Page Last Updated: Monday, 01 September 2008 20:19 EDT, © 2002, 2008

Topic: Analysis of messy data

Dr. Dean S. Hartley III

One of the techniques of operations research is the design of experiments to collect data for analysis. Various designs are available, such as fractional factorial, Latin hypercube, and response surface methodology (RSM). The purpose of these designs is to allow the identification of the factors (including interactions) that affect the results and to permit estimation of their influences. These methods assume that the controllable factors are actually controllable. Generally, these factors are actually controllable only within certain limits. Thus, the results may not be neat, orthogonal sets of independent data, but messy data. Additionally, many situations do not permit designed experiments. The data have been collected for some other purpose and have no structure at all. This results in very messy data.

Messy data can be analyzed, sometimes yielding valuable results. The first topic below includes a description of such a situation. The second topic describes industrial work that included analysis of messy data; however, the link does not directly refer to analysis of messy data.

Select one of the links below.

• Historical Validation of Attrition Algorithms: Performed for The Joint Staff/J-8, this work consisted of investigation of historical battle results data. A new, validated attrition algorithm has been developed. A macro model was developed, the Oak Ridge Spreadsheet Battle Model (ORSBM), which is being used in analysis and teaching. Published the book Predicting Combat Effects in 2001.

• Industrial Projects for Milliken & Co.: Wrote custom Manufacturing Resource Planning (MRP) systems. Wrote system for scheduling polyester yarn texturizing equipment in several mills. Wrote system for scheduling looms in several mills. Solved problem of entering the Dress Business.

Return to Website Entrance.