tmbstan

tmbstan integrates Template Model Builder (TMB) with Stan's no-U-turn sampler (NUTS) to perform Bayesian inference for complex hierarchical models and to evaluate the Laplace approximation for random effects.

Key Features:

• Integration with NUTS Algorithm: Leverages Stan's no-U-turn sampler (NUTS) to perform MCMC sampling for TMB models, including high-dimensional hierarchical models.
• Parallel Sampling: Supports parallel MCMC sampling to accelerate posterior estimation.
• Testing Laplace Approximation Accuracy: Uses NUTS sampling to test and assess the accuracy of the Laplace approximation for random effects in TMB models.
• Dual Paradigm Modeling: Enables fitting the same TMB model under both frequentist and Bayesian paradigms.
• Performance Comparison: Demonstrates performance comparable to Stan, typically within ±50% of Stan's speed for complex models.

Scientific Applications:

• Ecological modeling: Enables Bayesian hierarchical modeling commonly used in ecological research for complex natural-process models.
• Uncertainty quantification and prior incorporation: Facilitates incorporation of prior information and quantification of posterior uncertainty in statistical analyses.

Methodology:

Links TMB models directly into Stan and runs Stan's NUTS sampler for Bayesian inference, using NUTS samples to assess the Laplace approximation for random effects and bypassing ADMB modifications required by adnuts.