The log-likelihood ℓ(θ) measures how well each parameter value explains the data. Its maximum is the MLE; its curvature at the maximum, the observed Fisher information J, gives the standard error 1/√J. The left panel shows ℓ(θ) − ℓ(θ̂) (solid) and its quadratic approximation −J(θ − θ̂)2/2 (dashed); the parameter values above the dotted line at −1.92 form the 95% likelihood-ratio interval, the bracket below is the Wald interval θ̂ ± 1.96/√J. The right panel checks the fit: a hanging rootogram for Poisson counts, the likelihood itself for a binomial, and the de Finetti diagram for Hardy-Weinberg genotypes.
code/ch02_statistical_modeling.R (pois_gof, hwe_fit).