Correlational research is a type of study that measures two or more variables to find out whether, and how strongly, they are related, without manipulating any of them. Instead of changing conditions the way an experiment would, researchers observe variables as they naturally exist and use statistics to describe the relationship. It is one of the most common designs in psychology, public health, and the social sciences because it can reveal patterns in real-world data that experiments often cannot.
How Correlational Research Works
A researcher starts by identifying at least two variables they suspect are related. They then collect data on both variables from the same group of people (or the same set of observations) and run a statistical test to see whether the variables move together. Nothing is controlled, no treatment is assigned, and participants are not placed into groups. The data comes from how things already are in the world, not from conditions the researcher created.
Data collection typically takes one of three forms. Surveys and questionnaires are the most common: a researcher might ask hundreds of people about their sleep habits and their stress levels, then check whether the two are linked. Naturalistic observation is another route, where a researcher records behavior in its normal setting, like watching how often students check their phones during lectures and comparing that to their exam scores. A third approach uses archival data, pulling numbers from existing records such as hospital databases, census figures, or school transcripts. Health researchers increasingly rely on electronic health records from large populations, which can reveal relationships that tightly controlled experiments would miss.
Types of Correlation
The relationship between two variables falls into one of three categories.
- Positive correlation: Both variables move in the same direction. When one increases, the other increases too. Height and weight are a classic example: taller people tend to be heavier.
- Negative correlation: The variables move in opposite directions. As one rises, the other falls. Altitude and temperature illustrate this well: the higher you climb up a mountain, the colder it gets.
- Zero correlation: No consistent relationship exists between the two variables. The amount of tea a person drinks, for instance, has no meaningful connection to their intelligence.
Knowing which type you are dealing with matters because it shapes interpretation. A positive correlation between exercise and mood tells a different story than a negative correlation between screen time and sleep quality, even though both are useful findings.
Measuring the Strength of a Correlation
Researchers express the strength and direction of a correlation with a single number called a correlation coefficient, represented by the letter r. This value ranges from −1 to +1. A coefficient of +1 means a perfect positive correlation, −1 means a perfect negative correlation, and 0 means no correlation at all. In practice, perfect correlations almost never appear in real data.
General guidelines for interpreting the number look like this: values around 0.1 to 0.3 (positive or negative) are considered weak, 0.4 to 0.6 moderate, and 0.7 and above strong. These thresholds vary slightly by field. In psychology, an r of 0.7 is typically labeled strong, while in medicine the same value might be called only moderate. Context matters: a “weak” correlation in a lab study could still be meaningful in a large population.
The most widely used version is Pearson’s r, which measures linear relationships between two continuous variables. It assumes the data form a roughly straight-line pattern with no extreme outliers. When the relationship is not a straight line but still consistently moves in one direction (always rising or always falling), or when the data are ranked rather than measured on a continuous scale, Spearman’s correlation is a better fit. Spearman’s version is also more resistant to outliers, making it a practical choice when a few extreme data points could skew results.
To decide whether a correlation is statistically meaningful and not just a fluke of the sample, researchers check the p-value. The conventional threshold is p < 0.05, meaning there is less than a 5% probability the result occurred by chance alone. That cutoff is a widely used convention, not an absolute rule, and some fields use stricter thresholds for higher confidence.
Why Researchers Choose This Design
The most common reason is ethics. Many important questions cannot be tested with a controlled experiment because it would be harmful or impossible to manipulate the variable. You cannot randomly assign people to smoke for 20 years to study lung disease, or deliberately expose children to neglect to study its effect on development. Correlational research lets scientists investigate these relationships by studying people whose lives already differ in those ways.
Practicality is another driver. Experiments require tight control over conditions, and that control limits how well results generalize to everyday life. Randomized controlled trials, for instance, test treatments in carefully selected participants under ideal settings. Correlational studies that draw from general-population data often have higher external validity, meaning their findings are more likely to reflect what happens in the real world.
Correlational research also serves as a starting point. Before investing the time and money in a full experiment, researchers use correlational studies to check whether a relationship even exists. If no correlation shows up between two variables, there is little reason to design an expensive trial around them. And when multiple correlational studies from different angles all point toward the same pattern, they build converging evidence for a theory even without a single experiment.
Finally, correlation is used to evaluate measurements themselves. Researchers check the reliability and validity of tests and surveys by seeing whether scores correlate with other established measures of the same trait.
The Biggest Limitation: Correlation Is Not Causation
The most important thing to understand about correlational research is that it cannot prove one variable causes another. Two specific problems explain why.
The first is the directionality problem. If a study finds that people who exercise more report lower anxiety, the correlation alone cannot tell you which way the arrow points. It could be that exercise reduces anxiety, or it could be that people with lower anxiety find it easier to exercise. The data look identical in both scenarios.
The second is the third-variable problem. Two variables can appear closely linked not because one influences the other, but because a hidden third factor drives both. The classic example: ice cream sales and violent crime rates rise and fall together throughout the year. Ice cream does not cause crime, and crime does not cause ice cream purchases. Hot weather, a third variable, independently increases both. Failing to account for these confounding variables can lead to seriously misleading conclusions.
Because correlational studies do not manipulate variables or randomly assign participants, they have low internal validity. There is no built-in way to rule out alternative explanations the way a controlled experiment can. This does not make correlational findings useless. It means they need to be interpreted carefully and, ideally, supported by other types of evidence before anyone draws causal claims.
Correlational vs. Experimental Research
In an experiment, the researcher deliberately changes one variable (the independent variable) and observes whether another variable (the dependent variable) responds. Participants are randomly assigned to different conditions, which helps rule out confounding factors. This setup can support cause-and-effect conclusions.
In a correlational study, the researcher changes nothing. They measure variables as they already exist and look for statistical relationships. The tradeoff is straightforward: experiments offer stronger evidence for causation but are limited to artificial settings and smaller samples. Correlational studies sacrifice causal certainty but can study a wider range of topics, larger populations, and real-world conditions. Most research programs use both designs at different stages, with correlational work identifying promising relationships and experiments testing whether those relationships are truly causal.
Real-World Examples
Correlational research is behind many of the health and behavior findings you encounter in the news. The well-established link between smoking and lung cancer was first identified through large correlational studies that tracked smoking habits and disease rates across populations. Researchers could not ethically assign people to smoke, so they observed what happened naturally and found a strong, consistent relationship.
In psychology, studies on the relationship between social media use and depression in teenagers are almost always correlational. Researchers survey teens about their screen time and mental health, then look for patterns. These studies consistently find a negative correlation (more social media use is associated with worse mood), but the design alone cannot confirm that social media is the cause.
Public health surveillance relies heavily on correlational data pulled from electronic health records. By analyzing millions of patient records, researchers can spot relationships between medications and side effects, environmental exposures and disease clusters, or lifestyle factors and long-term outcomes, all without conducting a single experiment.