Decent exposure

Research data published online must be accessible to allow scrutiny by other academics if we are to prevent fraud, says Geoffrey Boulton

九月 13, 2012

A recent leader in Times Higher Education tackled the "murky matter" of research misconduct, arguing that UK institutions must be "fearless in shining a light on misconduct" ("Clarity begins at home", 23 August).

Institutions should indeed do what they can to discourage and expose fraud by their members. Many, however, lack the expertise to expose fraud in highly specialist research publications, and none has the resources to systematically scan them for evidence of fraud.

Historically, fraud and error have been most efficiently exposed when the evidence for published scientific claims has been accessible to rigorous scrutiny by peers. Open publication of theories and the data that support them has enabled this. The leader described sunlight as the best disinfectant to both deter and expose fraud, but the sun can only do its work if everything is out in the open.

The problem is that rapid and pervasive technological advances have changed the game. Although some argue that our unprecedented capacity to acquire, store, manipulate and transmit vast and complex data volumes places us on the verge of a second scientific revolution, it can also serve to make the data on which an argument is based inaccessible.

The evidence underlying a published scientific argument, including the full details of experiments and observations, all the data and an assessment of uncertainties, could once be contained between the covers of a journal article. Mega-, giga- or terabytes of data deny us that option. Fraudulent practice too frequently hides beneath an impenetrable data carapace, so that we need to find new ways of reasserting the historical values of openness. The authors of a Royal Society report, Science as an Open Enterprise, published in June, argue that where the thesis of a scientific paper depends on large data volumes that cannot be reproduced in the paper, the data must be concurrently accessible in a specified database so that other specialists in the same field can test its reproducibility and explore whether the thesis is supported by the data.

In the Jatinder Ahluwalia research misconduct case highlighted by THE ("Faking it", 23 August), while there is some reassurance that it was the difficulty of reproducing results that was crucial to exposure, greater openness and transparency could have prevented the situation getting to the stage that it did.

The genesis of the Royal Society report was partly in the furore that surrounded the leaking of emails from climate scientists at the University of East Anglia in 2009. In that case, allegations of scientific fraud were rejected by a series of independent inquiries, but greater openness and transparency in the work of those scientists would have left them less exposed to allegations in the first place.

We should recognise that error and fraud are very different, but the line can be blurred. Fraudulent papers must be retracted. Scientific papers that are merely wrong should not be retracted, but journals should be prepared to publish refutations. The progress of science benefits from knowing what is not true, while understanding how errors have arisen is part of the process of self-correction.

What of malpractice, particularly in the selective use of data to promote a favoured hypothesis? The Royal Society study addressed the publication bias reflected in the common failure to publish negative results from clinical trials. It is not fraud in the sense of publishing fake results, but can be equally harmful in confusing relationships between cause and effect. For example, the British Medical Journal recently showed that whereas 94 per cent of antidepressant trials in the US were positive according to the published literature, the Food and Drug Administration analysis of all trials showed that only 51 per cent were positive.

So we need "intelligent openness": making sure that data are readily accessible; that they are intelligible to those who wish to scrutinise them; and assessable and usable, with appropriate metadata to allow replication.

The rewards for attracting research funds and making scientific discoveries can create temptations that encourage bad practice or even fraud. Of 788 papers retracted by PubMed - a database of life science and biomedical abstracts and references - between 2000 and 2010, a quarter were fraudulent. When people are involved there is always the risk of wrongdoing, but requiring high standards of openness can improve the probability of identifying fraud and thereby deterring it.

Whether there is fraud, malpractice or honest error, we need a system that is robust enough to allow us to identify these problems to put them right. We should take action now to adapt the conduct and communication of science to the realities of our new data-rich world.

请先注册再继续

为何要注册?

  • 注册是免费的,而且十分便捷
  • 注册成功后,您每月可免费阅读3篇文章
  • 订阅我们的邮件
注册
Please 登录 or 注册 to read this article.