n.V1-4.05 | THE PEOPLE WHO DO THE WORK

01| Every model in this field runs on people — local, mostly unpaid, mostly uncounted, and mostly assumed to behave the same way at ten thousand as they did at ten. Three bodies of evidence say otherwise. What happens to a programme when it grows. What happens to knowledge when people teach people. And what happens to unpaid work when money arrives.

The voltage drop

02| The failure of successful pilots to survive expansion is not an impression. It has been studied and named.

"Between 50 and 90 percent of programs will lose voltage at scale."— John List (The Five Vital Signs of a Scalable Idea and How to Avoid a Voltage Drop, Behavioral Scientist)

03| The first reason is the hardest to accept, because it says the problem began before the scaling did:

"In more than half the cases where a government tries to scale an idea, it never had voltage, and when you try to replicate it, it doesn't work."— John List (ibid.)

04| The rest are mechanical rather than mysterious, and two of them apply to every community-led model ever designed:

"If your educational intervention pilot involves all of the good teachers, it likely won't work when it scales and the weaker teachers become involved.""If you are monitoring compliance in the pilot — that is, the experimenters are watching — failure to monitor during the rollout may result in a substantial voltage drop."— John List (ibid.)

05| Both describe one structural fact. A pilot recruits the most motivated people available, personally, in a small area, under observation. Expansion recruits the next-most-motivated, at distance, unobserved. Whatever the first group produced is not evidence about the fourth. List's prescription follows: build the constraints of the eventual rollout into the original test rather than discovering them afterwards, which means the first sites should not be the easy ones and the people running them should not be the founders.

The telephone game

06| Community-led models almost all rely on people teaching people, and the evidence on what that does to knowledge is decades old and consistent. Cascade training has been studied hardest in education and health, and the dominant critique is dilution — messages adulterated as they pass down the tiers, misinterpretation left behind. The mechanism is structural, not a matter of effort:

"Each layer in the cascade introduces potential for misunderstanding the original information, similar to a 'telephone game.' Only the first layer receives feedback from an actual expert, while subsequent layers must rely on their immediate trainers who might not have the expertise to correct nuanced mistakes."— Cascade-training literature review (A new conceptual framework for train-the-trainer professional development on a large scale, 2024)

07| The literature also names the conditions under which cascades hold: facilitators effectively engaged, confident in their ability to train, supported by supervision, and equipped with the leadership skills to hold a group in its local context. Every item on that list is a recurring per-person cost. Supervision in particular does not get cheaper as the cascade widens. This is the central tension in any peer-teaching model — the fix for degradation is the thing that makes growth linear again.

08| There is a second problem, less discussed. Detecting degradation means testing generation three against generation one, which almost nobody does, because the test can only produce bad news and the response to bad news is to slow down. We found no documented case of an organisation voluntarily slowing a peer-training cascade because a later generation tested worse than an earlier one.

What it costs to keep them

09| The closest thing to a community changemaker cadre run at national scale is the community health worker, and that field has forty years of evidence.

10| The pay question alone has been mapped across five models: volunteers with no financial incentive (Nepal's Female Community Health Volunteers), volunteers who sell health products (BRAC's shasthya shebika in Bangladesh), volunteers with financial incentives (India's ASHA), paid full-time workers (Iran's Behvarz), and mixed cadres working side by side (Ethiopia's Health Extension Programme). So the basic proposition — that local residents can be trained and deployed at scale — is not speculative. It has been done in the hundreds of thousands.

11| The evidence also names what breaks it:

"Lack of appropriate incentives, with resulting high rates of turnover are common challenges in large-scale CHW programs."— Review of community health worker programmes at scale (Human Resources for Health)

12| Ethiopian data from 2006 to 2008 put attrition at 11.6 percent in rural areas and 15.5 percent in urban. The urban figure is the one that matters for city work, because in a city the competing job is nearer.

13| Supervision is the other, and BRAC's arrangement shows what it actually costs: a programme organizer supervising 25 to 30 community health volunteers, two to three times a month. That is a paid professional per thirty community workers, present in person, indefinitely. Any model claiming that community delivery removes the professional should be read against that ratio.

14| What retains people without money is documented too. Research in Ethiopia found non-financial incentives — above all a career development model — key to motivating and retaining workers who receive no stipend, with 48 percent naming future training as a major non-financial incentive. A visible ladder is itself a retention instrument, and the evidence supports it.

