Table of Contents
- When to Use Linear Regression in SPSS
- Simple vs Multiple Linear Regression — Which Do You Need?
- Checking the 7 Assumptions Before Running Your Analysis
- Step 1 — Prepare Your Data in SPSS
- Step 2 — Run the Linear Regression in SPSS
- Step 3 — Interpret Every Output Table
- Step 4 — Write Up Your Results in APA Format
- Common Problems and How to Fix Them
- Frequently Asked Questions
How to Run Linear Regression in SPSS: Complete Step-by-Step Guide (2026)
Linear regression in SPSS looks straightforward — click a few menus, move variables into boxes, press OK. But most students who struggle with their regression assignment aren’t struggling with the clicks. They’re struggling with: which variables go where, which of the many output tables actually matter, what R² and Beta really mean, whether their assumptions are met, and how to write all of it up in APA format.
This is a complete, ungated guide to how to run linear regression in SPSS — covering both simple (one predictor) and multiple (two or more predictors) regression, with assumptions checking, full output interpretation, APA write-up templates, and a troubleshooting section for the errors students encounter most often.
💡 Before you start: Regression is only valid if your data meets its assumptions. Many guides skip this step and show you how to run the test — but if assumptions are violated and you do not address them, your results may be technically invalid even if the numbers look clean. Check assumptions first, every time.
When to Use Linear Regression in SPSS
Use linear regression when you want to predict or explain the value of one continuous outcome variable using one or more predictor variables. Classic examples include:
- Does study time predict exam scores?
- Do age, income, and education predict job satisfaction?
- Does advertising spend predict sales revenue?
- Do exercise frequency and diet quality predict blood pressure?
Linear regression answers: how much does the outcome change for every unit change in the predictor, and is that relationship statistically significant?
⚠️ Do not use linear regression if: your outcome variable is categorical (use logistic regression instead), you only want to measure the strength of a relationship without predicting (use correlation), or you are comparing group means (use t-test or ANOVA).
Simple vs Multiple Linear Regression — Which Do You Need?
| Type | Number of Predictors | Example Research Question | SPSS Menu Path |
|---|---|---|---|
| Simple Linear Regression | 1 predictor | Does study time predict exam score? | Analyze → Regression → Linear |
| Multiple Linear Regression | 2 or more predictors | Do study time AND sleep hours predict exam score? | Analyze → Regression → Linear (same menu — add more IVs) |
Both use the same SPSS procedure. The only difference is how many variables you move into the “Independent(s)” box. This guide covers both simultaneously — the steps and output are identical, with minor additions for multiple regression.
Checking the 7 Assumptions Before Running Your Analysis
According to the UCLA Office of Advanced Research Computing, linear regression requires seven assumptions to be met for results to be valid. Here is how to check each one in SPSS:
Assumption 1 — Continuous Outcome Variable
Your dependent variable must be measured at interval or ratio level (e.g., scores, measurements, counts). If it is categorical (yes/no, group membership), use logistic regression instead.
How to check: Look at your variable type in Variable View — it should be Scale, not Nominal or Ordinal.Assumption 2 — Linear Relationship
There must be a straight-line relationship between each predictor and the outcome. Non-linear relationships cannot be validly modelled with standard linear regression.
How to check: Graphs → Legacy Dialogs → Scatter/Dot → Simple Scatter → plot each IV against the DV. Look for a roughly straight-line pattern.Assumption 3 — No Multicollinearity (Multiple Regression Only)
In multiple regression, predictors should not be too strongly correlated with each other. High multicollinearity inflates standard errors and makes coefficients unstable.
How to check: In the regression dialog → Statistics → tick Collinearity diagnostics. In the output, VIF values above 10 (or Tolerance below .10) indicate a problem.Assumption 4 — No Significant Outliers
Extreme outliers in the residuals can heavily influence your regression line and distort results significantly.
