Online Program

Return to main conference page
Friday, May 31
Computational Statistics
Statistical Methods for Analyzing Large Scale or Massive Data
Fri, May 31, 1:30 PM - 3:05 PM
Grand Ballroom K

An Application of Linear Programming to Computational Statistics (306222)


*John M. Ennis, Aigora 
William J. Russ, The Institute for Perception 

Keywords: Optimization, Linear Programming, Multiple Statistical Comparisons, Letter Displays, Computational Efficiency

Linear programming is an optimization technique commonly applied in the field of operations research to solve well-specified but computationally intensive problems. One such problem that arises in the field of computational statistics is the problem of efficiently representing the results of a large number of multiple statistical comparisons in a table or letter display. In this presentation, we present newly developed algorithms employing linear programming to find letter displays that are minimal with respect to either: a) the number of distinct letters or b) the number of total letters used by the display. We then compare these algorithms with the previously state-of-the-art algorithms for letter display production to show that minimizations based on linear programming offer a dramatic improvement in computational efficiency. In conclusion, we describe the category of problems that can be solved using linear programming, and offer guidance on how statisticians facing computationally intensive problems might leverage linear programming to increase computational efficiency.