How to Interpret Logistic Regression: Odds Ratios, CIs, and Examples

Binary logistic regression models the log odds of an outcome. Exponentiating a coefficient produces an odds ratio. That odds ratio is not a risk ratio and should not be described as a percentage change in probability.

The model

For a binary outcome with probability $p$, the model is commonly written as:

$\log\left(\frac{p}{1-p}\right)=\beta_0+\beta_1X_1+\cdots+\beta_kX_k$

The coefficient $\beta_j$ is the change in log odds for a one-unit increase in $X_j$, holding other variables constant. The quantity $e^{\beta_j}$ is the odds ratio.

Interpreting a continuous predictor

Suppose the adjusted odds ratio for age is 1.04 per year, with a 95% confidence interval from 1.01 to 1.07.

A careful interpretation is:

Holding the other variables in the model constant, each one-year increase in age was associated with 1.04 times the odds of the outcome. The 95% confidence interval ranged from 1.01 to 1.07.

State the unit. An odds ratio per one year is different from an odds ratio per ten years.

Interpreting a categorical predictor

Suppose treatment B has an adjusted odds ratio of 0.70 relative to treatment A.

Participants receiving treatment B had 0.70 times the adjusted odds of the outcome compared with treatment A.

Name the reference category and clarify whether the outcome is desirable or adverse. “Lower odds” is ambiguous without the outcome definition.

Odds are not probabilities

Odds equal $p/(1-p)$. When an outcome is common, an odds ratio can look farther from 1 than the corresponding risk ratio. Avoid statements such as “40% more likely” unless the quantity actually supports that wording.

Predicted probabilities can be easier to understand. Report probabilities for meaningful covariate patterns or use marginal effects, while explaining how other variables were held or averaged.

Confidence intervals and p values

An odds-ratio confidence interval that excludes 1 corresponds to a conventional two-sided significance result at the matching level, subject to the model and test. The interval also shows compatible effect magnitudes and should not be reduced to a binary significant or not significant label.

See how to report p values, confidence intervals, and effect sizes.

Interactions change the interpretation

If the model includes $X_1X_2$, the effect of $X_1$ depends on $X_2$. Do not interpret the main-effect odds ratio as an overall effect. Present joint contrasts or predicted probabilities across meaningful combinations.

Important diagnostics

• Confirm the outcome coding and event category.
• Check independence or model clustering explicitly.
• Assess linearity in the logit for continuous predictors.
• Examine sparse cells, separation, and influential observations.
• Check multicollinearity and numerical stability.
• Assess calibration and discrimination for prediction tasks.
• Validate the model on appropriate data when making predictions.

Association is not causation

An adjusted odds ratio is conditional on the variables and functional forms in the model. It does not automatically remove confounding, selection bias, measurement error, or reverse causation.

Use the statistical test decision guide to confirm whether logistic regression matches the outcome, design, and estimand before interpreting software output.

Frequently asked questions

What does an odds ratio of 2 mean?

It means the modeled odds are twice as large for the stated comparison, conditional on the adjustment set. It does not necessarily mean the probability is twice as large.

How do I interpret an odds ratio below 1?

An odds ratio below 1 indicates lower odds relative to the specified unit or reference category. State the outcome and comparison explicitly.

What if the confidence interval includes 1?

The data and model do not exclude no odds association at that confidence level. Report the interval and compatible effect sizes rather than declaring proof of no effect.

Should I report predicted probabilities?

Often yes, especially when communicating to nontechnical audiences or when interactions make odds ratios difficult to interpret.