Sample Data Collection Logistics Management Analysis Report
Abstract
Annually, the U.S. Army Safety Center (USASC), Fort Rucker, Alabama, analyzes approximately 20,000 reports of accidents to identify cause factors, develop countermeasures and evaluate effectiveness of fielded countermeasures. Efficient analysis of such massive data requires use of methods of data reduction to reveal the essential targets. One of the methods used by the Safety Center is factor analysis. However, the nature of much of the accident data (binary) is not amenable to the Pearsonian Product Moment Correlation Coefficient that drives the factor analysis. A substitute coefficient (Jaccard Similarity Coefficient) has been employed by USASC but the theoretical foundation for using this procedure has not been established. Section 1 examines the feasibility of using similarity or matching/associative coefficient as a substitute for the correlation/covariance matrix in the factor analysis procedure. Also, examined is the definition of dichotomous scoring compared to the binary data made available to the USASC. From Section 1, it was determined that the method of data reduction had many mathematical pitfalls. A more efficient method should be utilized to reduce the accident data. Section 2 develops a methodology using accident data supplied by the USASC (Night Study data). This data is considered highly parsimonious in both the physical interpretation and mathematical complexity. The procedure which was investigated is the VARCLUS procedure contained in the SAS Institute INC. statistical package (SAS). It is felt that the parsimony mentioned above is minimized by using this procedure. The VARCLUS procedure was investigated because of its usefulness in interpreting large amounts of variables. VARCLUS is a variable-reduction method and it is also useful in determining if there is a relationship between variables.
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 01, 1991
- Accession Number
- ADA372990