Trust Density: The Metric Nobody Reports and Everybody Feels

Sixty-eight percent of Ugandan adults borrowed money from family or friends in the year before their 2024 Global Findex interview, and 57 percent bought food on credit. Eleven percent borrowed from a bank or similar institution. Those figures measure trust by what people did with money, and by that measure Ugandan trust is dense. A startup community can report the same thing about itself: count the pairs of members who have kept a promise with something at stake, divide by the pairs possible, and publish the ratio with its denominator.

Key Takeaways

• Stated and practiced trust diverge sharply in East Africa. Fewer than one Ugandan in ten told the World Values Survey that “most people can be trusted”, while 68 percent of adults borrowed from family or friends in 2024.
• The eleven ecosystem measurement frameworks reviewed by Morgan Clevenger and Chao Miao name trust once, inside a “relational dimension”. Their network indicators assume registered firms with ten or more staff.
• In a Kampala experiment, cooperation tracked how easily people could find one another: runners found 43.1 percent of same-group targets and 27.8 percent of others.
• Credit bureaus, network maps and referral logs each hold a slice of trust. Each records one kind of relationship and misses the peer obligations that carry a founder community.
• The Kept-Promise Rate reports the share of member pairs with a confirmed kept promise in a quarter, plus the share of those pairs that run through a single broker.

What Ugandans say about trust, and what they do with money

The oldest trust measure in social science is one question. “Generally speaking, would you say that most people can be trusted or that you need to be very careful in dealing with people?” In the World Values Survey wave that closed in 2004, 7.6 percent of Ugandans chose “most people can be trusted.” In the wave that closed in 2022, 9.5 percent of Kenyans and 11.9 percent of Ethiopians gave that answer, against 37 percent of Americans and 72 percent of Norwegians.

Afrobarometer asks the question differently and gets a different country. In its 2024 Uganda survey, 79 percent of respondents trusted their relatives “somewhat” or “a lot”, 64 percent their neighbors, 53 percent fellow citizens and 49 percent people from other ethnic groups. The Findex behavior is denser still. Beyond the family borrowing, 30 percent of Ugandan adults borrowed from a savings club in 2024 and 29 percent borrowed to start or run a business. Kenya shows the same shape: half of adults borrowed from family or friends and 55 percent bought food on credit, against 13 percent who borrowed formally.

These numbers come from different questions, years and samples, so they cannot be netted against one another. Side by side, they show the instrument deciding what it finds: a question about “most people” reads low where trust is carried by named people, and a question about last year’s borrowing reads high in the same society.

Edward Glaeser and three co-authors tested which kind of measure deserves weight. In their 2000 study “Measuring Trust”, the standard survey questions predicted trustworthy behavior in experiments far better than trusting behavior. What predicted trusting behavior was a person’s record of trusting outside the lab. A founder community’s trust is best read from that second kind of number: the record of what members have already risked on each other.

Eleven ecosystem frameworks and one mention of trust

Morgan Clevenger and Chao Miao’s chapter in Entrepreneurial Communities and Ecosystems (Routledge, 2022), edited by Clevenger and Michael Fortunato, reviews eleven ways scholars have proposed to measure an entrepreneurial ecosystem, from Neck and colleagues in 2004 to Theodoraki and colleagues in 2018. Their Table 3.3 lists the indicators: incubators, spin-offs, finance, policy, human capital, density, fluidity, connectivity, diversity, infrastructure. Trust appears by name once, inside Theodoraki’s relational dimension of “trust, norms, members’ obligations, and identification.” That placement fits the book’s own distinction, which treats community as the social body an ecosystem grows from.

The indicators that touch relationships show where the frame stops at the Ugandan border. Stam’s 2017 framework measures networks as “the percentage of businesses (with at least 10 employees) in a region that collaborate for innovation.” It measures finance as the share of SMEs that applied for bank finance and received it.

Run those two indicators in Kampala. The Uganda Bureau of Statistics’ 2009/10 business inquiry classed 96 percent of roughly two million businesses as informal, a category that includes having fewer than five employees. The figure is old and later censuses may have moved it, yet the shape is plain in Owino market. A count of ten-employee firms collaborating on innovation would miss the stall owners who extend each other stock on Tuesday and settle on Saturday. A bank-finance indicator would read 11 percent of adults and miss the 68 percent who borrowed on personal trust.

The book’s indicators assume firm registers and bank credit at the center. Here the trust that moves commerce runs through informal credit between people who know each other, so a Kampala hub’s metric has to come from behavior it can observe among its own members.

Mulago-Kyebando’s 148 runners

James Habyarimana, Macartan Humphreys, Daniel Posner and Jeremy Weinstein ran a series of games with a random sample of 300 residents of four adjacent Kampala parishes, Mulago I, II and III and Kyebando, to find out why ethnically similar neighbors cooperated more.

