A Study of Alternative Modeling Techniques of Attrition of First-term Navy Enlisted Sailors
Abstract
The purpose of this study was to examine enlisted Navy first-term attrition. Sailors who exit the Navy before completing the end of their first term of enlistment are a lost investment of direct costs (training) and indirect costs (force stability). The ability to predict individuals that are likely to attrite, and to accurately forecast the percentage of the force that is likely to leave early can potentially allow the Navy to proactively implement retention strategies. These strategies will result in maximization of training dollars invested in top-performing Sailors. This study uses modeling techniques and conditional probabilities in an attempt to generate an improved predictive ability of attrition. The best models to predict attrition were the Chi Square Automatic Interaction Detection (CHAID) (decision tree) and Logistic Regression. The models were then compared by looking at lift charts. Lift charts put the observations in the validation set into increasing or decreasing order on the basis of the score. After again examining the lift curves and overall accuracy, the CHAID model was chosen to be the best model in terms of accurately predicting first-term Sailor attrition. Variables that were found to be predictive included annual housing and subsistence allowance, base pay, dependent status, sea or shore duty, unemployment rates, the S&P 500, and members' present rating. It is recommended that groups within the U.S. Navy such as Manpower, Personnel, Training and Education (N1); Navy Recruiting Command (NRC); and others look at employing data mining methods and models to better predict attrition and capitalize on retaining top performing Sailors.
Document Details
- Document Type
- Technical Report
- Publication Date
- May 01, 2009
- Accession Number
- ADA499488
Entities
People
- J. S. Johnson
- Jerry C. Crabb
Organizations
- Navy Personnel Research, Studies, and Technology