Deception in research is morally problematic … and so too is not using it morally: reply to open peer commentaries on "the use of deception in public health behavioral intervention trials: a case study of three online alcohol trials".

We are very pleased that the explorationsoftheethical issues raised by our research practice (McCambridge et al. 2013a) have stimulated such thoughtful and diverse responses. Almost all indicate concern that it may be wrong to accept the moral hazards we are prepared to face, and we agree there are many reasons to take very seriously these issues, hence our attempts to do so. Most appear to accept that our research practices are of low risk and not likely to cause harm to participants, and the key issues center on informed consent. It is interesting that there are such differences in the particular concerns identified by the commentators, and we are struck by the breadth of their concerns.

Wang and Kitsis (2013) believe there should be an individual study requirement to measure the impact of deception when it is used, while we see a need to develop a body of research on this subject, though not necessarily in every study. Brief interventions opportunistically grab a few minutes of people's time, and our orientation is to stripping down rather than beefing up the research process, for reasons previously explained, though simple brief measures of the type proposed are appealing.

Plunk and Grucza (2013), Sisti and colleagues (2013), and similarly Wang and Kitsis (2013) take exception to our suggestion that debriefing may be morally problematic in certain circumstances. We had not intended to develop an argument about this here (we present data more fully elsewhere), merely to suggest a possible extension of this approach, and we do believe this worth considering further. We disagree with Plunk and Grucza that deception rather than debriefing should be considered the cause of any consequent harms. The latter requires the former, and given that the harms caused are directly consequent on debriefing, this procedure is unavoidably a component (i.e., not the sole) cause of them. We worry most about possible damage to public trust in science, a primary motivation of this proposal.

Schwab (2013) is, of course, correct that lying is more problematic than deception by omission, and telling lies requires stronger justification than incomplete disclosure. What we do not say, however, is designed to affect participants in particular ways for which the word deception seems appropriate, because the nature of the intent clearly matters.

For legal as well as ethical reasons, Zuraw (2013) distinguishes between situations in which deception in or after a consent procedure is more acceptable than studying individuals who are unaware that they are being studied. Although the latter is common in studies of health records, for example, we have a specific methodological agenda, namely, to evaluate the effects of research participation itself in biasing study outcomes. Eliminating or obscuring awareness of being studied is necessary for this particular enterprise if it is judged worth doing in the first place, and we make no case for the wider extension of this practice, as previously emphasizing the contextual specificity of our decision making.

O'Neil (2013) similarly distinguishes the use of deception in inducing participation to improve external validity or generalizability from methods that deceive participants to safeguard internal validity. We appreciate this distinction and are strongly influenced by the work of Donald Campbell, the original author of it (Campbell 1957) who overtime also extended thinking about it in more complex formulations (see, e.g., Cook and Campbell 1979). There is a widespread tendency in the research literature to think of internal and external validity as mutually exclusive, but they are related. In a behavioral intervention trial, the intervention effect estimate is influenced by study sample composition; excluding those with severe problems can lead to larger effect estimates because inference is restricted to a more healthy reference population (Humphreys 2000). The least biased estimate is the one from which to make inferences about the true effects in the population of interest. Internal and external validity are thus two sides of the same coin.

O'Neil and Zuraw argue that AMADEUS-1 (McCambridge et al. 2012) presents the most challenging ethical issues in ways that are clearer to us now than before. We acknowledge that increasing participation forms part of the rationale for this design specifically to avoid erecting artifactual barriers to intervention access as conventional trial designs do in this situation. There are also other methodological and substantive intervention effectiveness reasons to avoid bias. We agree it is important that the justification for studying people without consent be very clearly established. In AMADEUS-1 there was an explicit attempt to design out research participation effects prior to follow-up data collection to permit unbiased quantification of intervention effects. There was almost no difference in outcomes between the two groups recruited at baseline, giving further substance to our preexisting concerns that the small individual-level effects and their components were vulnerable to being biased by the research process, which motivated the study design in thefirst place (McCambridge et al. 2013b). Small effects widely obtained at an individual level are nevertheless importantat apopulation level (Rose 1992).

