Discriminant Analysis in SPSS Explained

Every researcher eventually faces this question: Which group does this case most likely belong to?

That is exactly what discriminant analysis SPSS helps you answer. It is a powerful multivariate statistical technique that classifies observations into predefined groups based on a set of continuous predictor variables.

Think of it as a sophisticated sorting mechanism. You already know the groups. You want to find out which combination of variables best separates them – and then use that knowledge to classify new, unknown cases.

For example, a bank might use discriminant analysis to classify loan applicants as high-risk or low-risk. A market researcher might use it to identify which customer attributes predict brand loyalty. A healthcare analyst might use it to separate high-risk patients from low-risk ones.

Moreover, discriminant analysis in SPSS makes this process accessible through a menu-driven interface – no complex coding required. This guide walks you through everything, from assumptions to full output interpretation.

When to Use Discriminant Analysis

Before running any analysis, confirm it is the right tool for your data and research question.

Use discriminant analysis SPSS when:

• Your dependent variable is categorical with two or more groups (e.g., buyer vs. non-buyer; customer segment A, B, or C)
• Your independent variables are continuous (metric predictors such as age, income, satisfaction score, or frequency of purchase)
• Your goal is either to describe which variables separate the groups or to predict group membership for new cases
• You have a reasonably large sample – at least 20 cases per predictor variable in the smallest group

Discriminant analysis is especially common in marketing, psychology, healthcare, finance, and social research. However, if your outcome variable is binary and your assumptions are not fully met, logistic regression is a strong alternative worth considering.

Discriminant Analysis vs. Other Classification Methods

Many researchers confuse discriminant analysis with similar techniques. Understanding the distinctions helps you choose correctly.

Method	Outcome Variable	Predictors	Best Use
Discriminant Analysis	Categorical (2+ groups)	Continuous	Classify and describe group differences
Logistic Regression	Binary (2 groups)	Mixed	Predict probability of group membership
Cluster Analysis	Not predefined	Continuous	Discover unknown natural groupings
MANOVA	Continuous	Categorical	Test mean differences across groups

Discriminant analysis is uniquely suited when you need both to understand which variables discriminate between groups and to classify new observations into those groups simultaneously.

Understanding how cluster analysis in data mining differs from discriminant analysis is especially important – cluster analysis discovers groups, while discriminant analysis classifies into groups you already know.

Assumptions of Discriminant Analysis

Every statistical method has assumptions. Violating them distorts your results. Check all of the following before running discriminant analysis in SPSS.

1. Multivariate Normality
Each predictor variable should be approximately normally distributed within each group. Use the Explore function in SPSS to check this with histograms and normality tests.

2. Homogeneity of Covariance Matrices
The variance-covariance matrices should be equal across all groups. SPSS tests this automatically using Box’s M Test. A non-significant result (p > 0.05) confirms that the assumption holds.

3. No Multicollinearity
Predictor variables should not be too highly correlated with each other. Check the within-groups correlation matrix in the output. Correlations above 0.90 signal a multicollinearity problem.

4. Linear Relationships
Discriminant analysis assumes linear relationships between predictors and the discriminant function. Scatter plots help you verify this visually before running the analysis.

5. Adequate Sample Size
A minimum of 20 cases per predictor variable in the smallest group ensures stable, reliable results. Smaller samples produce discriminant functions that fail to generalise to new data.

Step-by-Step: How to Run Discriminant Analysis in SPSS

Step 1 – Prepare Your Data

Open your dataset in SPSS. Make sure:

• Your grouping variable is numeric and categorical (e.g., 1 = Loyal, 2 = Occasional, 3 = Lapsed)
• Your predictor variables are continuous numeric variables
• Missing values are handled before running the analysis

If your grouping variable is stored as a string, go to Transform → Automatic Recode and convert it to a numeric categorical variable first.

Handling missing data carefully at this stage is critical. SPSS excludes cases with missing values by default, which can reduce your effective sample size significantly. Understanding how to delete missing data in SPSS before running discriminant analysis protects both your sample size and the reliability of your model.

Step 2 – Access the Discriminant Analysis Menu

Navigate to:

Analyze → Classify → Discriminant

The Discriminant Analysis dialog box will open. This is where you specify all your variables and analysis options.

