Health and risk communication: what the trials say about what readers understand

Year 3 / Module 33 Data visualisation in practice / Unit 8 / Health and risk communication: what the trials say about what readers understand

Health and risk communication: what the trials say about what readers understand

HCI3109 · Unit 8 of 10

The health and risk literature is the one place where visual communication has randomised evidence, and it shows that format decided in the design changes comprehension, which makes it the strictest test of the module's other claims.

Health and risk communication: what the trials say about what readers understand

Every other unit here argues that format matters, supporting the argument with perception studies, interaction studies or practice. This eighth of ten units finds the one area where the claim has been tested close to the way that would settle it. Studies there give the same numbers to comparable readers in different formats, and they measure the reader's understanding. That makes it the module's strictest test, because what it shows is narrower than the slogan that pictures help.

A chart carrying a risk presents a decision rather than a number. The first question is whether a risk is a share of a whole or a bare count. The second is whether a benefit is a proportional reduction or a change in the number helped. The evidence is stronger here, because it comes from randomised and nationally representative experiments and its reviewers report a heterogeneous literature. What studies measure is more often comprehension, liking or intention than the decision itself.

The founding claim is that representation, not capacity, decides understanding

Slovic (1987) gave the field its frame: risk perception examines the judgements people make when characterising and evaluating hazards. A risk, on that account, is something a reader judges, which makes its representation part of the object judged. Gigerenzer and Hoffrage (1995) showed that the representation does the work. They analysed several thousand solutions to Bayesian problems. With information in frequency formats, statistically naive participants derived up to 50 per cent of all inferences by Bayesian algorithms. The probability formats had hidden that competence.

Gigerenzer, Gaissmaier, Kurz-Milcke, Schwartz and Woloshin (2007) turned that into a diagnosis of the communicator. A statement that screening reduces the risk of dying by 25 per cent in fact means one fewer death in 1,000. Illiteracy is created by non-transparent framing. Gigerenzer and Edwards (2003) showed how far this reaches into practice. They gave experienced doctors a screening question with a 50 per cent sensitivity and a 3 per cent false positive rate; the doctors estimated the probability between 1 and 99 per cent, when the correct answer is about 5 per cent.

Gigerenzer and colleagues (2007) name a second misreading that restating the numbers does not repair. Many people do not know that higher survival rates under cancer screening do not imply longer life. Arithmetic is not the obstacle here. Survival is counted from the date of diagnosis, and screening moves that date earlier. A metric fixed this way can improve while deaths and lifespans do not move at all.

The winning formats are numerical before they are pictorial

Gigerenzer and colleagues (2007) asked for frequency statements in place of single-event probabilities, absolute risks in place of relative risks and natural frequencies. Fagerlin, Zikmund-Fisher and Ubel (2011) gathered ten methods reported to improve patients' understanding, from absolute risks as frequencies to pictographs and plain language. Trevena and colleagues (2013) advised that a format should reflect the reader's task. It should define a reference class over time, avoid variable denominators, and account for numeracy and graph literacy.

The strongest result is about a frame rather than a figure. Covey (2007) meta-analysed twenty-four articles reporting thirty-one experiments. Treatments were evaluated more favourably under the relative risk format than under absolute risk or number needed to treat.

The systematic reviews refuse the simple rule that pictures help

Ancker, Senathirajah, Kukafka and Starren (2006) reviewed work on health risk graphs. The features improving quantitative reasoning appear to differ from those most likely to change behaviour or intentions. Graphs showing only the numerator appeared to inflate perceived risk, the effect Stone, Yates and Parker (1997) measured directly, and readers preferred simplicity and familiarity not associated with accurate judgement. Lipkus (2007) found the evidence base too thin for confident recommendations, because formats had rarely been tested against one another on the same outcomes.

Icon arrays are the clearest positive result, and it is conditional

Galesic, Garcia-Retamero and Gigerenzer (2009) ran two experiments with fifty-nine older adults and 112 university students. Icon arrays raised risk-understanding accuracy for both low and high numeracy readers, but the risks shown were perceived as less serious. Arrays of 1,000 icons raised perceived seriousness and risk reduction relative to arrays of 100. Garcia-Retamero and Galesic (2010) then tested the format in national samples in the United States and Germany, adding icon arrays and bar graphs to numerical information in absolute or relative frames. Accuracy improved with either aid, and most when the aid depicted the entire population at risk.

