How to run a logistic regression

Introduction to Logistic Regression

Used for outcomes that are categorical with two options (binary).
Example outcome: Cure or Not Cure.
Predictors can be categorical, continuous, or a combination.

Running a Binary Logistic Regression

Access logistic regression through: Analyze > Regression > Binary Logistic.

Hierarchical Regression Approach

Steps similar to those used in multiple regression analysis.
Analysis done in blocks:
- Block 1: Enter intervention variable as the dependent variable.
- Block 2: Enter 'number of days of treatment'.
- Block 3: Enter interaction effect of variables.

Configuration Steps

Set reference category for categorical variables (e.g., intervention).
Under the save options, select options including standardized residuals.

Assessing Statistical Significance

Evaluate whether each model is statistically significant:
- Block 1: Check p-value for initial variable (intervention).
- Block 2: Assess p-value for variance explained with the second variable (number of days); if non-significant, do not proceed with this block.
- Block 3: Evaluate interaction effect; if p-value is non-significant, do not include this block in further analysis.

Rerunning Analysis with Significant Models

Close previous analyses and run logistic regression with only significant blocks.
Exclude blocks that have proven non-significant.

Final Outputs of Logistic Regression

Key output components include:
- Omnibus Test: Overall model significance.
- Model Coefficients: Indicate effect size (e.g., Nagelkerke R-square).
- Contingency Table: Displays observed vs. predicted outcomes.
- Variables in the Equation: Similar to regression coefficients, show standardized beta values, Wald test, and odds ratios (Exponential B).
- Observed Groups and Predicted Probabilities: Summary of model performance.

Conclusion

This guide provides a step-by-step process for conducting logistic regression using hierarchical methods. Further demonstrations will be provided in computer lab sessions.