What turnover research finds before an employee leaves

Turnover risk does not live in one universal list of predictors. It appears in the interaction between a person's employment history, the surrounding workplace and events that change how the job feels. Three studies make that point from very different settings.

Tenure is not one smooth risk curve

In the Ribes, Touahri and Perthame case study, new joiners and employees who had spent more than four years in the same role emerged as important groups. Performance, compensation scheme and manager time in position also carried signal. Remote location did not stand out in this particular job population.

The useful lesson is not that every employer has the same tenure thresholds. It is that early-tenure uncertainty and later role stagnation are different situations. A new hire may need clearer expectations, support and connection. A long-tenured employee in one role may need credible movement or renewed scope.

Departures can alter the local environment

AlKetbi and colleagues analyzed 519,860 public-registry employment records covering 121,883 professionals and 4,979 Hong Kong financial firms. When more than 30 percent of a person's peers had departed within six months, turnover probability was 23 percent higher relative to baseline. Adding network features improved predictive performance by 30.2 percent over a non-network baseline in the paper's walk-forward evaluation.

This is compelling predictive association, not proof that resignation spreads like an infection. Peer departures can change workload, information, confidence and outside opportunities. They may also reflect a common underlying shock. A practical team-level signal is therefore a cluster of departures that warrants listening and workload review, not a label placed on the people who remain.

A problem can strengthen fit when the response is aligned

Balthrop and Jung linked 466,236 employee communications to 45,873 employment spells across 21 trucking companies. Equipment-related problems were associated with longer job spells and improving sentiment after the report. The authors argue that driver and carrier incentives aligned around fixing the equipment, allowing the response to reinforce job fit and support.

Pay-related and other misaligned complaints behaved differently. The study does not show that hardship is good for retention. It suggests that a visible, solvable problem can become evidence about the employment relationship when the employer responds. The authors also identify self-reporting and unobserved employee differences as threats, and call for field experiments to address confounding.

What leaders can responsibly monitor

• Transitions: early-tenure friction, onboarding gaps and clarity about the role.
• Stagnation: long time in role without credible growth, expanded scope or recognition.
• Local shocks: clusters of departures, workload transfers and changes in team confidence.
• Response quality: whether reported problems receive a timely remedy and whether employees can see the follow-through.

These are workplace conditions to investigate at an appropriate group level. They are not a surveillance recipe. Retention work becomes more useful when the signal leads to a better system, not a secret score attached to a person.