Comment MATTHEW HORWOOD/GETTY MATTHEW Consider What Information — in What Format — Would Best Support Your Audiences’ Decisions

Home , False balance

Setting the agenda in research Comment MATTHEW HORWOOD/GETTY MATTHEW Consider what information — in what format — would best support your audiences’ decisions. Five rules for evidence communication

Michael Blastland, Alexandra L. J. Freeman, Sander van der Linden, Theresa M. Marteau & David Spiegelhalter

Avoid unwarranted certainty, e persuasive”, “be engaging”, There are myriad examples from the cur- “tell stories with your science”. rent pandemic of which we might ask: have neat narratives and partisan Most researchers have experts always been explicit in acknowledging presentation; strive to heard such exhortations many unknowns? Complexity? Conflicts of interest? inform, not persuade. times, and for good reason. Inconvenient data? And, importantly, their “BSuch rhetorical devices often help to land the own values? Rather than re-examine those message, whether that message is designed cases, we offer ideas to encourage reflection, to sell a product or win a grant. These are the based on our own research. traditional techniques of communications Our small, interdisciplinary group at applied to science. the University of Cambridge, UK, collects This approach often works, but it comes empirical data on issues such as how to com- with danger. municate uncertainty, how audiences decide

what evidence to trust, and how narratives be To Inform and Not Persuade.” The media affect people’s decision-making. Our aim is might urge us to aim for memorable sound to design communications that do not lead bites or go beyond the strength of the data: people to a particular decision, but help them be honest and aware of such traps. Quick tips for to understand what is known about a topic and to make up their own minds on the basis Offer balance, not false balance sharing evidence of that evidence. In our view, it is important to We can’t inform people fully if we don’t convey be clear about motivations, present data fully the balance of relevant evidence. The aim is to ‘inform but not persuade’, and and clearly, and share sources. We are all subject to a suite of psychological — as the philosopher of trust Onora O’Neill We recognize that the world is in an biases that mean we sometimes apply evidence says — “to be accessible, comprehensible, ‘infodemic’, with false information spread- to shore up our own beliefs, and find it difficult usable and assessable”. ing virally on social media. Therefore, many to accept evidence that goes against our ideas scientists feel they are in an arms race of and hypotheses. People also like to tell (and • Address all the questions and concerns of communication techniques. But consider the hear) stories that don’t meander through the target audience. replication crisis, which has been blamed in thickets of opposing opinions or pros and cons. • Anticipate misunderstandings; part on researchers being incentivized to sell But evidence communicators must challenge pre-emptively debunk or explain them. their work and focus on a story rather than on these instincts and offer evidence in the round. • Don’t cherry-pick findings. full and neutral reporting of what they have Partial presentation of evidence crops up • Present potential benefits and possible done. We worry that the urge to persuade or across scientific literature and in the public harms in the same way so that they can be to tell a simple story can damage credibility domain. Often, the argument made is that compared fairly. and trustworthiness. people can’t take in lots of information at once. • Avoid the biases inherent in any Instead, we propose another approach. We If you’re presenting written information, you presentation format (for example, use both call it evidence communication. can make it easier for them. Here’s a simple ‘positive’ and ‘negative’ framing together). tip from research in medical communication: • Use numbers alone, or both words and Inform, not persuade display the pros and cons in a table rather than numbers. Conventional communication techniques stating them in the text. Imagine a table com- • Demonstrate ‘unapologetic uncertainty’: might ‘work’ when the aim is to change paring proposed transmission-prevention be open about a range of possible people’s beliefs or behaviours. But should policies that lays out the projected harms and outcomes. that always be our aim? benefits of each policy in terms of mortality, • When you don’t know, say so; say what Early in the pandemic, we surveyed people morbidity, economics, environment and you are going to do to find out, and by across 13 countries from South Korea to mental health, breaking down subgroups and when. Mexico and asked what sources of information timescales. For your audiences, knowing what • Highlight the quality and relevance of the they trusted. We also asked them why. Their the key pros and cons are is crucial. underlying evidence (for example, describe answers show how sensitive people are to the We neglect people's interests at our peril. the data set). aims and interests of communicators. As soon as we are perceived to be ignoring or • Use a carefully designed layout in a clear “They sell news, not truth,” said one UK underplaying something our audience consid- order, and include sources. respondent about journalists; “I believe the ers important, our motivations — and hence Government are being guided by scientists our trustworthiness — will be questioned. spongiform encephalopathy. He adopted the and genuinely care about the population,” As one of the Australian participants in our following strategy: say what you know; what you said another; “WHO is paid by China,” replied COVID-19 survey in March said about their don’t know; what you are doing to find out; what a respondent from Japan. Friends and family reason for distrust of official information: “Are people can do in the meantime to be on the safe were often warmly described as having “no they hiding the full effects from us?” side; and that advice will change3. We check any- reason to lie”. one talking to the public about COVID-19 against These observations fit with the literature, Disclose uncertainties this list, especially the second point. which identifies expertise, honesty and good Part of telling the whole story is talking about New Zealand’s response to the pandemic intentions as the key to trustworthiness1. what we don’t know. has been praised. And the country’s Ministry Researchers need to demonstrate all three: The simplest argument for stating of Health web page on COVID-19 test results we cannot expect to be trusted on the basis uncertainty is that what we think we know is includes several paragraphs describing uncer- of expertise alone. constantly changing (wearing face coverings tainties, including the likelihood of a false neg- So how do we demonstrate good is an example). One of us (M.B.), writing with ative (meaning that a test says someone’s not intentions? We have to be open about our others in the medical journal BMJ, admitted infected when they actually are). The US Centers motivations, conflicts and limitations. that at some point, all three authors had been for Disease Control and Prevention page men- Scientists whose objectives are perceived wrong about COVID-19 (ref. 2). So, either we tions no such uncertainties. Neither does the as prioritizing persuasion risk losing trust. had better be certain, and right — or we should UK National Health Service website (despite During the COVID-19 crisis, one of us (D.S.) more humbly state our uncertainties. us raising the issue with them): it was deemed has frequently had to refuse journalists who When zoologist John Krebs became chair of too confusing. Even with a highly accurate test, tried to draw him away from his intention to the UK Food Standards Agency in the 2000s, thousands of people get false negatives and false stick to statistical evidence. As he told The he faced a deluge of crises, including dioxins assurance that could lead to risky behaviours. Times, “The banner across my T-shirt should in milk and the infectious cattle disease bovine When we trialled the wording with and

