Haute Lumière
Commerce · II.08 · MMXXVI · daylight
TEN CONCEPT BRIEFS · Chapter II.08 — Networks, Trust, and Topology
One page each. A reader who reads only these ten pages has the chapter.
BRIEF 1 — Generalised Trust, and Why It Is a Measurement
The idea. Generalised trust is the share of a population who answer yes to a single question: generally speaking, would you say that most people can be trusted? It is not a mood and it is not a virtue. It is a survey statistic with half a century of comparable readings behind it, and it moves slowly enough to be treated as infrastructure rather than weather.
The range. Across World Values Survey samples it runs from about 5 percent agreement at the low end to over 60 percent in the Nordic countries. That is a twelvefold spread in a single behavioural input to every transaction in an economy.
Worked example. Two firms of identical size buy the same component from a new supplier. In the high-trust economy the buyer sends a purchase order and pays on thirty days. In the low-trust economy the buyer wants a letter of credit, a pre-shipment inspection and a retention. The goods are the same. The difference in cost is entirely the cost of not knowing whether the other party will do what they said.
Why it matters. Kenneth Arrow's 1972 sentence is the whole discipline in one line: virtually every commercial transaction has within itself an element of trust, and much economic backwardness may plausibly be explained by the lack of mutual confidence. Once trust is a number, it can be budgeted for, substituted for, and bought.
You already know this because you have quoted two suppliers the same job and charged one of them a deposit.
BRIEF 2 — The Trust Dividend, With Its Error Bars
The idea. Higher generalised trust is associated with faster growth in income per head, at a magnitude large enough to matter and contested enough to require honesty.
Knack & Keefer (1997), 29 countries
10 pp more trust -> about 0.80 pp more annual growth
Zak & Knack (2001), 41 countries
15 pp more trust -> about 1.00 pp more annual growth
= 0.67 pp per ten pointsWorked example. A twenty-point move in trust — roughly the distance between a mid-table European country and a Nordic one — is 1.60 pp a year on the first slope and 1.33 pp on the second. Compounded over twenty-five years:
1.0160^25 = 1.4871 -> + 48.7 % income per head
1.0133^25 = 1.3925 -> + 39.3 % income per headThe honest part. Beugelsdijk, de Groot and van Schaik ran the coefficient through extreme bounds analysis in 2004. It held in the original 29-country sample and did not hold robustly in the extended one. The strongest causal evidence is elsewhere: Algan and Cahuc used the inherited trust of American descendants of immigrants — trust that crossed an ocean and cannot have been caused by the destination's current growth — and found a real effect.
Why it matters. Quote the range and a sceptic stays in the room. Quote the point estimate alone and you have handed them the easiest rebuttal in social science.
You already know this because you have watched a forecast quoted to two decimal places lose an argument that the same forecast quoted as a range would have won.
BRIEF 3 — Embeddedness
The idea. Economic action is carried out by people who are already in relationships, and those relationships change what the action costs and whether it happens at all. Mark Granovetter's 1985 argument cuts against two opposite errors: the under-socialised view, in which atomised agents transact on price alone, and the over-socialised view, in which people simply obey the norms they have internalised. Neither describes anyone you have met.
Worked example. A buyer has a contractual right to reject a late delivery and does not exercise it, because the supplier has delivered on time for nine years and had a fire. Under the under-socialised reading this is irrational. Under the over-socialised reading it is automatic. In fact it is a calculation — about future supply, replacement cost, and what the supplier will do next quarter — made possible only by the relationship that already exists.
The figure. Dyer and Chu's survey of 344 supplier–automaker relationships across three countries found procurement transaction costs roughly five times higher for the least trustworthy buyers than for the most trustworthy.
Why it matters. Embeddedness is where the discount lives. A firm that cannot see its own relational structure is carrying the cost of it and collecting none of the benefit.
You already know this because you have a supplier you would call at nine at night, and one you would only email.
BRIEF 4 — The Strength of Weak Ties
The idea. Information that is new to you arrives through people you do not see often, because the people you see often already know what you know.
