How do you run a logistic regression in SPSS?

Open Analyze, then Regression, then Binary Logistic. Move your 0-and-1 outcome into Dependent and your predictors into Covariates, click Categorical to declare any factor predictors and their reference category, request the Hosmer-Lemeshow test under Options, then click OK and read the Variables in the Equation table.

How do you interpret Exp(B) in logistic regression?

Exp(B) is the odds ratio. A value above 1 means the predictor increases the odds of the outcome, and below 1 means it decreases them, with the distance from 1 giving the effect size. Check the confidence interval: if it includes 1, the effect is not statistically reliable whatever the point estimate suggests.

What does the Hosmer-Lemeshow test tell you?

The Hosmer-Lemeshow test assesses how well the logistic model fits the data. Unusually, you want it to be non-significant: a p-value above .05 means there is no evidence of poor fit between the predicted and observed outcomes. A significant result signals that the model does not fit the data well.

How many cases do you need for a logistic regression?

A common rule of thumb is at least ten outcome events for each predictor in the model, counting the rarer of the two outcomes. Too few events relative to predictors produces unstable odds ratios with very wide confidence intervals, and in extreme cases causes complete separation, where a predictor perfectly splits the outcome.

How to run a logistic regression in SPSS

To run a logistic regression in SPSS, you open Analyze > Regression > Binary Logistic, enter a binary outcome and your predictors, declare any categorical covariates and their reference category, and read the odds ratios in the Variables in the Equation table. The model that holds up depends on checking model fit, watching for problems like separation, and interpreting Exp(B) as the odds ratio rather than a raw coefficient. This guide covers the procedure, the fit statistics, and the reporting.

When logistic regression is the right model

Binary logistic regression predicts a dichotomous outcome, such as pass or fail, relapsed or not, from one or more predictors that can be continuous or categorical. Linear regression is the wrong tool for a yes-or-no result because it can predict impossible probabilities below zero or above one; logistic regression instead models the log odds of the outcome, keeping predictions inside a sensible range. When your outcome is continuous, you would return to multiple regression in SPSS instead. The deeper meaning of the coefficients is unpacked in interpreting logistic regression.

Coding the variables before you run it

Two setup steps decide whether the output makes sense. First, code the dependent variable as 0 and 1, where 1 is the event you are predicting, because SPSS models the probability of the higher code. Second, declare every categorical predictor through the Categorical button and choose a sensible reference category (commonly the first or last), since each odds ratio is then read against that baseline. Skipping this step is the most common reason a logistic model is misread, because the odds ratios end up compared against the wrong group.

Running the model and the syntax to keep

Open Analyze > Regression > Binary Logistic, move the outcome into Dependent and the predictors into Covariates. Click Categorical to declare factor predictors, and under Options tick the Hosmer-Lemeshow goodness-of-fit test, Classification plots, and CI for Exp(B). The reproducible syntax is:

LOGISTIC REGRESSION VARIABLES outcome
  /METHOD=ENTER age treatment severity
  /CONTRAST (treatment)=Indicator(1)
  /PRINT=GOODFIT CI(95)
  /CRITERIA=PIN(.05) POUT(.10).

Keep an eye out for complete separation, where a predictor perfectly splits the outcome; SPSS signals it with a wildly large coefficient and a huge standard error, and it usually means you have too few events for that predictor.

The assumptions that still apply

Logistic regression drops the normality and equal-variance assumptions of linear models, but it keeps others. Observations must be independent; continuous predictors should have a linear relationship with the log odds, testable with the Box-Tidwell approach; predictors should not be strongly collinear; and you need an adequate number of events per predictor, with a common rule of thumb of at least ten outcome events for each predictor in the model. Thin data is the usual cause of unstable odds ratios, so this last point matters as much as any formal test.

Reading the logistic output in order

Several tables matter, and the order helps. The Omnibus Tests of Model Coefficients gives a chi-square test of whether your predictors improve on the null model. The Model Summary reports the minus two log likelihood and two pseudo R-squared figures, Cox and Snell and Nagelkerke, of which Nagelkerke is the more interpretable. The Hosmer-Lemeshow test is one you want to be non-significant, because a p-value above .05 indicates the model fits acceptably. The Classification Table shows how accurately the model sorts cases. Finally, the Variables in the Equation table carries the findings: B, the Wald test and its significance, and Exp(B) with its confidence interval. Reading these sits within working through the SPSS output viewer.

Interpreting the odds ratio

Exp(B) is the odds ratio, and it is the number you actually report. A value above 1 means the predictor raises the odds of the outcome; below 1 means it lowers them; and the distance from 1 gives the size of the effect, so an odds ratio of 2.0 means the odds double for each one-unit rise in that predictor. The confidence interval is decisive: if it includes 1, the effect is not statistically reliable, regardless of where the point estimate sits. For a categorical predictor, the odds ratio is always read against the reference category you set earlier, which is why that choice was so important.

Reporting logistic regression in APA style

Report the overall model fit, then each predictor as an odds ratio with its confidence interval. A clean passage reads: the model significantly predicted relapse, chi-square(3) = 24.6, p < .001, and explained about 28 percent of the variance (Nagelkerke R squared = .28), with acceptable fit on the Hosmer-Lemeshow test (p = .54). Each additional point of baseline severity raised the odds of relapse, odds ratio = 1.34, 95 percent confidence interval [1.12, 1.60], p = .002. Name your reference categories so the reader can place each odds ratio. The conventions are in reporting the model in APA style, and the interpretation deepens in interpreting logistic regression.