The denominator is the design decision the trials single out

Reyna and Brainerd (2008) named the two failures the unit keeps meeting. The first is a ratio bias, in which higher frequencies bias probability judgements. The second is denominator neglect, in which attention fixes on the number of events rather than the opportunities. Schwartz, Woloshin and Welch (2007) showed that a plain table can carry the denominator. The authors gave 274 participants a nine-row by two-column table on tamoxifen. On average they answered four of five comprehension questions correctly, and 71 per cent computed the absolute difference in cancer proportions between drug and placebo.

Comprehension is not a decision, and the evidence thins at the choice

The literature's measures are mostly comprehension, liking or intention, and the largest review has now measured the choice as well. Stacey and colleagues (2024) pooled 209 randomised trials of decision aids with 107,698 participants, finding high-certainty improvement in knowledge and risk-perception accuracy and moderate-certainty improvement in choices congruent with informed values. Zikmund-Fisher, Sarr, Fagerlin and Ubel (2006) asked 2,399 people to imagine themselves in one of four roles and then choose a treatment. Preferences for the risk-reducing option were much stronger in the professional roles: 73 per cent for the physician and 63 per cent for the medical director, against 48 per cent for the patient.

The contested question: does format decide comprehension, or only shift it?

The first position is that the trials prove the module's general claim where it could be proved. Gigerenzer and Hoffrage (1995) recovered a competence the probability format concealed. Galesic and colleagues (2009) and Garcia-Retamero and Galesic (2010) raised accuracy in the laboratory and in national samples. Gigerenzer and colleagues (2007) and Trevena and colleagues (2013) built it into advice and instruments.

The second holds that the same evidence supports a weaker claim. Covey (2007) traced most of the heterogeneity to the wording, which is evidence about a sentence rather than a design. Lipkus (2007) found the evidence base too thin for confident recommendations. Ancker and colleagues (2006) found that the features improving accuracy differed from those readers preferred, and that interpretation often required instruction. A benefit measured with a briefing may not survive without one.

The dispute is about the dependent variable. If the decision is, the second is stronger, because few studies measure one. What would settle it is a pre-registered trial that varies format while holding the numbers and task fixed. It would pre-specify the wording, so Covey's moderator is controlled. It would record numeracy and graph literacy. It would measure a decision with a downstream consequence rather than a comprehension score.

The two sides rest on different evidence. The strong side rests on randomised and nationally representative experiments (Gigerenzer and Hoffrage, 1995; Galesic and colleagues, 2009; Garcia-Retamero and Galesic, 2010). The weaker side rests on a meta-analysis of 31 experiments (Covey, 2007), a systematic review (Ancker and colleagues, 2006) and a review that found the base too thin (Lipkus, 2007). While the dispute is open, a designer should give absolute risks as frequencies, fix the denominator, and measure a decision rather than a comprehension score. The strongest objection is that the pooled result tests decision aids as a package rather than one format, so the claim that format changes understanding remains only partly settled.

Where this is heading

Trevena and colleagues (2013) close their primer with interactive web-based formats, which a reader on a metered connection may never load, Fagerlin and colleagues (2011) move towards plain language, and Ancker and colleagues (2006) set communication of uncertainty about risk as continuing research. The open question is whether better comprehension becomes a better decision. Stacey and colleagues (2024) have now answered part of it.

The sources this unit rests on

Each note states what the cited work did and found and what it adds to the argument. Identifiers and metadata were checked on 2 November 2026.

