Distribution Estimation by Computer Simulation.
Abstract
This document discusses the practice of estimating probability distribution functions by microcomputer simulation, in contrast to using mainframes. Three key issues are raised as commentary on the current state of the applied art. The first point is that there is a tendency to substitute arithmetic power (brute force) for analytic intelligence. Logically modeling a relationship and identifying the driving variables often yields more insight than voluminous computation, or at least, experimental effort can be greatly reduced by careful pre-analysis. A second observation is that the size of sampling error inherent in the simulation process is not fully appreciated. Sample sizes required for satisfactory levels of precision are frequently larger than intuitively expected by the analyst with minimal statistical training. Finally, the ability to execute many types of realistic simulations efficiently on desktop computers is currently limited by the relatively slow speed of these devices, particularly wehn the sampling is done in conjunction with spreadsheet software. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Apr 01, 1987
- Accession Number
- ADA180994
Entities
People
- Carl M. Harris
- William G. Marchal
Organizations
- George Mason University