Modeling Attrition in a Military Selection Context

Attrition, employee turnover, self-selection, and withdrawal all refer to an employee’s exit from an organization, or from an organization’s recruitment or selection process. When individuals with the desired knowledge, skills, abilities, and other qualities (KSAOs) attrit, it represents lost productivity to an organization (Barrick & Zimmerman, 2009). Therefore, organizations should seek a selection program that screens out unwanted characteristics while minimizing the voluntary withdrawal, or quitting, of those who would be a good organizational fit. A military selection context amplifies these two aims because of the limited number of qualified individuals relative to the organization’s personnel needs, and because of the high potential cost of a bad hire. However, there are few studies of attrition during a selection process, and even fewer in a military context that combines physical, cognitive, and personality components as relevant performance dimensions.

The purpose of the study was to model attrition from a military special operations selection through training program to determine what combination of physical abilities, cognitive abilities, and personality scales best predicts success. The study examined archival data from 748 candidate records spanning eight different classes during 2019. Secondary purposes of the study included comparing differences in attrition from the first week of the program to the remaining seven weeks, and comparing the predictive validity of a personality trait profile model to a model using personality scales T-scores. In conducting the analysis and modeling, exploratory factor analysis was conducted on the sample Jackson Personality Inventory-Revised (JPI-R) personality scales, finding both similarities and differences with previous study samples (Detwiler & Ramanaiah, 1996; Paunonen & Jackson, 1996).

The result of the study was a logit prediction model with a ROC AUC of .784, and an F1 score of .69, that incorporated three physical predictors, performance IQ, and three personality variables: JPI-R T-score for sociability, and two composites created from the factor analysis—a Conscientiousness Composite and an Openness Composite (negative relationship with candidate success). Models for week 1 attrition and attrition from weeks 2-8 differed from the 8-week attrition model, and from each other in the significance and the importance of the personality variables and of cognitive abilities. Physical predictors: run score, pushups score, and sit-ups score, were significant and strong predictors of success for each of the time periods. Verbal IQ was not significant for any time period, while performance IQ was significant in predicting 8-week success, and for success during the week 2-8 time period. Personality predictors varied the most by timeframe, although some component of Conscientiousness predicted strongly for each timeframe. Whereas Openness-related facets predicted for 8-week success and success from week 1 with a negative relationship, Openness factors were non-significant in weeks 2-8. In contrast, Anxiety, a related sub-facet of Neuroticism, predicted moderately (negative relationship) for success from weeks 2-8, but was non-significant for week 1 and for the 8-week program.

Unexpected findings included the sample’s different factor structure on the JPI-R, the dominance of the physical predictors in all models, and the strength of personality predictors relative to cognitive abilities. Implications for military and similar types of selection contexts, where selection through training includes a significant physical component, are discussed.