I've recently read a curious article on machine learning where the author used a very interesting dataset, which would normally take thousands of man-hours to recreate. Now, assuming I would like to either:

  • Recreate the author's results, or
  • Improve upon them

Does the author have any obligation to provide me with said dataset?

N.B. In this particular situation the author has declined my request

  • 2
    Did the author mention why he declined?
    – mdd
    Commented Apr 13, 2016 at 0:42
  • @mdiener his rationale was that building the dataset on my own would help me with research, which kind of makes sense, but would take up thousands of man-hours.
    – user14156
    Commented Apr 13, 2016 at 0:47
  • 4
    Perhaps he wants to make further publications? If, as you say, it takes thousands of hours to recreate, you cannot blame them for wanting to get further publications out of this before they give up this competitive edge. Some teams release the data after they run their course and this is indeed the preferred path, but I do not think they are even ethically required to give up their hard-earned advantage that early in the game. Particularly, since some people have curiously sloppy in acknowledging non-paper inputs (e.g. software) to their work. Commented Apr 13, 2016 at 1:43
  • 1
    Related: Requesting raw data from previously published research. I certainly understand the authors' reluctance to simply give away data that were expensive to collect. They very probably want to use it for future publications. It might be possible to collaborate with them if you have an idea and they have the data. For such a collaboration, you'd ideally already have some kind of relationship with mutual trust. Commented Apr 13, 2016 at 6:46
  • 1
    @CaptainEmacs well, the latest article on that particular dataset was published all the way back in 2011. I'm not sure if I would consider that an 'edge',
    – user14156
    Commented Apr 13, 2016 at 8:40

1 Answer 1


Inherently? No. While it might be good for science generally, they're under no obligation to do so.

That being said, some funding agencies require data be made available as a condition of their support, and some journals require it as a condition of publication. Absent that though, there's no obligation to do so.

You must log in to answer this question.