what is a good r squared value

A “good” R-squared value depends on context rather than a single magic cutoff.

Quick Scoop: Core idea

R-squared measures how much of the variation in your outcome your model explains (0 to 1, or 0% to 100%).

Higher is not always better; very high values can signal overfitting or issues with the data, especially in time series.

What counts as “good” varies a lot by field, data type, and purpose.

Typical “good” ranges by field

Here’s a rough feel for how different areas interpret R-squared:

[1][9][3] [9][1][3] [7] [7] [3] [3] [3][7] [7] [5] [5]

Field / context	Often acceptable	Often considered “good”
Social & behavioral sciences	0.10–0.30 can be meaningful because human behavior is noisy.	0.30–0.50 often seen as very good.
Clinical / medical studies	≈0.15 may be a reasonable threshold for a “meaningful” model in some proposals.	Higher is nice, but small values can still matter for complex human data.
Finance (e.g., factor models)	Below 0.40 often viewed as weak correlation.	Above 0.70 usually treated as a strong relationship.
Physical sciences & engineering	0.50–0.70 may be workable depending on the problem.	0.70–0.99 commonly expected for a “good” fit in many experiments.
Time-series (after differencing / stationarizing)	0.10–0.25 can still be useful when signals are weak.	“Good” is defined more by predictive performance and error diagnostics than raw R².

Why “it depends” is the real answer

In messy human domains, even an R-squared of 0.10 can reveal an important relationship, especially if predictors are statistically significant.

In controlled lab physics or engineering, people often expect R-squared much closer to 1 because measurement noise is lower and relationships are more stable.

Some authors explicitly argue that thresholds like 0.15 or 0.25 can be meaningful in areas such as clinical medicine or weak-signal time series.

A quick mental rule:

Noisy, human data → don’t be surprised by “good” models with R² between 0.1 and 0.4.

Clean, physical/engineering data → aim for R² above 0.7 if you’re calling it “good.”

Pitfalls: when a high R-squared is misleading

Even a very high R-squared can be bad news if:

You added lots of predictors just to boost R-squared.
- R-squared never goes down when you add more variables, even if they’re useless.

 * That’s why adjusted R-squared is often preferred; it increases only when new variables really help.

You’re working with time series and did not “stationarize” the data.
- For trending series, R-squared close to 1 can reflect shared trends, not real predictive structure.

You ignore residual diagnostics.
- Good models also have residuals that are roughly pattern-free and appropriately distributed, not just a high R-squared.

How to judge your own R-squared

To decide if your R-squared is “good,” ask:

What field am I in and how noisy is my outcome?
- Human behavior/clinical/finance → lower R² can still be useful.

 * Physics/engineering → expect higher R² for a model you call strong.

What is my goal?
- Explanation: Even modest R² can be valuable if it clarifies which variables matter.
- Prediction: Look at out-of-sample error (test set, cross-validation) rather than just R² on training data.

Am I checking adjusted R-squared and other metrics?
- Adjusted R² helps control for overfitting.

 * Combine it with RMSE, MAE, or other fit and validation checks.

A simple way to think about it:

“Good” R-squared is whatever gives you reliable, validated insight for your field and question, not a fixed magic number.

Mini example

Imagine you build a linear regression predicting exam scores from study hours and prior GPA in a university sample:

You get R² = 0.32 and adjusted R² = 0.30.
In an educational or behavioral context, that means about 30% of the variance in exam scores is explained, which is often considered quite solid.

If residuals look healthy and validation error is reasonable, you could honestly call this a “good” R-squared for that use case.

TL;DR:
There is no universal “good” R-squared value. In noisy human fields, 0.1–0.4 can be meaningful; in physical sciences and engineering, people often expect values above 0.7. Always interpret R² alongside your field norms, model purpose, adjusted R², and validation performance.

Information gathered from public forums or data available on the internet and portrayed here.