Predicting Army Post-Iet Attrition Using Logistic Regression and Time-Varying Covariates
Abstract
The Army is trying to reach a force of 500,000 by 2030. Within the next 10 years, the Army needs to play a balancing act of figuring out how many soldiers will retire, attrit, or not reenlist, and how many will leave for medical or other various reasons. Then the Army needs to figure out how many soldiers need to be recruited every year to reach the 500,000 goal. Because of factors such as lower recruiting goals, tightening labor markets, reduced incentives due to a tighter defense budget, and increasing obesity levels, it is getting harder to recruit prospective soldiers. In such an environment, military leaders need to know why soldiers attrit before their first term is complete, and the factors that contribute to this decision. This thesis uses multiple logistic regressions to determine if a soldier will attrit using personnel data from the Person-Event Data Environment database. We discovered that soldiers who attrit have more variables in common by year in contract than by their contract duration. Thus the models are by year in contract due to the changing nature of time-varying covariates. As the year in contract increases, the effects of demographic indicators generally decrease and the effects of medical-related indicators largely increase. This model can help Army G1 predict how many people will be in the military at a given timeknowledge that will also help leaders determine how to prevent attrition and increase the likelihood of success for soldiers.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jun 01, 2020
- Accession Number
- AD1114639
Entities
People
- Josephine H. Cammack
Organizations
- Naval Postgraduate School