neighbayes

Lead · Pre-release

neighbayes fits spatial econometric models by Bayesian estimation, some fifty model classes behind one formula interface: the cross-sectional family (SAR, SEM, SDM, SDEM, and SLX); spatial logit, probit, tobit, negative binomial, and zero-inflated models; fixed-, random-, and dynamic-effects panels; and origin–destination flows. Fitted models report direct, indirect, and total effects with their posterior intervals, and carry the Bayesian Lagrange-multiplier diagnostics. The log-determinant methods keep the Jacobian cheap enough for a model of every U.S. census tract. It builds on PyMC, and an optional JAX backend runs its Gibbs samplers on the sparse solvers of sparsax and the augmentation draws of augax.