Correlation r2, often called the coefficient of determination, measures how well a regression model explains variation in the outcome variable. It ranges from 0 to 1 and indicates the proportion of variance captured by the independent variables.
Understanding r2 helps analysts assess model fit, communicate results clearly, and make informed decisions about including or refining predictors in quantitative studies.
Quick Reference: Correlation r2 Essentials
| Metric | Interpretation | Range | Cautions |
|---|---|---|---|
| r2 Value | Proportion of variance explained | 0.0 to 1.0 | Higher is not always better |
| Adjusted r2 | Penalizes unnecessary predictors | 0.0 to 1.0, can be lower than r2 | Useful for model comparison |
| Predicted r2 | Cross-validation performance | 0.0 to 1.0 | Measures out-of-sample fit |
| Residual Plot | Visual check for patterns | Any pattern indicates issues | Guides model refinement |
Calculating Correlation r2 from Correlation Coefficient
In simple linear regression with one predictor, r2 equals the squared correlation between the observed and predicted values. This provides an intuitive link between correlation and explained variance.
To compute r2, square the Pearson correlation coefficient (r) when dealing with a single predictor. This transformation converts a strength-of-association measure into a proportion of variance explained.
For multiple regression, r2 generalizes to multiple predictors jointly, but squaring a single correlation is no longer sufficient. Use the regression output to extract the overall r2 instead of relying on pairwise correlations.
Adjusted r2 and Model Complexity
Why adjusted r2 matters
Adjusted r2 modifies r2 by accounting for the number of predictors and sample size. It increases only if a new predictor improves the model more than expected by chance.
When comparing models, adjusted r2 helps avoid overfitting by penalizing excess variables. This makes it a better fit metric than r2 when evaluating models with different numbers of predictors.
Predicted r2 and Cross-Validation
How predicted r2 works
Predicted r2 measures how well the model predicts new observations using cross-validation. It is computed by systematically leaving out each observation, fitting the model, and then assessing prediction accuracy.
High predicted r2 indicates strong generalizability, while a large gap between r2 and predicted r2 signals overfitting. Aim for aligned values to ensure robust performance.
Best Practices for Using Correlation r2
- Report r2 alongside adjusted r2 and predicted r2 for a complete picture.
- Inspect residual plots to validate model assumptions before interpreting r2.
- Use adjusted r2 when comparing models with different numbers of predictors.
- Treat r2 as one of several metrics, not the sole criterion for model quality.
- Contextualize r2 with subject-matter knowledge and practical significance.
FAQ
Reader questions
What does an r2 of 0.85 indicate about my model?
An r2 of 0.85 means that 85 percent of the variance in the outcome variable is explained by the model, suggesting a strong fit to the observed data.
Should I always aim for the highest possible r2?
No, chasing a very high r2 can lead to overfitting, especially with many predictors. Balance r2 with adjusted r2, predicted r2, and domain relevance.
Is r2 affected by outliers in the data?
Yes, outliers can inflate or deflate r2 by disproportionately influencing the regression line and residual sum of squares.
Can r2 be used to compare models with different response variables?
No, r2 is not comparable across different response variables because it depends on the variance of the specific outcome being modeled.