Step 3 – Assign Variables

In the main dialog box:

• Move your categorical outcome variable to the Grouping Variable box
• Click Define Range and enter the minimum and maximum group codes (e.g., Minimum: 1, Maximum: 3)
• Move all your continuous predictor variables to the Independents box

For variable entry method, beginners should select Enter Independents Together (standard method). This includes all predictors simultaneously and gives a comprehensive view of the full model.

Alternatively, choose Stepwise to let SPSS select only the predictors that contribute significantly to group separation. Stepwise is useful when you have many predictors and want to identify the most important ones.

Step 4 – Select Statistics Options

Click the Statistics button. Select:

• Means – Group means for each predictor variable
• Univariate ANOVAs – Tests whether each predictor significantly differs across groups
• Box’s M – Tests homogeneity of covariance matrices
• Within-groups correlation matrix – Checks for multicollinearity among predictors
• Unstandardised function coefficients – Needed for classification equations
• Fisher’s classification function coefficients – Used to compute group membership scores

Click Continue to return to the main dialog.

Step 5 – Set Classification Options

Click the Classify button. Select:

• Prior Probabilities – Choose All Groups Equal for exploratory research, or Compute from Group Sizes if groups differ substantially in size
• Display – Check Combined-groups plot and Separate-groups plot for visualisation
• Leave-one-out classification – This cross-validates your model’s accuracy by classifying each case using functions derived from all other cases. It gives a more realistic estimate of predictive accuracy

Click Continue.

Step 6 – Save Predicted Values

Click the Save button. Check:

• Predicted Group Membership – Adds a new column showing which group each case is classified into
• Discriminant Scores – Saves the function score for each case, useful for plotting
• Probabilities of Group Membership – Shows the probability of belonging to each group

Click Continue → OK to run the analysis.

Interpreting the SPSS Output

Discriminant analysis SPSS generates multiple output tables. Here is how to read each one systematically.

1. Tests of Equality of Group Means

This table shows whether each predictor variable significantly differentiates between groups.

• Look at the Wilks’ Lambda value and associated F-statistic for each predictor
• A significant F-value (p < 0.05) means that the predictor significantly discriminates between groups
• Variables with lower Wilks’ Lambda values have stronger discriminating power
• Wilks’ Lambda ranges from 0 to 1 – values closer to 0 indicate better discrimination

2. Box’s M Test

This tests whether the covariance matrices are equal across groups.

• A non-significant result (p > 0.05) confirms the assumption holds – proceed normally
• A significant result (p < 0.05) means covariance matrices differ. In that case, run a separate-groups covariance matrix analysis as a robustness check

3. Eigenvalues Table

Each discriminant function has an associated eigenvalue.

• Larger eigenvalues indicate better group separation
• The Canonical Correlation column shows the correlation between the discriminant scores and the grouping variable
• A canonical correlation of 0.70 or above indicates strong separation between groups
• The number of discriminant functions equals the number of groups minus one

4. Wilks’ Lambda – Multivariate Test

This table tests the overall significance of each discriminant function.

• The chi-square statistic and its p-value tell you whether the function significantly separates groups
• If p < 0.05, the discriminant function is statistically significant and useful

5. Standardised Canonical Discriminant Function Coefficients

This table shows the relative contribution of each predictor to the discriminant function.

• Coefficients with large absolute values indicate variables with greater discriminating power
• Standardised coefficients allow comparison across variables measured on different scales
• Positive coefficients increase the discriminant score; negative coefficients decrease it

This is equivalent to reading beta weights in a regression model. In fact, understanding correlation vs regression analysis helps you contextualise how discriminant function coefficients work in relation to other analytical frameworks.

6. Structure Matrix

The structure matrix shows the correlation between each predictor and the discriminant function. It identifies which variables are most strongly associated with each function.

• Loadings above 0.30 are generally considered meaningful
• This table helps you label and interpret what each discriminant function represents

7. Classification Results Table

This is the most practically important output in discriminant analysis in SPSS. It shows:

• How many cases are correctly classified into their actual group
• The overall classification accuracy percentage
• Results for both original and cross-validated classification

A classification accuracy of 75% or above is generally considered acceptable. However, always compare this against the baseline – what percentage would be correct by chance alone based on group sizes?

Real-World Example: Customer Segmentation

Imagine a retail bank wants to classify customers into three loyalty segments: Active, Passive, and Lapsed.