How to check: In the regression dialog → Save → tick Standardized residuals. After running, any case with a standardized residual above ±3 is a potential outlier — investigate individually.Assumption 5 — Normally Distributed Residuals
The residuals (differences between predicted and observed values) should be approximately normally distributed. This is not the same as requiring your variables to be normal.
How to check: In the regression dialog → Plots → tick Normal probability plot (P-P plot). Points should fall approximately along the diagonal line.Assumption 6 — Homoscedasticity
Residuals should show equal variance across all levels of the predictors — not fan out or narrow as predicted values increase.
How to check: In the regression dialog → Plots → set Y axis to ZRESID and X axis to ZPRED. The resulting scatterplot should show a random, even spread of points with no clear funnel or pattern.Assumption 7 — Independence of Observations
Each observation must be independent of the others — i.e., measuring different people or cases, not repeated measures on the same participant.
How to check: This comes from your study design, not SPSS. If you have repeated measures, use a repeated measures model instead.Step 1 — Prepare Your Data in SPSS
Open your dataset
Go to File → Open → Data and select your file (.sav, .csv, or .xlsx). If importing from Excel, SPSS will ask you to confirm variable names and data types.
Check Variable View
Click the Variable View tab at the bottom of the Data Editor. Confirm:
- Your dependent variable (outcome) is set to Scale measure
- Your independent variables (predictors) are also Scale for continuous data
- No variables have unexpected missing value codes
Check for Missing Data
Go to Analyze → Descriptive Statistics → Descriptives and check N for each variable. If N differs between variables, you have missing data that SPSS will handle by listwise deletion — make a note of this for your write-up.
Step 2 — Run the Linear Regression in SPSS
Open the Linear Regression Dialog
The Linear Regression dialog box will open with several boxes on the left (your variables) and assignment boxes on the right.
Assign Your Variables
- Move your outcome/dependent variable into the Dependent box
- Move your predictor/independent variable(s) into the Independent(s) box
- Leave the Method as Enter — this is the standard method for most assignments (all predictors entered simultaneously)
Click Statistics — Select These Options
Click the Statistics button and tick:
- ✅ Estimates (regression coefficients — always needed)
- ✅ Model fit (R² — always needed)
- ✅ Descriptives (means and SDs — good for reporting sample statistics)
- ✅ Collinearity diagnostics (VIF — needed for multiple regression)
- ✅ Confidence intervals (95% CIs for coefficients — increasingly required by APA)
Click Continue.
Click Plots — Set Up Residual Plots for Assumptions
Click the Plots button:
- Set Y to ZRESID and X to ZPRED → click Add
- Tick Normal probability plot
Click Continue.
Click OK to Run the Analysis
SPSS will generate output in a new Output Viewer window. You will see multiple tables — the next section explains exactly what each one means and which ones to focus on.
Step 3 — Interpret Every Output Table
SPSS produces up to six tables for linear regression. Here is what each one tells you and which values matter for your assignment:
Table 1: Model Summary
Model Summary Output
| Model | R | R Square | Adjusted R Square | Std. Error of the Estimate |
|---|---|---|---|---|
| 1 | .741 | .549 | .542 | 8.64 |
- R = the correlation between observed and predicted values. Higher is better.
- R Square (R²) = the proportion of variance in your outcome explained by your predictors. R² of .549 means your model explains 54.9% of the variance in exam scores. This is your primary model fit statistic.
- Adjusted R² = R² corrected for the number of predictors. Always report Adjusted R² for multiple regression (it penalises for adding predictors that don’t improve the model). For simple regression, R² and Adjusted R² will be very similar.
- Std. Error of the Estimate = the average amount your predictions are off by, in the units of your outcome variable. Lower is better.
Table 2: ANOVA Table — Is the Overall Model Significant?
ANOVA Output
| Sum of Squares | df | Mean Square | F | Sig. | |
|---|---|---|---|---|---|
| Regression | 5847.20 | 2 | 2923.60 | 39.14 | <.001 |
| Residual | 4706.80 | 63 | 74.71 | ||
| Total | 10554.00 | 65 |
- F statistic and Sig. = tests whether your overall model predicts the outcome significantly better than just using the mean. If Sig. < .05, your model is statistically significant overall.
