1 The Causal Question
The Tennessee Valley Authority (TVA) is one of the largest and most ambitious regional development programmes in US history. Created by Congress in 1933 as part of the New Deal, it built a network of dams and hydroelectric plants in the Tennessee River Basin, providing cheap electricity, flood control, and improved navigation to a region that was among the poorest in the United States. Federal investment ran to billions of 1933 dollars over two decades.
Did it work? More precisely: did the TVA cause lasting improvements in the economies of the counties it served, or were the observed improvements simply the continuation of pre-existing trends, or driven by the national recovery from the Great Depression?
Kline and Moretti [2014] used a difference-in-differences design to identify the causal effects of the TVA on local labour markets, comparing TVA counties to a matched comparison group of counties in the US South that did not receive TVA investment. Their paper is the most rigorous study of the TVA's economic effects and has become a landmark in the literature on place-based policies.
2 Historical Background
The TVA region—parts of Tennessee, Alabama, Mississippi, Georgia, North Carolina, Virginia, and Kentucky—was characterised in 1933 by subsistence farming, limited electrification, and high poverty rates. Farm incomes were less than half of the national average. The Tennessee River flooded regularly, disrupting agriculture and making navigation difficult.
The TVA's mandate was deliberately broad: it was not merely a hydroelectric utility but an institution for comprehensive regional development, providing technical assistance to farms, building roads, and developing recreational facilities in addition to its core infrastructure investments. By 1945, the TVA had built 16 dams and was the largest electricity producer in the United States.
The political economy of the TVA's geographic scope provides the variation that Kline and Moretti [2014] exploit: the TVA boundary was determined by the drainage basin of the Tennessee River, a geographic feature largely predetermined before the programme's economic effects began. Counties within the drainage basin received TVA investment; adjacent counties with similar pre-existing characteristics did not.
3 The Identification Strategy
3.1 Difference-in-Differences Design
Kline and Moretti [2014] implement a DiD design comparing TVA counties to control counties in the US South over the period 1940-1980. Their estimating equation is:
where Yct is the outcome in county c in Census year t, αc is a county fixed effect, δt is a year fixed effect, TVAc is an indicator for TVA counties, and Postt indicates years after TVA investment. The coefficient β identifies the average treatment effect on TVA counties relative to control counties, under the assumption that in the absence of TVA investment, outcomes in the two groups would have evolved in parallel.
3.2 Constructing the Comparison Group
The choice of comparison group is critical. The entire US South is not a valid control—Southern counties differ from TVA counties in historical, geographic, and economic dimensions that could affect their post-1933 trajectories independently of TVA investment. Kline and Moretti [2014] restrict the comparison group to counties in the Tennessee River's "neighbouring drainage basins"—counties in the region of Appalachia and the upper South that share the geographic, demographic, and economic characteristics of TVA counties but lie just outside the TVA's service area. This comparison group was explicitly considered as an alternative TVA service area during legislative debates, making its exclusion from the TVA essentially arbitrary.
They further validate the comparison group by testing pre-treatment parallel trends: in Census years before 1940, TVA and control counties show statistically indistinguishable trends in manufacturing wages, agricultural productivity, and population, providing credibility for the identifying assumption.
3.3 Heterogeneity and Spatial Spillovers
A methodological challenge is that the TVA may have generated agglomeration spillovers into neighbouring counties, biasing the DiD if some control counties received positive spillovers. Kline and Moretti test for such spillovers by estimating effects for control counties at varying distances from the TVA boundary. They find little evidence of positive spillovers in the immediate vicinity, suggesting that spillovers do not substantially bias the comparison group.
4 Key Findings
4.1 Manufacturing
The TVA had large positive effects on manufacturing employment and wages in the treatment region. By 1960, manufacturing employment per capita in TVA counties was approximately 30 percent higher than in comparable control counties, and manufacturing wages were approximately 40 percent higher. These are extraordinarily large effects, reflecting both the direct employment generated by TVA-related industries (particularly aluminium smelting, which is extremely energy-intensive) and the agglomeration effects from concentrating manufacturing activity.
