Until relatively recently it was hard for organisations like arXiv to assign DOIs- even if they wanted to. This is because, up until 2009, CrossRef was really the only major DOI registration agency and it has hitherto had a rule prohibiting the assignment of CrossRef DOIs to "preprints." The rationale for this was that, when CrossRef was founded, publishers thought that assigning separate DOIs to multiple versions of an article might make it harder to easily identify the version of record and that this, in turn, would confuse the citation record.
Note that arXiv could have decided to assign CrossRef DOIs to content that was not destined to be published. bioRxiv has done this for years. The problem is it is virtually impossible to predict what might be published. In practice this means that, despite its best efforts, bioRxiv ended up assigning several hundred CrossRef DOIs to items that eventually got published and were assigned a separate CrossRef DOI. It is understandable that arXiv didn't want to face this hassle.
arXiv and bioRxiv could have also gone to another registration agency (like DataCite) and assigned DOIs through them. Indeed- this is what FigShare and Zenodo have done. But again, the problem with this approach is that you can easily end up with multiple versions of the same content with different DOIs and no easy way to link them. You can even end up with the exact same version of the document with different DOIs. You can see this problem (amongst others) discussed here:
Which gets us back to CrossRef's prohibition on assigning DOIs to preprints. We have good news here- CrossRef will soon be lifting this prohibition. We have concluded that we will be much better able to clarify the citation record if we allow our members to assign DOIs to different versions of articles and for us to provide tools and guidelines for labelling and linking those versions. Although you should not interpret this as a change in their respective DOI policies, it is worth mentioning that both arXiv and bioRxiv were on the working group that helped us come to this decision. More details on the new policy will be coming out soon.
Disclosure- I am CrossRef's Director for Strategic Initiatives.
This post has been migrated from the Open Science private beta at StackExchange (A51.SE)