Privacy is often mentioned as an obstacle when implementing an open data policy, but never really elaborated on. Should you really bother about privacy when opening up your data? My answer: yes you should.
Alan Westin laid the foundation of our modern conception of information privacy, which focuses on the individual’s right to control what is known about him. The modern European right to information privacy still leans on the notion of privacy as a right to control one’s personal information. Article 8 of the Charter of Fundamental Rights of the European Union gives everyone the right “to the protection of personal data concerning him or her”. This fundamental right to information privacy is further elaborated by the EU Data Protection Directive. The concept of ‘processing personal data’ is the touchstone of this directive. Personal data should be processed fairly and for legitimate and specified purposes.
EU data protection is all about the protection of ‘personal data’. Personal data is “information relating to an identified or identifiable natural person” and an identifiable person is “one who can be identified, directly or indirectly, in particular by reference to an identification number or to one or more factors specific to his physical, physiological, mental, economic, cultural or social identity” (Article 2 of the EU Data Protection Directive). Personal data can thus be both directly and indirectly identifying.
Train times, the location of public toilets and the number of car accidents could all be open data. No open data provider will (hopefully) offer names, addresses, social security numbers, or other data that directly or indirectly identifies natural persons as open data. Open data is at the most anonymized or aggregated data that cannot be related to individuals. The Open Knowledge Foundation visualizes open data and “private data” as two non-overlapping subsets. Unfortunately, in reality this distinction is not so easy to draw.
Even when data has been anonymized or aggregated, data analysis techniques now allow us to re-identify individuals in such data (See Paul Ohm for an overview). For instance, when Netflix offered anonymized data for a contest for the best method to improve its movie recommendations, Arvind Narayanan and Vitaly Shmatikov showed that this data could in fact be used to identify Netflix subscribers.
In particular regarding open data, Andrew Simpson demonstrated that it is relatively easy to link statistical open data to individuals. In one case, names and addresses of councillors, and names, posts and salaries of senior public servants were uncovered by combining data from the British open data portal with other already available public data. The lack of consideration of other data in the public domain prior to publication of statistical open data thus led to the identification of individuals.
Combining datasets is at the core of de-anonymizing and de-aggregating data. Data that is non-identifiable today, may turn out be indirectly identifiable tomorrow. The more computing power and publicly available data, the easier it becomes to identify individuals in data. And when data can be related to individuals, data protection law kicks in.
What does this mean for open data providers? Open data providers should not just consider the identifiability of their open data in isolation. They should also take other publicly available data into account when selecting data that they want to offer as open data. That is a difficult task. Maybe open data is not such a great idea after all?
Or check out Opendatarecht.nl, a Dutch weblog on open data.