The figures. Granovetter asked professional and technical workers in Newton, Massachusetts how they had found their jobs. Of those who found work through a personal contact:
saw the contact often 16.7 %
saw the contact occasionally 55.6 %
saw the contact rarely 27.8 %
------------------------------------
less than twice a week 83.4 %The causal version. Rajkumar and colleagues ran five years of randomised variation in LinkedIn's recommendation algorithm across 20,000,000 people, producing about 2,000,000,000 new ties and about 600,000 new jobs — one job per 3,333 ties created. Moderately weak ties did the most work; the strongest ties did the least.
Worked example. You belong to a cluster of fifteen people who all know each other. One more tie inside that cluster reaches zero people you could not already reach. One bridge to a different cluster of fifteen reaches fourteen.
Why it matters. It converts a sentiment into a budget line. Relational time spent outside your own cluster has a measurable return and it is the first thing cut when a diary gets full.
You already know this because the last genuinely useful introduction you got came from someone you had not spoken to in a year.
BRIEF 5 — Dunbar's Number Is an Interval
The idea. The famous 150 was never a point estimate. Robin Dunbar's 1992 regression of primate group size on neocortex ratio predicted a human group size of 147.8 with a 95 percent confidence interval of 100.2 to 231.1 — an interval 131 people wide, stated in the original paper and dropped by almost everyone who has quoted it since.
The reanalysis. Lindenfors, Wartel and Lind redid it in 2021 with updated primate data and modern phylogenetic methods. Across specifications the intervals run from fewer than five people to more than five hundred. Their conclusion is that a single number cannot be derived this way.
What survives. The layered series is sturdier than the point:
5 · 15 · 50 · 150 · 500 · 1500 successive ratios about 3.0A ratio is a weaker claim than a point estimate, and weaker claims survive better data.
Worked example. A firm designs its org structure on "150 is the maximum for a self-managing unit." The honest statement is: somewhere between roughly 100 and 230, depending on interaction intensity and how much of it the work itself supplies. That is still a design constraint. It is just an honest one.
Why it matters. Every architecture built on a single number inherits a precision its source never claimed. Build on the interval and the design has somewhere to flex.
You already know this because you have seen a team of sixty work like one thing and a team of ninety split in two without anyone deciding.
BRIEF 6 — The Dyad Count and the Time Budget
The idea. A network of n people who all know each other holds n(n-1)/2 relationships, and each person maintains n-1 of them. That quadratic is why trust does not scale, and it can be priced in hours.
The budget. Dunbar's time-budget work puts human social interaction at about 20 percent of waking time. On a sixteen-hour waking day that is 3.2 hours, or 1,168 hours a year, for every relationship of every kind.
Worked example. Give each active tie two hours of genuine contact a year — one hour, twice, which is a floor rather than a norm.
n dyads ties/member hours/yr share of the 1,168 h
15 105 14 28.0 2.4 %
150 11,175 149 298.0 25.5 %
500 124,750 499 998.0 85.4 %
1500 1,124,250 1,499 2,998.0 256.7 %Why it matters. The layers stop being folklore and become arithmetic. At 150 a full network takes a quarter of your entire social life. At 1,500 it asks for more hours than a year contains — which is why the outer layer is acquaintance and not relationship, and why no amount of goodwill will change it.
You already know this because you have felt a group cross the size at which you stopped knowing what everyone was working on, and nobody announced it.
BRIEF 7 — Two Topologies, and Where They Cross
The idea. A trust network pays for assurance up front in relationship and almost nothing per transaction. A contract network pays almost nothing up front and something on every transaction. The costs depend on different variables, so there is a crossing point.
trust, per member per year C_t(n) = 150 x (n - 1)
contract, per member per year C_c(T) = 250 x T
150 (n - 1) = 250 T -> n = 1 + 1.667 T| transactions per member per year | trust cheaper below |
|---|---|
| 12 | n = 21 |
| 30 | n = 51 |
| 60 | n = 101 |
| 90 | n = 151 |
| 150 | n = 251 |
The cut. The trust curve depends on membership; the contract curve depends on transactions. So a high-trust community that does not trade with itself is the most expensive topology in economics. A 300-person association doing eleven pieces of business a year with itself carries 44,850 of maintenance per member to avoid 2,750 of contracting.
