GESIS Training Courses
user_jsdisabled
Search

Scientific Coordination

Dr.
André Ernst

Administrative Coordination

Janina Götsche

Advanced Bayesian Statistical Modelling in R and Stan

About
Location:
Online via Zoom
 
Course duration
Tu-We: 9:00-17:00 CEST
We+Th: 9:00-12:15 CEST
 
General Topics:
Course Level:
Format:
Software used:
Duration:
4 days
Language:
Fees:
Students: 300 €
Academics: 450€
Commercial: 900 €
 
Keywords
Additional links
Lecturer(s): Dr. Denis Cohen

About the lecturer - Dr. Denis Cohen

Course description

Please note: There is an additional session on 24.11. Check the schedule for times
 
Statistical models are widely used in the social sciences for measurement, prediction, and hypothesis testing. While popular statistical software packages cover a growing number of pre-implemented model types, the diversification of substantive research domains and the increasing complexity of data structures drive persistently high demand for custom modeling solutions. Implementing such custom solutions requires that researchers build their own models and use them to obtain reliable estimates of quantities of substantive interest. Bayesian methods offer a powerful and versatile infrastructure for these tasks. Yet, seemingly high entry costs still deter many social scientists from fully embracing Bayesian methods.
 
This workshop offers an advanced introduction to Bayesian statistical modeling to push past these initial hurdles and equip participants with the required skills for custom statistical modeling,. Following a targeted review of the underlying mechanics of generalized linear models and core concepts of Bayesian inference, the course introduces participants to Stan, a platform for statistical modelling and Bayesian statistical inference. Participants will get an overview of the programming language, the R interface RStan, and the workflow for Bayesian model building, inference, and convergence diagnosis. Applied exercises allow participants to write and run various model types and to process the resulting estimates into publication-ready graphs.


Target group

Participants will find the course particularly useful if:
  • They seek an advanced introduction to applied Bayesian statistical modeling
  • They have used statical models but want to understand the underlying mechanics of estimation, inference, and the post-estimation processing into meaningful quantities of interest
  • They want to move beyond pre-implemented models and craft custom solutions that meet the idiosyncratic requirements of specific empirical applications


  • Learning objectives

    By the end of the course participants will:
  • Know Bayesian fundamentals and know how to implement the workflow for Bayesian statistical model building
  • Know how to translate formal statistical models into code using Stan to start crafting their own custom models
  • Know how to process estimates from generalized linear models into substantively meaningful quantities of interest (without relying on pre-implemented software)
  • Know how to use distributional summaries as a flexible framework for reporting inferential uncertainty
  •  
     
    Organizational structure of the course
    This course is organized in six sessions à 180 minutes, each split approx. 50-50 into a lecture and a hands-on lab session with a 15-minute break in between. A 90-minute lunch break divides morning and afternoon sessions.
     
  • Lectures will alternate between classical lecturing, interactive Q&A sessions, and regular applied code demonstrations and empirical illustrations.
  • Labs allow participants to work collaboratively on applied problem sets in small groups in Zoom breakout rooms. The problem sets typically comprise of a sequence of problems as typically encountered in real-world empirical work. Hints and solutions will be provided.
  • During the lab sessions, the instructor will be available in the main meeting to assist with any questions that may arise.
  •  
    Note: The first five sessions will be taught as a 2.5-day workshop. The final half-day session will take place two weeks later. This allows participants to work on the development and implementation of their own models. We will then discuss participants' work and experiences in the final session. Participants can send their code and a short report detailing their model and potential problems encountered during implementation in advance of this session in preparation of individual consultations.


    Prerequisites

  • Working knowledge of the software environment R
  • Working knowledge of (generalized) linear models
  • Basic knowledge of linear algebra (recommended)
  • Basic knowledge of probability theory (recommended)
  •  
    Software requirements
    This workshop requires installations of recent versions of R, RStudio, and RStan. Throughout the second half of the workshop, we will use Stan via its R interface RStan. Setting up RStan can be somewhat time-consuming as it requires the installation of a free-of-charge C++ compiler. Therefore, workshop participants should follow these instructions on the Stan Development Team's GitHub to install and configure the rstan package and its prerequisites on their operating system before the workshop. If you do not have administrator privileges on your machine, please approach your system administrator in advance of the workshop.


    Schedule