Binary Logistic Regression in SPSS: Step-by-Step (Run & Interpret)

Binary Logistic Regression in SPSS: Step-by-Step (Run & Interpret)

Logistic regression is one of the most-used techniques in applied statistics — and one of the most misunderstood. If your outcome variable is yes/no, pass/fail, churned/stayed, or diseased/healthy, ordinary linear regression is the wrong tool. Binary logistic regression is the right one, and SPSS runs it in just a few clicks.

In this tutorial, you’ll learn how to run a binary logistic regression in SPSS using the point-and-click menus, reproduce it with paste-ready syntax, and interpret every table in the output — including the model fit statistics and the odds ratios that reviewers actually look at.

When to use binary logistic regression

Use binary logistic regression when:

  • Your dependent (outcome) variable is binary — two categories only, such as admitted/not admitted, clicked/didn’t click, relapsed/didn’t relapse.
  • Your independent variables can be anything — continuous (age, test score), categorical (gender, treatment group), or a mix.
  • Your observations are independent — each row of your dataset is a separate, unrelated case.

If your outcome has more than two categories, you need multinomial (unordered categories) or ordinal (ordered categories) logistic regression instead — both available in SPSS, but different procedures.

A common mistake: coding the outcome as text like “Yes”/”No” and getting surprised by which category SPSS treats as the event. SPSS models the probability of the higher-coded category. So code your outcome numerically (e.g., 0 = no, 1 = yes), and verify which category is the event in the “Categorical Variables Codings” and “Case Processing Summary” tables.

Running binary logistic regression in SPSS (menu steps)

Follow these clicks exactly:

  1. Analyze → Regression → Binary Logistic…
  2. Move your binary outcome variable into the Dependent box.
  3. Move your predictors into the Covariates box. If a predictor is categorical, click Categorical…, move it to the right, choose a contrast (the default, Indicator, is usually fine), click Change, then Continue.
  4. Method: keep Enter (all predictors entered at once) unless you have a reason to use stepwise methods.
  5. Click Options… and tick Hosmer–Lemeshow goodness-of-fit, Classification plots, and CI for exp(B) (confidence intervals for the odds ratios). Click Continue.
  6. Click Save… if you want predicted probabilities — tick Probabilities. Useful for diagnostics and ROC curves. Click Continue.
  7. Click OK.

Paste-ready SPSS syntax

Run the same analysis from the Syntax Editor (opens with File → New → Syntax, paste, then select all and click the green Run arrow):

LOGISTIC REGRESSION VARIABLES outcome
  /METHOD=ENTER predictor1 predictor2 predictor3
  /CONTRAST (cat_predictor)=Indicator
  /PRINT=GOODFIT CI(95)
  /CRITERIA=PIN(0.05) POUT(0.10) ITERATE(20) CUT(0.5).
  • VARIABLES outcome — your binary dependent variable.
  • /METHOD=ENTER — enter all predictors in one block (use FSTEP for forward stepwise if needed).
  • /CONTRAST (cat_predictor)=Indicator — treats the categorical predictor with indicator (dummy) coding. Drop this line if all predictors are continuous.
  • /PRINT=GOODFIT CI(95) — adds the Hosmer–Lemeshow test and 95% confidence intervals for the odds ratios.
  • CUT(0.5) — the classification cut-off; the default 0.5 is standard for balanced outcomes.

Checking the assumptions

Logistic regression doesn’t assume normally distributed residuals or homoscedasticity (unlike linear regression), but it does have its own requirements:

  1. Binary outcome — exactly two categories.
  2. Independent observations — no repeated measures or clustered data (use multilevel models instead).
  3. Linearity of continuous predictors with the logit — check with a Box–Tidwell approach or scatterplots of predictors against predicted probabilities.
  4. No severe multicollinearity — run collinearity diagnostics via a linear regression with the same predictors (Analyze → Regression → Linear → Statistics → Collinearity diagnostics) and look for VIF values well under 10.
  5. Adequate sample size — a rule of thumb is at least 10–15 events (cases in the smaller outcome category) per predictor. Small samples produce unstable odds ratios.

How to interpret the output, table by table

Omnibus Tests of Model Coefficients. The chi-square for the model tells you whether the model with your predictors beats the intercept-only (null) model. You want a significant p-value here.

Model Summary. SPSS reports two “pseudo R-squared” values: Cox & Snell and Nagelkerke. They are not true R-squared values and can’t be interpreted as “proportion of variance explained” — but higher is better. Report both; Nagelkerke is the more commonly cited of the two.

Hosmer and Lemeshow Test. A non-significant p-value (p > .05) indicates the model fits the data well. A significant result suggests poor fit — reconsider your predictors or model specification.

Classification Table. Shows how well the model classifies cases at the chosen cut-off. Report the overall percentage correct, and note the sensitivity/specificity split — a model that predicts one category far better than the other may need attention.

