R-squared vs Adjusted R-squared in SPSS: Explained

Open the Model Summary table after any regression in SPSS and you will see two numbers sitting side by side: R-squared vs adjusted R-squared. They look almost identical, they usually differ only in the second decimal place, and it is tempting to ignore one of them. That is a mistake. The gap between them is telling you something important about your model — usually, whether you have added predictors that are not earning their place. This guide explains what each number means, how to get both from SPSS output, and which one to report in your thesis.

What Is R-Squared?

R-squared (R²) is the proportion of the variance in your dependent variable that your regression model explains. If you are predicting exam scores from hours studied and sleep, and R² = .62, your model explains 62% of the variation in exam scores. The remaining 38% is unexplained — it comes from variables you did not measure, measurement error, and plain noise.

R-squared is calculated from the sums of squares in the ANOVA table of your regression output:

R² = Regression Sum of Squares ÷ Total Sum of Squares

It always falls between 0 and 1, and it has one famous flaw: it never goes down when you add a predictor. Add a useless variable to your model — the number of letters in a participant’s name, say — and R-squared will stay the same or creep up slightly, even though the variable adds nothing real. It rewards complexity without asking whether the complexity is justified. That is exactly the problem adjusted R-squared was designed to fix.

What Is Adjusted R-Squared?

Adjusted R-squared applies a penalty for every predictor you add. It adjusts R-squared for the number of predictors (p) relative to your sample size (n), using this formula:

Adjusted R² = 1 − [(1 − R²) × (n − 1) ÷ (n − p − 1)]

Because of that penalty, adjusted R-squared can stay flat or even fall when you add a weak predictor — which is the honest answer. It can also, unlike R-squared, be negative, which happens when a model fits worse than simply predicting the mean of the dependent variable for everyone.

Three consequences follow from the formula:

  • With few predictors and a large sample, R-squared and adjusted R-squared are nearly identical. The penalty is tiny because (n − 1) ÷ (n − p − 1) is close to 1.
  • With many predictors and a small sample, the gap widens. Ten predictors on 40 participants will show a noticeably lower adjusted R-squared — a warning that the model may be overfitting.
  • Adjusted R-squared is the fairer number for comparing models that use different numbers of predictors. R-squared will always favour the bigger model; adjusted R-squared will not.

How to Get R-Squared and Adjusted R-Squared in SPSS

Both values appear automatically in every linear regression you run in SPSS. There is nothing extra to switch on. Here is the click-by-click route.

Step 1: Open the Linear Regression dialog

Go to Analyze > Regression > Linear. Move your dependent variable into the Dependent box and your predictor(s) into the Independent(s) box.

Step 2: Check the statistics you want (optional but useful)

Click Statistics. Make sure Estimates and Model fit are ticked — they are ticked by default. If you want to compare models later, tick R squared change as well; it is useful for hierarchical regression. Click Continue.

Step 3: Run the analysis

Click OK. SPSS runs the regression and opens the Output Viewer with several tables. The one you need is the second table, titled Model Summary.

How to Interpret the SPSS Model Summary Table

The Model Summary table has five columns. Here is what each one means:

  • R — the multiple correlation coefficient: the correlation between the values your model predicts and the actual values. It is the square root of R-squared.
  • R Square — the proportion of variance explained, as described above.
  • Adjusted R Square — R-squared after the penalty for the number of predictors.
  • Std. Error of the Estimate — the typical size of a prediction error, in the units of your dependent variable. Smaller is better, but it only makes sense compared with the scale of your outcome.

Below the table, SPSS prints a footnote naming the predictors in the model — useful when you run several models and need to tell them apart.

Always check the ANOVA table underneath as well. It tells you whether the model as a whole is statistically significant. A high R-squared from a non-significant model is not something to report with confidence, and the ANOVA table’s F-test and p-value are what establish that the model explains more than chance would.

R-Squared vs Adjusted R-Squared: A Worked Example

Imagine you are predicting thesis marks (out of 100) for 60 students. You run three models:

  • Model 1 — one predictor, hours studied per week. R² = .48, adjusted R² = .47. The gap is tiny: one predictor, decent sample.
  • Model 2 — three predictors: hours studied, sleep, and previous GPA. R² = .62, adjusted R² = .60. The extra predictors genuinely help, and the penalty is still small.
  • Model 3 — the same three predictors plus four weak ones (age, commute time, number of siblings, favourite colour coded as a number). R² = .64, adjusted R² = .59. R-squared went up, as it always does — but adjusted R-squared went down. The four extra variables added noise, not signal, and adjusted R-squared caught it.

