Three winters ago in Japan, my wife and I prepared for our first-ever skiing adventure. Excited but frugal, we decided at the last moment to save a few yen by buying universal-size ski gloves. They looked fine in the store—stretchy material, adjustable straps. Hope, as it often does, overrode experience.
On the slopes the next day, reality arrived quickly. For my wife, the gloves were snug and functional. For me, they slipped constantly. My fingers went numb within minutes. Every run became an exercise in pocket-warming, my skiing reduced to a cycle of cold and frustration. We had bought the same product, yet we had purchased two entirely different experiences.
We recognise this flaw immediately in everyday life. We accept that bodies differ and comfort is contextual. Yet in development policy, we routinely suspend this wisdom. Programmes are designed for the average farmer, the typical household or the standard village. Benefit sizes and compliance conditions are calibrated to a statistical middle—a phantom mean that feels reassuringly neutral on a spreadsheet.
The paradox is not that results turn out uneven; that should surprise no one. The real puzzle is why we are surprised at all. We know people live under different constraints, yet we build economic systems as if difference were a complication rather than the central design problem. This is a failure of economic imagination. What if the most important variable is not missing from our models, but hiding in plain sight—in the simple fact that no two people experience the same policy in the same way?
The Blister and the Blueprint
A shoe designed for the average foot does not merely fit poorly; it injures. For most, it creates small, accumulating harms—blisters that make every step painful. A few lucky feet will find the fit tolerable. From a distance, the designer might declare the shoe a success. The average wear pattern looks fine; complaints are scattered and easy to dismiss in a final report.
This is how many policies fail quietly. They are blueprints drawn for an imaginary typical resident—a statistical ghost that haunts the plans. When that blueprint meets reality, the mismatch shows up as friction: a condition that is easy for some and binding for others; a benefit that transforms one household and barely registers for another. The policy does not collapse. It rubs.
Why? Because the value of a resource is never fixed. It depends on who you are and what you already have. The same amount of money or support enters different lives at different pressure points. For someone living close to the edge, a resource acts as relief—preventing loss and stabilising the present. For someone with more room to manoeuvre, that same resource becomes leverage: something to be invested and multiplied into future gains.
Neither response is irrational. Both are sensible reactions to different architectures of constraint. What differs is not intelligence or effort, but the economic meaning of the intervention. When policies are built around an average person who does not exist, they mistake these meaningful differences for statistical noise. In reality, those differences are the mechanism itself.
In economics, this is the problem of heterogeneity and diminishing marginal utility. A dollar does not generate the same benefit for everyone because people face different constraints and starting points. When policies are judged by averages alone, they conceal these varied responses—differences that often carry more policy weight than the mean itself. This is why modern development research asks not only whether a policy works, but for whom, and under what conditions.
Two Lives, One Number
Imagine a local government rolling out a well-intentioned policy: a one-time $1,000 investment grant, allocated by lottery to entrepreneurs across a village. The aim is straightforward—stimulate productive activity. Six months later, evaluators compare average business incomes between recipients and non-recipients. The result looks promising. On average, treated entrepreneurs earn about $565 more. The bar chart is reassuring (figure 1): the programme worked.
But beneath that tidy average, the story fractures. Mina and Sarita—entrepreneurs in a Nepali hill district—both received the same $1,000, yet they inhabit different architectures of constraint. Mina’s economic life is built on thin margins: a rain-fed plot, volatile prices, debt that compounds faster than her crops grow. Sarita’s architecture has more load-bearing walls: a tea shop with steady foot traffic and predictable turnover.
For Sarita, the grant is catalytic capital. She upgrades her cooking setup, lowers operating costs, and expands her menu. Her monthly income rises by over $1,100—the grant multiplies. In the data, this shows as a rightward shift in the income distribution for shopkeepers—the signature of accumulation (figure 2). Growth arrives with dispersion (figure 3).
For Mina, the grant is protective liquidity. Investing in new seeds would mean gambling with her family’s food security. Instead, she pays down a high-interest loan. Her business income rises only slightly. By conventional metrics, the effect looks negligible. But something critical changes: the likelihood of a catastrophic month falls. The grant acts as a buffer.
Average these two trajectories, and you get that $565—arithmetically correct, but economically deceptive. The average records a midpoint that neither entrepreneur actually experienced. It hides that the same policy functioned as a boost for one and a buffer for the other. This is a story of different economic velocities—speed and direction depend on the human architecture through which resources flow (figure 4).
The Trap of the Grand Design
The failure, if it can be called that, is rarely a lack of competence. It is a habit of design. Faced with immense complexity, policymakers understandably reach for simplification: a single benefit size, uniform rules, a single metric of success. These are institutional reflexes shaped by budget cycles and audit trails. The problem begins when simplification hardens into assumption—when the average becomes a stand-in for reality.
Designing for the mean is seductive because it produces clean systems. Uniform grants are easier to explain, administer, and justify. On paper, the policy travels smoothly from blueprint to rollout. In practice, that smoothness is achieved by sanding away the very differences that shape how people respond. The result is rarely collapse. It is quiet under performance.
This is why programmes so often generate ambiguous results. The averages move a little, but the variance grows. Beneath the surface, the same $1,000 acts as seed capital for Sarita and an insurance premium for Mina. When success is defined narrowly—as income growth alone—one of those functions vanishes from the ledger.
The trap, then, is not technical; it is interpretive. When we optimise for a single outcome like growth, we misread protection as failure. We unintentionally privilege those already positioned to convert resources into multiplication. None of this requires bad intent. It emerges naturally from systems designed around uniformity.
The alternative is not chaos. but architectural awareness. This begins with designs that ask not just “what should this achieve?”, but “for whom, and under what constraints?”. It means building targeted adaptability. Measurement, too, can widen its lens, valuing not only how much income increased, but how much risk fell or pressure was relieved.
Seen this way, the lesson of heterogeneity refines rather than critiques policy ambition. The grand design does not fail because it aims high. It falters when it forgets that people enter systems not as averages, but as lives already in motion.
Seeing the People in the Data
To see heterogeneity clearly is not to abandon rigour; it is to practise a deeper form of it. The work of development economics begins not with averages, but with attention—the willingness to notice that the same resource can mean survival in one life and acceleration in another. When we learn to see that difference, the data does not become messier. It becomes more honest.
This recognition is the cornerstone of the human architecture of development. Before we debate markets, institutions, or reforms, we must first acknowledge that policies do not act on spreadsheets. They move through lives already shaped by risk, obligation, and hope. The average is not a mistake—but it is never the whole story.
Our task, therefore, shifts. It is no longer about finding the single best policy for a representative person. It is about building systems that can respond to a spectrum of human rationalities, each sensible within its own constraints.
This sets the stage for February, where we will explore how those constraints shape what we call irrational choices—and why they are often anything but.
Useful Readings
For readers who wish to explore the simulated data, replicate the figures, or adapt this exercise, all code and results are available in the accompanying GitHub project repository. To see how these ideas appear in the research, the following papers are a good place to start.
Banerjee, A. V., Breza, E., Duflo, E., & Kinnan, C. (2019). Can microfinance unlock a poverty trap for some entrepreneurs? (NBER Working Paper No. 26346). National Bureau of Economic Research.
Edmonds, E. V., & Schady, N. (2012). Poverty alleviation and child labor. American Economic Journal: Economic Policy, 4(4), 100–124.