- If the overall model is NOT significant (Sig. > .05), do not interpret individual coefficients — the model as a whole fails to predict the outcome.
- The df values give you the numbers for APA reporting: F(regression df, residual df) = F(2, 63) in this example.
Table 3: Coefficients — The Most Important Table
Coefficients Output
| Variable | B | Std. Error | Beta | t | Sig. | 95% CI Lower | 95% CI Upper | VIF |
|---|---|---|---|---|---|---|---|---|
| (Constant) | 18.42 | 4.21 | 4.37 | <.001 | 10.01 | 26.83 | ||
| Study Hours | 4.12 | .58 | .621 | 7.10 | <.001 | 2.97 | 5.27 | 1.84 |
| Sleep Hours | 1.87 | .74 | .221 | 2.53 | .014 | .40 | 3.34 | 1.84 |
| Column | What It Means | When to Use It |
|---|---|---|
| B (Unstandardised coefficient) | Change in outcome for every 1-unit increase in this predictor, holding others constant. Study Hours B = 4.12: each additional study hour predicts a 4.12-point increase in exam score | Use for prediction equations and describing effect magnitude in the original units |
| Beta (Standardised coefficient) | The same effect expressed in standard deviations. Allows fair comparison across predictors regardless of different measurement scales | Use to compare which predictor has the strongest effect: Study Hours (β = .621) is stronger than Sleep Hours (β = .221) |
| Sig. (p-value) | Whether this individual predictor significantly predicts the outcome independently of the others | Report for each predictor — both should be below .05 for significance |
| 95% Confidence Interval | The range within which the true B value is likely to fall 95% of the time. If it does not cross zero, the effect is significant | Include in write-up per APA 7th edition recommendations |
| VIF | Variance Inflation Factor — measures multicollinearity. VIF below 5 is acceptable; above 10 is problematic | Check for multiple regression — if VIF is high, predictors are too correlated and coefficients are unreliable |
⚠️ Writing the regression equation: Exam Score = 18.42 + (4.12 × Study Hours) + (1.87 × Sleep Hours). Use the B values (not Beta) to write the prediction equation. Beta values are only for comparing predictor importance.
Step 4 — Write Up Your Results in APA Format
Here is a complete APA 7th edition write-up for the multiple regression above — use this as your template and substitute your own values:
A multiple linear regression was conducted to examine whether study hours and sleep hours predicted exam scores. Preliminary assumption checks confirmed linearity via scatterplot inspection, normal distribution of residuals via P-P plot, homoscedasticity via residual plot, and no multicollinearity (VIF = 1.84 for both predictors).
The overall model was statistically significant, F(2, 63) = 39.14, p < .001, and explained 54.9% of the variance in exam scores (R² = .549, Adjusted R² = .542). Study hours was the strongest significant predictor (β = .621, B = 4.12, 95% CI [2.97, 5.27], p < .001), with each additional study hour associated with a 4.12-point increase in exam score. Sleep hours also made a significant independent contribution (β = .221, B = 1.87, 95% CI [.40, 3.34], p = .014).
A simple linear regression was conducted to examine whether study hours predicted exam scores. The model was statistically significant, F(1, 64) = 71.40, p < .001, and study hours explained 52.7% of the variance in exam scores (R² = .527). Study hours significantly predicted exam performance (β = .726, B = 4.87, 95% CI [3.72, 6.02], p < .001), with each additional study hour associated with a 4.87-point increase in exam score.
Common Problems and How to Fix Them
❌ Problem: “I cannot find the Linear Regression option in the menu”
You may be in the wrong SPSS module, or the menu may be hidden. Ensure you are in the Data Editor window (not the Output Viewer). The path is Analyze → Regression → Linear.
