A regression model predicting college GPA from SAT score has r^2 = 0.25. A student interprets this as: 'SAT score is a poor predictor of college GPA because only 25% of the variation is explained.' An admissions officer counters that in the context of predicting human behavior, r^2 = 0.25 is meaningful. Which perspective is more statistically sound?
- ANeither is correct; r^2 should not be used to evaluate models
- BThe admissions officer is correct; the usefulness of r^2 depends on context, and in behavioral sciences, r^2 = 0.25 can represent a meaningful relationshipCorrect
- CBoth are incorrect; r^2 = 0.25 means the model is wrong
- DThe student is correct; r^2 < 0.50 always indicates a poor model
Why B is right
The admissions officer's perspective is more statistically sound. In behavioral and social sciences, human behavior is influenced by many factors, so any single predictor explaining 25% of the variance is substantial. There is no universal threshold for 'good' r^2 — it depends on the field, the complexity of the phenomenon, and the purpose of the model.
Why the other 3 choices miss
- A
- Incorrect. r^2 is a standard and useful measure for evaluating the explanatory power of a regression model. It should be interpreted in context, but it is a valid tool.
- C
- Incorrect. A model with r^2 = 0.25 is not 'wrong.' It explains 25% of the variation, with the remaining 75% due to other factors. The model captures one meaningful component of a complex outcome.
- D
- Incorrect. There is no universal rule that r^2 < 0.50 is 'poor.' In some fields (physics), r^2 > 0.99 is expected, while in others (psychology, education), r^2 = 0.25 represents a strong finding.