The Model With No Middle

Corridors move goods. Arthur Lewis forgot the towns.

The machines at the Dire Dawa Industrial Park, in eastern Ethiopia's corridor to the coast, run day and night, and they need almost nothing from the people who tend them. The factory belongs to a Chinese yarn manufacturer. The cotton arrives by container ship from Australia and the United States, and is trucked up from Djibouti along the highland road. The spindles turn. Every few hours, a woman walks the aisle and replaces one that has run out.

That is the job. That is most of the job there is.

I visited in November 2025. The floor was vast, clean, and nearly empty of people. The noise was tremendous. I had to lean close to my guide to hear him explain the production capacity, the export volumes, and the investment figures. He was proud of all of it, and there was something to be proud of: a functional factory, making a product that would reach global markets, operating in one of Africa’s fastest-growing economies. By the metrics that development economists first reach for, this was a success story.

I kept looking at the cotton bales. Ethiopia grows its own cotton, in the Awash Valley, two hundred kilometers to the west. The factory didn’t want it. The Chinese manager was direct about why: the quality wasn’t good enough. The fiber specifications that global yarn markets require aren’t what Ethiopian smallholders currently produce. You could hear this as a verdict on Ethiopian agriculture. I heard it as a description of a missing investment, the extension services, the improved seed varieties, the supply chain infrastructure between the farm and the factory gate, that nobody built when they planned the export processing zone.

The city officials had insisted we make the drive. The Industrial Park sits twenty minutes outside Dire Dawa, past the edge where the city’s streets give way to flat scrubland. I understood the insistence once we arrived. This was the thing they were proudest of: evidence that their city was open for business, connected to global markets, moving up the value chain. They pointed at the machinery, the export figures, the Chinese investment. All of it real.

Driving back into the city afterward, Dire Dawa announced itself mainly through Coca-Cola. Large signs on low buildings, red and white cutting through the dust. Coca-Cola has figured out Dire Dawa: found the market, built the distribution, put its name where people would see it. The yarn factory’s road led to Djibouti. The city behind us was connected to almost none of it.

Dire Dawa was built by a railway. The French laid it in 1902, connecting Addis Ababa to the coast, and the city grew up around the logic of transit and trade. Standing in it now, you could feel the memory of that logic without seeing much evidence of its current equivalent. The Industrial Park was the official answer to that absence. Twenty minutes outside town, largely automated, importing its inputs from the other side of the world, exporting its output to Asia.

Arthur Lewis would have recognized the factory. He would not have recognized the math.

In 1954, the St. Lucian economist W. Arthur Lewis published a paper that gave development economics its spatial grammar. The model was elegant: a traditional agricultural sector with surplus labor, a modern industrial sector in the city that absorbs it, and structural transformation as the arc connecting the two. Wages in the traditional sector are held down by surplus labor. The modern sector grows by drawing in that surplus, paying slightly above subsistence, accumulating capital, and growing further. Eventually, the surplus is exhausted, wages equalize, and the economy has transformed.

The model won Lewis the Nobel Prize in 1979. It also became the default blueprint for industrial policy across the developing world, which is a different thing from winning a prize and considerably more consequential.

Three things went quietly wrong.

The first is that the Lewis model is aspatial. It specifies no geography between the farm and the factory. The implicit assumption is that surplus labor walks to work. The entire economy of secondary towns, periodic markets, small processors, truckers, traders, and rural credit agents has no place in the model. Lewis drew a two-sector diagram, and development planners built a two-sector world.

The second problem was identified almost immediately by Bruce Johnston and John Mellor in 1961. Rising agricultural productivity, they argued, does more than release labor. It generates consumption linkages: farm households with rising incomes spend money on goods and services, most of which are produced in nearby towns rather than distant cities. Causality runs both ways, and much of it happens in the middle, in small-town economies. Lewis saw agriculture as a reservoir to drain. Johnston and Mellor saw it as an engine, and the towns around it as the transmission.

The third problem was the one nobody planned for. The manufacturing absorption Lewis assumed never arrived at the scale his model required. Dani Rodrik documented this as “premature deindustrialization”: African cities urbanized at rates Lewis’s model would predict, but without the industrial payoff. The surplus labor arrived. The factories, to the extent they came at all, brought their own cotton.

