Informative and non-informative priors in bayesian statistics: Application in cytogenetic biological dosimetry of mixed n + γ radiation
Warsaw Medical Physics Meeting 2015
Iwona Pacyniak1, Krzysztof W. Fornalski2, Maria Kowalska1
1Centrl Laboratory for Radiological Protection, Warsaw
2PGE EJ 1, Warsaw
Mixed radiation n+g comes mainly from nuclear reactors. These two types of radiation have different properties and also different biological effectiveness. In case of exposure of people to this kind of radiation it is very important to assess not only the total absorbed dose but also neutron and gamma contribution. It is possible by using iterative method and also bayesian statistics as a new approach in this field. An appropriate prior information (generally expressed as a probability distribution) used for calculation of absorbed dose as a probability distribution, is very important in the bayesian method.
If researcher has some earlier information about phenomenon, then it is possible to use informative prior containing specific information about the variable for which the distribution is calculated. In case of lack of preliminary precise information, non-informative prior might be used. This distribution does not specify the exact variable and define it in a very general way. Results obtained in experiments confirm non-informative priors as an accurate distributions to be used.
25 minutes 12 seconds