Bayesian techniques for imputation of missing occupational dose/exposure data

It has been proposed that occupational exposure / dose limits change the distribution of occupational exposures/doses. The often used log normal model represents the distribution of doses quite well for low doses. In dose ranges that approach occupational limits dose distributions deviate from a log normal and tend to be more normal.Kumazawa and Numakunai [7] have developed the hybrid log normal model to account for this shift and applied this new model to several data sets using classical statistics to estimate parameters. One of these parameters is the hybridization parameter which is used to investigate whether a distribution is predominantly normal or log normal. Weapplied the Bayesian approach to the hybrid log normal model. We Obtained predictive densities to impute missing occupational exposures/doses and obtained marginal posterior densities for the parameters, including the hybridization parameter. The marginal posterior density of the hybridization parameter quantifies the remaining uncertainty about the parameter in a quantitative probabilistic fashion and is thus an improvement over published estimation techniques [7]. We also obtained a predictive density using the bivariate normal model which compares quite well to imputation results with a Bayesian linear regression technique used by Dolan [1].