Purpose: To present the impact in coverage of different methods for Poisson confidence intervals and the impact in dose coverage of different uncertainty factors. A detailed explanation of the uncertainty sources in the Bayesian method is also presented. Materials and methods: The exact coverage of uncertainty Poisson confidence intervals and the dose uncertainty interval coverage were performed by simulations using R-based scripts. Results: The Poisson exact calibration interval via the Modified Crow and Gardner method resulted in coverage quite close to the nominal level of confidence; additionally, the method retains the shortest property of Crow and Gardner, and gains the property of a lower limit strictly increasing in the mean of dicentrics. The unlimited simultaneous calibration interval seems to be the method of choice to preserve the coverage at 95% under parametric and nonparametric conditions but is a conservative method. When samples came from a Poisson distribution, the ISO propagation of errors and Bayesian approaches seem to be the closest to the 95% coverage. Conclusions: The Modified Crow and Gardner method should be preferred over the Garwood method for Poisson exact confidence intervals. The unlimited simultaneous calibration interval did not lose its property to preserve the coverage at 95% applying a regression coverage factor of value 2.02 at the point of doses studied in the simulation.
- confidence intervals
- Dicentric assay