15| One limit runs across all of it. Every one of these cadres is single-domain. A shasthya shebika does health. An ASHA does health. No evaluated programme at scale has trained community members as generalists across health, livelihoods, infrastructure and advocacy. The single-domain design is not an accident — it is what makes training tractable and quality measurable. Whether a generalist version works has no evidence either way.

The workforce nobody counts

16| Underneath all of this sits a labour force that no funding cycle touches, and its size was just revised upward by a factor of two.

"Globally, some 34.5 per cent of working-age people, or 2.1 billion individuals, engage in volunteer work each month, according to new data-collection efforts and improved measurement tools that are providing a better understanding of the scale of volunteer engagement worldwide."— UNV & ILO (2026 State of the World's Volunteerism Report)

17| The 2022 edition put it at 862 million. The population did not double; the measurement did. UNV is explicit that the 2025 survey applied a more comprehensive approach based on ILO guidance and captured volunteer work in all its forms, producing significantly higher estimates than national surveys. The right statement is not that volunteering grew. Most of it was never counted, and a better instrument found it. Africa records the highest monthly rate anywhere, at 58.5 percent of working-age individuals.

18| The most consequential number is the split:

"Most volunteer work continues to be carried out directly, as opposed to through organizations. Globally, 25 per cent of working-age individuals are estimated to engage in direct volunteer work, compared to only 11.7 per cent who engage in organization-based volunteer work."— UNV & ILO (2026 SWVR)

19| Roughly two thirds of participation sits outside any organisation at all — not outside the funded sector, outside organisations of any kind. Every coordination mechanism, funding channel, training programme and reporting requirement in this industry attaches to the smaller share.

20| Why nobody had the number: fewer than a third of UN member states have measured volunteer work in national statistics since 2008, predominantly high-income ones, at an average of two recordings per country across fifteen years. Of those, only 40 cover total volunteer activity; the rest count organisation-based work alone, which the report says plainly will underestimate the true scale.

21| And one definitional decision deserves more attention than it gets. Volunteer work is restricted to activity recognised as producing goods or services. The report names what falls outside:

"Human activities that are inherently personal (i.e. which cannot be assigned to others) are considered to be outside the scope of work and are, therefore, not classified as volunteer work. Examples of voluntary non-work activities include protesting, boycotting, going on strike and donating blood."— UNV & ILO (2026 SWVR)

22| So the 2.1 billion excludes protest, boycott and strike. It counts the person who repairs a neighbour's roof and not the person who organises the tenants against the landlord. The activity most central to poor-led movements — mobilising, contesting, demanding — is definitionally absent from the world's headline measure of unpaid civic contribution.

What happens if you pay them

23| Which raises the design question everyone eventually reaches: if this workforce is the permanent layer, would money produce more of it?

24| The literature splits inside a single domain. Richard Titmuss argued in 1970 that paying blood donors would reduce supply by converting a gift into a transaction. Mellström and Johannesson tested it in three arms — donate unpaid, receive SEK 50, or choose between taking the SEK 50 and giving it to charity. For men, no significant difference. For women, willingness to donate fell by almost half when payment was introduced — and the effect vanished entirely when the payment could be redirected to charity. Same money, different meaning. Lacetera, Macis and Slonim then examined nearly 14,000 American Red Cross blood drives plus a field experiment and found incentives raised donations with no crowding-out at all.

25| Meanwhile the community health worker reviews point the other way: salaried status, supervision and training are repeatedly identified as determinants of success, with attrition the standing problem in volunteer models. The strongest single trial in that literature used neither salary nor volunteering but a sales margin.

26| The tension is real rather than apparent. Sustained, supervised, accountable frontline work appears to require payment. Discrete acts of giving appear vulnerable to being spoiled by it. If both hold, the variable that matters is the nature of the task, not the presence of money — and the question is never "pay or don't pay" but what the payment is understood to be. A stipend framed as recognition, a commission framed as a margin, and a wage framed as a price for labour are three different interventions with potentially opposite effects on people who were already contributing.

27| Nobody has tested this in community development. Not one trial we located randomises framing while holding monetary value constant, and measures what happens to the people who were contributing before the scheme arrived. Given how many programmes now pay community members stipends, that is a remarkable hole.