What Lewis’s model erased was the middle. Not the middle class, though that too, but the spatial middle: the towns, the markets, the processing clusters, the agricultural value chains that connect smallholder farms to urban consumers. Thomas Reardon at Michigan State has spent two decades mapping this “hidden middle,” the traders, processors, and logistics firms between farm and consumer, and finding it is both larger and more important than the development literature acknowledges. In many African countries, the rural nonfarm economy has grown to become the predominant source of household income. It lives in secondary towns. It is not in Lewis’s model.

---

The development economists Paul Dorosh and James Thurlow ran an economy-wide model of Ethiopia that separated cities, towns, and rural areas, which sounds obvious, but almost nobody does it. Their 2013 finding was that the linkages between agricultural production and small towns are larger than those between agriculture and major cities. In their model, redirecting urban growth toward towns rather than cities produces broader-based growth and greater poverty reduction. Africa’s actual urbanization pattern, which concentrates investment in primate cities and export processing zones, weakens exactly the linkages that drive structural transformation.

The Dire Dawa Industrial Park is a case study in Dorosh and Thurlow’s problem. It was designed to attract foreign investment, generate exports, and employ labor. It does the first two. On the third, the automation and the Australian cotton have taken care of most of it. What it was never designed to do is build agricultural linkages, develop the surrounding small-town economy, or turn the Awash Valley’s cotton farmers into suppliers capable of meeting global fiber specifications. Those investments don’t fit on the Industrial Park’s org chart. They weren’t in the plan because they weren’t in the model.

---

Three thousand kilometers to the southwest, the Lobito Corridor is under construction.

The corridor connects the Congolese copper and cobalt belt through Zambia to the Angolan port of Lobito, rehabilitating a railway line built by the colonial economy for extraction and left to rust by the postcolonial economy. The United States, the European Union, and a coalition of development finance institutions have committed billions of dollars. The stated vision has expanded since the original mining-logistics framing: agriculture, regional integration, and industrial development along the route.

The vision is genuine. The planning is still catching up.

The towns along the Lobito corridor in eastern Angola, the Zambia Copperbelt periphery, and southern DRC sit in agricultural hinterlands with significant underexplored potential. Zambia’s Eastern Province grows maize, soybeans, and groundnuts. The DRC’s Katanga plateau has soil and rainfall that agricultural economists have noted for decades, with little follow-up. These towns have markets, traders, and small processors who constitute exactly the hidden middle Reardon describes. They have been operating without reliable transport connections, without cold storage, without the kind of infrastructure that allows agricultural surplus to become agricultural commerce.

The corridor will give them access to the railway. Whether it gives them anything else depends on decisions being made right now, while the concrete is still wet.

The risk is not that Lobito will fail. The risk is that it succeeds on its own terms: moves copper efficiently, generates export revenue, attracts mining investment, and builds almost no linkage to the agricultural economies it passes through. A Lewis corridor. Extraction up, inputs down, towns watching from the side of the road.

The alternative requires designing the towns as part of the investment, which corridor planning frameworks have never had to do because Lewis didn’t include the towns in the model. It means asking what the maize farmers in Zambia’s Eastern Province need to become reliable suppliers to processors in Ndola. It means cold storage at the railheads, not just track. Extension services timed to the opening of transport links. Credit instruments that work for smallholders who are three days from the nearest bank. Investment in the fiber specifications, to borrow the Dire Dawa manager’s frame, before the factory arrives and finds the local supply isn’t good enough.

None of this is technically difficult. It’s institutionally difficult because it requires coordinating agricultural ministry investments with infrastructure ministry timelines with private sector location decisions across three national governments with different planning cycles and different political economies. Lewis gave everyone a justification for not doing it: in his model, the modern sector handles itself, and the traditional sector exists to feed it labor. The towns in between don’t need a plan because they’re not in the diagram.

---

I think about the women at the Dire Dawa factory. Not with sentimentality, they have jobs, which is more than most, and the work is steady. But they are servicing a supply chain that begins in Australia and ends in Asia, and their city is twenty minutes away and almost entirely uninvolved. The Awash Valley cotton farmers are two hundred kilometers away and also uninvolved. The Industrial Park generates export statistics and a ribbon-cutting photograph, and very little else that connects to the economy surrounding it.

The city officials were proud of it. Of course they were. It is visible, measurable, and large. The investments that would actually develop Dire Dawa’s economy, the ones that would build agricultural linkages, upgrade cotton quality, develop small processors, thicken the town’s commercial life, are none of these things. They’re diffuse, slow, institutionally unglamorous, and they don’t appear in the model.

The Lobito corridor is a $10 billion opportunity to make the same choice, at scale, across three countries.

The cotton is still coming from Australia.