One game measured the network directly. The researchers photographed 148 randomly chosen targets and sent 148 runners, each with a photo, a name and a parish, to find them. A runner who found the target within three hours earned 20,000 shillings, with the reward falling by 1,000 shillings an hour after that. Forty-nine of the 148 succeeded, just over a third. Runners sent to find someone from their own ethnic group succeeded 43.1 percent of the time, against 27.8 percent for others.

The other games ruled out the warm explanations. The researchers found no evidence that co-ethnics shared tastes or felt more generous toward each other. Cooperation rested on norms and institutions that made it easier to sanction someone who failed to contribute. And people who were easy to locate offered more in the giving game even when their partner came from a different group.

For an ecosystem builder, that finding defines the thing to measure. Trust density is the ease with which members can find one another and hold one another to a promise, which is a different quantity from how warmly they speak of each other.

Records that already hold trust, and what each one misses

Several instruments already capture part of this behavior in East Africa, each built for a different relationship.

The credit record comes closest to behavioral trust. Uganda’s bureaus collect loan amounts, payments and group membership, and dfcu Bank tells its customers that a good repayment record may substitute for “big securities like land titles.” Borrowers must consent before their history is shared. The record is still a lender’s view.

Endeavor’s Kenyan map traces flows of people, capital and information; its 2018 predecessor found donor-funded support organizations to be the most influential actors in the Kenyan ecosystem. It shows who connects to whom and leaves open whether those connections have been tested.

Gharad Bryan, Dean Karlan and Jonathan Zinman paid South African borrowers to refer friends for loans and found no evidence that referrers picked better borrowers, yet large effects from referrers pressing their friends to repay. A referral count tells a hub who will lean on whom, the same enforcement path the Kampala runners revealed.

Perverse incentives are real incentives

Onora O’Neill gave the 2002 BBC Reith Lectures on trust. In the third lecture she argued that public services had been loaded with “performance indicators chosen for ease of measurement and control rather than because they measure accurately what the quality of performance is.” Hospitals chased waiting-list targets and schools chased easy passes. Her summary was five words: “Perverse incentives are real incentives.”

A published trust number for a hub carries exactly that risk. Members can confirm favors to each other to lift the rate. A hub funded on community metrics can count handshakes as promises. A funder can rank hubs on the figure and teach every hub to manage it. The risk is real, so a trust report should be designed on the assumption that someone will try to game it.

The same lecture also points to the design. She observed that people who tell pollsters they distrust doctors still book operations, so “our action often provides evidence that we still trust.” She also asked for “intelligent accountability”, an account given to people with “sufficient time and experience to assess the evidence.” For a founder community, those people are the members themselves.

Trusted with very little

Every behavioral trust measure above counts small things: a crate of stock paid for on Saturday, a referral honored, an advance repaid. Jesus put the same unit at the center of trust: “Whoever can be trusted with very little can also be trusted with much.” The next lines name what is being handled, “someone else’s property”, and ask who will trust a person with “true riches” if she has been careless with that. O’Neill’s warning lands on exactly this ground, because once a count exists, people start keeping promises for the count. Luke 16 places the reason for faithfulness with a Master who sees the transaction that was never logged, such as the overpayment a trader hands back when nobody would have noticed. Founders who keep those promises give a logged rate its meaning. The rate samples faithfulness; the unlogged promises decide whether the sample tells the truth.

The Kept-Promise Rate

The Kept-Promise Rate gives a hub one number to report each quarter, with two companions that keep it honest. The rate is the count of member pairs who completed at least one kept promise with something at stake during the quarter, divided by the number of possible pairs. A kept promise is any obligation between two members that was met as agreed and confirmed by both: stock or a service supplied on credit and paid, a referral made and the customer served, an advance repaid, a joint bid delivered. Both sides confirm, which answers the gaming O’Neill predicts, since one member cannot inflate the count alone.

Picture a cohort of 30 founders; the figures here are invented. It has 435 possible pairs. If 61 pairs confirm a kept promise in the quarter, the rate is 14 percent. The first companion is the broker share: the fraction of those 61 pairs that include the single most-connected member. If 27 of them include the program manager, 44 percent of the community’s tested trust runs through one person, and it will leave when she does. The Kampala runners showed trust following the paths along which people can be found, and paths that all cross one desk make a fragile community. The second companion is the method in plain words: who counted, which promises qualified, and how members consented.

The reporting order follows O’Neill. Members see the cohort figure first, as one aggregate with no ranking of individuals, and only then does it go to a funder. A rising rate with a falling broker share means trust is spreading between founders; with a rising broker share, the hub is doing the trusting on everyone’s behalf.

No AVODA cohort has a kept-promise rate on record. When we publish the first one, it will carry its denominator and its broker share.

Frequently Asked Questions

What is trust density in a startup community?

Trust density is the share of possible relationships in a community that have carried a real obligation and held. In practice, count member pairs who kept a confirmed promise with something at stake in a period, such as credit repaid or a referral honored, and divide by all possible pairs among members.

