One headline says a supplement works. Another says a study found “no significant difference.” It is tempting to treat the second headline as a definitive rejection—or to dismiss it because it does not fit the first. Both reactions can miss useful information.

A better question is: how large was the estimated difference, how uncertain was it, and would that difference matter to the decision you are making? This guide gives you a way to read a result without needing to perform the statistics yourself.

“Not significant” is not a synonym for “identical”

A conventional significance test often asks whether the data are sufficiently incompatible with a model that assumes no difference. Failing to cross a threshold such as p<0.05 does not establish that the groups are identical. The American Statistical Association also emphasises that a p-value does not measure the size or practical importance of an effect. [1]

So when an article reports a “non-significant” result, look for the actual estimated difference. The label alone cannot tell you whether the estimate is tiny, substantial but imprecise, or somewhere between those possibilities.

Read the estimate and its uncertainty together

A confidence interval helps describe the precision of an estimate under the analysis assumptions. A wide interval can remain compatible with several very different interpretations. A narrower interval may give a clearer picture. Neither removes concerns about study design, measurement or selective reporting. [2]

Here is a deliberately invented example. It does not describe a real supplement, exercise or trial. Suppose two studies estimate an additional 0.5 kilograms on a strength test:

Both intervals include zero. Treating both as merely “no effect” would throw away the difference in precision. Conversely, neither example tells you the probability that a particular customer will improve. These are illustrative comparisons of average effects, not personalised forecasts. [2]

Can research show that differences are too small to matter?

Yes, when that question is addressed appropriately. Equivalence testing starts with a justified range of differences considered too small to be important, then tests whether the evidence supports effects within that range. It is a different task from simply failing to find a statistically significant difference. [3]

The range must make sense for the question. A difference that is irrelevant for one decision may matter for another. A well-designed equivalence study can therefore be informative without proving that two options are exactly identical in every respect.

You do not need to run an equivalence test while shopping. Just notice the distinction between “we did not detect a difference” and “we tested whether important differences could reasonably be ruled out.” If the paper only did the former, the headline should not quietly claim the latter.

Do not turn uncertainty into a sales argument

“The study did not prove it fails” is not positive evidence that a product works. Equally, “the result was significant” does not tell you that a benefit is large enough to justify the price, effort or inconvenience. Decisions need more than a threshold. [1]

For a hypothetical optional product costing £25 a month, you might ask what improvement would make £300 a year worthwhile to you. The arithmetic is simple; the value judgement is personal. These prices are invented for the example. A possible small benefit may feel worthwhile to one person and unnecessary to another.

Write your priority before reading the marketing. Is it less time spent planning, a measurable training outcome, or simply a product you enjoy using? A study of one outcome cannot settle a different priority.

A result still has boundaries

Statistics describe the comparison made in the study. They do not extend its follow-up period or change who participated. Before applying a result, check the population, dose or programme, comparison group and measured outcome. Our guide to who was actually studied walks through that separate question.

The same reading habit applies to reassurance about unwanted effects. A study that found no difference in one measured outcome over a limited period does not establish the absence of every possible problem. For a concrete example, see the direct creatine and hair-loss trial, where the outcome and study duration are central to interpreting the result.

A four-question reading card

1. What was compared? Name both options and the outcome.
2. How big was the estimated difference? Look beyond the significant/non-significant label.
3. How precise was it? Find the interval and the authors' limitations.
4. Would it change my decision? Consider the size of a worthwhile benefit, the burden and the wider evidence.

This card is an editorial aid, not a research-quality scoring system. If the information is missing, write down the missing question rather than inventing an answer. You can ask for clarification or postpone an optional purchase.

The aim is to feel more capable of making a choice, not to turn every meal or workout into a research project. When you want help organising the practical side, explore WeightStart and assess what its planning and support offer against your actual needs.

Sources and reading notes

Updated 2 October 2026. Numerical examples and the reading card were created for this article. The references below explain research interpretation; they are not trials demonstrating a WeightStart outcome.

1. American Statistical Association. Statement on Statistical Significance and P-Values: official release and six principles. 7 March 2016.
2. Greenland S, Senn SJ, Rothman KJ, et al. Statistical tests, P values, confidence intervals, and power: a guide to misinterpretations. European Journal of Epidemiology. 2016;31:337–350. doi:10.1007/s10654-016-0149-3. Open-access methodological essay.
3. Lakens D. Equivalence Tests: A Practical Primer for t Tests, Correlations, and Meta-Analyses. Social Psychological and Personality Science. 2017;8(4):355–362. doi:10.1177/1948550617697177. University-hosted full text.