How it works

An AI translates the study

We don't read the papers ourselves. We built an AI algorithm that takes a study and translates it into numbers a non-statistician can understand.

Why we do it

The goal is to avoid getting carried away by exaggerated relative numbers and to think about effect size more critically. It also makes reading a bit easier, because you get a list of criticism from people who know what they are doing.

We find the criticism, you check it

We gather published criticism for you, but you are responsible for checking the claims. You can comment critically on our work and help us make it better.

Where the fact box format comes from

The fact box was developed by Gerd Gigerenzer's group at the Harding Center for Risk Literacy. It shows what happened to the same number of people with and without a treatment, in absolute numbers, with benefits and harms side by side. We follow their published guide and methods paper. Our own additions are the switch between 100 and 1,000 people and the marking of the difference. The Harding Center did not produce, review or endorse this site.

1. McDowell M, Rebitschek FG, Gigerenzer G, et al. A simple tool for communicating the benefits and harms of health interventions. MDM Policy Pract. 2016;1(1). doi:10.1177/2381468316665365
2. Harding Center for Risk Literacy. Methods paper, version 2.2. 2021. hardingcenter.de
3. Aubertin P, Hinneburg J, Hille L, et al. Fact Boxes: What gets through? A focus group study. Z Evid Fortbild Qual Gesundhwes. 2022;168:96–105. doi:10.1016/j.zefq.2021.12.011 Our reason for trying a page on a few readers before use.

Studies that report a chance

Some studies give their main result as a chance: for example, a 94% probability that deaths were higher with one drug. These are Bayesian analyses. The chance says how sure the direction is. It does not say how big the difference is: a difference can be very likely and still small.

So for these studies every row of the fact box answers two questions separately: how big the difference was, in people out of 100 or 1,000, and how sure the direction is, as a chance in %. These are the choices behind that display, and what each one rests on.

• The absolute numbers stay, with the chance beside them. A high chance of a small difference is not evidence of a large one (Goligher 2024; Uslan and Smith 2026).
• The chance is given for more than one size of difference. In one trial the chance of any benefit was 88% to 99%, and the chance of a large benefit 0% to 48%, depending on the starting assumptions (Goligher 2018).
• We do not say which size matters. No fixed chance counts as proof. What is acceptable depends on cost and on the alternatives, so that is left to the reader (Goligher 2024).
• The chance is drawn as 50 dots. People judged chances more precisely from dots they could count than from a curve. With 100 dots they stopped counting; 50 dots gave the best decisions (Kay 2016; Fernandes 2018).
• The dots carry no marker for the average. Adding averages to such pictures led people to discount the uncertainty (Kale 2020).
• A word never stands without its number. The same word means very different chances to different people, and stating uncertainty in numbers cost less trust than stating it in words (Hashim 2024; Andreadis 2021; van der Bles 2020; EFSA 2018 and 2019; Budescu 2014).
• The words follow a published scale. “Very likely” starts at 90% and “likely” at 66%. This is the scale of the Intergovernmental Panel on Climate Change, which the European Food Safety Authority adapted (Mastrandrea 2010; EFSA 2018).
• The dots are explained as a differential diagnosis. The true diagnosis is not known yet, so the doctor gives each differential diagnosis a chance, and together the chances make 100%. No paper here: this wording was tried on one reader, after two others had failed.

Two limits. The display studies tested lay people on everyday decisions, such as when a bus would arrive, not doctors reading a study. And the dots are drawn from a curve fitted to each study's printed result, not from the study's own calculations.

1. Andreadis K, Chan E, Park M, et al. Imprecision and preferences in interpretation of verbal probabilities in health: a systematic review. J Gen Intern Med. 2021;36(12):3820–3829. doi:10.1007/s11606-021-07050-7
2. Budescu DV, Por H-H, Broomell SB, Smithson M. The interpretation of IPCC probabilistic statements around the world. Nat Clim Chang. 2014;4(6):508–512. doi:10.1038/nclimate2194 Read as a university press release only.
3. EFSA Scientific Committee. Guidance on uncertainty analysis in scientific assessments. EFSA J. 2018;16(1):5123. doi:10.2903/j.efsa.2018.5123
4. European Food Safety Authority. Guidance on communication of uncertainty in scientific assessments. EFSA J. 2019;17(1):5520. doi:10.2903/j.efsa.2019.5520
5. Fernandes M, Walls L, Munson S, Hullman J, Kay M. Uncertainty displays using quantile dotplots or CDFs improve transit decision-making. Proc CHI. 2018. doi:10.1145/3173574.3173718
6. Goligher EC, Tomlinson G, Hajage D, et al. Extracorporeal membrane oxygenation for severe acute respiratory distress syndrome and posterior probability of mortality benefit in a post hoc Bayesian analysis of a randomized clinical trial. JAMA. 2018;320(21):2251. doi:10.1001/jama.2018.14276 Read as an abstract only.
7. Goligher EC, Heath A, Harhay MO. Bayesian statistics for clinical research. Lancet. 2024;404(10457):1067–1076. doi:10.1016/S0140-6736(24)01295-9
8. Hashim MJ. Verbal probability terms for communicating clinical risk – a systematic review. Ulster Med J. 2024;93(1):18–23. europepmc.org
9. Kale A, Kay M, Hullman J. Visual reasoning strategies for effect size judgments and decisions. IEEE VIS. 2020. arXiv:2007.14516. arxiv.org
10. Kay M, Kola T, Hullman JR, Munson SA. When(ish) is my bus? User-centered visualizations of uncertainty in everyday, mobile predictive systems. Proc CHI. 2016. doi:10.1145/2858036.2858558
11. Mastrandrea MD, Field CB, Stocker TF, et al. Guidance note for lead authors of the IPCC Fifth Assessment Report on consistent treatment of uncertainties. Intergovernmental Panel on Climate Change; 2010. ipcc.ch
12. Uslan DZ, Smith EA. Cefepime and mortality—a dosing problem, not a drug problem. JAMA Netw Open. 2026;9(9):e2632904. doi:10.1001/jamanetworkopen.2026.32904
13. van der Bles AM, van der Linden S, Freeman ALJ, Spiegelhalter DJ. The effects of communicating uncertainty on public trust in facts and numbers. Proc Natl Acad Sci USA. 2020;117(14):7672. doi:10.1073/pnas.1913678117