Dealing With Data 2018: Summary reflections

The annual Dealing With Data conference has become a staple of the University’s data-interest calendar. In this post, Martin Donnelly of the Research Data Service gives his reflections on this year’s event, which was held in the Playfair Library last week.

One of the main goals of open data and Open Science is that of reproducibility, and our excellent keynote speaker, Dr Emily Sena, highlighted the problem of translating research findings into real-world clinical interventions which can be relied upon to actually help humans. Other challenges were echoed by other participants over the course of the day, including the relative scarcity of negative results being reported. This is an effect of policy, and of well-established and probably outdated reward/recognition structures. Emily also gave us a useful slide on obstacles, which I will certainly want to revisit: examples cited included a lack of rigour in grant awards, and a lack of incentives for doing anything different to the status quo. Indeed Emily described some of what she called the “perverse incentives” associated with scholarship, such as publication, funding and promotion, which can draw researchers’ attention away from the quality of their work and its benefits to society.

However, Emily reminded us that the power to effect change does not just lie in the hands of the funders, governments, and at the highest levels. The journal of which she is Editor-in-Chief (BMJ Open Science) has a policy commitment to publish sound science regardless of positive or negative results, and we all have a part to play in seeking to counter this bias.

Photo-collage of several speakers at the event

A collage of the event speakers, courtesy Robin Rice (CC-BY)

In terms of other challenges, Catriona Keerie talked about the problem of transferring/processing inconsistent file formats between heath boards, causing me to wonder if it was a question of open vs closed formats, and how could such a situation might have been averted, e.g. via planning, training (and awareness raising, as Roxanne Guildford noted), adherence to the 5-star Open Data scheme (where the third star is awarded for using open formats), or something else? Emily earlier noted a confusion about which tools are useful – and this is a role for those of us who provide tools, and for people like myself and my colleague Digital Research Services Lead Facilitator Lisa Otty who seek to match researchers with the best tools for their needs. Catriona also reminded us that data workflow and governance were iterative processes: we should always be fine-tuning these, and responding to new and changing needs.

Another theme of the first morning session was the question of achieving balances and trade-offs in protecting data and keeping it useful. And a question from the floor noted the importance of recording and justifying how these balance decisions are made etc. David Perry and Chris Tuck both highlighted the need to strike a balance, for example, between usability/convenience and data security. Chris spoke about dual testing of data: is it anonymous? / is it useful? In many cases, ideally it will be both, but being both may not always be possible.

This theme of data privacy balanced against openness was taken up in Simon Chapple’s presentation on the Internet of Things. I particularly liked the section on office temperature profiles, which was very relevant to those of us who spend a lot of time in Argyle House where – as in the Playfair Library – ambient conditions can leave something to be desired. I think Simon’s slides used the phrase “Unusual extremes of temperatures in micro-locations.” Many of us know from bitter experience what he meant!

There is of course a spectrum of openness, just as there are grades of abstraction from the thing we are observing or measuring and the data that represents it. Bert Remijsen’s demonstration showed that access to sound recordings, which compared with transcription and phonetic renderings are much closer to the data source (what Kant would call the thing-in-itself (das Ding an sich) as opposed to the phenomenon, the thing as it appears to an observer) is hugely beneficial to linguistic scholarship. Reducing such layers of separation or removal is both a subsidiary benefit of, and a rationale for, openness.

What it boils down to is the old storytelling adage: “Don’t tell, show.” And as Ros Attenborough pointed out, openness in science isn’t new – it’s just a new term, and a formalisation of something intrinsic to Science: transparency, reproducibility, and scepticism. By providing access to our workings and the evidence behind publications, and by joining these things up – as Ewan McAndrew described, linked data is key (this the fifth star in the aforementioned 5-star Open Data scheme.) Open Science, and all its various constituent parts, support this goal, which is after all one of the goals of research and of scholarship. The presentations showed that openness is good for Science; our shared challenge now is to make it good for scientists and other kinds of researchers. Because, as Peter Bankhead says, Open Source can be transformative – Open Data and Open Science can be transformative. I fear that we don’t emphasise these opportunities enough, and we should seek to provide compelling evidence for them via real-world examples. Opportunities like the annual Dealing With Data event make a very welcome contribution in this regard.