In AMADEUS-1, participants were unaware they were being studied when they accessed routine practiceoramodified version of it. Participants volunteered to take part in a cross-sectional survey, for follow-up purposes, missing the explicit focus on alcohol (a decision for which there was a strong methodological justification, based on experimental data; Bendtsen et al. 2012). We cannot know whether those who accessed the intervention and did not later participate in the survey would have chosen to do so if they were offered that choice. This group were responding to an e-mail from the student health care service encouraging them to think about drinking, an invitation they accepted.

AMADEUS-1 was aimed at enhancing effectiveness of routine service provision. We cannot tell whether the commentators would be less concerned by this context, but we are. The deception involved in many ways may be similar to other health services research in which routine data or service developments are evaluated. We do not mean to suggest that this removes the moral dilemmas, just that they are less unusual than they might first seem. We also wonder whether not having a formal consent procedure is really so different to offering one in which, to use Schwab's term, we would be lying as extensively about the nature of the study as occurred in the other two studies (Kypri et al. 2010; Kypri et al. 2011).

Not doing research that involves taking ethical risks to better understand the nature of the problems caused by alcohol and how to intervene with this behavior should not be an ethically comfortable position. Hendershot and colleagues (2013) identify important historical advances in understanding in this field as they eloquently articulate a range of benefits that flow from constraining individual autonomy for population health benefit. There is also a contemporary need that can be concisely stated less dryly than we did in the original article: Alcohol kills approximately 5 million people every year, and these numbers are growing rapidly (Lim et al. 2012), as the alcohol industry targets emerging economies in low and middle income countries in ways which will drive new epidemics of death and disease, if allowed to continue (Casswell 2013). Think of the awful tragedies of Syria, Iraq, Afghanistan, and all the other major wars of the early 21st century; alcohol kills more than all of them combined. Just like these wars, alcohol ruins childhoods and kills young adults in ways that tobacco does not, as well as slowly accumulating health problems over the life course with sustained heavy drinking and dependence.

We know so little about this behavior and about how people can be helped to avoid the death and disease it causes. This is partly because our research investigations are so clumsy that conventional practice in behavioral intervention trials in public health interferes with what we are trying to study, in ways we do not understand clearly (McCambridge in press). If careful use of deception helps to advance understanding of this behaviour and how people may be helped to change for themselves, surely there is a moral imperative to use this knowledge to prevent or ameliorate avoidable suffering on such a scale.

We know well the challenges involved in trying to address this particular set of corporate induced diseases (Babor and Robaina 2013; Jahiel 2008), and helping people to manage their lives in ways that protect them from the effects of other health-compromising activities and commodities is centraltomuchofpublic health today (Childress et al. 2002). The overall impact of the commentaries has been to deepen our appreciation of the moral complexities of these issues, shaping our ongoing considerations of them but without having dramatically altered our thinking. In part, this reflects a distinction between our perspective and that of most of the commentators, and that is what should be the proper unit or frame of analysis of ethical issues. Do the rights and welfare of individual research participants take precedence over the needs of populations to benefit from this research? Pragmatism involves trying to balance both (Brendel and Miller 2008), an approach that clearly involves taking ethical risks with the rights of research participants here that need to be carefully considered, which seems particularly appropriate for public health (Buchanan and Miller 2006). Nonetheless, the core purpose of this research is to do so in ways that deliver benefits to populations or society. Greater attention to authorizing deception or proxy consent at the population rather than the individual level is an attractive possibility.

The metaphor of the tip of an iceberg of unknown size being visible above the waterline seems appropriate for evaluating the use of deception in research. Our research community decided long ago that it was acceptable to interfere with informed consent to do the types of studies we now do online. We worry about what might happen to our participants as a result of our actions, for which we accept responsibility. Our efforts to be vigilant toward issues they raise generate few concerns, though this does not mean that we are right. We are taking new steps probably using more deception than our colleagues have done in the past and are humbled when we consider how little we know compared to how much we need to know about alcohol, public health, and human behavior more generally. Our determination to gain understanding in these studies and to apply what we find has, we believe, a strong moral foundation. The need for public health research ethics guidance seemed clear to us before and following this process of publication and review remains so, and where the commentators discuss this issue, they also agree. This is a long-term project and one that should not be restricted to professional ethicists or researchers alone.