The calculated >1-MeV pressure vessel fluence is used to determine the fracture toughness and integrity of the reactor pressure vessel. It is therefore of the utmost importance to ensure that the fluence prediction is accurate and unbiased. In practice, this assurance is provided by comparing the predictions of the calculational methodology with an extensive set of accurate benchmarks. A benchmarking database is used to provide an estimate of the overall average measurement-to-calculation (M/C) bias in the calculations (<M/C>). This average <M/C> is used as an ad-hoc multiplicative adjustment to the calculations to correct for the observed calculational bias. However, this average only provides a well-defined and valid adjustment of the fluence if the M/C data are homogeneous; i.e., the data are statistically independent and there is no correlation between subsets of M/C data.

Typically, the identification of correlations between the errors in the database M/C values is difficult because the correlation is of the same magnitude as the random errors in the M/C data and varies substantially over the database. In this paper, an evaluation of a reactor dosimetry benchmark database is performed to determine the statistical validity of the <M/C> adjustment to the calculated pressure vessel fluence. Physical mechanisms that could potentially introduce a correlation between the subsets of M/C ratios are identified and included in a multiple regression analysis of the M/C data. Rigorous statistical criteria are used to evaluate the homogeneity of the M/C data and determine the validity of the <M/C> adjustment.

For the database evaluated, the M/C data are found to be strongly correlated with dosimeter response threshold energy and dosimeter location (e.g., cavity versus in-vessel). It is shown that because of the inhomogeneity in the M/C data, for this database, the benchmark data do not provide a valid basis for adjusting the pressure vessel fluence.

The statistical criteria and methods employed in this analysis are generic and may be applied in benchmarking applications where the M/C comparisons are used to determine an adjustment of the calculations.