The Logit Link
What you'll learn: How the logit link transforms probabilities into a scale suitable for modeling binary outcomes in generalized linear models.
When you're modeling binary outcomes (yes/no, success/failure, 0/1), you're really predicting probabilities—values that must stay between 0 and 1. But linear predictors (like the right-hand side of a regression equation) can produce any value from negative infinity to positive infinity. The logit link bridges this gap.
What the Logit Link Does
The logit link transforms a probability p into log-odds:
logit(p) = log(p / (1 - p))
This is the natural logarithm of the odds of success. If the probability of success is 0.75, the odds are 0.75/0.25 = 3 (meaning success is 3 times as likely as failure). Taking the log gives us approximately 1.099.
Why This Works
The logit function maps probabilities (restricted to 0–1) onto the entire real line (−∞ to +∞). This makes it perfect for connecting to a linear predictor. Instead of predicting p directly (which could go outside [0,1]), we predict logit(p), which can be any real number.
Think of it like this: probabilities are "cramped" into a narrow range. The logit link "stretches" them out so we can use our familiar linear regression tools, then we can transform back to probabilities when interpreting results.
The Binomial Connection
The logit link is the canonical link for binomial (binary) outcomes, meaning it naturally pairs with the exponential family representation of the binomial distribution. This makes estimation efficient and interpretation meaningful.
Key Takeaway: The logit link transforms probabilities to log-odds, allowing linear predictors to model binary outcomes without constraint violations.