logo
Loading...
Back to all articles
RB 1017 minNov 20, 2026

How to Promote an Observational Study Without Implying Cause

By Tarun Singhal

Share:
How to Promote an Observational Study Without Implying Cause

Your cohort paper is out, and the result is genuinely interesting: people with more of one thing had less of another. In the paper you wrote "associated with" every time, adjusted for the confounders you could measure and spent a paragraph on what you could not.

Then the press office drafts a headline, a colleague shares it, and by the third retelling the association has become an instruction. Promoting an observational study well mostly means keeping the verb you chose.

Why Observational Results Drift Into Causes

Readers want to know what to do. A sentence that says "X is associated with Y" invites them to turn it into "do more X", and every shortening step makes that easier.

The drift has been measured. In a BMJ study of 462 press releases on biomedical and health research issued by 20 leading UK universities in 2011, Sumner and colleagues compared each release with its journal article and with 668 news stories. For correlational results (cross-sectional and longitudinal observational data), 33% of press releases made the main claim more strongly causal than the paper did.

What happened next is the part that matters for you. When the press release contained an exaggerated causal claim, 81% of the related news stories did too. When it did not, the figure was 18%. The study itself is observational, so it shows an association rather than proof that releases cause news exaggeration, but the pattern is hard to ignore: the news tended to inherit the wording it was handed.

The same pattern shows up at the far end of the pipeline. The CLAIMS systematic review in PLOS ONE looked at the 50 academic articles linking an exposure to a health outcome that were most shared on Facebook and Twitter in 2015, along with 64 media articles about them. Reviewers judged that 34% of the academic studies and 48% of the media articles used language too strong for the study's strength of causal inference, and that 58% of the media articles misreported the question, results, intervention or population. The authors say their sample was built to reflect what got shared, so it is not representative of all research.

A Wording Ladder for Your Drafts

Sumner and colleagues rated statements on a scale of increasing determinism. Their examples, all about wine and cancer, make a useful ladder to check promotional copy against.

LevelExample wording from the studyWhen it fits an observational result
Correlational"Drinking wine is associated with increased cancer rates"Your default for a cohort, case-control or cross-sectional finding
Ambiguous"Drinking wine linked to cancer risk"One step up from correlational on the scale; add context so it is not read as cause
Conditional causal"Drinking wine might increase cancer risk"Only alongside a clear statement that the design cannot show cause
Can cause"Drinking wine can increase cancer risk"Usually beyond what one observational study supports
Unconditionally causal"Drinking wine increases cancer risk"Not a summary of a correlational result

Source: the coding scale in Sumner and colleagues, BMJ 2014. The right-hand column is this guide's advice, not part of the study.

The same paper explains why the bottom rung is hard to reach: a correlation between wine and a disease could arise because wine raises the risk, because the disease changes drinking, or because drinking travels with some other factor that raises the risk. That sentence, adapted to your own variables, is often the most useful caveat you can add.

What to Keep Next to the Result

The STROBE statement is the reporting guideline for cohort, case-control and cross-sectional studies. The STROBE Explanation and Elaboration paper describes it as a checklist of 22 items, and the quotations below come from that paper. Several items translate directly into what a post, slide or press release should carry.

  • Item 1(b): an "informative and balanced summary of what was done and what was found". Name the design in your first sentence, so readers know they are looking at an association.
  • Item 16(a): give unadjusted and confounder-adjusted estimates with their precision, and make clear which confounders were adjusted for and why. In a short format, one line on what you adjusted for is enough.
  • Item 16(c): "If relevant, consider translating estimates of relative risk into absolute risk for a meaningful time period." Give the absolute numbers alongside the relative one, so readers can see how many people are affected.
  • Item 20: give "a cautious overall interpretation" that considers limitations, multiplicity of analyses, results from similar studies and other relevant evidence.
  • Item 21: discuss generalisability. Say who was studied, so readers do not assume it applies to everyone.

Adjusted is not the same as causal

The Explanation and Elaboration paper makes a point worth repeating in your own words when you promote the study: readers may think adjusted results equal the causal part of the association, but adjusted results are not necessarily free of random error, selection bias, information bias or residual confounding. Saying "after adjusting for age, income and baseline health" is good practice. Implying that adjustment settled the question is not.

