Error Predictions in Accounting Populations.
Abstract
The purpose of this thesis was to examine robustness, mean relative tightness, coverage, and power of selected unmodified and modified Cox and Snell and Stringer error limit bounds. The simulation was performed by repetitive sampling from an accounting population with various known book value and error distributions. Additional modifications to the selected modified Cox and Snell bounds was done by incrementally loosening the bounds by 5, 10, and 20 percent in a search for bounds with better performance characteristics. There were several conclusions that could be made from this research. The modified Cox and Snell bounds can achieve high coverages for accounting populations with low error amount intensity (EAI) with significant increases in mean relative tightness over unmodified Cox and Snell bounds. The Stringer and Cox and Snell bounds can still achieve high coverages with significant improvements in mean relative tightness when the nominal confidence level is lowered from 90% to 85%. Only minor changes in prior probability settings materially affect the performance of Cox and Snell bounds. And, in accounting populations with low EAI, the selected Cox and Snell bounds are conservative.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 01, 1986
- Accession Number
- ADA174457
Entities
People
- Blaine F. Webber
Organizations
- Air Force Institute of Technology