The Economist: “….The anonymisation of a data record typically means the removal from it of personally identifiable information. Names, obviously. But also phone numbers, addresses and various intimate details like dates of birth. Such a record is then deemed safe for release to researchers, and even to the public, to make of it what they will. Many people volunteer information, for example to medical trials, on the understanding that this will happen.
But the ability to compare databases threatens to make a mockery of such protections. Participants in genomics projects, promised anonymity in exchange for their DNA, have been identified by simple comparison with electoral rolls and other publicly available information. The health records of a governor of Massachusetts were plucked from a database, again supposedly anonymous, of state-employee hospital visits using the same trick. Reporters sifting through a public database of web searches were able to correlate them in order to track down one, rather embarrassed, woman who had been idly searching for single men. And so on.
Each of these headline-generating stories creates a demand for more controls. But that, in turn, deals a blow to the idea of open data—that the electronic “data exhaust” people exhale more or less every time they do anything in the modern world is actually useful stuff which, were it freely available for analysis, might make that world a better place.
Of cake, and eating it
Modern cars, for example, record in their computers much about how, when and where the vehicle has been used. Comparing the records of many vehicles, says Viktor Mayer-Schönberger of the Oxford Internet Institute, could provide a solid basis for, say, spotting dangerous stretches of road. Similarly, an opening of health records, particularly in a country like Britain, which has a national health service, and cross-fertilising them with other personal data, might help reveal the multifarious causes of diseases like Alzheimer’s.
This is a true dilemma. People want both perfect privacy and all the benefits of openness. But they cannot have both. The stripping of a few details as the only means of assuring anonymity, in a world choked with data exhaust, cannot work. Poorly anonymised data are only part of the problem. What may be worse is that there is no standard for anonymisation. Every American state, for example, has its own prescription for what constitutes an adequate standard.
Worse still, devising a comprehensive standard may be impossible. Paul Ohm of Georgetown University, in Washington, DC, thinks that this is partly because the availability of new data constantly shifts the goalposts. “If we could pick an industry standard today, it would be obsolete in short order,” he says. Some data, such as those about medical conditions, are more sensitive than others. Some data sets provide great precision in time or place, others merely a year or a postcode. Each set presents its own dangers and requirements.
Fortunately, there are a few easy fixes. Thanks in part to the headlines, many now agree that public release of anonymised data is a bad move. Data could instead be released piecemeal, or kept in-house and accessible by researchers through a question-and-answer mechanism. Or some users could be granted access to raw data, but only in strictly controlled conditions.
All these approaches, though, are anathema to the open-data movement, because they limit the scope of studies. “If we’re making it so hard to share that only a few have access,” says Tim Althoff, a data scientist at Stanford University, “that has profound implications for science, for people being able to replicate and advance your work.”
Purely legal approaches might mitigate that. Data might come with what have been called “downstream contractual obligations”, outlining what can be done with a given data set and holding any onward recipients to the same standards. One perhaps draconian idea, suggested by Daniel Barth-Jones, an epidemiologist at Columbia University, in New York, is to make it illegal even to attempt re-identification….(More).”