I think the way it is currently is confusing because the reference is the same as the value for "disputed by", so its unclear if the statement is both in support of and disputes the same value. Gstupp (talk) 16:50, 11 July 2017 (UTC)

One should note that the binary supported/not-supported dichotomy may not good in many cases. Often the life science results are statistical with a mean and standard error. A result that is not significant is not necessarily not supporting the claim. In the long term it would be better if Wikidata could represent numeric results for further meta-analysis. I made some experiments over at test Wikidata, see, e.g., https://test.wikidata.org/wiki/Q1741. — Finn Årup Nielsen (fnielsen) (talk) 16:40, 12 July 2017 (UTC)

I'd like to see some more research on "negative results" in (history of) sciences and humanities: types, example, relevance etc. -- JakobVoss (talk)

@Andrawaag: Can you copy the meaning CIViC database (Q27612411) gives for "Does Not Support"? Looking at what the hover-over says, that term is not used clearly in the database and includes "inconclusive" as possible reason. Inconclusive data does neither negate nor dispute some statement. Therefore, if I read the above and the CiViC website, I am not sure "disputed by" is the correct term. Lack of support is not necessarily the same as a contradiction. The ultimate solution for negation, IMHO, would be if Wikidata had an option to just negate a statement, not by introducing a negation of the statement, but more like a qualifier on the statement property, equivalent to the logical NOT (logical negation (Q190558)). That would only be used if that statement does not automatically follow from other statements, by means of reasoning. For example, if something is a planet (Q634), it automatically is not a human (Q5), if Wikidata specifies they (or some superclasses) are mutually exclusive (mutually exclusive events (Q2625281)). --Egon Willighagen (talk) 07:18, 14 July 2017 (UTC)

There is two kind of statements : refuted statements and unrefuted one. This is valid for theories. I propose a class « refuted theory » a subclass of « theory » - and to make it a subclass of former entity (Q15893266)    to easily discard them in searches about entities that are destroyed/dead nowdays. for statement, the right kind of model is to use the deprecated rank. It’s definition is « something who was believed to be true / asserted but has been disproven or not believed anymore ». author  TomT0m / talk page 17:34, 22 August 2017 (UTC)

@Andrawaag, Egon Willighagen: negation as a qualifier fits in with the wikidata data model reasonably well and prevents us having to duplicate every property with its negative. For some statements "no value" works as a negation (eg. spouse (P26) indicates the person was never married) but that doesn't work when you are trying to negate a statement with both a property and a value. However, I'm concerned that qualifiers can be somewhat hidden - would it be possible for the UI to treat a negation qualifier (or other mechanism) specially so it is clear that it is a negation of the statement, not just a qualification of it? ArthurPSmith (talk) 15:47, 23 August 2017 (UTC)

I don’t think so, this would be giving a qualifier a meaning, and Wikibase is ignorant of the meaning of any qualifier. If that occurs, that would imply to modify the wikidata data model itself, and a better solution would occur, like a dedicated rank. However we already have a rank for false stuffs, see my comment above. Do we have a convincing usecase to motivate the introduction for a rank for stuffs that were never believed to be true ?

We also have a property to model stuffs like « non human entity » : disjoint union of (P2738) . Say we divide the world into concrete and abstract entities. This statement

⟨ entity ⟩ disjoint union of (P2738) ⟨ list of values as qualifiers (Q23766486)    ⟩
list item (P11260) ⟨ abstract entity ⟩
list item (P11260) ⟨ concrete entity ⟩

does the trick. author  TomT0m / talk page 16:28, 23 August 2017 (UTC)

---

The above discussion is preserved as an archive. Please do not modify it. Subsequent comments should be made in a new section.