Correlational Research | Definition & When To Use
Correlational research explores how two or more variables are statistically related. Importantly, these variables are measured as-is, without manipulation.
Correlational research can be helpful when it is unethical to manipulate variables (e.g., withholding medical treatment from someone violates research ethics) or impossible to do so (e.g., you cannot manipulate someone’s age).
Unlike experimental research, correlational research cannot establish causation. You can characterize how variables are related, but you cannot prove that changes to one cause changes to the other.
Common statistical methods to calculate correlation include Pearson’s r and regression analysis.
They collect data from 100 students, recording their GPA and how many hours of sleep they get, on average, each night.
The researcher finds that students who sleep more tend to have higher GPAs (in other words, there is a strong, positive correlation between the two).
Because this is a correlational study, the researcher cannot conclude that more sleep causes higher grades. There may be other variables that are influencing these results (e.g., perhaps students with better time management sleep more and also do better in school).
When should you conduct correlational research?
Correlational research involves measuring variables as they occur naturally, without attempting to manipulate them. This is distinct from experimental research, which involves manipulating an independent variable and measuring the resulting change in a dependent variable.
Researchers conduct correlational research when it is unethical to manipulate a variable, when a variable cannot be manipulated, or when it does not make sense to manipulate a variable. The following examples illustrate each of these scenarios.
For example, consider a researcher who is interested in how substance abuse impacts academic performance in high school students. It would be extremely unethical for them to force anyone (and especially minors) to consume illegal substances.
The research could instead conduct a correlational analysis. They might collect data on high school students’ existing patterns of substance use and explore how these patterns are correlated with their academic performance.
Sometimes, regardless of the ethical implications, it is not possible to manipulate variables of interest. The following example explains when this might be the case.
Imagine a researcher is interested in how someone’s ability to navigate an unfamiliar environment changes as they age. Neither of these variables can be directly manipulated. The researcher would instead have to assess someone’s navigation ability and explore how this was correlated with age.
In other cases, there are no independent and dependent variables—in other words, you do not expect a causal relationship between the two variables you are measuring. In these situations, conducting an experiment makes no sense. An example of this is validating a new measurement instrument.
When you check if something measures what it’s supposed to, you are assessing its construct validity. This can be done by using correlation to compare your new measure to an existing, well-established measure (your ground truth). If the two are strongly correlated, this provides evidence that they are measuring the same thing.
What is a correlation?
The purpose of correlational research is to characterize the correlation between two variables. A correlation is simply the relationship between two or more variables quantified using statistics.
A correlation can be described in terms of its direction and strength.
Direction
The direction of a correlation defines how one variable changes with respect to the other.
If increases in one variable are associated with increases in a second variable, the two are positively correlated. If you plot these two variables as points on a graph, the line you draw through them would have a positive slope (i.e., it points up and to the right).
If increases in one variable are associated with decreases in the second (or vice versa), the two are negatively correlated. A line drawn through these variables on a graph would have a negative slope (it points down and to the right).
If there is no relationship between two variables, you can say that they are uncorrelated. There would be no obvious pattern when you plot them as points on a graph.
| Type of relationship | Example | Description |
|---|---|---|
| Positive correlation | Outdoor temperature and the number of people at the beach. | As temperature increases, so does the number of people at the beach. |
| Negative correlation | Hours of screen time before bed and sleep quality. | As hours of screen time before bed increases, sleep quality decreases. |
| Uncorrelated | A person’s height and how many siblings they have | There is no relationship between these two variables. |
Strength
Correlation can also be described as strong, moderate, or weak. You can think of the strength of a correlation as how consistent the relationship between two variables is:
- If a change in one variable is consistently and predictably accompanied by a change in the second, the correlation is strong.
- If a change in one variable is generally, but not always, accompanied by a change in the second, the correlation is moderate.
- If a change in one variable is kind of, but not really, accompanied by a change in the second, the correlation is weak.
However, direction and strength are distinct properties. You can have a weak, positive correlation or a strong, negative one.
Positive correlations are also not better than negative correlations—they’re simply two different ways variables can change with respect to each other.
Correlation example
The following image depicts correlations of varying strengths and directions.
Correlation vs causation
Correlation can tell you how one variable changes with respect to another variable, but it cannot tell you that one variable changes because of another variable. In other words, correlational research cannot be used to prove causation.
Correlational research cannot be used to identify cause-and-effect relationships because of factors like the directionality problem and the third variable problem. Instead, researchers must use techniques like experimental research or longitudinal studies.
The directionality problem
To establish a cause-and-effect relationship, you must know which variable is doing the influencing (the directionality of the relationship). In correlational research, because both variables are measured at the same time, it is impossible to determine whether one variable is affecting the other or vice versa.
Note that directionality is different from the direction of a correlation. Directionality concerns whether one variable changes because of another variable; the direction of a correlation concerns whether two variables tend to increase or decrease together.
You find that there is a strong, positive correlation between time outdoors and happiness; in other words, people who spend more time outdoors are also happier.
It may be tempting to conclude that being outside causes happiness. However, because this is correlational research, you cannot establish directionality. An equally possible option, given your data, is that happier people are more motivated to leave their homes and spend time outside.
Using correlational research, you cannot distinguish between the possibilities that (1) spending more time outdoors causes happiness or (2) happiness causes people to spend more time outdoors. You can simply say that there is a positive relationship between these variables.
The third variable problem
Sometimes, a correlation between two variables can be influenced by a confounding third variable. The two variables may seem like they are related, but this relationship is actually caused by a third, unmeasured variable. This is called the third variable problem.
At first glance, it may seem that simply growing up around more books increases a child’s academic outcomes.
However, a potential third variable is the parents’ level of education. More educated parents may be more likely to own many books and also to support their child’s learning in different ways. It’s not the books themselves improving the academic performance; it’s the broader educational environment created by their parents.
