Field Experiments Research Papers - Academia.edu (original) (raw)

“Practitioners” — whether business managers, lawmakers, clinicians, or other actors — are constantly innovating, in the broad sense of introducing new products, services, policies, or practices. In some cases (e.g., new drugs and medical devices), we’ve decided that the risks of such innovations require that they be carefully introduced into small populations, and their safety and efficacy measured, before they’re introduced into the general population. But for the vast majority of innovations, ex ante regulation requiring evidence of safety and efficacy neither does — nor feasibly could — exist. In these cases, how should practitioners responsibly innovate?

Most commonly, innovation is ad hoc and intuition-driven. Sometimes, a practitioner will attempt to rigorously determine the effects of a novel practice by comparing it to an alternative possible innovation or to the status quo. Not infrequently, if those subject to such A/B testing (as marketers and data scientists refer to it) or experimentation (as scientists in other fields call it) are fully informed about it, the results will be badly biased. In those cases, the practitioner may undertake the exercise more or less in secret, at least initially.

Practices that are subject to A/B testing generally have a far greater chance of being discovered to be unsafe or ineffective, potentially leading to substantial welfare gains. Yet the conventional wisdom is that “human experimentation” is inherently dangerous and human experimentation without informed consent is always unethical.

Facebook recently learned this lesson the hard way after the public learned about a 2012 experiment it conducted to determine the effects of News Feed, an innovation the company had launched in 2006 that marked a major shift in how now 1.44 billion people allocate their time and in the way they observe and interact with others. Academic studies have suggested two contradictory hypotheses about the risks of News Feed: that exposure to friends’ positive posts is psychologically risky (through a social comparison mechanism) and that exposure to negative posts is psychologically risky (through an emotional contagion mechanism). But these contradictory studies were mostly small and observational. The company alone was in a position to rigorously determine the mental health effects of its service, and to do so relatively cheaply. And for one week in January of 2012, it conducted an experiment in which it attempted to do just that.

How should we think about Facebook’s decision to conduct an experiment? Reaction was in fact swift and fierce. Criticism by both the public and some prominent ethicists centered on the fact that the 700,000 or so users involved had not consented to participate in what appeared to be a study designed to psychologically harm users by manipulating their emotions. Critics charged Facebook with exploiting its position of power over users, treating them as mere means to the corporation’s ends, and depriving them of information necessary for them to make a considered judgment about what was in their best interests. Some demanded federal and state investigations and retraction of the published results of the experiment.

But this considerable discussion paid scant attention to the experiment’s relationship to Facebook’s underlying practice of algorithmically curating users’ News Feeds and its risks and uncertainties which, after all, were imposed on 1.44 billion users without their knowledge or consent. In this article, using the Facebook emotional contagion experiment and, to a lesser extent, the OkCupid matching algorithm experiment, as case studies, I explore two frames through which we can think about these and similar corporate field experiments. The first frame is the familiar one used by ethicists and regulators to govern human subjects research. Contrary to popular belief, this frame, articulated in the Belmont Report and codified in the federal Common Rule, appropriately permits prima facie duties to obtain subjects’ informed consent to be overridden when obtaining consent would be infeasible and risks to subjects are no more than minimal — criteria, I argue, that there are good reasons to believe applied to the Facebook experiment.

The second frame contextualizes field experiments against the backdrop of the underlying practice they’re designed to study. Foregrounding the experimenter’s role as a practitioner, it asks how she ought to responsibly innovate and about the appropriate role of experiments in that innovation process. Experiments involving a tight fit between the population upon whom (no more than minimal) risks are imposed and the population that stands to benefit from the knowledge produced by a study may not only be ethically permissible; where they are conducted by the innovator who is both the proximate cause and cheapest avoider of any innovation-related costs, these experiments may be ethically laudable or even obligatory.

Like Rubin’s vase, where viewers vacillate between seeing a vase or two opposing faces in profile, each of these two frames becomes salient by bringing some aspect of the overall situation to the foreground: either the experiment or the practice whose effects it tests. But almost everyone saw the Facebook experiment through the first framework of human subjects research every time, and never through the second framework of responsible innovation. Why?

Using the OkCupid experiment as a mini case study, I dub the “A/B illusion” the widespread tendency to view a field experiment designed to study the effects of an existing or proposed practice as more morally suspicious than a practitioner’s alternative of immediately implementing an untested practice. The A/B illusion, which the Common Rule lamentably fosters, can cause us to overregulate research and underprotect a practice’s end-users. For instance, given probative but inconclusive evidence that News Feed psychologically harms users through exposure to negative and/or positive posts, Facebook’s unique position to establish the effects of its innovation, and the relative ease with which it could do so, most criticisms of the Facebook experiment reflect the A/B illusion and should be inverted: It is not the practitioner who engages in A/B testing but the practitioner who simply implements A who is more likely to exploit her position of power over users or employees, to treat them as mere means to the corporation’s ends, and to deprive them of information necessary for them to make a considered judgment about what is in their best interests.