Ancker, J. S. et al. (2006). Design Features of Graphs in Health Risk Communication: A Systematic Review. Journal of the American Medical Informatics Association, 13(6), 608-618. doi:10.1197/jamia.m2115 Reviewed research on health risk graphs; found the features improving reasoning differ from those changing behaviour and that numerator-only graphs inflate perceived risk. Curbs the assumption that graphics beat text.
Covey, J. (2007). A Meta-analysis of the Effects of Presenting Treatment Benefits in Different Formats. Medical Decision Making, 27(5), 638-654. doi:10.1177/0272989x07306783 Meta-analysed twenty-four articles reporting thirty-one experiments; relative risk formats produced more favourable evaluations of treatment than absolute risk or number needed to treat, with heterogeneity explained by the wordings chosen. Shows the frame moves the reader.
Fagerlin, A., Zikmund-Fisher, B. J. and Ubel, P. A. (2011). Helping Patients Decide: Ten Steps to Better Risk Communication. JNCI Journal of the National Cancer Institute, 103(19), 1436-1443. doi:10.1093/jnci/djr318 A commentary presenting ten methods reported to improve patients' understanding, from absolute risks as frequencies to pictographs and plain language. Carries the advice into practice.
Galesic, M., Garcia-Retamero, R. and Gigerenzer, G. (2009). Using icon arrays to communicate medical risks: Overcoming low numeracy. Health Psychology, 28(2), 210-216. doi:10.1037/a0014474 Ran two icon-array experiments with older adults and students; icon arrays raised accuracy for low and high numeracy readers, the risks shown were perceived as less serious, and 1,000-icon arrays raised perceived seriousness. The clearest positive result.
Garcia-Retamero, R. and Galesic, M. (2010). Who profits from visual aids: Overcoming challenges in people's understanding of risks. Social Science & Medicine, 70(7), 1019-1025. doi:10.1016/j.socscimed.2009.11.031 Compared adding icon arrays and bar graphs to numerical information in absolute or relative frames, in national samples in Germany and the United States. Either aid improved accuracy, most when it depicted the whole population at risk. The benefit depends on what the picture counts.
Gigerenzer, G. and Hoffrage, U. (1995). How to improve Bayesian reasoning without instruction: Frequency formats. Psychological Review, 102(4), 684-704. doi:10.1037/0033-295x.102.4.684 Analysed several thousand solutions to Bayesian problems; found that with information in frequency formats, statistically naive participants derived up to 50 per cent of all inferences by Bayesian algorithms. The founding demonstration.
Gigerenzer, G. and Edwards, A. (2003). Simple tools for understanding risks: from innumeracy to insight. BMJ, 327(7417), 741-744. doi:10.1136/bmj.327.7417.741 Analysed how doctors present statistical information, using a screening question with 50 per cent sensitivity and a 3 per cent false positive rate; doctors estimated the correct answer of about 5 per cent between 1 and 99 per cent. Argues the problem lies in the representation.
Gigerenzer, G. et al. (2007). Helping Doctors and Patients Make Sense of Health Statistics. Psychological Science in the Public Interest, 8(2), 53-96. doi:10.1111/j.1539-6053.2008.00033.x Reviewed evidence on statistical illiteracy across patients, journalists and physicians, using the example that a 25 per cent reduction in the risk of dying of breast cancer means one fewer death in 1,000, and argued illiteracy is created by non-transparent framing.
Lipkus, I. M. (2007). Numeric, Verbal, and Visual Formats of Conveying Health Risks: Suggested Best Practices and Future Recommendations. Medical Decision Making, 27(5), 696-713. doi:10.1177/0272989x07307271 Reviewed the empirical literature on conveying risk magnitudes; found few overall recommendations could be made, citing inconsistent testing on the same outcomes and a shortage of randomised comparisons of formats. The evidence base is thin.
Reyna, V. F. and Brainerd, C. J. (2008). Numeracy, ratio bias, and denominator neglect in judgements of risk and probability (the published title uses the American spelling). Learning and Individual Differences, 18(1), 89-107. doi:10.1016/j.lindif.2007.03.011 Reviewed research on numeracy in health judgement, showing many adults fail simple ratio problems and that higher frequencies bias probability judgements; named ratio bias and denominator neglect. Names the failure this unit designs against.
Schwartz, L. M., Woloshin, S. and Welch, H. G. (2007). The Drug Facts Box: Providing Consumers with Simple Tabular Data on Drug Benefit and Harm. Medical Decision Making, 27(5), 655-662. doi:10.1177/0272989x07306786 Tested comprehension of a nine-row by two-column table about tamoxifen with 274 participants; four of five comprehension questions were answered correctly on average and 71 per cent computed the absolute difference between drug and placebo. A table doing a chart's work.
Slovic, P. (1987). Perception of Risk. Science, 236(4799), 280-285. doi:10.1126/science.3563507 Framed risk perception as the examination of the judgements people make when characterising and evaluating hazards, in order to anticipate public responses and improve communication. Established risk as something judged rather than received.
Stacey, D. et al. (2024). Decision aids for people facing health treatment or screening decisions. Cochrane Database of Systematic Reviews, 2024(1), CD001431. doi:10.1002/14651858.cd001431.pub6 Pooled 209 randomised trials with 107,698 participants. Found high-certainty gains in knowledge and accurate risk perceptions, and moderate-certainty gains in choices matching informed values.
Stone, E. R., Yates, J. F. and Parker, A. M. (1997). Effects of numerical and graphical displays on professed risk-taking behaviour. Journal of Experimental Psychology: Applied, 3(4), 243-256. doi:10.1037/1076-898x.3.4.243 Gave participants paired fictitious products and compared numerical probabilities with a graph showing only the number harmed; the graph without the part-to-whole relation raised what people would pay for the safer product. The primary experiment behind the numerator claim the review reports.
Trevena, L. J. et al. (2013). Presenting quantitative information about decision outcomes: a risk communication primer for patient decision aid developers. BMC Medical Informatics and Decision Making, 13(Suppl 2), S7. doi:10.1186/1472-6947-13-s2-s7 Built a primer from a consensus group of fourteen, identifying eleven components of risk communication and advising that a format should reflect the reader's task, define a reference class over time and avoid variable denominators. Turns the denominator into a requirement.
Zikmund-Fisher, B. J. et al. (2006). A matter of perspective: choosing for others differs from choosing for yourself in making treatment decisions. Journal of General Internal Medicine, 21(6), 618-622. doi:10.1111/j.1525-1497.2006.00410.x Asked 2,399 people from a stratified internet sample to imagine themselves as a patient, a physician, a medical director or a parent, then to choose a treatment; preferences for the risk-reducing option were much stronger in the professional roles. The role, not the chart, moved the choice.

