Abstract
In mid-2014 a story broke about Facebook conducting psychological experiments on the newsfeed of almost 700,000 of its users without consent or IRB approval. The aim of the research was to test the hypothesis that users would experience emotional contagion – that is, if they perceived others as posting more negatively they would post more negatively, and if they perceived others as posting more positively they would post more positively. They tested this by manipulating the newsfeed users saw showing them either more negative posts from their friends or more positive ones. The effect was significant, albeit small; altered newsfeeds demonstrated a slight emotional contagion effect.
This kind of A/B testing is common in online websites, and some wondered why there was any controversy. Others were concerned about the absence of any consent other than the user agreement you agree to when you sign up to Facebook (and did not at the time include research explicitly). Although the effect was small, some were worried about the potential for harm given the sheer numbers involved – what if one of the users who had a more negative newsfeed had a serious mental illness? Concerns were also raised about whether ethical approval ought to have been sought as well as jurisdictional issues: where were these 700,000 users located, and was approval required in each jurisdiction?
The experiment, as a collaboration between academic researchers at Cornell, and Facebook, existed in a grey area of federal regulation. It was allegedly designed by the Cornell researchers, but the protocol was carried out exclusively by Facebook employees. The Cornell researchers then assisted with data analysis, which the Cornell Institutional Review Board (IRB) deemed a “pre-existing data-set,” and not in need of substantive ethical review. As a private entity not in receipt of federal research funds, Facebook is exempt from review under the Common Rule. Though Facebook later claimed it reviewed the experiment through its own in-house oversight mechanisms, the exact structure of this review remains a mystery.
The Facebook study became the public face of ethical issues arising from the balooning of private entities, and private funding, conducting research with human subjects that are exempt from (at least, in the US) ethical review regulations. The year before the Facebook study appeared, personal genomics company 23andme suspended marketing of its test kits when the FDA issued a notice that they had failed to demonstrate the validity of their tests as medical devices. Despite this setback, the company has collected large volumes of patient data for research pur-poses, with minimal oversight. In early 2015, concerns were raised again about online health research with the release of Apple’s ResearchKit framework, which allows researchers to create apps to collect information from participants for research purposes. As research with large data sets becomes easier to pursue, and more profitable, we should expect an increase in the number of experiments being conducted partly or totally online. This new trend, and the regulatory grey area it inhabits, is largely unexplored in the research ethics literature.
In this special issue of Research Ethics we engage with these issues and others in large online research like the Facebook emotional contagion experiment.
In this issue
boyd engages with explaining the public reaction to this research and sug-gests that public discomfort with big data is at the heart of it; she suggests we should not distinguish between academic and corporate practices.
Flick suggests that there was a failure of appropriate ethical oversight here and suggests some solutions to this.
Shaw discusses consent issues in the Facebook study.
Selinger and Hartzog argue that a neglected ethical issue regarding debate around the study is how user identities were coopted.
Recuber discusses how power is naturalized and utilized in this and other big data studies.
Gertz suggests that IRBs and external review will do little to limit or control corporate power.