Some Sample-Size Considerations in Material Accountancy Verification

Sample sizes are discussed for variables measurements necessary for independent verification of a material balance using the statistical approaches adopted by the International Atomic Energy Agency (IAEA). It is shown that sample sizes for variables measurements in the attri- butes mode can be reduced by adopting better(but still conservative) approximations for detection probabilities. An expression is derived for these sample sizes. This method of calculating sample sizes retains the valuable characteristic that detection probabilities are a function only of the total amount of the falsification, not of the distribution of the falsification among strata. In addition, the standard approach to determining detection probabilities for a diverter strategy of falsification through small biases involves determining a parameter called the \"variance-inflation factor\", which is the maximum amount by which the diverter chooses to inflate his random error. The choice of this number is problematic, since it is under adversary control. A method is described whereby minimal sample sizes adequate to detect a strategy of random error inflation can be calculated.