This is the practical rule in one picture: if adjusted R-squared drops when you add predictors, those predictors are not pulling their weight. If it rises, they are. When the two numbers are close, your model is parsimonious. When they are far apart, you probably have too many predictors for your sample size — a common problem in thesis data with 40–60 participants and a long list of variables.

SPSS Syntax for R-Squared and Adjusted R-Squared

If you prefer syntax, or you need a reproducible record of your analysis, the following runs a linear regression and produces the same Model Summary table. Replace the variable names with your own:

REGRESSION
  /MISSING LISTWISE
  /STATISTICS COEFF OUTS R ANOVA
  /CRITERIA=PIN(.05) POUT(.10)
  /NOORIGIN
  /DEPENDENT thesis_mark
  /METHOD=ENTER hours_studied sleep previous_gpa.

The R subcommand in the STATISTICS line is what asks SPSS to print R, R-squared, and adjusted R-squared in the Model Summary table. Run the syntax with Run > All, or highlight it and click the green play button.

Which One Should You Report?

The honest answer is: report both, and let adjusted R-squared do the comparing. In practice:

  • Report R-squared when you describe how much variance a single model explains. It is the number most readers expect to see.
  • Report adjusted R-squared alongside it whenever you compare models, justify adding predictors, or work with a small sample. Reviewers and examiners increasingly expect it.
  • Never report only the higher number to make a model look better. If adjusted R-squared is much lower than R-squared, say so and explain why — usually too many predictors for the sample size, or one or two predictors contributing almost nothing.
  • Do not compare R-squared across different dependent variables. An R² of .70 for predicting exam scores and an R² of .30 for predicting life satisfaction are not comparable — the outcomes have different amounts of explainable variance.

Reporting in APA Style (7th ed.)

APA style expects you to report the overall model test and the variance explained together. A typical sentence for the worked example above would read:

Hours studied, sleep, and previous GPA together significantly predicted thesis marks, F(3, 56) = 30.55, p < .001, R² = .62, adjusted R² = .60.

Note the details: the F-test comes from the ANOVA table (degrees of freedom are the regression df and the residual df), R² and adjusted R² come from the Model Summary table, and both are reported to two decimal places. If you are comparing models, add the change in R-squared: “Adding sleep and previous GPA significantly improved the model, ΔR² = .14, F(2, 56) = 10.32, p < .001.” You can find ΔR² in the Model Summary table when you tick R squared change in the Statistics dialog, and its F-test in the same table’s Sig. F Change column.

When your regression is hierarchical — predictors entered in blocks — report the R-squared and adjusted R-squared for each block, so the reader can see what each set of variables added.

Key Takeaways

  • R-squared is the proportion of variance your model explains; it never decreases when you add predictors, so it flatters big models.
  • Adjusted R-squared penalises each extra predictor, so it is the honest number for comparing models with different numbers of predictors.
  • Both appear in the Model Summary table of any SPSS linear regression (Analyze > Regression > Linear) — no extra settings needed.
  • A large gap between the two numbers usually means too many predictors for your sample size; if adjusted R-squared falls when you add variables, those variables are not helping.
  • Report both in APA style, with the F-test from the ANOVA table: F(df regression, df residual), p, R², and adjusted R².

Frequently Asked Questions

Can adjusted R-squared be higher than R-squared?

No. Adjusted R-squared is always equal to or lower than R-squared, because the adjustment only ever subtracts. If the two are equal, that usually means you have one predictor and a large sample, where the penalty rounds to nothing.

Can adjusted R-squared be negative?

Yes. It happens when your model fits worse than predicting the mean for everyone — typically with predictors that have essentially no relationship to the outcome, especially in a small sample. A negative adjusted R-squared is a clear sign the model is not useful.

Which is more important, R-squared or the p-value?

They answer different questions. The p-value from the ANOVA table tells you whether the model explains more than chance. R-squared tells you how much it explains. A model can be significant with a tiny R-squared (common with large samples) or non-significant with a large R-squared (common with tiny samples). Report both.

Should I use R-squared or adjusted R-squared when choosing between two models?

Use adjusted R-squared. R-squared will always prefer the model with more predictors, even if the extra ones are useless. Adjusted R-squared only rewards predictors that improve fit enough to outweigh the penalty.

Does a high R-squared mean my model is good?

Not on its own. A high R-squared with violated assumptions — non-linearity, heteroscedasticity, influential outliers — can be misleading, and a model can overfit a small sample badly while showing a flattering R-squared. Always check the residual plots and the ANOVA significance test alongside it.

Need this done on your data? Get a free quote → /services/