Firth Logistic Regression

Firth Logistic Regression is a specialized statistical method designed to solve two common problems in medical research: small sample sizes and complete separation (perfect prediction).

Why use Firth instead of Standard Logistic?

Fixes Convergence Failure: Solves "complete separation" errors where standard software crashes or gives infinite odds ratios.
Unbiased Estimates: Reduces bias in coefficients when n < 100 or events are rare.
Always Finite: Guarantees finite parameter estimates and confidence intervals.

Mathematical Concept

Standard logistic regression estimates parameters (β) by maximizing the Likelihood function L(β). Firth (1993) proposed maximizing a penalized likelihood:

L*(β) = L(β) · |I(β)|^1/2

Where |I(β)| is the determinant of the Fisher Information matrix. This penalty term effectively shrinks the coefficients towards zero, reducing the bias caused by small samples.

Interpretation Guide

Interpreting Firth regression is similar to standard logistic regression, but the Odds Ratios (OR) are more conservative and reliable.

Statistic	Interpretation
Coefficient (β)	Log-odds change. Use `exp(β)` to get OR.
Odds Ratio (OR)	If OR > 1: Risk factor increases outcome probability. If OR < 1: Protective factor.
p-value	Likelihood ratio test based on penalized log-likelihood.

When to Choose This Method

You have fewer than 10-20 events per variable.
Your standard logistic regression shows extremely large Odds Ratios (e.g., OR > 50) or massive Standard Errors.
You are studying a rare disease or outcome.

Ready to analyze your data?

Run Firth Logistic Regression online for free without installation.

Start Analysis Now