Why it matters. It tells you which counterparties belong on which curve, and the answer is usually not the one seniority would give.
You already know this because you have watched a networking group consume everyone's evenings and produce four introductions.
BRIEF 8 — The Transaction Sector
The idea. The work of transferring ownership — measuring, contracting, enforcing, insuring, inspecting, brokering — is itself a large industry, and trust is the thing that shrinks it.
The figure. John Wallis and Douglass North measured the American transaction sector and found it rose from roughly 25 percent of GNP in 1870 to something over 45 percent by 1970. At the tail, the World Bank's last published enforcing-contracts series put the cost of taking a commercial dispute through the courts at about 21.5 percent of the claim in high-income OECD countries, over roughly 589.6 days, and about 41.6 percent of the claim across Sub-Saharan Africa.
Worked example. A routine contracted transaction in a well-run firm might carry 250 of contracting cost — template, credit check, purchase-order match, collections provision. On a 10,000 order that is 2.5 percent. The enforcement figure is four-fifths of an order of magnitude higher because enforcement is the tail, not the mean, and most transactions never reach it.
Why it matters. The transaction sector is where the trust dividend is paid. It is also where it is measurable, which means a firm can find its own figure without waiting for anyone's national statistics.
You already know this because you have seen a deal cost more in legal fees than the thing being bought.
BRIEF 9 — The Price of Closure, on Both Sides
The idea. A closed high-trust network produces exclusion, and the exclusion is expensive for the insiders as well as the outsiders. Both figures can be computed, and they are not the same size.
The insider's cost. Suppose delivered prices are spread uniformly over plus or minus 10 percent around the index. The expected best price from k independent candidates is 0.9 + 0.2/(k+1).
open to 40 qualified suppliers 0.9 + 0.2/41 = 0.904878
closed to 8 network insiders 0.9 + 0.2/ 9 = 0.922222
------------------------------------------------------
the price of closure 1.7344 % of purchase spend
on a purchase budget of 20,000,000 346,883 per yearThe outsider's cost. Resnick, Zeckhauser, Swanson and Lockwood sold identical goods under an established reputation and under a fresh identity. The established identity earned 8.1 percent more. A competent newcomer with 500,000 of revenue pays 40,500 a year for being new.
insider 1.7344 % of spend
outsider 8.1000 % of revenue
ratio 4.67 xWhy it matters. Both sides pay and the outsider pays about four and two-thirds times as much. Put both figures on one page and opening the network stops being an ethical argument and becomes a procurement decision.
You already know this because you have been the new supplier who had to discount to get the first order, and you have also been the buyer who only ever called the same three people.
BRIEF 10 — Over-Embeddedness, and Portable Reputation
The idea. More embeddedness is not better. Brian Uzzi found firm failure risk to be U-shaped in the proportion of embedded ties: too few and you are transacting with strangers at full cost; too many and you cannot see the market, cannot switch, and go down with your partners. The optimum is interior.
Worked example. A firm concentrates 70 percent of a critical category with one trusted supplier. Transaction costs are beautiful. Then the supplier's plant floods. The firm has no qualified alternative because every alternative was let lapse, and re-qualifying takes two quarters. The relationship was an asset and the concentration was a liability, and they were the same decision.
The fix, in two parts. Cap concentration by category and keep two live alternatives in every critical input. Then make reputation portable: give a counterparty a delivery record they can show another buyer. It looks like generosity and it is the opposite — a network whose reputational asset cannot leave has to be joined, which raises its entry price, which shrinks the number of candidates k, which costs the insiders the 1.7344 percent.
Why it matters. Portability is how you keep the selection pool wide and the relationships deep at the same time. Those are usually treated as a trade-off. They are not.
You already know this because you have kept a supplier you had outgrown, and you know exactly why you did it.