MerQuR+Statistics Glossary

Statistics Glossary → Profile likelihood confidence interval

Profile likelihood confidence intervalconcept

A method that derives a parameter's confidence interval from the profile likelihood curve, in which the other parameters are re-estimated to maximize the likelihood at each value of that parameter.

The parameter of interest θ is fixed at a given value and all other parameters are re-estimated under that constraint; the resulting maximum log likelihood is the profile log likelihood of θ. The confidence interval consists of all θ values that the likelihood ratio test does not reject: 2 [ℓ(θ̂) - ℓ_p(θ)] ≤ χ²₁(1 - α); for the 95% level the threshold is about 3.84.

The Wald interval (estimate ± 1.96 × SE) assumes a symmetric parabola around the peak of the likelihood. With small samples, rare events or a parameter near a boundary (e.g. a variance close to zero), this assumption breaks down and the Wald interval misleads. The profile interval follows the true shape of the likelihood, can be asymmetric and is unaffected by reparameterization. The price is that the model must be refitted for every candidate value.

An everyday example

A rare side effect occurs in 3 of 200 patients. The Wald method may suggest an interval so symmetric that its lower limit falls below zero, although a proportion cannot be negative. The profile interval, following the true shape of the likelihood, stays above zero and is asymmetric, stretching further to the right.

In MerQur

In Hierarchical Bayesian Regression, the 'Profile CI for variance components' option gives profile likelihood confidence intervals for σ_u and σ_ε.

Analyses where this concept appears

Hierarchical Bayesian Regression

Related concepts

Maximum likelihood (ML) estimationVariance components