✅ Fix: Click on the Data Editor window first, then navigate the menus from there.❌ Problem: My R² is very low (below .10)
A very low R² suggests your predictors explain very little variance in the outcome. This is a valid finding in many social science contexts — it does not mean you ran the analysis wrong.
✅ Fix: Report it honestly and discuss possible reasons in your interpretation — unmeasured variables, non-linear relationships, or the predictors genuinely having little relationship with the outcome.❌ Problem: The overall model is significant but none of the individual predictors are
This happens most often with multicollinearity — when two or more predictors are strongly correlated, each appears non-significant individually even though together they predict the outcome.
✅ Fix: Check VIF values. If VIF is above 5-10, multicollinearity is the issue. Consider removing one of the correlated predictors or combining them.❌ Problem: My residual plot shows a clear funnel shape (wider spread as fitted values increase)
This indicates heteroscedasticity — a violation of the homoscedasticity assumption. Your standard errors may be inaccurate.
✅ Fix: Try transforming your outcome variable (log transformation is most common) or use robust standard errors. Report the violation in your write-up and how you addressed it.❌ Problem: My P-P plot shows points clearly deviating from the diagonal line
This suggests residuals are not normally distributed — a violation of the normality assumption. The severity depends on how far points deviate and your sample size (large samples are more robust to this).
✅ Fix: Check for outliers first (they often cause non-normality). For large samples (N > 100), minor deviations are usually acceptable. Report what you found and justify proceeding.❌ Problem: I don’t know whether to report B or Beta
Use B (unstandardised) for the prediction equation and for describing how much the outcome changes per unit of predictor. Use Beta (standardised) for comparing which predictors are relatively more important. Most APA write-ups report both.
✅ Fix: Report both in your coefficients table, highlight Beta when discussing relative predictor importance, and use B when writing the regression equation.Still Stuck on Your SPSS Regression Assignment?
Our PhD statisticians run your regression correctly, check all seven assumptions, interpret every output table, and write your results section in APA format — delivered by your deadline.
Get SPSS Regression Help →Frequently Asked Questions
Go to Analyze → Regression → Linear. Move your outcome variable into the Dependent box and your predictor into the Independent(s) box. Click Statistics and tick Estimates, Model fit, and Confidence intervals. Click Plots, set Y to ZRESID and X to ZPRED, and tick Normal probability plot. Click OK to run the analysis. The key output tables are the Model Summary (R²), the ANOVA table (overall model significance), and the Coefficients table (individual predictor effects).
Simple linear regression uses one predictor variable to predict one outcome. Multiple linear regression uses two or more predictors. In SPSS, both use the same Linear Regression procedure — the only difference is that you move more than one variable into the Independent(s) box for multiple regression. Multiple regression also requires you to check for multicollinearity using VIF values from the Collinearity diagnostics option.
R² (R-squared) represents the proportion of variance in your outcome variable that is explained by your predictor(s). An R² of .45 means your model explains 45% of the variance in the outcome. For multiple regression, always report Adjusted R², which accounts for the number of predictors and does not artificially inflate when you add more variables. In social sciences, R² above .30 is generally meaningful; above .50 is considered strong.
The appropriate response depends on which assumption is violated and how severely. For outliers: investigate and consider removing if genuinely erroneous data. For non-normality: check outliers first; large samples are relatively robust. For heteroscedasticity: consider log-transforming the outcome variable. For multicollinearity: remove one correlated predictor or combine predictors. Always report assumption violations and how you addressed them — attempting to hide violations is academically problematic.
Report: the test name and purpose; whether assumptions were met; the overall model F statistic with both degrees of freedom, p-value, R², and Adjusted R²; and for each significant predictor: the standardised beta coefficient (β), unstandardised coefficient (B) with 95% CI, t-value, and p-value. Example: “Study hours significantly predicted exam scores (β = .62, B = 4.12, 95% CI [2.97, 5.27], t = 7.10, p < .001).”
Leave a Reply