METHOD GUIDE
Multiple Regression in SPSS
Collinearity, Residual Checks and Which Coefficients to Report
A model with several predictors returns a coefficient for each, and how far each can be trusted depends on collinearity, the residuals and how the predictors were chosen. Collinearity statistics and residual plots are read first, and only then is any single coefficient described.
IN SHORT
What the Coefficients Mean and What the Output Cannot Settle
A multiple regression models a continuous outcome from two or more predictors at once, so that each coefficient describes the expected difference in the outcome for a one-unit difference in that predictor with the others held constant. The model is used to predict and to describe association, and causal wording needs a design that supports it. To run multiple regression in SPSS, open Analyze, Regression, Linear, enter the outcome as Dependent and the predictors as Independent(s) with the Enter method, and request confidence intervals and collinearity statistics under Statistics. OnlineSPSS, a specialist statistical consultancy for academic, clinical and professional research, carries out and reviews this analysis through an in-house PhD-level statistician.
ANALYSIS SUPPORT
Analysis Support for This Method
Which predictors enter, and in what order, is a research decision before it is a dialog setting. Where the model is built in stages, the staged version answers the question directly.
Multiple Linear Regression
The multi-predictor model fitted within your study design, with the collinearity and residual checks, the choice of predictors and the reporting of coefficients agreed for your data.
Hierarchical Regression
Predictors entered in theoretically ordered blocks, for when the question is how much the fit changes once a new set of variables is added to the model.
BEFORE YOU RUN IT
Model Decisions to Make Before Reading Any Coefficient
Predictor choice, coding and sample size shape what the coefficients mean, so they are decided from theory and not from the output.
Predictors From Theory Predictors are chosen from the research question and prior work. Letting a selection method pick them from the same data inflates the fit and biases the estimates, and reporting only the final p-values does not repair it.
Coding of Categories A categorical predictor with several levels needs indicator variables with a stated reference group, which the Linear dialog expects to be prepared in advance. A single column of group codes assumes equal spacing.
Cases per Predictor The sample must be large enough for stable estimates, and no fixed ratio settles it. With few cases per predictor the coefficients vary widely and R squared is optimistic, which adjusted R squared only partly corrects.
Missing Data The Linear dialog excludes cases listwise by default, so a few missing values spread over many predictors can shrink the sample. The N in the output is compared with the intended sample, and the pattern of missingness is considered.
IN SPSS
How to Run Multiple Regression in SPSS
This outline uses IBM SPSS. The dialog labels and the options offered vary by version, so compare them with your own.
- Prepare the predictors, including indicator variables for categorical ones, and inspect each against the outcome.
- Choose Analyze, Regression, Linear, place the outcome under Dependent and all predictors under Independent(s), and keep the Enter method.
- Under Statistics, tick Confidence intervals, Model fit, Collinearity diagnostics, and R squared change if blocks are used.
- Under Plots, plot standardized residuals against standardized predicted values, and add the partial plots.
- Under Save, request Cook’s distance and Mahalanobis distance, so that influential cases can be examined.
- Read the Model Summary and ANOVA first, then the Coefficients table with its collinearity statistics.
READING THE OUTPUT
Where the Coefficients Table and the Collinearity Output Fit Together
The SPSS collinearity diagnostics appear in two places, as Tolerance and VIF columns in the Coefficients table and as a separate Collinearity Diagnostics table with condition indices.
| Output element | What it tells you | What to check |
|---|---|---|
| Model Summary | R, R Square, Adjusted R Square, the standard error of the estimate and, if requested, R Square Change | That Adjusted R Square is reported beside R Square, and change statistics are used only for blocks planned in advance |
| ANOVA | An F test of whether the predictors as a set improve on predicting the mean | That it is reported with both degrees of freedom and is not read as showing that every predictor matters |
| Coefficients: B and 95.0% Confidence Interval for B | Each predictor’s slope in raw units, adjusted for the others, with its interval | That the interval is reported with B and that the units are stated |
| Coefficients: Beta | Each slope in standard-deviation units | That it is used for rough comparison within one sample, since it depends on the spread and shifts when predictors overlap |
| Tolerance and VIF | How much of each predictor’s variance is shared with the others | That values past a rule of thumb, such as a VIF above 5 or 10, prompt a closer look and do not decide anything alone |
| Collinearity Diagnostics | Eigenvalues, Condition Index and Variance Proportions | That a high condition index matters when two or more predictors also carry large variance proportions on it |
| Residual and partial plots | The pattern of residuals against predicted values, and each predictor’s adjusted relation with the outcome | That curves, funnels and single points are read before the coefficients are interpreted |
| Saved distances (Cook’s, Mahalanobis) | Which cases influence the fit or lie far from the others on the predictors | That cases standing out are checked for data errors before any are excluded, and the decision is stated |
REPORTING
Reporting the Model and Choosing Which Coefficients to Show
The model-level statistics open the write-up, and the coefficient table then carries the interpretation, predictor by predictor.
