linking a dataset in the metadata of an article with a DOI registered with CrossRef can be done following the documentation of CrossRef via a related_item with the following XML snippet:
<rel:program name="relations"> <rel:related_item> <rel:description> some description </rel:description> <rel:inter_work_relation relationship-type="isSupplementedBy" identifier-type="doi">10.prefix/suffix </rel:inter_work_relation> </rel:related_item> </rel:program>
Optionally, the authors could also cite their dataset so that it will end up in the references section of the metadata as well. This would also emphasize that a data publication in as valuable as an article published with one of the scientific publishers, rather an solely being an supplement.
For the way back i.e. with DataCite metadata an XML snippet may look like this:
<relatedIdentifier relatedIdentifierType="DOI" relationType="IsSuplementTo">10.prefix/suffix </relatedIdentifier>
With DataCite Schema 4.4 a relatedItem was introduced - same term as with CrossRef. The introduction tells:
Also, a new property, relatedItem, is introduced which contains information about a resource related to the one being registered, e.g., a journal article, conference paper, or a book series. It can be used to describe a text citation where relatedIdentifier cannot be used because the related resource does not have an identifier.
Further down, the definition follows:
Information about a resource related to the one being registered, e.g., a journal or book of which the article or chapter is part.
The allowed values, examples and other constrains state:
Can be used to provide series information or a text citation where the related resource does not have an identifier. However, it is also optional to provide an identifier here.**
All this together is anything but clear. While the introduction provides a journal article as an example, the definition uses an other concept, which is journal (or book) which has articles. Being familiar with CrossRef metadata, one might pick the relatedItem to link to an article published with an scientific publisher, while this is actually not the way to go following this guidance document.
This guidance document clearly states, that the relatedItem should be used if the related source does not have an identifier. The other way around, relatedIdentifier should be used, if the resource has an identifier. Both are summarised as the second use case.
Nevertheless, calling something that is related but does not have a PID a relatedItem, while CrossRef already calls things that are related and may have a PID a related_item, will certainly cause confusion among users.