4.2 Agriculture
The TVA's effects on agriculture are more nuanced. Agricultural productivity (output per farm worker) increased substantially, consistent with the TVA's investments in soil conservation, fertiliser programmes, and extension services. But agricultural employment declined, as the combination of higher wages in manufacturing and mechanisation of farming drew workers out of agriculture. The net effect on agricultural wages is positive: farm workers who remained in agriculture earned more. But the TVA accelerated the structural transformation away from agriculture, which had ambiguous welfare effects for workers with limited mobility.
4.3 Persistence and Decay
An important finding is that the manufacturing employment gains were substantial through 1960 but had decayed considerably by 1980. This pattern is consistent with a "big push" interpretation: the TVA created agglomeration economies that persisted for several decades but eventually dissipated as wages in the TVA region converged toward those in the rest of the South. The persistence of agricultural productivity gains through 1980, while manufacturing gains faded, suggests that the TVA's human capital and agricultural extension investments may have been more durable than the manufacturing agglomeration it created.
5 Benefit-Cost Analysis
Kline and Moretti [2014] use their estimates to conduct a partial benefit-cost analysis. The welfare gains from higher wages for workers in TVA counties must be weighed against (1) the direct cost of TVA investment, and (2) the potential displacement of economic activity from other US regions.
If the manufacturing gains in TVA counties came partly at the expense of other US regions—if agglomeration is a zero-sum game of attracting firms from elsewhere—then the social return to TVA investment could be much lower than the private returns to TVA counties. The paper uses a structural model of local labour markets to calibrate this displacement effect, finding that the national social return to TVA investment was positive but substantially lower than the local return to TVA counties. This distinction between the local return to a place-based policy and the national social return is central to the modern policy debate about enterprise zones, opportunity zones, and regional development subsidies. Policies that visibly "work" for specific places may not increase aggregate welfare if they merely relocate economic activity [Neumark and Simpson, 2011].
6 Limitations
General Equilibrium. The DiD design captures the local effect of TVA investment but cannot measure the displacement of economic activity from other US regions. The benefit-cost calculations require structural assumptions about the extent of displacement.
Mechanisms. The paper documents effects on employment and wages but cannot cleanly separate the channels: cheap electricity, improved transportation, flood control, and agricultural extension all occurred simultaneously. Disentangling which component drove the manufacturing effects would require additional variation.
External Validity. The TVA was an unusually large and sustained investment in a very specific historical context—the Great Depression, when factor markets were severely depressed and the counterfactual without TVA investment was grim. Modern place-based policies operate in different economic environments, limiting the direct applicability of the TVA evidence to contemporary policy debates.
7 What We Learn
Three lessons emerge. First, sufficiently large and sustained public investment can generate lasting agglomeration economies in lagging regions—the TVA's manufacturing effects persisted for three decades. Second, the distinction between private and social returns to place-based policy matters enormously for welfare analysis; a policy that is locally beneficial may be nationally neutral or negative. Third, rigorous quasi-experimental evaluation of historical programmes can recover causal effects that OLS would confound, but only when the identifying variation is credibly exogenous—here, the pre-determined geography of the Tennessee River drainage basin.
References
- Kline, P. and Moretti, E. (2014). Local economic development, agglomeration economies, and the big push: 100 years of evidence from the Tennessee Valley Authority. Quarterly Journal of Economics, 129(1):275-331.
- Neumark, D. and Simpson, H. (2011). Place-based policies. NBER Working Paper, w18964.
- Callaway, B. and Sant'Anna, P. H. (2021). Difference-in-differences with multiple time periods. Journal of Econometrics, 225(2):200-230.
- Rambachan, A. and Roth, J. (2023). A more credible approach to parallel trends. Review of Economic Studies, 90(5):2555-2591.