Academic Support · Research Methods
Logistic Regression in SPSS: How to Read Odds Ratios
An SPSS binary logistic regression table can answer a useful research question, but only if the outcome, reference groups and odds ratios are read correctly. This guide shows a worked interpretation and the checks to make before writing your results.
When is binary logistic regression appropriate?
Use binary logistic regression when the outcome has two coded states, such as whether a respondent completed a programme (1) or did not (0), and you want to examine its association with one or more predictors. First confirm which state SPSS models. A reversed event code reverses the interpretation of every odds ratio.
Before fitting the model, document the research question, predictor coding, category reference groups, missing values and sample selection. The coefficient for a continuous predictor describes a one-unit change in that predictor while other included variables are held constant. A one-unit change in age, weekly hours or a five-point scale can mean very different things.
Read the output in the right order
| SPSS item | What it tells you | Common mistake |
|---|---|---|
| Outcome coding | Which event is being modeled | Describing the opposite outcome |
| B | Change in log odds for the specified comparison | Calling it a probability change |
| Exp(B) | Odds ratio for the specified unit or category | Calling it a percentage-point change |
| 95% CI for Exp(B) | Range of odds ratios compatible with the model and data | Reporting only the p value |
| Reference category | Group against which a category is compared | Assuming the comparison direction |
IBM’s binary logistic regression documentation describes these outputs. UCLA’s odds-ratio explanation provides further interpretation examples.
A worked example: odds are not probability
Illustrative data only: Imagine a GCC training study whose outcome is programme completion (1=yes). Suppose weekly study hours has an adjusted Exp(B) of 1.30, a 95% confidence interval of 1.08–1.56 and p=.005. The careful statement is: For each additional weekly study hour, the odds of completion were 30% higher, adjusting for the other variables in this model. This is an association, not evidence that an extra hour caused completion.
It does not mean the probability rose by 30 percentage points. If the comparison probability were 20%, its odds would be 0.20 ÷ 0.80 = 0.25. Multiplying those odds by 1.30 gives 0.325; converting back gives 0.325 ÷ 1.325, or about 24.5%. At that starting point, the illustrative probability difference is about 4.5 percentage points. A different starting probability produces a different percentage-point change.
Now suppose an employment-group comparison has Exp(B)=0.72, 95% CI 0.45–1.15 and p=.17. The estimate points toward lower odds than the stated reference group, but the interval includes 1. Describe the estimate and its uncertainty; avoid a confident claim of a group difference.
Checklist before you write the results section
- State the event coded 1 and the reference group for each categorical predictor.
- Check sample size, exclusions and missing values against the study protocol.
- Record the unit of each continuous predictor and whether it was rescaled.
- Report the odds ratio and confidence interval together; use the p value as supporting information.
- Separate association from causation and odds from probability.
- Discuss model assumptions, design limits and potentially important unmeasured variables.
- Make sure tables and prose describe the same model and coding.
Choose the right interpretation before drafting a results paragraph
| Question | Check in the analysis | What to write |
|---|---|---|
| Which event is modeled? | Inspect the dependent-variable encoding and the outcome named in the research question. | “The modeled event was programme completion (1=yes).” |
| What does the predictor compare? | Confirm a continuous variable’s unit or the categorical reference group in SPSS. | “The comparison is each additional weekly hour, holding other included predictors constant.” |
| How precise is the estimate? | Read Exp(B) with its confidence interval and check the analysis sample. | “The adjusted odds ratio was 1.30 (95% CI 1.08–1.56).” |
| Can probability be stated? | Specify a valid starting probability and the other predictor values for a model scenario. | “At an illustrative 20% baseline, the model scenario corresponds to about 24.5%, not 50%.” |
The last number is a teaching calculation using odds multiplied by 1.30, not a prediction from an actual SPSS dataset. Do not apply it to a study population without a fitted model and an appropriate baseline. UCLA’s SPSS odds-ratio explanation shows why odds and probability use different denominators.
Model-checking decisions that change the interpretation
A clear Exp(B) table does not by itself establish a useful model. Before submitting a dissertation or paper, check how missing observations changed the analysis sample, whether sparse categories create unstable estimates, and whether the model form is appropriate for continuous predictors. If a predictor is rescaled or the reference category changes, its odds-ratio wording must change with it. Compare the reported table with the final saved SPSS output rather than an earlier exploratory run.
For example, “age” measured in years and “age” measured in decades produce different one-unit odds ratios from the same underlying relationship. A ten-year increase is not a one-year increase. Likewise, an odds ratio for Group A relative to Group B should not be casually described as Group B relative to Group A. State the category order explicitly.
Frequently asked questions
Does Exp(B)=1.30 mean a 30% higher probability?
No. It means 30% higher odds for the specified comparison, conditional on the other terms in that model. Translate to probability only for a stated baseline and model scenario.
Does a confidence interval crossing 1 prove no association?
No. It indicates the interval includes both lower and higher odds than the reference at the stated confidence level. Consider the estimate, precision, study design and sample size.
Can Labeeb choose the model and write my thesis results for me?
Labeeb offers research consultation, methodological support and interpretation assistance. The researcher should approve the analysis, verify their own data and remain responsible for authorship and conclusions.
Check the interpretation before submission
If you are deciding whether this is the right method, read Labeeb’s statistical-test selection guide. For help checking an approved analysis plan or explaining results accurately, see our data analysis consultation service. You remain responsible for your research decisions and final authorship.