Fitting “standard” distributions to data is necessarily good: Dogma or myth?

  • Authors:
  • Bennett L. Fox

  • Affiliations:
  • -

  • Venue:
  • WSC '81 Proceedings of the 13th conference on Winter simulation - Volume 1
  • Year:
  • 1981

Quantified Score

Hi-index 0.00

Visualization

Abstract

This paper questions the conventional wisdom that “standard” distributions such as the normal, gamma, or beta families should necessarily be used to model inputs for simulation and proposes an alternative. Relevant criteria are the quality of the fits and the impact of the input distributions on effective use of variance reduction techniques and on variate-generation speed. The proposed alternative, a “quasi-empirical” distribution, looks good on all these measures.