Turnover is expensive. This is especially true in the military when the turnover occurs shortly after military training. At that early point in a Soldier’s career, the military has invested significantly in his or her training with limited return on investment in on-the-job performance, in addition to other “sunk costs” (e.g., resources expended recruiting, hiring, and onboarding the new Soldier). Many civilian sector organizations incur similar losses when new employees leave early. Accordingly, a number of organizations are investing heavily in talent analytics to model employee turnover to identify how best to reduce turnover. However, talent analytics will be only as effective as the key assumptions an organization makes about turnover and how it treats turnover in its models.

HumRRO has studied military turnover, its nature and key drivers, extensively. For example, we conducted one of the largest studies of early career turnover in the military, tracking upwards of 100,000 Soldiers through their first 4 years in the Army. Several important findings stand out from our research that non-military organizations could learn from when modeling turnover.

Lesson #1: The nature of turnover changes over time.

  • What we found:
    Performance and medical/physical factors accounted for approximately 80% of all turnover in the first 6 months of service. Beyond 6 months, turnover resulting from counterproductive or deviance-related issues emerged as more prevalent (approximately 60% of all turnover occurring between 2 and 3 years of service and nearly 50% of turnover thereafter). In sum, the type of Soldier who leaves in training is not the same type of Soldier who leaves post-training.
  • What it means for non-military organizations:
    Employees who turn over early in high-turnover, civilian occupations (e.g., teacher, bank teller) may similarly leave for different reasons than employees who exit several years post-hire. Turnover is a dynamic, time-varying behavior that occurs for different reasons depending on when it occurs in the employee lifecycle. Modeling the nature of turnover (or turnover profiles) across the employee lifecycle is the critical first step to effectively managing turnover. Otherwise, organizations are likely to expend valuable resources trying to hit a moving target.

Lesson #2: No single, key driver best predicts turnover.

  • What we found:
    Although there were some commonalities in the key drivers of turnover (e.g., cognitive ability was strongly related to all turnover types, physical conditioning was related to both performance and medical/physical turnover), the more prevalent finding was that different drivers best predicted different types of turnover. In particular, the key drivers most predictive of turnover shortly after joining the military differed markedly from those that were most predictive of post-training turnover (when counterproductive or deviance-related issues emerged as the primary reasons for turnover).
  • What it means for non-military organizations:
    Modeling turnover as a single, binary outcome obscures or grossly underestimates the key drivers of turnover. Shifting from simply predictive to prescriptive talent analytics that inform useful interventions to reduce turnover requires treating turnover as a multidimensional, multifaceted behavior.

Lesson #3: The relations between key drivers and turnover can change over time.

  • What we found:
    In a few cases, the magnitude of the relation between a key driver and a certain type of turnover increased or decreased over time. In sum, the predictive effects of the primary key drivers of turnover varied across time. One potential explanation for this pattern is that other situational- or event-based drivers may also play a role in influencing turnover.
  • What it means for non-military organizations:
    The best predictor of a given type of turnover can change over time. Factoring this feature in when modeling turnover enables an organization to select not only the best intervention, but also (and more importantly) its optimal implementation point. Further, organizations would be well-advised to consider measuring and modeling key situational- or event-based drivers that precipitate turnover, in addition to the individual-level drivers we have traditionally focused on.

Talent analytics hold significant promise for modeling and generating insights into employee turnover, among other important outcomes. However, organizations can all too easily make the wrong assumptions about turnover that seriously limit or undercut the value of talent analytic-generated solutions. As the old adage goes, “Garbage In-Garbage Out” (G-I-G-O). As our experience modeling military turnover illustrates, avoiding the G-I-G-O trap starts with a well-informed understanding of the target and how best to treat or model it.

About the Authors:

Michael Ingerick

Director of Military Personnel Research

Rodney A. McCloy

Principal Staff Scientist