The best manager is the one who didn't volunteer.

A site loses four engineers in six months. Each exit interview names something different: pay, commute, a project, a partner's move. The HR record shows what the interviews did not: all four reported to the same manager, who was promoted two years ago because he was the best engineer on the team.

Who becomes a manager

In a 2026 study in the Quarterly Journal of Economics, Weidmann, Vecci, Said, Bhalotra, Adhvaryu, Nyshadham, Tamayo and Deming ran a preregistered experiment in which some managers were chosen by self-selection and others assigned at random. The volunteers did worse. What predicted a good manager was a measurable skill at economic decision-making, not personality, not demographics, not confidence. Validated against a retail chain, one standard deviation of manager quality was worth about 4.1 million dollars a year per store, around 25% of sales.

The earlier study of promotion among 131 firms' sales staff explains why the volunteers get the job: firms promote on current performance in the current role, which predicts worse management, and the people most eager to manage are often the ones with the most to gain from the title.

What a manager does to attrition

Hoffman and Tadelis, in a large technology firm with a 95% survey response rate, found that a manager's people-management skill was the strongest predictor of whether their reports stayed. Moving from the 10th to the 90th percentile predicted a 60% fall in turnover, including the turnover the firm regretted. The effect did not show up in most non-attrition outcomes. The firm rewarded those managers with better ratings and promotions, so the signal was there; it was simply not the one used to choose them.

Span of control, the number people reach for first, explains less. A 2025 cohort of 87 Swedish healthcare first-line managers found that the number of subordinates did not predict manager turnover; stress indicators did. A Danish hospital study found wider spans associated with intention to quit and worse work-life balance, but not with burnout. The size of the team matters less than who runs it.

What the record can measure

Manager effects on attrition need no survey. The HR system holds every reporting line with its dates, every resignation with its date, and every manager change. Grouping resignations by the manager at the time, above a minimum team size, adjusted for site, tenure and job family, gives each manager's share of their team's turnover against what the same team would have produced elsewhere. A change of manager is one of the factors the turnover literature treats as a shock, which is why the practice we write into the Retention Coworker includes a skip-level check-in after any change of manager.

This is also where the limits matter. A manager effect in the record is an aggregate pattern that tells HR where to look. It is not a verdict on a person; the Weidmann study shows how wrong the obvious verdicts are. Names stay with HR business partners, managers see teams, and the first response to a high-turnover team is a conversation, not a rating.

What to change in the next promotion round

1. Measure the candidate on the job they are moving to, not the one they hold. Decision-making under uncertainty is testable.
2. Stop treating willingness as evidence. The experiment says it predicts worse.
3. Read manager-level attrition, adjusted and in aggregate, before confirming any management appointment.
4. After every manager change, schedule the skip-level conversation. It is the cheapest insurance the research supports.