No, correlational research does not show cause and effect. It can reveal that two variables are related, and it can measure how strongly they move together, but it cannot prove that one variable causes the other to change. That distinction matters more than it might seem, because confusing correlation with causation leads to real misunderstandings in health, science, and everyday decision-making.
What Correlational Research Actually Measures
In a correlational study, researchers collect data on two or more variables without manipulating any of them. They observe what naturally occurs and then measure whether the variables move together. The strength of that relationship is expressed as a correlation coefficient, a number between -1 and +1. A value close to +1 means the variables rise and fall together. A value close to -1 means one goes up while the other goes down. A value near zero means there’s no consistent linear relationship.
What this tells you is association. If hours of sleep and test scores have a positive correlation, it means people who sleep more also tend to score higher. But the correlation alone can’t tell you that sleeping more caused better scores. Maybe less-stressed students both sleep better and study more effectively. Maybe healthier students do both. The data pattern looks the same regardless of the underlying reason.
Two Problems That Block Causal Claims
The Third Variable Problem
A confounding variable (sometimes called a lurking or third variable) is something related to both variables you’re studying that creates the illusion of a direct link between them. The classic example: ice cream sales and home break-ins both rise in summer. If you only looked at the numbers, you’d see a positive correlation. But ice cream doesn’t cause crime. Warmer weather independently drives both. Without controlling for temperature, the data is misleading.
This happens constantly in health research. One study found a correlation between drink preference and body weight. People who preferred beer weighed more on average than people who preferred wine. But gender was a confounding variable. Men in the sample were more likely to prefer beer and also weighed more on average than women. The apparent link between drink choice and weight was at least partly an artifact of who was choosing what.
The Directionality Problem
Even when two variables genuinely influence each other, correlational data can’t tell you which direction the influence runs. The American Psychological Association defines the directionality problem as a situation where two variables are known to be related but it’s unclear which is the cause and which is the effect. If depression and social isolation are correlated, does isolation lead to depression, or does depression cause people to withdraw? Correlational data alone can’t answer that.
Why Experiments Can Show Causation
The reason experimental research can establish cause and effect comes down to two features that correlational studies lack: manipulation and random assignment.
In an experiment, researchers deliberately change one variable (the independent variable) and then measure whether a second variable (the dependent variable) changes in response. Because the researcher controls what changes, they can establish the sequence of events. They know the cause came before the effect.
Random assignment handles the third variable problem. By randomly sorting participants into groups, researchers ensure that characteristics like age, gender, health status, and personality are distributed roughly evenly across conditions. This means any difference in outcomes between groups can be attributed to the variable that was manipulated, not to some hidden factor. Randomized controlled trials are considered the gold standard for causal claims precisely because, when properly implemented, randomization balances both observed and unobserved participant characteristics across groups.
Where Correlational Research Still Has Value
None of this means correlational research is useless. Many important questions can’t be tested experimentally. You can’t randomly assign people to smoke for 30 years to see if it causes cancer. You can’t randomly assign children to experience poverty. For ethical and practical reasons, correlational and observational designs are sometimes the only option.
Longitudinal studies, which follow the same people over years or decades, help narrow the gap. By tracking who was exposed to something first and who developed an outcome later, researchers can at least establish the sequence of events. This doesn’t eliminate confounding variables entirely, but it rules out reverse causation. If a behavior consistently precedes a health outcome across thousands of people over many years, that’s stronger evidence than a single snapshot in time. The tradeoff is that longitudinal designs still struggle to fully separate the back-and-forth influence between exposure and outcome, especially when the delay between the two is long.
Epidemiologists also use a set of criteria, originally proposed by Sir Austin Bradford Hill, to evaluate whether a correlation likely reflects a true causal relationship. These include the strength of the association, whether the finding has been replicated consistently across different populations, whether there’s a dose-response pattern (more exposure leads to more effect), whether the timing makes sense, and whether the relationship is biologically plausible. No single criterion proves causation, but when multiple criteria are met, confidence grows substantially. This framework is how researchers built the case against smoking decades before a randomized trial would have been ethical.
Why This Confusion Shows Up Everywhere
If the distinction between correlation and causation seems academic, consider how often it’s ignored in real life. A study of 130 prominent health news stories found that 49% made causal claims based on non-randomized study designs. Headlines regularly say a food “raises” or “slashes” disease risk when the underlying study only found a correlation. News coverage almost never mentions whether the evidence actually supports a strong causal claim or notes the limitations of correlational data.
This has real consequences. Misleading health claims create public confusion and erode trust in science. When last year’s headline says coffee prevents heart disease and this year’s says it causes anxiety, the contradiction often reflects two correlational studies being presented as though they proved opposite things. Neither proved anything causal. Researchers who study science communication recommend that journalists use cautious language for correlational findings: words like “may,” “could,” “is associated with,” or “is linked to” instead of definitive causal language.
You can apply the same filter when reading health news. If a study didn’t randomly assign people to different conditions, treat the finding as a clue, not a conclusion. Look for whether the story mentions confounding variables, how large the study was, and whether the result has been replicated. A single correlational study is a starting point for investigation, not the end of one.
Spotting Spurious Correlations
Sometimes correlations are not just non-causal but completely meaningless. Researcher Tyler Vigen maintains a database of spurious correlations that illustrate this point vividly. American cheese consumption correlates with the volume of U.S. edible fish imports. Google searches for “cat memes” correlate with automotive recalls for airbag issues. These variables have nothing to do with each other, but with enough data points and enough variables, statistical coincidences are inevitable.
These examples are funny, but they highlight a serious point. A correlation coefficient doesn’t know whether a relationship makes sense. It just measures whether two sets of numbers move in similar patterns. Interpretation requires human judgment, domain knowledge, and ideally a well-designed experiment to test whether the relationship holds up under controlled conditions.