Report the model first: R² = ___, adjusted R² = ___, F(df1, df2) = ___, p = ___, N = ___. Then give a table with B, its standard error and 95% CI [___, ___], the standardized beta and p for each predictor, naming the reference group of every categorical predictor. The Linear output prints no interval for R squared, so any interval comes from another program, for example a bootstrap. State the largest VIF or the range of tolerances and how collinearity was judged, which residual checks were made, and any influential case and its handling. For a model built in blocks, add the change in R squared with its F test. Describe coefficients as adjusted differences, and avoid the phrase effect of unless the design supports a causal reading.
COMMON MISTAKES
Habits That Make a Regression Table Hard to Defend
Predictor selection and the reading of single coefficients among overlapping predictors account for most reviewer objections to a regression table.
Stepwise Selection as Default Letting the program add and remove predictors by p-value gives optimistic fit and unstable coefficients. If selection is used it is declared and checked on new data, otherwise predictors follow theory.
VIF Used as a Verdict Dropping a predictor because its VIF crosses a rule of thumb can remove the variable the question is about. High values are a reason to look at the structure, and at how wide the intervals are.
Beta as Importance Standardized betas are read as the relative importance of predictors, yet they depend on the sample's variances and change when predictors overlap. They support only a rough comparison.
Unnamed Reference Group The coefficient of an indicator variable is a difference from a reference category. Without the reference named, the sign and the size of the coefficient cannot be interpreted.
Residuals Ignored A good R squared and significant coefficients can sit beside curved or funnel-shaped residuals. The residual and partial plots are examined before the model is described.
Controlled-For Language Saying a predictor was controlled for implies adjustment is complete, though omitted and mismeasured variables remain. The wording reports adjusted associations and does not claim that confounding was removed.
Model already fitted? Independent Statistical Review checks the collinearity handling, the coefficients you report and their wording before you submit.
FREQUENTLY ASKED QUESTIONS
Questions About Multiple Regression Output
On collinearity numbers, standardized coefficients and the sample size SPSS actually used: questions that surface once a model with several predictors has run.
What do the SPSS collinearity diagnostics tell me beyond VIF?
The Collinearity Diagnostics table shows condition indices, and a high index matters when two or more predictors also carry large variance proportions on the same dimension. VIF and tolerance compare each predictor with all the others, while the condition index points to the specific combination of predictors that is nearly redundant. The cutoffs often quoted, such as 30 for the index and .50 for the proportions, are rules of thumb, not tests.
Can I compare predictors by their standardized beta in SPSS?
Only roughly, and only within one sample. A standardized beta in SPSS expresses each slope per standard deviation, which puts predictors on a common scale, but it depends on the sample’s variances, shifts when predictors are correlated and does not carry over to other samples. If the aim is the relative importance of predictors, say so and use a method built for that question, such as dominance analysis or relative weights; the squared part correlation, which equals the change in R squared when a predictor is entered last, is one descriptive view.
Why is the N in my regression smaller than my dataset?
The Linear dialog drops every case with a missing value on any variable in the model by default. The total degrees of freedom in the ANOVA table plus one equal the number of cases actually used. A pairwise option exists, but it can give coefficients based on different samples, so the first step is to report how many cases were lost and why.
Can OnlineSPSS check the collinearity and the coefficients in my model?
Yes. An in-house PhD-level statistician can review the collinearity statistics, the residual checks and the coefficients you plan to report, or carry out the model on your data from the start. Once the model is settled, Results Reporting & Interpretation covers the tables and the wording. The research question and the conclusions remain with you.
How do you protect my research data and confidentiality?
Your data and project details are handled confidentially, and project information is not published. Researcher names and institutions are not used publicly without permission. An NDA is available where required and can be arranged before sensitive materials are shared. Read more about confidentiality and data security.
What happens when I request an assessment, and how long does it take?
You share your research question, design and relevant materials, such as the dataset or the output you have, and a statistically trained Research Coordinator reviews the request and responds the same day. The analysis, deliverables, timeline and one fixed project fee are then confirmed before any work begins, so the timeline follows from the agreed scope.
Still have a question? Find more answers in our full FAQ.
NEXT STEP
Is Collinearity Clouding
Your Coefficients?
List the predictors with how each was measured, the sample size and your Coefficients table. From those, the assessment confirms how collinearity will be judged and which coefficients the reported model will carry.
No Commitment Required
Strictly Confidential
Fixed Project Fee
In-House PhD-Level Statisticians
Same-Day Response
Serving Researchers Since 2014