Variables in the Equation — the odds ratios

This is the table that matters most:

  • B — the regression coefficient (log-odds).
  • Exp(B) — the odds ratio, the number you’ll actually report. Exp(B) > 1 means the odds of the event increase as the predictor rises; Exp(B) < 1 means the odds decrease.
  • Sig. — the p-value for each predictor’s contribution.

For example, if Exp(B) = 1.35 for study hours, each additional hour is associated with 35% higher odds of the outcome — holding other variables constant. Always report the 95% confidence interval alongside it.

Odds vs probability: the two-minute version

SPSS reports log-odds and odds ratios, which confuse beginners. Probability is the chance of the event (0 to 1); odds are the ratio of that chance to the chance of the event not happening. An odds ratio of 2.0 doesn’t mean “twice as likely” in the probability sense — it means the odds are doubled. When reporting to a non-technical audience, convert with a concrete example: if the baseline odds are 1:4 (a 20% probability) and the odds ratio is 2, the new odds are 2:4, or about 33% probability. The percent-change shortcut is handy too: (Exp(B) − 1) × 100 gives the percentage change in odds per one-unit increase in the predictor.

Troubleshooting: warnings SPSS gives you

  • “Estimation terminated at iteration number 20 because maximum iterations have been reached.” The model failed to converge — usually because of separation or too few events. Simplify the model, check crosstabs for empty cells, and increase ITERATE only as a last resort.
  • Huge standard errors on some coefficients. Another symptom of separation or multicollinearity. Drop or combine problematic predictors.
  • Constants or unexpected reference categories. SPSS picks the reference category of a categorical predictor based on your Contrast settings (Indicator uses the first or last category). Always check the “Categorical Variables Codings” table so the odds ratios are compared against the reference group you intended.

Rare events and small samples

Logistic regression gets shaky when the event is rare or the sample is small — the classic rule is at least 10–15 events per predictor, and rare-event outcomes (say, fewer than 5% of cases) deserve extra caution. If your data falls short, options include collecting more data, reducing the number of predictors, or collapsing sparse categories. Report your sample size and event counts transparently; reviewers check this first.

Beyond binary: multinomial and ordinal logistic regression

If your outcome has more than two categories, binary logistic regression no longer applies. SPSS offers two direct extensions: multinomial logistic regression (Analyze → Regression → Multinomial Logistic) for unordered categories like chosen transport mode or dropout reason, and ordinal logistic regression (Analyze → Regression → Ordinal) for ordered outcomes like Likert-scale responses. Both follow the same interpret-first logic — odds ratios, model fit, classification accuracy — so everything you learned here transfers.

Reporting in APA style (7th ed.)

A binary logistic regression was performed to examine the association between [predictors] and [outcome]. The model was statistically significant, χ²(df, N = n) = [value], p = [value], Nagelkerke R² = [value], correctly classifying [xx.x]% of cases. [Predictor] was significantly associated with [outcome] (B = [value], SE = [value], Wald = [value], p = [value]), with an odds ratio of [Exp(B)] (95% CI [[lower], [upper]]), indicating that [plain-language interpretation]. The Hosmer–Lemeshow test indicated adequate model fit, χ²(df, N = n) = [value], p = [value] (p > .05).

Key Takeaways

  • Binary logistic regression predicts a two-category outcome from any mix of continuous and categorical predictors.
  • Code the outcome 0/1 and confirm which category SPSS models as the event.
  • In SPSS: Analyze → Regression → Binary Logistic; add /PRINT=GOODFIT CI(95) in syntax for fit tests and odds-ratio confidence intervals.
  • Report the Omnibus chi-square, Nagelkerke R², Hosmer–Lemeshow p, classification accuracy, and each predictor’s odds ratio with its 95% CI.
  • Check sample size (10–15 events per predictor) and multicollinearity before trusting the odds ratios.

FAQs

What is the difference between binary and multinomial logistic regression?

Binary logistic regression handles outcomes with exactly two categories. Multinomial logistic regression extends this to outcomes with three or more unordered categories (e.g., chosen transport: car, bus, train). SPSS runs both from the Regression menu family.

Why are my odds ratios enormous (or tiny)?

Extremely large or small odds ratios often signal separation — one predictor (or combination) nearly perfectly predicts the outcome — or a very small number of events per predictor. Check your crosstabs and sample size; penalized approaches or simplifying the model may help.

Can I include categorical predictors in logistic regression?

Yes — declare them via the Categorical button (or /CONTRAST in syntax). SPSS creates dummy variables automatically and you can set the reference category.

Should I use stepwise selection?

Stepwise methods are convenient but controversial: they can capitalize on chance and produce unstable models. Prefer theory-driven model specification (Enter) and report your full model.

How is logistic regression different from linear regression?

Linear regression predicts a continuous outcome and assumes normally distributed, homoscedastic residuals. Logistic regression predicts a binary outcome’s probability via the logit function and has its own set of assumptions (see the section above).


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