How to deal with journal requirements to share data when data comes from industry and industry partner would not permit sharing?

Question

Many authors, especially in manufacturing, use data from real manufacturing companies (not from a lab experiment) for their research. They are asked to report only the results from their analysis on those datasets in journal papers without publishing the underlying data due to confidentiality agreements with the companies. However, I see that the journals are increasingly asking the authors to deposit the raw data so that the research is reproducible. On the other hand, the companies are not ready to make their internal data go public but they are ok with publishing the summary statistics on the datasets. As a result, authors face difficulty in publishing their research results in journal papers.

Any strategies to handle this situation effectively and convince editors about the non-availability of the datasets to other researchers?

Answer 1

The recommendation I always give is to clearly state where the data was obtained from and what strategy was used to select it. If it’s data owned by a commercial partner who is not willing to allow open sharing of the data, you’re unlikely to get them to change their mind, but for the end goal of supporting the reproducibility of the research, clearly stating your sources is a good second best. That will give other researchers the opportunity to approach the company directly and sign their own NDA to get access to the data.

Answer 2

I am going to question the premise:

You ask: Any strategies to handle this situation effectively and convince editors about the non-availability of the datasets to other researchers?

I'm going to say: You can't and you shouldn't.

It isn't that the journal editors don't believe you when you say "This data is not able to be released for other researchers." They don't really care why it isn't available, but they do believe that without that data, it is not the kind of article they want to publish.

Some options which may be valid depending on circumanstances:

• Find a journal that is Ok publishing without data. (For understandable reasons these might not be as good as the more stringent ones)
• Create a new synthetic dataset, that has similar properties to your real data, and present your primary results on that. (And mention as a secondary point your results on the real data)
• Work out what it would take to make the company happy to release the data:
  • Perhaps removing identifiers (E.g. for personal identifying data k-anonymity is a common technique)

See also the related question: Can you publish studies based on confidential customer data that comes from private companies?

Answer 3

@Lyndon White makes a good point about the journal reserving the right to decide what kind of articles they would like to publish. Though I'm in agreement, I would like to raise one point of caution.

Publishing is a commercial enterprise, and is driven by economic considerations and profit margins more than increasing reproducibility. The fact that these aspects are actively discussed on ASE is a sign that gradually researchers are becoming aware of how skewed the market is.

In this scenario, a journal could demand data on the grounds of scientific reproducibility, and make this data available to subscribers only. This feature could increase subcriptions from rival industries as well as research labs. In effect, the journal would then be selling data from one source to another. The authors would be willing to turn a blind eye because the journal is prestigious, while the source of data (eg. industry) would grow increasingly distrustful of this journal, widening the industry-academia gap. I find it difficult to see this as a good sign.

The only solution that comes to mind is the erection of mutual non-disclosure agreements between the journal and the author, that permit the author to decide (atleast in part) who gets to see/use the data. This has its own flaws, but it could be a start.