The Rose Garden Event: A Hierarchical Bayesian Approach to Modeling Positive Coronavirus Tests

• Activity Number: 353 - Research and Educational Tools
• Type: Contributed
• Date/Time: Thursday, August 12, 2021 : 10:00 AM to 11:50 AM
• Sponsor: Section on Statistics and Data Science Education
• Author(s): Jedidiah Harwood* and Eric A. Suess
• Companies: CSU East Bay and CSU East Bay
• Keywords: Bayesian; Statistics Education; MCMC; Hierarchical model; Gibbs Sampling

Abstract

We present an example of a Bayesian hierarchical model, for an advanced undergraduate Bayesian Statistics class, with a focus on explaining the variability in coronavirus testing. This demonstration was applied to data from various Trump events throughout 2020, where testing and social distancing were not enforced. Using R and the JAGS software for Gibbs Sampling, we developed predictive, posterior distributions for the total number of people who would have tested positive for coronavirus at each (and every) event. Additionally, we developed posterior distributions for the coronavirus test sensitivity and specificity. The MCMC chains appeared to properly converge to the posterior distributions, and was further suggested so, by the Gelman-Rubin statistic.