Categorical › Logistic regression

Conditional logistic (matched case-control)

Conditional logistic regression estimates predictor effects within matched sets in a matched case-control study, eliminating confounding between sets.

What is Conditional logistic (matched case-control)?

In matched case-control designs (each case matched to one or more controls on age, sex, hospital, etc.), the matching factors are confounded with the case-control distinction in a way ordinary logistic can't disentangle. Conditional logistic conditions on the matched-set totals, removing the stratum-level intercept and estimating only effects that vary within sets.

Mechanically, conditional logistic likelihood is the same as a stratified Cox PH likelihood with event time = 1 for cases and = 2 for controls (or via the equivalent ClogLog parameterisation).

Use this whenever your design pre-matched cases to controls. Treating matched data with unconditional logistic + matching variables as covariates is wrong — within-set effects and between-set effects get mixed.

When should I use Conditional logistic (matched case-control)?

Matched case-control studies (1:1, 1:k, or variable k:1).
Sibling-pair, clinic-matched, or twin studies where the matching is part of the design.

What data does it need?

Binary outcome (1 = case, 0 = control) + matching stratum column + predictors.

What does it report?

β, SE, z, p, OR + 95% CI, AIC, concordance.

What does it assume?

Independent matched sets.
Within-set predictors vary (effects of matching variables themselves are not estimable).
No effect modification by stratum unless modelled explicitly.

How do I interpret the result?

Coefficients are interpretable just like ordinary logistic — exp(β) is an OR — but the OR is conditional on the matched set rather than marginal.

Predictors constant within every matched set (e.g. the matching variable itself) drop out of the likelihood and can't be estimated.