Caveats do not seem to cost you coverage

Researchers and press officers often worry that a caveat will kill the story. In the same BMJ study, caveats were rare in press releases, and the authors found no evidence that they reduced news uptake. They also found little evidence that exaggeration increased uptake. The numbers for caveats were small, and the authors call for more research, but the fear behind leaving caveats out is not supported by their data.

A Worked Example

Say, as a hypothetical, that you followed 4,000 adults for five years and found that 6% of those who volunteered regularly received a new depression diagnosis, compared with 8% of those who did not. After adjusting for age, income, baseline health and prior depressive symptoms, the association held, with a confidence interval that came close to no difference. You could not measure personality or social support.

  1. Drifted post: "Volunteering cuts depression risk by 25%. Want better mental health? Start helping others." It turns an association into a cause, gives only the relative difference, drops the uncertainty and adds advice the study cannot support.
  2. Better post: "In our five-year cohort study of 4,000 adults, people who volunteered regularly were less likely to receive a new depression diagnosis: 6% compared with 8%, or about 2 fewer per 100 people. The association held after adjusting for age, income and baseline health, but the estimate is uncertain, and because this is an observational study we cannot say volunteering caused the difference. People who volunteer may differ in ways we could not measure, such as personality or social support. The full paper is linked below."

The second post keeps the design, the absolute numbers, the adjustment, the uncertainty and the main alternative explanation. It is longer, and it is the version a journalist can quote without getting it wrong. If you are briefing your institution's communications team, the guide on working with your university press office covers how to review a release before it goes out, and if your study is a survey, see promoting a survey study for the sampling side.

A Checklist Before You Share

  1. Name the study design (cohort, case-control or cross-sectional) in the first sentence.
  2. Use "associated with" unless your paper makes a stronger, defended claim.
  3. Check every headline and post against the wording ladder above.
  4. Give the absolute difference over a stated time period, not only a relative one.
  5. Say what you adjusted for, and name the most important thing you could not measure.
  6. Signal uncertainty in plain words or with the confidence interval.
  7. Say who was studied, so the result is not read as applying to everyone.
  8. Leave out advice to readers that your design does not support.
  9. Read the press release draft against your abstract, line by line, before you approve it.

Where Amplify Fits

The hard part of promoting an observational study is keeping "associated with", the absolute numbers and the main caveat attached in every format people actually see. Amplify turns the PDF you upload into a video abstract (16:9, about 3 minutes), a vertical short (9:16, 45 to 60 seconds), an audio brief and a one-page print-ready infographic. They are generated from your paper and nothing else, with no open-web filler, and every artifact carries the title, the authors and a link back to the paper of record.

Each format arrives as three variants you can edit line by line or regenerate, so you can check the wording against your abstract and fix any verb that has climbed the ladder. A subject-matter reviewer reads every draft before it can leave your workspace, every claim is traced to a page in the PDF, and nothing publishes until you approve it. If you share a video abstract with a journal, check its specifications and AI policy first, and ask the editor if you are unsure.

The embeddable widget puts every format beside the paper on your faculty profile or lab site, where readers can read, listen and chat with the paper without leaving the page. Analytics show reads, listens, questions asked, languages and which format moved people, which is useful for grant reports and broader-impact sections.

An observational study can be one of the most useful things a field produces, as long as it is retold as what it is. Give people the careful version first, and they have something accurate to repeat.

See what Amplify makes from one observational study

#Observational Studies#Epidemiology#Psychology#Research Communication#Press Releases

Written by

Tarun Singhal

Related articles

How to Promote a Randomized Controlled Trial Without Spin
RB 101Nov 17, 20267 min

How to Promote a Randomized Controlled Trial Without Spin

A practical guide to sharing the results of a randomized controlled trial so that posts, press releases and talks say what the trial found. It covers what spin looks like, what to keep next to the result, a worked example and a checklist.

By Tarun Singhal

How to Promote a Survey Study Without Overstating the Numbers
RB 101Nov 14, 20267 min

How to Promote a Survey Study Without Overstating the Numbers

A practical guide to sharing survey findings so the headline number travels with the context it needs. It covers what to disclose, how margins of error really work, opt-in samples, a worked example and a checklist.

By Tarun Singhal

How to Promote a Methods Paper So Other Labs Actually Use It
RB 101Nov 11, 20266 min

How to Promote a Methods Paper So Other Labs Actually Use It

A practical guide to promoting a methods, protocol or software paper by making the method easy to find, try and cite. It covers a citable protocol, a code DOI, a CITATION.cff file, a worked example and a checklist.

By Tarun Singhal