The Practical Application of Variance Reduction Techniques in Probablistic Assessments

Abstract

The use of variance reduction techniques in probabilistic risk assessments is examined, using an actual assessment model as an example. Variance reduction techniques may allow probabilistic risk assessments to be performed more efficiently. The efficiencies of stratified sampling, Latin hypercube sampling, and antithetic sampling are compared with simple random sampling. Stratified sampling appears to be less efficient than simple random sampling, and is not recommended for use in probabilistic assessments. The efficiency improvement offered by antithetic and particularly, Latin hypercube sampling can justify their use under certain circumstances.