From result to evidence: how a conclusion is built

A surprising result sparks curiosity, but a scientific conclusion needs more than surprise. It must connect a question with observations, explain how they were obtained, and show which alternatives remain possible. Science does not always advance by following a linear recipe; it does require that claims can be tested against evidence and specific criticism.

Turning an idea into a question

Saying that one material is better leaves a lot undefined. Better for what use, compared with which alternative, and according to what criterion? A testable question could compare the energy stored by two designs under the same procedure. The OpenStax introduction to scientific reasoning helps distinguish a testable explanation from a claim that does not specify what observation could call it into question.

A hypothesis does not become reliable just because it was formulated before the experiment, but defining it clearly limits the possibility of changing the question after seeing the data. It also makes it possible to identify results that do not fit expectations. Those results can be informative even if they do not produce a spectacular headline.

Designing a comparison

Imagine a fictional test of two coatings. If we test one at one temperature and the other at a different temperature, the result mixes material and conditions. We need to decide which variables to control, which to vary, and how to distribute the test units. The NIST experimental design handbook places these decisions in relation to the study's objective.

Control does not mean freezing all of reality. It means constructing a comparison that makes it possible to answer a defined question. Then we will need to study whether the result holds in different situations. A highly controlled experiment can clarify a mechanism and still leave its usefulness in a factory or natural setting unresolved.

Measuring is also defining

Before discussing decimal places, we should identify the quantity and the unit. Duration can mean time until failure, time until losing a fraction of performance, or time during which a condition was maintained. These are different definitions that can produce different numbers even when every operation is correct.

The International System described by NIST provides a common language of units. But using the same unit does not guarantee measuring exactly the same thing. Two tests expressed in hours may have used incompatible end criteria.

Uncertainty needs an explanation

Repeating a measurement helps study its variability, although it does not automatically eliminate systematic error. Calibration, resolution, and procedure can also contribute to uncertainty. NIST distinguishes components evaluated statistically from those evaluated using other information. The classification describes how they are evaluated, not a simple distinction between random and systematic causes.

A figure accompanied by an uncertainty provides more information if it explains how it was calculated and what scope it has. Writing many decimal places does not demonstrate precision. The NIST guide to communicating uncertainty emphasizes the need to describe the method and relevant components.

Association, explanation, and scope

If we observe two variables changing together, alternative explanations may still exist. A third variable may influence both; there may also be participant selection or changes in procedure. A causal interpretation needs an argument connecting the design to the alternatives it aims to rule out.

In our coating example, observing less deterioration does not yet demonstrate that the material works in any climate or lasts for a specific number of years. It may support a more limited conclusion: under the tested conditions and according to that criterion, the measured behavior was different. The precision of this scope is a strength of the explanation.

Reading the study and its limitations

When consulting a publication, we can look for the question, the methods, the number of independent units, and how the data were analyzed. It is also useful to distinguish observed results, the authors' interpretation, and proposed applications. A press release may accurately summarize a study, but it does not replace the details needed to evaluate its conclusions.

Peer review provides a process of examination, not a guarantee of infallibility. Convergence among observations, methods, and teams strengthens a conclusion. This guide offers questions for reading research; it does not certify any study or claim that these texts have undergone human review. Each SciClue article shows its sources to facilitate verification and keep the limits of what is explained visible.

What we leave for someone else to check

The report Reproducibility and Replicability in Science, by the National Academies, examines how repeating and cross-checking research strengthens confidence and what obstacles can prevent this. It also cautions that a discrepancy should not automatically be interpreted as fraud: it may reveal the limitations of a procedure or open up a new question.

A practical way to apply this idea is to distinguish what is needed to review a calculation from what is needed to obtain new observations. These are related tasks, but each answers a different question. A results file without units or instructions may seem transparent and still be insufficient to check the analysis.

The following resource is a reading guide, not a reproducibility certificate. When considering a specific study, the question what could another person check with what is available? helps identify missing evidence. Access restrictions should also be explained: it is not always possible to publish sensitive data or repeat an unrepeatable event.