Correlation and causation: how to read a science headline

The difference between correlation and causation

A correlation describes how two variables are related. Causation adds a claim: changing one produces a change in the other. The Australian Bureau of Statistics explains why correlation alone does not establish that cause.

When you find phrases such as associated with or related to, keep that scope in your notes. If you later write causes or improves, you are adding something that still needs justification. The verb matters as much as the number.

A fictional example: reading on a commute and a quiz

Imagine a made-up survey in which people who say they read while commuting also score higher on a knowledge quiz. This is not a real study or a PaperFlow result. It is a scenario for practice.

From that description, you could note an association between two measures: reported reading and quiz results. You do not yet know whether reading on the commute explains the difference. Perhaps people already had different levels of interest in the topic, or their journeys left them different amounts of reading time.

Write two separate sentences: what was observed, and what explanation you would like to investigate. That separation lets you follow an interesting idea without treating it as confirmed.

Ask what else could explain the pattern

The Royal Statistical Society, in its guide to statistics on social media, highlights two possibilities: a third factor may influence both variables, and some apparent connections can occur by chance.

In our example, you could ask about prior knowledge. You could also reverse your story: do people who already enjoy the topic choose to read more about it? These are hypotheses to check. Naming them does not establish that any one is the right explanation.

Try writing down an alternative before sharing the headline. If you do not know whether the study examined it, leave the question open and look for that point in the methods or discussion.

Check how the groups were formed

Look for the comparison researchers made. Did they observe choices people had already made? Did they assign activities? Was there a group to compare the outcome with? The NCCIH explanation of study design shows how a controlled trial compares groups receiving different conditions.

Continuing our invented example, one design question would be how to compare two reading activities while keeping other conditions clear. That requires describing selection, assignment and measurement. Saying an experiment was conducted does not replace checking those details.

Your note can stay with what you found: they used a survey, compared groups, or I could not identify the design. A label is not enough to automatically score the quality of the work.

Look at what happened first and what explanations were tested

The UK Analysis Function guide highlights timing and alternative explanations when examining a causal claim. Two things happening together does not settle those questions.

Return to our scenario: if reading and knowledge were measured only once, what information would tell you which came first? If the paper says it accounted for prior knowledge, look for how that was measured and included in the comparison.

You do not have to solve a statistical analysis on your phone. You can identify missing information and avoid a conclusion that depends on it. If you do not have the paper, our guide to finding a news story's source helps you start.

A striking number does not answer every question

The NCCIH distinguishes statistical significance from practical importance: a difference may be statistically significant yet have little practical usefulness. It also notes that a single study rarely provides a definitive answer.

When a news story highlights a figure, check what it measures and what it is compared with. In the fictional example, you would ask what the quiz assessed and how much the results differed. A figure about performance on that task does not by itself justify a promise about learning any subject.

Save the specific measure alongside the sentence. That keeps the distance between the data and the interpretation visible.

Practise rewriting the headline without adding a cause

Return to the invented claim at the start. With only the survey information, a more careful version would be: in this survey, people who reported reading on their commute scored higher on a knowledge quiz.

The rewrite keeps who was observed, what they reported and what was measured. It does not claim that adopting the activity will produce the same result in someone else. It also does not dismiss the association: it may raise a question worth studying.

You can use this editorial note for your next reading. It helps organise questions; it is not a validated reliability scale:

• Claim: what exactly does the headline say?
• Comparison: who or what was observed, and what was it compared with?
• Timing: what happened first, and how do we know?
• Alternatives: what other explanations were examined?
• Open question: what information is missing before I use a causal verb?

Try it during your next break

Choose a reading from PaperFlow's open library, follow its source and look for the verb describing the finding. Note whether it reports a relationship, a comparison or a causal conclusion, and which part of the method you would need to check to understand it.

You can stop there and return later. The aim of this practice is to leave a clearly framed question alongside the link. You do not have to turn every spare moment into a complete research appraisal.