Comment without the explicit uncertainties around the attempts to sow doubt (known as prebunking), repeating: researchers should not take up test result, we found that the uncertainties they resist being swayed by misinformation or the reins of rhetoric blindly or feel that they did not seem to undermine trustworthiness. disinformation6–8. should always harness the tools used by the However, the wordings did affect people’s per- Prebunking requires anticipating potential media and entertainment industries to shape ception of whether the test recipient should misunderstandings or disinformation attacks, people’s emotions and beliefs. Nor should they isolate if they got a negative result. In other and that means understanding the concerns of assume that they are apolitical, unbiased and words, people correctly interpreted the mes- the audience. Read public forums and popular utterly objective — all of us have values, beliefs sages without having their trust undermined news sources. Consider what decisions your and temptations. Even if we choose to be an by an upfront description of uncertainties. audiences are making and what informa- ‘honest broker’, the first person we need to be Other research finds little downside in tion — in what format — would best support honest with is ourselves. expressing findings as a range (such as these, from whether to wear a face covering to In our research across ten countries, we ‘between x and y’) rather than an exact num- see how people’s willingness to be vaccinated ber4. Often, the degree of uncertainty is part “The urge to persuade or against COVID-19 correlates with their back- of the core message. In January 2018, the ground levels of trust in science, scientific BBC News website announced that over the to tell a simple story can researchers and doctors, alongside their worry three months to the previous November, damage credibility and about the virus and their general beliefs in the “UK unemployment fell by 3,000 to 1.44 trustworthiness.” efficacy of vaccines. million”. Left unmentioned (because the UK Trust is crucial. Always aiming to ‘sell the Office of National Statistics made it hard to science’ doesn’t help the scientific process or find) was the fact that the margin of error was whether to take a vaccine. Consider the costs the scientific community in the long run, just ±77,000. (We are heartened that the Office of and benefits as they see them. as it doesn’t help people (patients, the public National Statistics has since started reporting When we developed a web tool about or policymakers) to make informed decisions ranges more prominently, and we have seen treatments for women newly diagnosed with in the short term. That requires good evidence journalists follow this lead.) breast cancer, we read the comments on patient communication. Ironically, we hope we’ve forums. This revealed that people wanted to persuaded you of that. State evidence quality know the magnitude of survival benefit and For more on evidence communication, see Audiences also judge the credibility of of possible harms. For example, one woman Supplementary information. information based on the quality of the under- said that a 1% improvement in survival was not lying evidence, more than its clarity, the usual worth the side effects of the drug tamoxifen (we priority for a communications department. paraphrase to preserve confidentiality). The The authors Here’s a sign of how readily audiences pick information we ended up presenting was more out cues for quality of evidence. In a study to complex and what people wanted to know. Michael Blastland is a board member learn what formats work best for presenting and Alexandra L. J. Freeman is executive medical data, we used a version of the phrase What next? director at the Winton Centre for Risk and “x out of 100 people suffered this side effect”, The field of evidence communication has Evidence Communication, University of and about 