n.V1-6.02 | WHAT IT COSTS TO BE GIVEN SOMETHING

01| Between a programme that exists and a person who has it, there is a distance. The sector has consistently under-priced that distance, and where it has tried to shorten it, it has usually tried the cheapest thing first.

02| The cheapest thing is language. Change what the programme calls people, soften the vocabulary, remove the words that sting. Bhargava and Manoli tested exactly this at scale. Among 35,050 people who had failed to claim substantial Earned Income Tax Credit benefits despite having already been notified in writing that they were eligible, messaging designed to reduce perceived stigma produced no measurable effect on claiming. Neither did indemnification against errors, nor time-required estimates, nor envelope framing. What worked was mechanical: complex materials cut take-up by 6 percentage points, a longer worksheet by 4, and simply displaying the amount of money at stake raised it by 8.

03| Read that against Robert Walker's seven-country work on the shame of poverty and there is no contradiction, only a location. Shame is real, widespread and consequential — and messaging aimed at dissolving it did not move behaviour in the one large experiment built to detect it. The reconciliation is that shame is produced by structural features of how programmes treat people — visibility, disclosure, conditionality, bureaucratic posture — and is not much moved by what those programmes say about it. Adopting a term is not an intervention. Whatever the renaming is meant to accomplish has to be carried by something the person actually experiences.

The gap is two gaps

04| The single most important fact about take-up is that it is not uniform, and treating it as uniform has produced twenty years of misdirected argument. SNAP served 88 percent of eligible individuals in 2022 — the highest rate in the fifty-year history of estimating it. Roughly one in four households eligible for federal rental assistance receives it. Of 11.8 million children eligible for child care subsidies in 2022, about 1.8 million — 16 percent — received one.

05| Three programmes: 88 percent, 25 percent, 16 percent. Any argument that treats low take-up as a general property of the safety net is describing something that does not exist.

06| The distinction underneath is the one that decides what can be done. SNAP is an entitlement — everyone eligible who applies receives it, so the shortfall is a claiming gap and navigation support can close it. Housing vouchers and child care subsidies are capped appropriations — the slots are fixed below the number of eligible people, so the shortfall is a rationing gap. Improving navigation there moves a person up a queue without changing how many are served. No coordination layer can navigate somebody into a voucher that does not exist. An argument built on aggregate counts of "unserved eligible" people conflates the two, and the mix determines the size of the opportunity.

Doing it with them

07| Where the gap is a claiming gap, one finding replicates cleanly and it is not about information. Bettinger, Long, Oreopoulos and Sanbonmatsu randomised low-income families into hands-on help completing the federal student aid application, personalised aid information alone, or nothing. The assistance arm was substantially more likely to submit, to enrol in college the following autumn, and to receive more aid. The information-only arm showed no significant effect on submission at all.

08| Finkelstein and Notowidigdo found the same ordering in a different domain with 30,000 elderly people likely eligible for food assistance and not enrolled. Over nine months, 6 percent of the control group enrolled, 11 percent of those given information, and 18 percent of those given information plus application assistance. Information roughly doubled it. Assistance roughly tripled it.

09| Two programmes, two populations, one result: telling people helps somewhat, doing it with them helps far more. That is a stronger and more specific finding than the general complaint about fragmentation, because it names the active ingredient — hands-on completion, not provision of facts. What neither trial establishes is durability. Both measure enrolment, over months rather than years, and neither follows whether assisted enrollees stay enrolled or whether enrolment produces the outcomes the programme exists for.

What the cost is made of

10| Herd and Moynihan decompose the burden into three parts: learning costs — finding out the thing exists and whether you qualify; compliance costs — queues, forms, documentation, money paid to intermediaries; and psychological costs — stigma, loss of autonomy in dealings with the state, the stress of the process itself. The taxonomy earns its place by separating what a single intervention treats identically. Information addresses the first. Assistance addresses the second. The Bhargava and Manoli null suggests neither addresses the third.

11| Their sharper argument is that these burdens are frequently not accidental. Where burden functions as a rationing device, reducing it is a political act rather than a technical improvement, and will be resisted as one.

12| If burden suppressed applications at random it would be merely wasteful. It does not. Deshpande and Li used closures of Social Security field offices as a natural experiment and found a persistent 16 percent decline in disability recipients in surrounding areas, concentrated among applicants with moderately severe conditions and low education. Applications fell only 10 percent — meaning the closures reduced targeting accuracy, screening out disproportionately eligible people. Finkelstein and Notowidigdo cut the other way: those induced to enrol by their intervention were less sick and received lower benefits than those already enrolled. Reducing burden brought in people with less to gain.

13| These are not contradictory. They are different programmes with different screening properties, and together they kill any simple story. Burden does not uniformly screen out the neediest, and removing it does not uniformly reach them. Under what conditions application cost screens on need rather than on capability is unresolved, and it is the question that decides whether navigation help is progressive or regressive.

The thing nobody has measured

14| One proposition sits underneath all of this and has no evidence behind it in either direction. Organisations that control access to resources set conditions for access, and conditions are a form of rule. Reporting requirements shape what a person says about themselves. Eligibility criteria shape what they present as their situation. On that reading, the agencies distributing help are governing the people who receive it.

15| It is not implausible. It is also not measured. There is a substantial literature on aid conditionality, a substantial literature on participation, and ethnographic writing on how people present themselves to gatekeepers — and nothing that establishes these organisations as governing actors in the sense that the criminal-governance literature establishes for armed groups. The honest treatment is to state the proposition, mark it untested, and say what would test it: whether recipients alter their self-presentation to meet programme criteria and by how much, whether conditions function as rules with sanctions, and whether people experience them as authority. Until then it stays an intuition, and an intuition should not be dressed as a finding.