Teacher attrition in hard schools

Why do the schools that most need experienced teachers keep losing them fastest?

The pattern is depressingly reliable: the schools serving the most disadvantaged intakes have the highest staff turnover, the youngest staff, and the most vacancies filled late. The obvious reading is that the work there is harder and people leave hard work.

But that should not satisfy us, because hard work is normally paid for. Difficult jobs attract a premium — night shifts, remote postings, dangerous trades — and the premium is what stops those posts emptying. So the question is not really why teachers leave. It is why nothing arrives to stop them.

In an ordinary labour market, if two employers offer identical pay and one has worse conditions, workers flow to the pleasant one until the unpleasant one raises its wage. That differential is not generosity; it is what clears the market.

Now look at how teachers are paid. In most systems the salary is set by a schedule — years of service and qualifications — applying across a district or a nation. Two schools ten minutes apart, with utterly different conditions, offer the identical figure. The variable that would ordinarily equilibrate the posts is, by design, held constant.

If pay cannot adjust, something else must, and the only quantity left free is who takes the job. A teacher with options exercises them; a teacher without — newly qualified, newly arrived, out of area — takes what is available. The hard school is not staffed by people who chose it over the alternative; it is disproportionately staffed by people for whom it was the alternative. Sorting does the work a wage would do elsewhere.

But composition alone would give a stable, if unfair, arrangement — a school permanently staffed by novices at a constant rate. What makes it get worse rather than merely stay bad?

Here is the loop. Each departure creates a vacancy, and late vacancies are filled by whoever is still available, which skews novice. Novices need induction, mentoring, and someone to ask — supplied by exactly the experienced staff whose numbers are falling. The school's capacity to absorb newcomers declines as its need for it rises. Coherence goes with them: who has taught what, which pupil needs what, why the department sequences topics as it does. That is knowledge held collectively and lost individually. And a school losing coherence is one where behaviour and workload get worse — its conditions degrade. Conditions were the input we started from, so the loop closes, and closes as a reinforcing one.

Is that right, or merely tidy? The rival explanation is simpler: teachers leave because the children are hard, and nothing about the school can change that. These make different predictions. Work on transfer decisions — Ladd's analyses, and Boyd and colleagues' on New York City — points one way: once working conditions are measured properly, particularly leadership, discipline and safety, and the influence teachers have over their own practice, the apparent effect of pupil demographics falls sharply. Teachers appear to be leaving badly-run schools rather than poor children — and since those schools disproportionately serve poor children, the two look identical unless you separate them.

That is hopeful, because leadership is a policy variable and intake is not. It is also where honesty is most required: the evidence is observational, resting on teachers' own reports and on controlling for what happened to be measured, and a school's conditions and its results are entangled in ways no survey unpicks.

One result pushes harder. Ronfeldt and colleagues examined turnover in New York and found it predicted lower achievement even for pupils whose own teacher had not left. That is what the individual-quality story cannot absorb: if turnover mattered only because good teachers depart, pupils remaining with their teacher should be unaffected. That they are not is evidence for an institution-level harm rather than a sum of individual losses.

Three refinements.

First, "hard school" has been doing two jobs and should be split: the intake, which a school cannot change, and the conditions, which it substantially can. The evidence suggests the second is what teachers respond to. That does not make the first irrelevant — intake shapes conditions — but it locates the intervention somewhere reachable.

Second, some turnover is healthy. A school with none cannot replace a poor match, and beginners must start somewhere. The harm is specifically in rate: churn fast enough that induction cannot keep pace and nobody is left who remembers how the place works.

Third, on evidence quality. I quote no effect sizes, because published magnitudes vary enormously with the measure — leaving the profession, the school, or the district are three different numbers routinely reported as one. Retention bonuses targeted at hard-to-staff schools have been evaluated with reasonably credible designs and appear to move retention modestly; I recall the direction as positive and the size as small, and go no further. That literature tests the compensating-differential reading cleanly: if a modest payment moves retention at all, pay rigidity really is part of the mechanism.

How could this be refuted? The load-bearing claim is that conditions dominate intake. The refuting observation would be a school whose leadership and conditions changed markedly while its intake did not, with turnover unmoved — or the reverse. Principal-succession studies are the natural place to look. The Ronfeldt spillover claim has its own refutation available: if the harm to non-switching pupils disappears once you condition on the departing teachers' effectiveness, it was compositional after all and there is no institutional loss to explain.

• Why targeted retention payments move retention less than a flat wage differential would predict.
• How a school rebuilds collective knowledge after it has been lost.

Every term the collection defines is gathered in the glossary.