EXPERIMENTAL COMPARISON OF MONTE-CARLO SAMPLING TECHNIQUES TO EVALUATE THE MULTIVARIATE NORMAL INTEGRAL
Abstract
The present study compares techniques for estimating manpower requirements where a number of individually varying skills, performance potentials, background and behavioral factors must be considered. The specific objective was to evaluate two different numerical methods for estimating probability when a multivariate normal model (for example, one involving scores on a battery of tests) can be assumed. In a series of simulation experiments in which random vector observations were generated, probability estimates were computed by each of the two methods. Probability regions on which the experiments were based were chosen to have a variety of properties. The precision of the two methods was compared from the magnitudes of the variances of the probability estimates over independent samples. Results indicated that when the probability region is very small, the more complex of the two methods (importance sampling) is superior; but when the sampling approximation is poor, the precision of the probability estimates favors the simpler Monte-Carlo procedure. The computational procedures developed appear to be practical methods of estimating probability based on multiple scores for individuals in a sample population.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jun 01, 1969
- Accession Number
- AD0695672
Entities
People
- Elizabeth N. Abbe
Organizations
- United States Army Research Laboratory