Check your understanding & apply it · 4 questions, 1 essay

1A leaflet states that a treatment reduces the risk of dying by 25 per cent. Using the unit's evidence, which prediction about how readers judge the treatment is best supported?

2A page reports that '3 people died' without stating the size of the population, and readers take the risk to be large. Which diagnosis does the unit's evidence support?

3A screening programme wants a single aid that will raise risk understanding in a population with a wide range of numeracy. Using the unit's evidence, which judgement about the aid is defensible?

4A programme claims: 'The trials prove that format decides comprehension.' Which judgement does the unit's reading of the evidence support?

5Essay A national bowel screening programme has asked for a web page that tells people aged 60 to 74 that the screening test reduces the risk of dying of bowel cancer by 25 per cent. The page will be read on phones by people with a wide range of numeracy, in several community languages, and the programme wants more people to take the test. You have the underlying numbers: in a comparable population, about 3 people in 1,000 die of bowel cancer without screening and about 2 in 1,000 die with it. Using this unit's sources, describe the page you would build, naming the format decisions you would make and the evidence behind each, say what you would refuse to do and why, predict before release which misunderstanding is most likely, name the one test you would run, and then state the strongest counter-argument to your design and answer it.

I would keep the 25 per cent figure off the page as the headline, because it is a relative risk reduction and Covey (2007) found that relative risk formats produce more favourable evaluations of treatments than absolute risk or number needed to treat formats, which means the figure moves the reader while telling them little about what will happen to them. Following Gigerenzer and colleagues (2007), I would present the same information as natural frequencies with an explicit denominator, that about 3 people in 1,000 die of bowel cancer without screening and about 2 in 1,000 die with it, so the reader can see that the 25 per cent reduction is one fewer death in 1,000. I would state the reference class and the time frame, as Trevena and colleagues (2013) require, and avoid the 1 in x form and any change of denominator between sentences. For the visual, I would use an icon array of 1,000 figures showing the whole population at risk rather than only those affected, because Garcia-Retamero and Galesic (2010) found the largest accuracy gains when the aid depicted the entire population at risk, and I would accept that a larger array changes how serious the risk feels, since Galesic and colleagues (2009) found arrays of 1,000 icons raised perceived seriousness relative to arrays of 100. I would refuse the 25 per cent headline and refuse any figure without its denominator, because Reyna and Brainerd (2008) describe denominator neglect as the failure this design is there to prevent, and I would refuse decoration that increases liking without improving accuracy, which Ancker and colleagues (2006) found were different properties. Before release I would predict that readers take the screening as preventing most bowel cancer deaths rather than one in 1,000, the error Covey's wording finding and Reyna and Brainerd's denominator neglect both predict. The test I would run is a comprehension check that asks readers to recover both the absolute number helped and the denominator and to state the case against taking the test, not a rating of how clear the page looked. The strongest counter-argument is that a bare icon array and a frequency sentence may be less persuasive to people who avoid numbers, so fewer may take the test. I would answer it by noting that Lipkus (2007) found few formats for conveying risk had been tested against one another on the same outcomes, so persuasion is an open question rather than an established benefit of opacity, and that the programme's aim of informed uptake is not served by a figure that inflates the benefit.