Radiation Protection Dosimetry Advance Access originally published online on December 14, 2007
Radiation Protection Dosimetry 2008 129(4):411-418; doi:10.1093/rpd/ncm483
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Maximum likelihood estimates of mean and variance of occupation radiation doses subjected to minimum detection levels
1 Health Physics Division, Bhabha Atomic Research Centre, Mumbai 400 085, India
2 Radiation Safety Systems Division, Bhabha Atomic Research Centre, Mumbai 400 085, India
3 Health, Safety and Environment Group, Bhabha Atomic Research Centre, Mumbai 400 085, India
* Corresponding author: ddatta{at}barc.gov.in
Received September 14, 2006, amended October 26, 2007, accepted October 26, 2007
Data collection and its analysis in the field of nuclear safety is an important task in the sense that it powers the improvement of safety as well as reliability of the plant. Thus, occupational exposure data analysis is presented to measure the safety or reliability of radiation protection of a given facility. It also is required as a basic input in making decisions on radiation protection regulations and recommendations. A common practice in radiation protection is to record a zero for observation below minimum detection limit (MDL) doses, which leads to an underestimation of true doses and overestimation of the dose–response relationship. Exposure data (both external and internal) are collected by monitoring each individual and this kind of monitoring generally is graded as low-level monitoring. So, in such low-level monitoring, the occurrence of exposure below MDL invites statistical complications for estimating mean and variance because the data are generally censored, i.e observations below MDL are marked. In Type I censoring, the point of censoring (e.g. the detection limit) is ‘fixed’ a priori for all observations and the number of the censored observations varies. In Type II censoring, the number of censored observations is fixed a priori, and the point of censoring vary. The methodology generally followed in estimating mean and variance with these censored data was the replacement of missing dose by half the MDL. In this paper, authors have used the maximum likelihood estimation (MLE) approach for the estimation of mean and standard deviation. A computer code BDLCENSOR has been developed in which all these MLE-based advanced algorithms are implemented. In addition to the MLE-based method, an expectation maximisation algorithm has also been implemented. The code is written using Visual BASIC 6.0. The paper describes the details of the algorithms adopted for handling such censored data to estimate bias free mean and standard deviation.