Latin hypercube sampling for the simulation of certain nonmonotonic response functions [electronic resource].
- Albuquerque, N.M. : Sandia National Laboratories, 1980.
Oak Ridge, Tenn. : Distributed by the Office of Scientific and Technical Information, U.S. Dept. of Energy.
- Physical Description:
- Pages: 14 : digital, PDF file
- Additional Creators:
- Sandia National Laboratories
United States. Department of Energy. Office of Scientific and Technical Information
- The Latin Hypercube Sampling, LHS, plan was presented by McKay, Beckman and Conover (Technometrics, 21, 239 - 245(May 1979)) as an alternative to simple random sampling, SRS, in Monte Carlo studies. For each realization of random input vector X, suppose that a computer code produces a scalar output Y = h(X). The referenced paper proves that, when h is monotonic in each of its arguments, some common summary statistics on Y are more efficient under LHS than under SRS. This monotonicity restriction on h excludes some interesting response functions. For example, suppose X is a vector of a mission's sensor measurements, and Y = h(X) is one when the mission is successful and zero otherwise. Function h is not monotonic in a common case; it is not monotonic if success occurs when X is near the true value, and failure occurs when X departs sufficiently from this value in any direction. Conditions on h, general enough to include this setting, are given under which LHS produces more efficient estimates than SRS.
- Published through SciTech Connect.
American Statistical Society meeting, Houston, TX, USA, 11 Aug 1980.
Diegert, K.V.; Diegert, C.
- Funding Information:
View MARC record | catkey: 14358022