How do surveys measure trust in Uganda and Kenya?

The World Values Survey asks whether “most people can be trusted”; under 10 percent of Ugandans (wave ending 2004) and Kenyans (wave ending 2022) said yes. Afrobarometer asks about specific groups: in 2024, 79 percent of Ugandans trusted relatives and 53 percent fellow citizens. Wording drives the answer, so behavioral data is a stronger base.

Can repayment data measure trust between founders?

Partly. Uganda’s credit bureaus record repayment history, including group members, and since 2022 SACCOs and microfinance lenders report to them. That record captures trust between a lender and a borrower. Obligations between founders, such as stock on credit or shared orders, need the community’s own two-sided record.

How do you stop a trust metric from being gamed?

Count only promises both parties confirm, report one aggregate for the cohort with no individual rankings, show members the figure before funders, and publish the method beside the number. Onora O’Neill’s warning about perverse incentives applies; accountability to people who can judge the evidence is the safeguard.

How often should a hub report trust density?

Quarterly suits most founder communities, because trade credit and referrals between small firms turn over within weeks. Report the rate, the number of possible pairs and the broker share together. A quarter shows repeat transactions and catches trust concentrating on one person early.

Next Step

Settle the rules before anyone starts tallying: which promises qualify, who confirms them, and what members see before any funder does. Joshua can work through those three questions with your team on a strategy call. After it, mark each pair on the member list against last quarter’s dealings, divide by the possible pairs, and note how many of those pairs run through one person.

Related Reading

• Chama Savings Groups: Covenant Economics Africa Built
• Alumni Networks Are the Product; Cohorts Are the Funnel
• AI for SACCOs: Credit Decisions You Can Actually Defend
• Accelerator Outcomes Measurement: Survival and Revenue

Sources and Evidence

1. Morgan R. Clevenger and Michael W-P Fortunato (eds.), Entrepreneurial Communities and Ecosystems: Theories in Culture, Empowerment, and Leadership, Routledge, 2022: chapter 3 by Clevenger and Chao Miao, Table 3.3 and the Metrics section (pp. 72 to 77), including Theodoraki’s relational dimension and Stam’s network and finance indicators (DOI 10.4324/9781351045711).
2. Our World in Data, “Share of people agreeing with the statement ‘most people can be trusted'”, Integrated Values Surveys: question wording; Uganda 7.6 percent (wave ending 2004), Kenya 9.5 and Ethiopia 11.9 percent (wave 2017 to 2022), United States 37 and Norway 72 percent.
3. Afrobarometer Dispatch AD1210, “Ugandans embrace unity in cultural diversity, but perceptions of ethnic bias are on the rise”, July 2026: Figure 1, trust in people, Uganda 2024 (relatives 79, neighbors 64, other citizens 53, other ethnic groups 49 percent).
4. World Bank Global Findex 2025, borrowed from family or friends, Uganda and Kenya, 2024: 68.3 and 49.6 percent of adults; companion indicators fin22a (formal borrowing 10.8 and 12.6), fin22c, fin22e and fin22f from the same database.
5. Glaeser, Laibson, Scheinkman and Soutter, “Measuring Trust”, Quarterly Journal of Economics, 2000: survey trust questions predict trustworthy behavior better than trusting behavior; past trusting behavior predicts trust.
6. Habyarimana, Humphreys, Posner and Weinstein, “Why Does Ethnic Diversity Undermine Public Goods Provision? An Experimental Approach”, 2006 working version of the APSR 2007 article: 300 subjects in Mulago-Kyebando, Kampala; 49 of 148 runners found targets, 43.1 versus 27.8 percent; sanctioning and findability.
7. Endeavor Insight and Endeavor Kenya, “Mapping the Kenyan Entrepreneurship Network”, 2026: more than 100 founder interviews (April to May 2025), over 730 companies, network flows of people, capital and information; 2018 finding on donor-funded support organizations.
8. dfcu Bank, “Credit Reference Bureau Awareness”: data collected including group members, consent rights, the 2022 regulations adding SACCOs and microfinance lenders, National Identity Number as identifier from 1 February 2024.
9. Bryan, Karlan and Zinman, “Referrals: Peer Screening and Enforcement in a Consumer Credit Field Experiment”, AEJ: Microeconomics, 2015: South Africa; no evidence of screening, large enforcement effects.
10. Onora O’Neill, Reith Lectures 2002, Lecture 3, “Called to Account”, BBC transcript: performance indicators, “Perverse incentives are real incentives”, action as evidence of trust, intelligent accountability.
11. Nankya Winifred Mukalazi, Uganda Bureau of Statistics, “Data collection challenges in Establishment Surveys: The Case of Uganda Business Inquiry 2009/10”: 96 percent of about 2 million businesses informal; informal attributes include fewer than five employees.
12. Luke 16:10-12, New International Version, Bible Gateway: “Whoever can be trusted with very little can also be trusted with much.”