Tutorial “Bayesian Semiparametric Regression”, Berlin, Dec 13-14, 2023
Semiparametric regression models overcome some of the restrictions of
classical forms of regression models such as (i) the linearity of
covariate effects, (ii) the independence of observations, or (iii) the
focus on specific types of response distributions and on modelling the
conditional mean alone. In this sense, semiparametric regression forms
an overarching model class comprising various special cases such as
generalized additive models (GAMs), models with random effects, spatial
regression models, or generalized additive models for location, scale,
and shape (GAMLSS). Bayesian inference based on Markov chain Monte Carlo
(MCMC) simulation techniques provides a particularly attractive way of
statistically treating such models and developing extensions. This is,
for example, due to the modularity of MCMC that allows to flexibly
combine different blocks and algorithms for different types of model
parameters.
This tutorial will combine lectures with practical exercises on the
implementation of Bayesian semiparametric regression utilizing the novel
probabilistic programming environment Liesel
(https://liesel-project.org). Liesel is developed with the aim of
supporting research on Bayesian inference based on MCMC simulation in
general and semiparametric regression in particular. Its three main
components are (i) an R interface (RLiesel) for the configuration of
initial semiparametric regression models, (ii) a graph-based
model-building library where the initial model graph can be manipulated
to incorporate new research ideas, and (iii) an MCMC library for
designing modular inference algorithms combining multiple types of
well-tested MCMC kernels.
In the tutorial, we will build on Liesel and discuss (i) general
principles of Bayesian inference with MCMC, (ii) Bayesian additive
models, and (iii) Bayesian distributional regression. We will combine
theoretical background information with hands-on work on applications
for all course parts. For participating, knowledge of the principles of
Bayesian inference, familiarity with linear and generalized models, and
some experience in statistical programming with Python or R will be
beneficial.
The instructors for the tutorial will be Thomas Kneib (University of
Göttingen), Paul Wiemann (Texas A&M University), and Hannes Riebl
(University of Göttingen). Thomas Kneib is a Professor of Statistics and
has contributed to the field of Bayesian semiparametric regression with
new statistical methodology, as well as the development of software and
applications in various contexts. Paul Wiemann and Hannes Riebl are
postdoctoral researchers and Liesel’s leading developers.
The tutorial is part of the CMStatistics 2023 and is coorganized by the
COST Action HiTEc (see
http://www.cmstatistics.org/CMStatistics2023/tutorials.php for details).