Course contentsShow
← Data Science
Lesson 533 of 2,14514. Simple Linear RegressionFree lesson

Interpreting R-Squared Values

Understand R-squared ranges from 0 to 1; learn what low vs high values mean for predictive power.

Interpreting R-Squared Values

What You'll Learn

You'll learn how to interpret R-squared values on their 0-to-1 scale and understand what different values tell you about your model's predictive power.

Understanding the R-Squared Scale

Now that you know what R-squared is and how to calculate it, the critical question becomes: what does the number actually mean?

R-squared always falls between 0 and 1 (or 0% to 100%). This bounded scale makes interpretation intuitive:

Low R-Squared Values (closer to 0)

  • R² = 0: Your regression line explains none of the variance—it predicts no better than simply using the mean.
  • R² = 0.15: Only 15% of variance is explained; 85% remains unexplained. Your model captures very little of the pattern.
  • Interpretation: Weak predictive power. The independent variable doesn't help much in predicting the dependent variable.

High R-Squared Values (closer to 1)

  • R² = 0.85: Your model explains 85% of the variance—most of the variation is captured by your regression line.
  • R² = 1.0: Perfect fit. Every point lies exactly on the line (rare in real data).
  • Interpretation: Strong predictive power. The independent variable does an excellent job predicting the dependent variable.

The Middle Ground

An R² of 0.50 means your model explains half the variance—a moderate relationship exists, but substantial unexplained variation remains.

Context Matters

What counts as "good" depends on your field:

  • Physics/chemistry: R² > 0.9 often expected (tight physical laws)
  • Social sciences: R² = 0.3–0.5 may be quite respectable (human behavior is complex)
  • Finance: R² = 0.1 might still be valuable (markets are noisy)

Key Takeaway: R-squared ranges from 0 (no explanatory power) to 1 (perfect prediction). Higher values indicate better fit, but what's "good enough" depends heavily on your domain and research context—not just the number itself.