Scientists aren't sharing data

Empirical Study of Data Sharing by Authors Publishing in PLoS Journals
by Caroline J. Savage, Andrew J. Vickers

The PLoS data sharing policy states that “publication is conditional upon the agreement of authors to make freely available any materials and information associated with their publication that are reasonably requested by others for the purpose of academic, non-commercial research.” - That's a pretty clear guideline. The authors of this article requested data from ten researchers who published in PLoS but they were only able to get data from one of them.

Click here to see the chart that illustrates the responses they got from researchers.

The people who gathered the requested data don't appear to be doing research for private industry, I'm assuming that they are doing research at universities or for non-profits. So in theory they are working for the greater good of scholarship. Perhaps I'm wrong about this, but I do think that data sharing in a field like private pharmaceutical research is very different from sharing data that's gathered at a university.

This article by Savage and Vickers reinforces the one insurmountable problem with sharing data - it takes a long time to annotate data so that other people can understand it. Somehow the one person who sent them data had it all annotated ahead of time. Otherwise, they were able to do it within a few hours, but that doesn't seem possible.

The other problem with sharing data is that it is uncharacteristic of scientists, so to expect this of them is a shift in how they think about their work. One of the scientists that was contacted responded that they just didn't know this was a guideline of submitting to PLoS and that if they had known ahead of time they never would have submitted. It turns out that their data was not supposed to be released to a third party.

As if this article isn't already enough bad news, Savage and Vickers mention a previous similar study done in which requests for data were sent to authors who published in American Psychological Association (APA) journals. The APA study had a 25% return on data requests. The reason why this is discouraging is that the APA only requires researchers to share data to "verify the substantive claims through reanalysis", which means that when researchers are asked for their data, they're essentially being put on trial. When Savage and Vickers requested data, they claimed to be looking for a new hypothesis based on the data, so it was clearly for their own research. Also, the APA data sets were merely survey responses while Savage and Vickers were requesting data from medical studies. Not surprisingly, it's easier to acquire raw data from surveys than data from health sciences, but this is likely a reflection of how long it takes to annotate complicated data. It's just unfortunate that data which is detrimental to health can't be shared as easily.

In order for digital curation to function, researchers need to deposit their data into repositories. If they aren't willing to share data with their fellow researchers, there isn't much hope for them to hand the data over to strangers.