Predictor variables collected include:

• Years as a customer
• Number of products held
• Average monthly transaction value
• Customer satisfaction score
• Days since last interaction

After running discriminant analysis in SPSS, the output reveals:

• Two discriminant functions are extracted (3 groups − 1 = 2)
• Function 1 explains 91.4% of total variance (p < 0.001)
• Customer satisfaction score and number of products held show the largest standardised coefficients
• Overall classification accuracy: 82.3%

The bank now knows which two variables most powerfully predict the loyalty segment. Moreover, it can classify new customers into segments immediately using Fisher’s classification functions saved in the output.

This kind of segmentation insight also connects directly to data analysis and interpretation in quantitative research – discriminant analysis is one of the most powerful tools for moving from raw survey data to actionable segment profiles.

Reporting Discriminant Analysis Results

When writing up results, follow this structure:

Method section:
State the type of discriminant analysis used (simultaneous or stepwise), the number of cases, the grouping variable, and the predictor variables included.

Results section:

• Report Box’s M test result and confirm the assumption status
• Report Wilks’ Lambda, chi-square, df, and p-value for each function
• Report eigenvalue and canonical correlation for each function
• Report standardised coefficients and structure matrix loadings
• Report classification accuracy for original and cross-validated results

Example reporting sentence:
“A discriminant analysis was conducted to determine which predictor variables best differentiated between three customer loyalty segments. Box’s M test was non-significant (p = 0.18), confirming homogeneity of covariance matrices. Two discriminant functions were extracted. Function 1 was statistically significant (Wilks’ λ = 0.31, χ² = 84.6, p < 0.001) and accounted for 91.4% of the discriminating variance. Overall classification accuracy was 82.3%.”

Discriminant Analysis and Multivariate Methods

Discriminant analysis belongs to the broader family of multivariate techniques. Mastering it opens doors to more advanced analytical work.

For researchers handling complex datasets with many interdependent variables, learning how to perform multivariate analysis in SPSS builds on the same foundational logic as discriminant analysis – understanding variance, group separation, and the relationship between multiple predictors simultaneously.

In addition, discriminant analysis pairs well with factor analysis in SPSS. Researchers often reduce a large number of correlated predictors into a smaller set of factors first, then use those factor scores as cleaner inputs into a discriminant model.

Common Mistakes to Avoid

Even experienced analysts make these errors in discriminant analysis SPSS:

• Ignoring Box’s M significance – If Box’s M is significant, do not proceed without using separate covariance matrices
• Using categorical predictors – Discriminant analysis requires continuous predictors. Use logistic regression for categorical predictors
• Overfitting with too many predictors – More predictors than cases per group inflates accuracy artificially
• Reporting only original classification accuracy – Always report cross-validated accuracy. Original accuracy is optimistically biased
• Ignoring the structure matrix – Many researchers focus only on standardised coefficients. The structure matrix gives a complementary and often clearer picture of each variable’s contribution
• Skipping assumption checks – Running discriminant analysis without verifying normality and homogeneity of covariance produces unreliable results

Practical Tips for Better Discriminant Analysis Results

Follow these best practices every time you run discriminant analysis in SPSS:

• Always run descriptive statistics first – Check group means and standard deviations before touching the Classify menu
• Use stepwise selection for exploratory work – It identifies the strongest predictors automatically and removes redundant ones
• Save discriminant scores – Plot them to visualise how well groups actually separate in the discriminant space
• Use a holdout sample – Split your data. Build the model on 70–80% of cases. Validate on the remaining 20–30%
• Compare against baseline accuracy – A model classifying 80% correctly is impressive only if chance accuracy would be 33% (for three equal groups). Always benchmark properly
• Document all decisions – Record which variables were included, which method was used, and which assumptions were checked

For researchers building end-to-end analytical workflows – from survey programming through data processing to multivariate classification – discriminant analysis in SPSS represents the analytical stage where cleaned, well-structured survey data finally yields clear, predictive group insights.

Conclusion

Discriminant analysis in SPSS is one of the most powerful and practical tools in the quantitative researcher’s toolkit. It bridges the gap between description and prediction – revealing which variables separate groups and enabling classification of new observations with measurable accuracy.

The process is structured and systematic. Define your groups. Prepare your data. Check assumptions. Run the analysis. Interpret the output layer by layer – from Wilks’ Lambda to the classification matrix.

Moreover, SPSS makes every step accessible through point-and-click menus. You do not need advanced mathematical knowledge to run or interpret discriminant analysis correctly. You need methodological discipline and a clear research question.

Frequently Asked Questions