4% of all participants took the time been growing over several decades, mainly Cambridge, UK. Sander van der Linden is to write in an open response box that a sample stemming from researchers in medical com- a board member at the Winton Centre and size of 100 people was not enough5. This was a munication, but there is still much we don’t director of the Cambridge Social Decision- misunderstanding due to our choice of words. know about its effects, or best practice. If one Making Lab, University of Cambridge, UK. We did not literally mean 100 people, but it is is not trying to change belief or behaviour, it’s Theresa M. Marteau is a board member at the notable that the participants were not scien- hard even to know how to measure success. Winton Centre and director of the Behaviour tific researchers or even students: they were Like all engagement efforts, many of the and Health Research Unit, University of representative UK residents (120 of the 1,519 effects of a message are moderated greatly Cambridge, UK. David Spiegelhalter is chair respondents who left unsolicited comments by non-verbal cues and the relationships of the Winton Centre. overall mentioned sample size). between communicator and audience. But e-mails: [email protected]; As scientists, we tend to underestimate the these challenges are why we think it important [email protected] sophistication of our audiences’ sensitivity to to consider alternative approaches (see ‘Quick cues of quality and how these affect trust. In tips for sharing evidence’). Supplementary information accompanies this practical terms, overtly stating that a piece of In some fields, such as conservation science Comment: see go.nature.com/3pivy6v evidence is of high or low quality is unsubtle or public health, researchers might, depend- but definitely noticed by a significant propor- ing on the circumstances, feel that they should 1. White, M. P. & Eiser, J. R. in Trust in Risk Management: Uncertainty and Scepticism in the Public Mind. Ch. 4, tion of a non-specialist audience. People in our become advocates of their subject, advancing 95–117 (Taylor & Francis, 2010). surveys also ask to know the size and source of their positions with ‘every trick in the book’. 2. Smith, G. D., Blastland, M. & Munafò, M. Br. Med. J. 371, data sets, so that they can gauge relevance to Indeed, all researchers are “partisan advo- m3979 (2020). 3. Champkin, J. Significance 10, 23–29 (2013). them. Such information should be provided. cates of the validity and importance of their 4. van der Bles, A. M., van der Linden, S., Freeman, A. L. J. work”, according to a recent study9. There is a & Spiegelhalter, D. J. Proc. Natl Acad. Sci. USA 117, 7672– Inoculate against misinformation continuum from ‘informing’ to ‘persuading’ — 7683 (2020). 5. Brick, C., McDowell, M. & Freeman, A. L. J. R. Soc. Open Many will worry that following these key and researchers should choose their position Sci. 7, 190876 (2020). principles — especially exposing complexities, on it consciously. Political and professional 6. Roozenbeek, J. & van der Linden, S. Palgrave Commun. 5, uncertainties or unwelcome possibilities — communicators often have aims and obliga- 65 (2019). 7. Maertens, R., Roozenbeek, J., Basol, M. & van der will let ‘merchants of doubt’ or bad actors tions that push them towards persuasion, Linden, S. J. Exp. Psychol. Appl. https://doi.org/10.1037/ warp their message. But there are other ways whereas scientists should feel more free to xap0000315 (2020). to guard against this. Research on climate judge what is appropriate. 8. van der Linden, S., Leiserowitz, A., Rosenthal, S. & Maibach, E. Glob. Challenges 1, 1600008 (2017). change, COVID-19 and other topics shows that Many researchers do an excellent job 9. Leng, G. & Leng, R. I. The Matter of Facts: Skepticism, if people are pre-emptively warned against of engaging with the public. Still, it bears Persuasion, and Evidence in Science (MIT Press, 2020).