Why Do Some Countries Produce So Much More Output Per Worker Than Others?
📄 Summarized from the full manuscript · Human-reviewed for faithfulness before publication
In brief
In 1988 a US worker produced in ten days what an average worker in Niger produced all year. This 1999 paper traces that 35-fold gap: some of it comes from more machinery and more schooling, but most is an unexplained "productivity" residual. Across 127 countries, the authors show that gap tracks a measure of "social infrastructure" -- whether a country's institutions let people keep the fruits of their own work rather than losing them to corruption, expropriation, or crime -- and use historical factors like distance from the equator and colonial language ties to argue the relationship is causal, not just correlation.
What this paper finds — and why it matters
Hall and Jones ask why output per worker differs so enormously across countries – more than 35-fold between the United States and Niger in 1988 – and decompose the answer into physical capital intensity, educational attainment (human capital), and a residual productivity term, finding capital intensity accounts for a factor of about 1.5 and human capital a factor of about 3.1 of the U.S.-Niger gap, leaving a factor of about 7.7 as the unexplained productivity residual. Their central contribution is to push one level deeper: across 127 countries, they find a powerful, close empirical association between output per worker and a measure of “social infrastructure” – the government policies and institutions that determine whether individuals capture the social returns to their productive effort as private returns, or instead see that output diverted through expropriation, corruption, confiscatory taxation, or private predation. Because social infrastructure might itself be a consequence rather than a cause of high output (richer countries could simply afford better institutions), the authors instrument for social infrastructure using measures plausibly tied to historical Western European influence – distance from the equator, the Frankel-Romer predicted trade share, and the shares of the population speaking English or a Western European language – and still find a strong, precisely estimated effect: their main instrumental-variables specification implies that a difference of 0.01 in their social-infrastructure index is associated with a 5.14% difference in output per worker, a result that survives tests of the instruments’ validity and is robust to alternative instrument sets and to using only the 79 countries with complete data. The paper’s headline finding – echoed in its summary schema “Output per Worker = f(Inputs, Productivity) = f(Social Infrastructure)” – is that long-run differences in capital accumulation, education, and productivity, and hence in living standards, are driven in large part by a common, more fundamental cause: the institutional and policy environment within which people invest, create, and produce.
Summary of a classic paper, AI-assisted and human-reviewed. See the linked original for the authoritative claims and full conditions.
Questions & answers
Q1. How large is the output-per-worker gap the paper sets out to explain, and how much of it does standard growth accounting attribute to capital and education?
The paper’s illustrative example is that 1988 output per worker in the United States was more than 35 times higher than in Niger – an average U.S. worker produced in about ten days what an average Niger worker produced in a year. Using an aggregate-production-function decomposition in the tradition of Mankiw, Romer, and Weil (1992), the authors attribute a factor of about 1.5 of this gap to differences in physical capital intensity and a factor of about 3.1 to differences in educational attainment, leaving a factor of about 7.7 – the largest single piece – as an unexplained total-factor-productivity residual.
Q2. Why do the authors focus on cross-country differences in levels of output per worker rather than growth rates?
The authors argue that levels, not growth rates, are what is most directly relevant to welfare (since welfare depends on the level of goods and services consumed), and cite evidence (Easterly, Kremer, Pritchett, and Summers 1993) that cross-country growth rates are weakly correlated across decades, suggesting growth-rate differences are often transitory, whereas models of cross-country technology diffusion (e.g., Parente and Prescott 1994; Barro and Sala-i-Martin 1995; Eaton and Kortum 1995) imply all countries eventually grow at a common long-run rate, making persistent differences in levels the economically interesting object to explain.
Q3. What does the paper mean by “social infrastructure,” and what two components does its empirical index combine?
“Social infrastructure” is defined as the institutions and government policies that determine the economic environment within which individuals accumulate skills and firms accumulate capital and produce output – in the authors’ framing, infrastructure that “gets the prices right” so that individuals capture the social returns to their actions as private returns, protecting the output of productive units from diversion such as thievery, expropriation, confiscatory taxation, and corruption. The paper’s empirical social-infrastructure index combines a measure of government antidiversion policies (a “Government Antidiversion Policies,” GADP, index) with a measure of the economy’s openness to international trade, treating government as both the potentially most efficient provider of protection against diversion and, in practice, frequently a primary agent of diversion itself.
Q4. Why do the authors worry that a simple cross-country correlation between social infrastructure and output per worker might not reflect causation, and how do they address this?
The authors are explicit that causation could run in both directions, or even predominantly from output to infrastructure – for example, poor countries may simply lack the resources to build effective institutions – so a raw correlation would not establish that better social infrastructure causes higher output. To address this, they instrument for social infrastructure using variables plausibly correlated with social infrastructure for historical reasons but not directly caused by a country’s current output: distance from the equator (as a proxy for the historical reach of Western European influence and colonization patterns), the Frankel-Romer measure of predicted trade share (based on geographic determinants of trade), and the fractions of the population speaking English or another Western European language at birth.
Q5. What is the paper’s main instrumental-variables result, and how precisely is it estimated?
The main specification, using all four instruments (distance from the equator, Frankel-Romer trade share, English-speaking share, Western-European-language share), finds that a 0.01 difference in the social-infrastructure index is associated with a 5.14 percent difference in output per worker, with a standard error of 0.508 – a precisely estimated effect. Overidentification tests do not reject the model’s exclusion restrictions (e.g., the orthogonality of distance from the equator with the error term), and a test of equal coefficients on the two components making up the social-infrastructure index (the GADP policy index and the openness measure) is also not rejected.
Q6. Is the main result sensitive to which instruments or which sample of countries is used?
No – the paper reports the result is robust to using a more limited instrument set (distance from the equator and the Frankel-Romer instrument alone), to restricting the sample to the 79 countries with no imputed data, and the authors report (without presenting full results) that dropping any single instrument at a time does not overturn the finding, so no one instrument is driving the result. An OLS specification (without instrumenting) gives a smaller coefficient (3.289) and rejects the coefficient-equality test, consistent with the endogeneity concern the instrumental-variables strategy is designed to address.
Q7. What are the paper’s headline conclusions, in the authors’ own summary?
The paper summarizes its findings as: (1) many predictions of growth theory can be fruitfully examined using cross-sectional levels of income rather than only growth rates; (2) differences in physical capital and educational attainment explain only part of the enormous cross-country variation in output per worker, leaving a large productivity residual; (3) differences in social infrastructure across countries cause large differences in capital accumulation, educational attainment, and productivity, and therefore in income; and (4) the extent to which countries have adopted favorable social infrastructure is partly explained by historical Western European influence, and the finding that social infrastructure causes income differences is robust to measurement-error and endogeneity concerns via the instrumental-variables strategy.
Key terms in this paper
Definitions below follow the paper's own usage.
- Social infrastructure
- the institutions and government policies that determine the economic environment within which individuals accumulate skills and firms accumulate capital and produce output, in particular the extent to which that environment protects the output of productive activity from diversion (private predation such as thievery, or public diversion such as expropriation, confiscatory taxation, and corruption) rather than itself being a vehicle for diversion; the paper's central explanatory variable, measured by an index combining government antidiversion policies and openness to trade.
- Diversion
- activity, private (thievery, squatting, organized-crime protection rackets) or public (expropriation, confiscatory taxation, corruption, predatory regulation), that redirects the output of a productive unit away from the people who produced it; the paper frames a favorable social infrastructure as one that protects against diversion, noting the paradox that government is both the potentially most efficient provider of such protection and, in practice, often a principal source of diversion itself.
- Instrumental variables for social infrastructure
- the paper's strategy for addressing the possibility that output per worker causes (rather than merely correlates with) social infrastructure, using distance from the equator, the Frankel-Romer predicted-trade-share measure, and the shares of the population speaking English or a Western European language -- variables chosen for their plausible historical connection to the spread of Western European institutions and their claimed lack of direct causal connection to current output per worker.
- Levels accounting versus growth accounting
- the paper's methodological choice to explain cross-country differences in the *level* of output per worker (via an aggregate production function decomposed into capital, human capital, and a productivity residual, then related to social infrastructure) rather than differences in growth *rates*, motivated by the argument that growth-rate differences across countries are often transitory while level differences are what persists and matters most directly for welfare.