Binary Logistic Regression Analysis using SPSS: What it is, How to Run, and Interpret the Results.

Binary logistic regression is the go-to method when the outcome is dichotomous—coded as 0/1—so researchers can estimate how one or more predictors shift the odds of belonging to one group versus another. Instead of modeling a continuous dependent variable like linear regression, logistic regression predicts group membership by calculating the probability of “success” relative to “failure,” producing results in odds ratios. That makes it useful for questions such as whether customers will return, whether applicants are admitted, whether candidates win elections, or whether job applicants are selected.

The transcript frames logistic regression as a tool for assessing the strength and direction of relationships between independent variables and a binary dependent variable. The dependent variable is coded with one category as 1 (the “target” group) and the other as 0 (the reference group). For example, private bank choice can be coded as 1 and public bank choice as 0. Predictors can include factors like technology, interest rates, value-added services, perceived risks, and reputation. The model estimates how changes in these predictors affect the log-odds of choosing the target group, and the odds ratio translates those effects into an interpretable metric.

Assumptions differ from linear regression: logistic regression does not require a linear relationship between dependent and independent variables, and predictors do not need to be interval-scaled or normally distributed. The dependent variable must be dichotomous, with mutually exclusive and exhaustive categories—each case belongs to exactly one group. Sample size guidance is also emphasized: larger samples are needed than in linear regression, with commonly cited rules such as at least 50 cases per predictor (Field, Lee, and Blank, 2000) or at least 30 observations per independent variable.

Running the analysis in SPSS follows a standard workflow: Analyze → Regression → Binary Logistic. The dependent variable is assigned as the “Preferred Choice,” while independent variables are placed as covariates. Options can include the Hosmer–Lemeshow goodness-of-fit test, confidence intervals for odds ratios, and related diagnostics.

Interpretation starts with the output structure. The Case Processing Summary confirms the number of observations (341 respondents in the example) and shows how the dependent variable is coded (public sector bank = 0, private sector bank = 1). Block 0 provides a baseline model with no predictors. Block 1 adds predictors and is evaluated using the Omnibus test of model coefficients (significant improvement over the null model) and the Hosmer–Lemeshow test (a good fit when the result is not significant, i.e., significance ≥ 0.05). A contingency table further checks whether observed and predicted classifications align.

Model performance is assessed via pseudo R-square measures (e.g., Nagelkerke’s R-square as an adjusted Cox & Snell statistic) and, crucially, the classification table. In the example, the model correctly classifies about 75.7% of cases, with sensitivity reported around 96.4% for predicting private bank choice and specificity around 17.8% for predicting public bank choice—indicating strong detection of the target group but weaker identification of the reference group.

Finally, the odds ratio table identifies which predictors significantly affect the outcome. Odds ratios greater than 1 increase the odds of choosing the target group as the predictor rises; odds ratios less than 1 decrease those odds. Confidence intervals that do not cross 1 signal statistical significance. In the example, value-added services shows odds greater than 1 (with a 95% confidence interval entirely above 1), while perceived risk shows an odds ratio below 1 (with a confidence interval entirely below 1), meaning higher perceived risk reduces the likelihood of choosing private banks.