A Method for Comparing and Combining Distance Dependent Calculated and Measured Values of the DS02 Dosimetry System for Japanese Atomic Bomb Survivors

H.M. Cullings¹; D.L. Preston²; M. Hoshi³; and S. Fujita¹ (¹Radiation Effects Research Foundation; ²Hirosoft International Corp.; ³Hiroshima University)

The DS02 Report (RERF 2005) will include distance dependent estimates of uncertainty for all neutron and gamma ray components of its calculated (C) kerma for the atomic bombs in Hiroshima and Nagasaki; it also analyzes a large body of related measurements (M’s) made over several decades on materials exposed to the bombs. Two important questions are how to determine whether the M’s “agree with C,” and, if so, how to quantify how much the M information reduces the uncertainty of usable values. A good approach is to define a model to compare continuous functions of distance rather than attempt to define the statistics of many simultaneous comparisons of discrete quantities. The questions can then be restated as “how to test if M and C are consistent with the same model parameter estimates,” and “how to combine the M’s and C to obtain more precise estimates.” We suggest a useful vehicle is the regression model developed at RERF for estimating doses to survivors at distances beyond the 2.5 km limit of the core DS02 system: a spline (arbitrary knots at 1 & 2 km) of linear regressions of the logarithm of kerma on slant distance and the logarithm of slant distance. This simple and flexible model accommodates both exponential attenuation with a gradually changing relaxation length and a power term for the inverse square of distance. Point estimates of model parameters were obtained by regressing on C, with excellent fits. For free-in-air gamma kerma, we describe a method to re-estimate variance-covariance for the model parameter estimates that gives confidence bands consistent with any specified distance dependent uncertainty estimates for C, and we give results of a regression on the M’s weighted by the inverses of their estimated variance. We show examples of statistical tests of equality of parameter estimates for C and M, combined estimates, and comparisons of the estimated distance dependent uncertainty of C, M, and combined/modeled values.

Return to Session list.