Sampling Strategies for Destructive Tests


  • W. G. Warren


Monte Carlo simulation, statistical methods, cost efficiency, concomitant variables


Two strategies are considered whereby the full sample need not be destroyed to obtain estimated of low-order percentiles. The properties are examined by Monte Carlo methods. One strategy, based on a proof-loading concept, appears to have practical value, particularly when used in conjunction with a non-destructively-determined concomitant variable. Even when the concomitant variable has a fairly low correlation with the property of interest, the number of samples destroyed is appreciably reduced. This number is, however, a random variable. The second strategy, which again uses a concomitant variable, fixes the number destroyed, but in other respects is less satisfactory; for reliable results the number destroyed will be uncomfortably high unless the variables are very highly correlated. The use of a subjective ranking rather than a measured concomitant variable, to reduce the number of specimens destroyed, is considered. It is seen that, under some circumstances, subjective ranking, or even allocation into ordered groups, will be cost-efficient.


Click, F. P. A cheap sequential procedure for non-parametric tolerance limits or conservative estimation of small percentiles in stress studies. Am. Statist. Assoc. (in press).nKendall, M. G., and W. R. Buckland. 1971. A dictionary of statistical terms. 3rd ed. Oliver and Boyd, Edinburgh.nTeichroew, D. 1956. Tables of expected values of order statistics and products of order statistics from samples of size 20 and less from the normal distribution. Ann. Math. Statist. 27:410-426.n






Research Contributions