Analysis of unbalanced occupational exposure data using a Bayesian random effects model

Abstract

Krishnamoorthy and Mathew (2002) made use of a one-way random effects model to analyse lognormally distributed data. They proposed the use of generalized confidence intervals and generalized p-values (frequentist methods) when a balanced data set (an equal number of measurements for each observational unit) was available. Their method was demonstrated on occupational exposure data. Harvey and van der Merwe (2014) developed a Bayesian approach, using objective prior distributions, to analyse the same situation as analysed by Krishnamoorthy and Mathew (2002). Krishnamoorthy and Guo (2005) subsequently extended the generalized confidence intervals and generalized p-value approaches to “unbalanced” data sets (unequal number of measurements for each observational unit). Similarly, in this article the purpose is to extend the Bayesian approach in Harvey and van der Merwe (2014) to unbalanced data. Several non-informative priors are evaluated. Occupational exposure data is used for comparison. The Bayesian approach developed is applicable to any one-way random effects model with lognormally distributed unbalanced data. Results from this article indicate that the Bayesian approach is comparable to the frequentist approaches and indeed offers additional modelling ability.