Why a Knock on the Door Beats a Dozen Robocalls

After the pandemic shut schools, a problem that had always simmered boiled over. Across the United States, the share of students who were chronically absent — missing at least 10 percent of school days, roughly 18 days a year — nearly doubled. Automated calls, warning letters, and truancy threats did little. In 2021, Connecticut tried something older and simpler. It sent school staff to knock on the doors of its most absent students, in its 15 poorest districts, and talk to them and their families face to face. The program was called LEAP: Learner Engagement and Attendance Program.

It worked. Attendance rose 4 to 5 percentage points. Chronic absenteeism fell 13 to 15 points. Disciplinary incidents dropped too. Why does a home visit succeed where a dozen phone calls fail?

The model: relational trust

The best-evidenced explanation is relational trust, a concept developed by sociologists Anthony Bryk and Barbara Schneider from a decade studying Chicago's public schools in the 1990s. Their finding, set out in Trust in Schools (2002): the schools that improved were not the ones with more money or better programs, but the ones where the adults and the families trusted each other — where each side believed the other was acting in the child's interest. Trust was the hidden infrastructure on which everything else ran. Schools in the top quartile of relational trust were far more likely to show real gains in reading and maths; those that stayed low-trust had almost no chance.

A home visit is, mechanically, a trust-building act. That is its whole theory of change — and, as we'll see, the data is cleaner on that than on the more ambitious claim that visits build lasting habits of self-discipline in the child.

The mechanism, step by step

Here is how a knock on the door turns into a changed habit.

1. It signals that someone noticed. A robocall is a system talking. A person at your door is a human who chose to come. That distinction is the intervention. It converts an anonymous institution into a known individual — the raw material of trust.
2. It surfaces the real obstacle. Chronic absence rarely has one cause. A phone call cannot discover that a child is minding a younger sibling, or afraid of a bully, or without clean clothes. A conversation at home can.
3. It rebuilds the broken link between home and school. Once a teacher and a parent know each other, absence becomes something they manage together rather than something the school polices and the family resents.

Notice what this does not yet claim: that the child has acquired a durable new habit of showing up. That stronger claim is exactly where the evidence gets thin.

The other view: was it the program, or the students?

Any evaluation of a program people opt into faces the same trap, and naming it is the most useful thing here.

The threat is selection bias: the possibility that the students who got the visit were already different — in ways that explain the outcome — from those who did not. If districts quietly visited families who were readier to re-engage, the program would look effective even if the knock did nothing.

The researchers attacked this in two layers. The first is selection on observables — differences you can see in the data, like income or race, which you can statistically match away. The harder one is selection on unobservables — differences you cannot measure, like how motivated a family already was.

Here is the key move. Students were visited at different times over the rollout. The raw data appeared to show the effect growing — attendance begetting more attendance, a compounding habit. But that pattern can be manufactured by when people enter: if families visited later were systematically different (say, harder cases reached last, or keener ones reached first), comparing early-visited to late-visited students confounds the program with the timing. Using methods designed for staggered rollouts, the researchers separated the two.

The result splits cleanly. The size of the effect held up: home visits really did improve attendance and cut absenteeism, broadly, across gender, race and income. But the apparent growth over time vanished. The honest effect is constant, not compounding.

This matters for the model. A real, durable boost is worth a great deal — it vindicates relational trust. But the appealing idea that the child internalises the habit and attendance snowballs on its own was partly a statistical mirage. The relationship, not a newly minted virtue in the student, is doing the work — which is why it may need maintaining.

Where else this shows up

The pattern — a human contact outperforming an automated one, precisely where trust is the missing ingredient — repeats far beyond schools.

• Debt collection. Agencies find a respectful call from a named person recovers more than a stack of threatening letters, because it restores a relationship instead of triggering avoidance.
• Medicine. Patients with chronic conditions adhere to treatment far better when a nurse calls or visits than when they get reminder texts. Compliance becomes partnership rather than policing.
• Voter turnout. Decades of randomised experiments by political scientists Alan Gerber and Donald Green show the same ranking every time: a face-to-face conversation on the doorstep reliably lifts turnout; a mailer or robocall barely moves it.

The common thread: humans respond to being seen by another human in a way no notification reproduces. Systems scale cheaply; relationships do not. The expensive, un-scalable thing wins exactly when trust is what's broken.

What to watch

The real test of LEAP is persistence. A visit this year is one thing; whether the gain survives into years when no one knocks again is another — and the "constant, not compounding" finding is a warning that it might not.

Watch two signals. First, whether states that copied the model report similar effect sizes under independent evaluation; replication is how you tell a real effect from a local fluke. Second, whether attendance gains eventually show up in test scores — the study found behaviour changed but could only suggest learning followed.

Expectation: ~70% that at least one rigorous replication confirms an attendance effect of broadly similar size; under 40% that the gains fully persist two or more years after visits stop without further contact. The relationship, not the single knock, is what has to be maintained.

Distilled from NBER Working Papers