Random effects
in
AD Model Builder

Technical details

• Model specification in C++ like language
• Hyper-parameters (variance components etc.) estimated by maximum likelihood
• Marginal likelihood evaluated by the Laplace approximation or importance sampling
• ADMB-RE calculates exact derivatives using Automatic Differentiation
• All the useful features of ordinary AD Model Builder are available

Why choose ADMB-RE?

• Flexibility: In principle you can implement any random effect you can think of
• Convenience: Computional details are transparent. Your only responsability is to formulate the loglikelihood
• Computational efficiency: ADMB-RE is up to 50 times faster than winBUGS
• Robustness: With exact derivatives you can fit highly nonlinear models
• Convergence diagnostic: The gradient of the likelihood function provides a clear convergence diagnostic, while with MCMC judging convergence is difficult.