Guides / multiple regression spss
How to run multiple regression in SPSS
The Linear Regression dialog will give you an answer with default settings. It will not give you the diagnostics that tell you whether to trust it — and four of those are a single checkbox each.
Go to Analyze > Regression > Linear, put your outcome in Dependent and your predictors in Independent(s). Click Statistics and tick Estimates, Confidence intervals, Model fit, Collinearity diagnostics and Durbin-Watson. Click Plots and put ZRESID on the Y axis and ZPRED on the X axis. Click OK and read the Model Summary, ANOVA and Coefficients tables.
Step by step
The click path
Your outcome must be continuous. For a binary outcome use Analyze → Regression → Binary Logistic instead.
Leave Method on Enter unless you have a specific reason. Stepwise selection is data-driven and widely criticised — it capitalises on chance and produces models that do not replicate.
Confidence intervals and collinearity are both off by default and both matter.
This single plot checks linearity, homoscedasticity and outliers at once.
Adds a column to your dataset flagging influential cases.
SPSS offers Forward, Backward and Stepwise selection. All three choose predictors based on p-values in your particular sample, which inflates R², biases coefficients away from the null and produces models that fail on new data. Enter your predictors deliberately, chosen from theory.
Output
Reading the output
| Table | What to read |
|---|---|
| Model Summary | Adjusted R Square — not R Square, once you have more than one predictor |
| ANOVA | Whether the model as a whole beats predicting the mean. Sig. should be below .05 |
| Coefficients | B, its confidence interval, Beta, t, Sig., and the VIF column |
In the Coefficients table, B is the unstandardised coefficient: the change in the outcome per one-unit change in that predictor, holding the others constant. Beta is the standardised version, useful only for comparing predictors measured on different scales.
R Square rises whenever you add a predictor, including random noise. Adjusted R Square penalises for the number of predictors. If adding a variable raises R Square but lowers Adjusted R Square, that variable is not earning its place.
Diagnostics
Checking assumptions
The ZRESID against ZPRED scatterplot is the single most informative output and takes one checkbox.
| Pattern in the plot | Means | Response |
|---|---|---|
| Formless band around zero | Healthy | Proceed |
| Funnel widening to the right | Heteroscedasticity | Bootstrap, or transform the outcome |
| Curved band | Non-linearity | Add a quadratic term, or transform |
| Isolated points far out | Outliers | Check Cook's distance; investigate those cases |
Send the dataset and your research question. A named statistician fits the model, runs the full diagnostics and returns annotated output with the coefficients interpreted in plain language.
Get a fixed quoteVIF
Multicollinearity
Ticking Collinearity diagnostics adds Tolerance and VIF columns to the Coefficients table.
| VIF | Reading |
|---|---|
| Below 5 | No practical concern |
| 5–10 | Worth attention |
| Above 10 | Usually treated as serious |
Collinearity damages the interpretation of individual coefficients rather than the model's overall prediction. Coefficients become unstable, swinging in size or sign when another predictor is added, and standard errors inflate. If your question is predictive it may not matter; if it is about which predictor matters, it may invalidate the analysis.
Dummy coding
Categorical predictors
SPSS Linear Regression does not accept categorical predictors directly. A variable with three categories needs two dummy variables.
Create a 0/1 variable for each category except one.
That omitted category is the reference group, and every coefficient is interpreted against it.
Include all of them or none — entering some of a set makes the coefficients uninterpretable.
Every coefficient for a dummy variable means “compared with the reference group”. A results table that omits which category that was cannot be interpreted.
Pitfalls
Five mistakes
1. Using stepwise selection
Data-driven predictor selection inflates R² and produces models that do not replicate.
2. Reporting R Square in a multiple model
Use Adjusted R Square.
3. Not requesting the residual plot
One plot checks three assumptions. It is off by default.
4. Testing normality of the outcome
The assumption concerns the residuals.
5. Causal language
Regression on observational data estimates association. “Predicted” is a statistical term, and readers will take it literally.
Reporting
How to report it
| Element | How to write it |
|---|---|
| Overall model | F(2, 107) = 18.44, p < .001, adjusted R² = .24 |
| A predictor | b = 3.96, SE = 0.42, β = .48, t(107) = 9.43, p < .001, 95% CI [3.13, 4.79] |
| Non-significant predictor | b = 0.21, SE = 0.38, β = .04, t(107) = 0.55, p = .583 |
Send your dataset. A named statistician fits the model, runs the diagnostics and returns APA tables you can use directly.
See SPSS data analysisAnswers
Frequently asked questions
How do I run multiple regression in SPSS?
Analyze > Regression > Linear. Put your continuous outcome in Dependent and predictors in Independent(s), leave Method on Enter, then tick Confidence intervals and Collinearity diagnostics under Statistics and set up the ZRESID against ZPRED plot under Plots.
Should I use stepwise regression in SPSS?
Generally no. Stepwise selects predictors on p-values in your particular sample, which inflates R squared, biases coefficients and produces models that fail to replicate. Enter predictors deliberately based on theory.
Should I report R Square or Adjusted R Square?
Adjusted R Square whenever you have more than one predictor. R Square rises with every variable added, including irrelevant ones, so it overstates fit. If adding a predictor raises R Square but lowers Adjusted R Square, that predictor is not earning its place.
What VIF value indicates multicollinearity in SPSS?
Below 5 is generally untroubling, 5 to 10 warrants attention, and above 10 is usually treated as serious. Collinearity harms the interpretation of individual coefficients rather than overall prediction, so it matters most when your question is which predictor is responsible.
How do I include a categorical predictor in SPSS regression?
Create dummy variables using Transform > Recode into Different Variables — one 0/1 variable for each category except one, which becomes the reference group. Enter the whole set together, and state in your write-up which category was the reference.
Does multiple regression assume the outcome is normally distributed?
No — it assumes the residuals are. A skewed outcome can produce perfectly normal residuals once predictors are accounted for. Use the normal P-P plot of standardised residuals SPSS produces rather than testing the raw variable.
Keep reading