The author analyzes the predictors of costly first-term military attrition using administrative data for all accessions across four service branches in fiscal years 2002 through 2013. The analysis shows who accesses, who attrites and when, and what observable characteristics are associated with attrition in the first 36 months of service. The predictive power of recruitment and accession data in efforts to mitigate attrition is also documented.
Predicting 36-Month Attrition in the U.S. Military
A Comparison Across Service Branches
- Who accesses to the Army, Navy, Air Force, and Marine Corps?
- Who attrites from these service branches during the first 36 months?
- When do attriters typically attrite?
- Can recruitment and accession data be used to predict who will attrite?
First-term attrition—in which a new enlisted recruit does not complete his or her first contract—is a costly and ongoing issue across all military service branches, costing, on average, thousands of dollars per enlistment and millions of total dollars per year. Past research has shown that attrition is strongly associated with several characteristics of recruits that are observable at the time of recruitment, or at least by the time of accession. Comparison across studies is difficult because different studies focus on different services, use different sets of variables, or use samples from different time periods. The author of this report provides a comparative analysis of the predictors of attrition.
The analysis relies on data consisting of all enlisted accessions between fiscal years 2002 and 2013 in the Army, Air Force, Marine Corps, and Navy, for a total of 2,189,024 accessions from 2,034,045 unique individuals, and shows who accesses, who attrites, when they attrite, and what observable characteristics are associated with attrition at various points during the first 36 months of service. The analysis also documents the predictive power of the data to distinguish attriters from nonattriters to assess the value of recruitment and accession data in developing policies to mitigate attrition. To highlight promising avenues for future research, the author hypothesizes potential mechanisms behind attrition, based on observed similarities and differences across services and over the course of the first term.
Average recruit characteristics and attrition patterns vary among the services
- The Army has the highest overall attrition rate, the Marine Corps the lowest.
- For all services, the attrition rate is highest prior to month 6 and levels out by month 7, staying roughly constant after that.
- By the end of 36 months, total attrition varies from 18.5 percent in the Marine Corps to 29.7 percent in the Army.
The marginal effects of demographic characteristics, such as gender, marriage, and high school completion, show patterns that vary across services and over time
- Women are more likely to attrite in the Army than in the other services.
- Recruits without a high school diploma or equivalent are more likely to attrite in the Navy.
- Married recruits are more likely to attrite in the first year but less likely to attrite from then on.
- These patterns show that different recruits have different risk periods for attrition, implying that personal characteristics may interact with individual experiences to produce different rates of attrition at different points during the first term.
It is unlikely that simple policies aimed at screening candidates based on their probability of attrition will be cost-effective
- Predictive algorithms will screen out too many nonattriters.
- A major cause of attrition seems related to factors that either are unobservable (or not recorded at enlistment) or occur after accession.
- These factors might be called fit or a taste for military life.
- Findings of further research must strive to link postrecruitment taste factors to the characteristics that can be observed during recruitment.
- Further research may not be able to predict who will attrite but may help determine when a recruit is at greatest risk of attrition.
Table of Contents
Data Overview and Descriptive Statistics
Probit Regression Sensitivity and Specificity in Predicting Attrition
Marginal Effects of Recruit Characteristics
Main Regression Results