Use logistic regression when your data asks a binary "yes or no" question. It identifies how specific factors influence the likelihood of an event occurring, such as whether a user converts or a part passes inspection. This method provides clear odds ratios to help you understand the impact of your inputs.
Understanding Binary Outcomes
Logistic regression is designed for situations where the answer is binary. Whether you are analyzing clinical trial results, manufacturing quality checks, or marketing conversions, this method translates complex relationships into actionable probabilities.
Unlike linear methods that predict continuous numbers, this approach focuses on the log-odds of an event. By calculating the odds ratio for each factor, you can clearly communicate which variables are driving your results and by how much.
Evaluating Model Performance
Lattice provides a complete diagnostic suite to verify your model's reliability. Beyond just coefficients, you get a 2x2 confusion matrix that calculates accuracy, precision, and recall using a default threshold of 0.5.
The AUC (Area Under the Curve) is included to help you assess the model's overall discrimination power. This helps you understand how well your predictors differentiate between the two outcome groups, allowing you to gauge the model's readiness for real-world deployment.
Interpreting Results with Confidence
Every analysis returns the 95% Wald Confidence Interval for your odds ratios, providing the range of uncertainty for your estimates. This ensures that your findings are presented with the necessary context for academic or operational decision-making.
Lattice handles the mathematical heavy lifting, providing McFadden's pseudo-R² to help you assess the quality of the fit. This gives you a clear sense of how much variance your predictors explain, helping you avoid over-interpreting results in models with poor predictive power.
Data Requirements and Best Practices
To get the most accurate results, ensure your response variable is binary. If you do not specify a positive class, the platform will automatically select one and provide a warning; we recommend always explicitly defining your positive class to ensure the odds ratios align with your specific research or business goals.
If you encounter a 'perfect separation' error, this typically indicates that your predictors are too good at predicting the outcome—meaning one variable might perfectly divide your data. In these cases, consider whether certain variables need to be combined or if the data requires a different regularization approach.
1 · Intent → method
An LLM picks regression_logistic from a fixed catalog.
2 · Method → numbers
Deterministic Python engine runs the math. Same input → same output.
3 · Numbers → plain language
A second LLM translates the result into your domain’s vocabulary.
What does the Odds Ratio (OR) tell me in logistic regression?
The Odds Ratio represents the change in the odds of your outcome happening for every one-unit increase in a predictor. An OR greater than 1 means the odds increase, while an OR less than 1 means the odds decrease.
How do I know if my logistic regression model is performing well?
Look at the AUC (Area Under the Curve) and McFadden's pseudo-R². An AUC above 0.8 is generally considered good, while a McFadden's pseudo-R² above 0.2 indicates a good model fit for your binary data.
Tool input schema
Schema for regression_logistic not exported yet (run pnpm export:registry).