Home/Insights/Methodology reference/Correlational research design
What Is Correlational Research: Definition, Types, and Examples
It measures whether two variables move together. That is a genuine finding and it is not a causal one, however strong the coefficient.
What a correlation is
A correlation quantifies the strength and direction of a linear association, from −1 to +1. Pearson’s r assumes linearity and interval data; Spearman’s rank correlation does not and is appropriate for ordinal data or monotonic non-linear relationships.
Correlational designs are used where manipulation is impossible, unethical or impractical — which is most of epidemiology and much of psychology.
Types of correlational research
Correlational studies are usually described in two ways: by the direction of the relationship they find, and by how the data are collected. Neither involves assigning anyone to a condition; the researcher measures variables as they occur.
Types by direction of the relationship
| Type | What it means | Coefficient |
|---|---|---|
| Positive correlation | As one variable increases, the other tends to increase | Between 0 and +1 |
| Negative correlation | As one variable increases, the other tends to decrease | Between 0 and −1 |
| No (zero) correlation | No consistent linear relationship between the variables | Close to 0 |
A coefficient near zero rules out only a linear relationship; a curved relationship can still exist, which is why the data should be plotted.
Types by how the data are collected
| Method | How it works | Main limitation |
|---|---|---|
| Naturalistic observation | Behaviour or outcomes are recorded in their usual setting without intervening | Observer influence and limited control over what is measured |
| Survey or questionnaire | Participants report on the variables of interest, often at a single time point | Self-report bias and dependence on who chooses to respond |
| Archival or secondary data | Existing records, registries or datasets are analysed | Variables are limited to what was originally recorded |
General classification of correlational designs.
Examples of correlational research
Each example below measures two or more variables without manipulating them. In every case the design can show whether the variables move together, but not which one, if either, causes the other.
Illustrative examples of correlational questions
| Field | Question | Variables measured | What it can and cannot show |
|---|---|---|---|
| Public health | Is screen time associated with sleep duration in adolescents? | Daily screen hours; hours of sleep | An association; not whether screens shorten sleep or poor sleep increases screen use |
| Education | Is class attendance associated with exam scores? | Attendance rate; final exam score | A relationship; motivation may drive both |
| Clinical research | Is body mass index associated with systolic blood pressure? | BMI; blood pressure readings | Strength and direction; confounders such as age and diet remain |
| Psychology | Is perceived social support associated with depressive symptoms? | Support scale score; symptom scale score | An association; the direction of influence is unknown |
Illustrative examples written for this guide, not drawn from specific published studies.
Why causation does not follow
An association between X and Y is consistent with X causing Y, Y causing X, a third variable causing both, selection into the sample, or chance.
The confounding case is the one that misleads most often, because the confounder is frequently unmeasured and therefore invisible in the analysis.
The traps specific to correlation
Restriction of range. Measuring a narrow slice of the range attenuates the correlation. A test that predicts performance across the full population may appear useless within an already-selected group.
Ecological fallacy. A correlation at group level does not hold at individual level. Countries with higher average intake of something may have higher disease rates without any individual relationship existing.
Outliers and non-linearity. A single extreme point can create or destroy a correlation. Plot the data; a coefficient reported without a scatterplot hides the shape.
Multiplicity. A correlation matrix of twenty variables contains 190 coefficients, of which around ten will reach p < 0.05 by chance alone.
Reporting it honestly
Give the coefficient with a confidence interval and the sample size. Describe the association in associational language throughout — "was associated with", not "led to" or "resulted in".
In practice
Fix, flag or query.
The distinction that keeps an edit defensible: what an editor may change silently, what must be shown to the author, and what only the author can answer.
- Never use: causal verbs for a correlational result.
- Always: plot the data before trusting a coefficient.
- Correct for: multiplicity when reporting a correlation matrix.
Reporting standards
- STROBE Statement, for observational analyses
- American Statistical Association, statement on p-values and multiplicity
Reporting guidelines named here are hosted by the EQUATOR Network. Writing to the checklist for your design before submission is the cheapest available improvement to a methods section.
Related