PDFs of the presentations are now available in the Edinburgh Research Archive (ERA). Videos from the day will be published on MediaHopper in the coming weeks.

Other resources

Martin Donnelly
Research Data Support Manager
Library and University Collections
University of Edinburgh

Share

EDINA’s ShareGeo Open content into DataShare

Many fascinating datasets can be found in our new ShareGeo Open Collection: http://datashare.is.ed.ac.uk/handle/10283/2345  .

This data represents the entire contents of EDINA’s geospatial repository, ShareGeo Open, successfully imported into DataShare. We took this step to preserve the ShareGeo Open data, after the decision was taken to end the service. Not only have we maintained the accessibility of the data but we also successfully redirected all the handle persistent identifiers so that any existing links to the data, including those included in academic journal articles, have been preserved, such as the one in this paper: http://dx.doi.org/10.1007/s10393-016-1131-y .

Similarly, should the day ever arrive when DataShare was to be closed, we would endeavour to find a suitable repository to which we could migrate our data to ensure its preservation, as per item 13 of our Preservation policy.

We were able to copy the content of almost all metadata fields from ShareGeo to DataShare. The fact both repositories use the Dublin Core metadata standard, and both were running on DSpace, made the task a little easier. The University of Edinburgh supports the Dublin Core Metadata Initiative. DataShare’s metadata schema can be found at https://www.wiki.ed.ac.uk/display/datashare/Current+metadata+schema setting out what our metadata fields are and which values are permitted in them.

Our EDINA sysadmin (and developer) George was very helpful with all our questions and discussions that took place while the team settled on the most appropriate correspondence between the two schemas. The existing documentation was a great help too. George then produced a Python script to harvest the data, using OAI-PMH to get a list of ShareGeo items, then METS for the metadata and bitstreams. He then used SWORD to deposit them all in DataShare.

The team took the opportunity to use DSpace’s batch metadata editing utility and web interface to clean up some of the metadata: adding dates to the temporal coverage field and adding placenames and country abbreviations to the spatial coverage field, to enhance the discoverability of the data.

For example “GB Postcode Areas” can be found using the original handle persistent identifier: http://hdl.handle.net/10672/51 as well as the new DOI which DataShare has given it – DOI: 10.7488/ds/1755. Each of the 255 items migrated to our ShareGeo Open Collection contains a file called metadata.xml which contains all the metadata exactly as it was when exported from ShareGeo itself. I have manually added placenames in the spatial coverage field (which was used differently in ShareGeo, with a bounding box i.e. “northlimit=60.7837;eastlimit=2.7043;southlimit=49.8176;westlimit=-7.4856;”). Many of these datasets cover Great Britain, so they don’t include Northern Ireland but do include Scotland, England and Wales. In this case I’ve added the words “Scotland”, “England” and “Wales” in Spatial Coverage (‘dc.coverage.spatial’), even though these are already implicit in the “Great Britain” value in the same field, because I believe doing so:

  • enhanced the accessibility of the data (by making the geographical extent clearer for users unfamiliar with Great Britain) and…
  • enhanced the discoverability of the data (users searching Google for “Wales” now have a chance of seeing this dataset among the hits).

James Crone who compiled this “GB Postcode Areas” data is part of EDINA’s highly renowned geospatial services team.

Part of James’ work for EDINA involves producing census geography data for the UK DataService. He has recently added updated boundary data for use with the latest anonymised census microdata (that’s from the 2011 census): see the Boundary Data Selector at https://census.ukdataservice.ac.uk/get-data/boundary-data .

Pauline Ward is a Research Data Service Assistant for the University of Edinburgh, based at EDINA.

Detail from GB Postcode Areas data, viewed using QGIS.

Detail from GB Postcode Areas data, viewed using QGIS.

Share