Correlation or Regression? The Difference, When to Use Each and How to Report It
When a Pearson or Spearman correlation is enough and when you need linear regression, what each one does and doesn't tell you, how to read the SPSS output and how to report it in APA.
Part of the series: Statistical Analysis for a Seminar Paper or Thesis: The Complete Guide
Correlation and regression answer related questions, so they get mixed up. In short: correlation tells you whether two variables move together and how strongly. Regression predicts one variable from one or more others and tells you by how much.
When to use correlation
When the hypothesis is "X and Y are related", with no direction. Example: "hours of sleep are positively related to exam scores".
- Pearson (r): two continuous variables, a linear relationship, no extreme outliers.
- Spearman (rs): ordinal data, skewed distributions or a small sample.
r runs from -1 to 1; in absolute value 0.1 is weak, 0.3 moderate, 0.5 strong. In SPSS: Analyze → Correlate → Bivariate.
When to use regression
When there is a predictor and an outcome, or you want several predictors together and each one's unique contribution. Example: "sleep and motivation will predict the score, controlling for year of study".
Three things matter in the output (Analyze → Regression → Linear):
- R²: how much of the outcome's variance the model explains.
- The model's F and p: whether the model as a whole is significant.
- β (standardised coefficient) and p per predictor: each variable's unique contribution with the others held constant.
Multiple regression also needs a multicollinearity check (VIF below about 5) and roughly normal residuals.
What neither of them tells you
Neither proves causation. "Sleep predicts scores" in a correlational study means they are related, not that sleep causes high scores. Causation needs an experimental design.
Reporting in APA
Correlation: "Hours of sleep correlated positively with exam scores, r(98) = .34, p < .001."
Regression: "The model explained 21% of the variance in scores, F(2, 97) = 12.86, p < .001, R² = .21. Sleep predicted scores uniquely (β = .29, p = .002), as did motivation (β = .24, p = .011)."
Full regression tables: Reporting regression in APA. For a yes/no outcome use logistic regression.
Common mistakes
- Running a regression when the hypothesis is only a relationship - a correlation is enough.
- Reporting b (unstandardised) without saying so, or comparing b across predictors on different scales.
- Inferring causation from a correlational study.
- Skipping the correlation table of all study variables before the hypotheses.
FAQ
Can I use both? Yes, and it's common: a correlation table with the descriptives, then a regression for the prediction hypothesis.
Where does it fit? Step 4 of 7 - Statistical analysis for a seminar paper: the complete guide. Need mediation or moderation (PROCESS)? Tell us.
Tell us what you need, and we'll come back with an expert in Statistical analysis and a quote.
Last updated: