Development accounting is a simple tool with potentially powerful insights. It asks whether the measured differences in inputs between developing and developed countries can account for the gap in GDP per worker. Twenty years ago, Francesco Caselli famously argued that the best estimate was “no”. This column revisits that conclusion in light of two decades of intensive research. The authors show that inputs account for a much higher share than previously thought – between 55% and 70%, as opposed to 30% before. An expanded measure of human capital that incorporates education quality and work experience drives much of this result.
Policymakers in developing countries face a challenging budget situation. On the one hand, they have a limited ability to levy taxes. On the other hand, they have a long list of seemingly worthy policy proposals: upgrading the airport in the capital city, expanding secondary schools to underserved rural regions, or using pay to recruit better teachers, among many examples. Prioritising among these proposals is key. Which delivers the most development for the dollar?
Development accounting provides one way to attack this question. It does not investigate the deep roots of a nation’s poverty. Nor does it provide the cost-benefit calculus for a specific project. Rather, it asks an intermediate question: can low development be understood as a lack of the inputs to production? If so, which inputs matter most? To answer this question, development accounting measures cross-country differences in the inputs to production. It scales each input by an estimate of its importance for GDP. If the scaled differences cannot account for the differences in GDP per worker, then the rest is attributed to gaps in the productivity with which countries use inputs.
Twenty years ago, Caselli (2005) reviewed the status of the literature on development accounting. He reached a stark conclusion. Could inputs account for GDP per worker variation? His memorable answer was “no, way no”. His best estimate was that they accounted for perhaps one-third of the variation. The rest was left to productivity. Caselli did not view his conclusion as a success, but rather as a humbling reminder of how much the economics profession still needs to learn. This finding helped kick off a generation of research trying to understand why productivity might vary so much across countries and how poor countries could raise it (Prescott 1998, Restuccia and Rogerson 2008, Hsieh and Klenow 2009).
In a new paper (Lagakos and Schoellman 2026), we revisit this conclusion in light of two decades of additional work in the field. Updating to the most recent data, which covers a larger set of countries, changes little. Inputs account for only 30% of GDP per worker variation using the classic setup. However, improved and expanded measures of inputs have reshaped this conclusion.
A natural guess is that developing countries are unproductive because workers simply do not work much. New data allow us to measure the hours of labour rather than the number of workers per country (Bick et al. 2016, 2018, Gethin and Saez forthcoming). Workers in developing countries work longer hours – they are leisure poor and consumption poor. That makes it even harder to understand why they produce so little output.
There have been several advances in the measure of physical capital beyond the classic method of adding up past investment expenditures. Researchers have made new estimates of capital services in place of capital stocks and incorporated natural capital, which is the cropland and mineral deposits that raise output without requiring investment. These new measures do not change the main conclusion. Capital is vastly more abundant in developed than in developing countries. Yet capital–output ratios vary little between them. This finding suggests that rich countries invest a lot because they have high human capital or productivity, not the other way around. It points away from investment-oriented policy as a promising avenue.
The main revision to the classic conclusion comes from improved measurement of human capital – the value of the workforce’s skills and knowledge. The earliest measures focused on the average years the workforce had spent in school. Under this measure, human capital was judged to be moderately important, accounting for just 23% of GDP per worker differences.
The past two decades have incorporated additional investments into the measure of a nation’s human capital. Start with education quality – how much students learn in school, rather than how long they sit there. Students in many developing countries score more than two standard deviations below the OECD mean (Angrist et al. 2021a, 2021b). The test score distributions in developing and developed countries barely overlap (Gust et al. 2022). Adjusting for quality implies that human capital accounts for 36% of GDP per worker differences.
Investments in workers happen beyond schooling as well. Recent estimates show that workers roughly double their wages over their careers in the richest countries, but only increase them by 50 percent in developing countries (Lagakos et al. 2018b, Jedwab et al. 2023). Migrants display the same pattern: experience acquired in developed countries is more valued in the US labour market than experience acquired in developing countries (Lagakos et al. 2018a). Workers in developed countries are healthier, too. For example, people in developed countries are on average 15 cm taller than those in developing countries, reflecting better nutrition and health care and lower disease exposure (Weil 2007). Summing these components (both literally and figuratively) suggests that human capital accounts for 66% of cross-country income differences.
The experience of international migrants provides a check on this figure. A migrant takes her portable human capital to her new country, but swaps the workplace, infrastructure, and laws when she moves. If her wages rise by as much as the gap in GDP per worker, we would conclude that the country where she works is the main determinant of her productivity. Instead, research finds that migrants from the poorest countries to the US triple their wages, against a GDP per worker gap of a factor of 32 (Hendricks and Schoellman 2018). Moving countries explains one-third of the gap, leaving the other two-thirds to human capital – a strikingly similar conclusion to when we sum the components of human capital.
Put together, this research suggests that inputs can account for a much larger share of GDP per worker variation – 55% to 70%. The range is wide, but the conclusion is not. Inputs are more important for understanding GDP per worker differences than we previously thought. Less of the gap is left to unexplained differences in productivity.
Despite these advances, some important points remain unsettled. The first is the source of human capital differences. The interpretation of test score differences or wage growth over the career is nuanced. Recent research finds that parents and culture play a role in both (Singh 2020, De Philippis and Rossi 2020, Ek 2024). Even if skills are important, we do not yet understand what leads workers in some countries to accumulate so much more of them.
The second is the role of public capital. Pritchett (2000) argued that public investment in poor countries buys more bridges to nowhere, more school buildings started but never finished, and more funds siphoned off before they reach a project. The latest estimates of the effectiveness of government investment offer support for this idea (Devadas and Pennings 2018). Whether this matters depends on the role that public capital plays in output. This critical parameter is also one that we know little about. Public capital could account for anywhere from nearly 0% to 29% of GDP per worker differences.
Third, recent research thinks about the interaction between skills and technology, rather than treating them as isolated inputs. Developed countries have much higher shares of skilled workers than developing countries, but similar skill premiums. The explanation is that they must use skilled labour much more efficiently. Evidence from migrants and from trade attributes most of this to more skill-biased technology in developed countries, with a smaller role for higher-quality skilled workers (Rossi 2022a, 2022b, Malmberg 2025, Caselli and Ciccone 2018). The fact that developed countries have more skilled workers, higher-quality skilled workers, and more skill-biased technology is surely not an accident. Researchers are working to disentangle which forces drive these three critical components.
Development accounting cannot tell policymakers which project to finance. But it does offer guidance over whether inputs can account for low GDP per worker in developing countries – and if so, which ones do the job. The answer that it provides has changed over the last two decades. Inputs seem increasingly important, with human capital doing much of the work. Renewed attention to the low levels of skills in developing countries seems warranted.
Source : VOXeu
