Glossary
Plain-language definitions of the research terms used across Unmythed. Articles link here the first time a term appears.
- Absolute and relative risk
- Absolute risk is your actual chance of something happening, such as 2 in 10,000. Relative risk compares two groups, such as “twice as likely”. A big relative change can be a tiny absolute one, which is why headlines often quote relative risk.
- Anecdotal evidence
- Personal stories and individual experiences. They can suggest questions worth studying, but they can’t show whether something works, because we can’t tell what would have happened otherwise.
- Blinding
- Keeping participants, and ideally researchers too, from knowing who received the real treatment. It stops expectations from influencing the results. When both sides are unaware, the study is “double-blind”.
- Confirmation bias
- Our tendency to notice, remember and trust information that fits what we already believe, and to overlook information that doesn’t.
- Conflict of interest
- When a researcher or author has something to gain from a particular result, such as funding from a company that sells the product being tested. It doesn’t prove a study is wrong, but it should be disclosed and considered.
- Confounding
- When a third factor affects both things being compared, creating a link that isn’t cause and effect. For example, people who take vitamins may also exercise more, so healthier outcomes might come from the exercise.
- Control group
- The group in a study that doesn’t get the treatment being tested, used for comparison. Without one, it’s impossible to know whether the treatment made any difference.
- Correlation and causation
- Correlation means two things tend to change together. Causation means one actually causes the other. Correlation alone doesn’t show causation.
- Dose–response
- When more of something produces a bigger effect. A consistent dose–response pattern makes a cause-and-effect link more believable.
- Effect size
- How big a difference something makes, not just whether there is one. A result can be statistically significant but too small to matter in real life.
- Illusory correlation
- Seeing a relationship between two things that aren’t actually related, usually because the examples that fit are more memorable than the ones that don’t.
- In vitro and animal studies
- Research done on cells in a dish (in vitro) or in animals. These are valuable early steps, but many results don’t carry over to people.
- Meta-analysis
- A statistical method that combines the results of several studies to get a more reliable overall answer. It’s only as good as the studies that go into it.
- Observational study
- Research that watches what people already do, rather than assigning a treatment. Useful for spotting patterns, but vulnerable to confounding, so it rarely proves cause and effect on its own.
- Peer review
- The process where other experts assess a study before a journal publishes it. It catches many problems, but not all, so peer-reviewed doesn’t automatically mean correct.
- Placebo and placebo effect
- A placebo is a dummy treatment, such as a sugar pill. The placebo effect is the real improvement people sometimes feel simply from expecting to get better. Good trials compare treatments against a placebo.
- Preprint
- A study shared publicly before peer review. Preprints spread findings quickly but haven’t yet been checked by other experts.
- Publication bias
- The tendency for studies with exciting or positive results to be published more often than studies that found nothing, which can make a treatment look better than it is.
- Randomised controlled trial (RCT)
- A study where participants are randomly assigned to receive a treatment or a comparison, such as a placebo. Randomisation balances out other differences between groups, making RCTs the strongest single type of study for testing cause and effect.
- Replication
- Repeating a study, ideally by a different team, to see whether the result holds up. Findings that can’t be replicated are unreliable.
- Sample size
- The number of people or observations in a study. Small samples are more easily thrown off by chance.
- Statistical significance and p-values
- A p-value estimates how likely results at least this extreme would be if there were no real effect. By convention, p < 0.05 is called “significant”. It doesn’t tell you how large or important an effect is, or the chance that a claim is true.
- Systematic review
- A review that gathers all the research on a question using set, transparent rules, then assesses its quality. It’s among the strongest forms of evidence, unlike ordinary reviews, which can pick and choose studies.
Want to put these to use? Try the ten-question guide to checking a claim.