An econometric study covering 63 countries between 2005 and 2021 establishes that a 1% increase in new robot installations per 10,000 workers is associated with a 0.037% to 0.039% decline in unemployment. The study estimates an average association between robot installations and unemployment, without identifying an institutional mechanism.
The essentials
- The study reports an average association between more new installations and lower unemployment across its full sample (ScienceDirect, panel 63 countries, 2005-2021).
- A 1% increase in new robot installations per 10,000 workers is associated with a 0.037% to 0.039% decline in unemployment in the full sample.
- This institutional conclusion does not follow from the published estimates.
- The study provides no forecast timeline for emerging countries.
- Two signals would allow monitoring of this tipping point: the evolution of the robot stock per 10,000 workers and the sectoral unemployment rate in manufacturing industries.
The robot that creates jobs in Indonesia destroys them in Germany
When the International Federation of Robotics publishes its annual figures, it measures stocks. It does not measure effects. This is a 2024 article on a panel of 63 countries covering 2005-2021, with analyses by sex, age, and education level.
The published result concerns a 1% increase in new robot installations per 10,000 workers across all 63 countries. The study does not document an inversion of the effect within an OECD group. The study estimates a statistical association over the observed period, without establishing that it is stable year after year.
The usual reading of the technology-employment debate assumes a uniform effect. Robots destroy jobs, or they create them, depending on which school of thought you belong to. The study analyzes heterogeneity by sex, age, and education level, without estimating a density threshold. The article provides no nonlinear threshold for inversion of the employment effect.
This law of diminishing returns to automation is not a theoretical surprise. It was foreseeable. Its empirical demonstration on a global panel is too.
The role of institutions
In his Nobel lecture published in 2025, Daron Acemoglu shows why identical technologies produce different prosperity trajectories across countries. Inclusive institutions orient technical progress toward uses that broaden economic participation; extractive institutions channel it toward capital rents.
The 2024 study applies this framework to automation across a panel of 63 countries covering 2005-2021. In Indonesia, Vietnam, or Mexico, robots arrive in economies where industrialization can contribute to rural occupational mobility, even though manufacturing is not uniformly the principal current vector of access to the middle class. At this stage, automation can increase productivity and sometimes employment, but it can simultaneously reduce the share of labor in added value and accentuate wage gaps, without gains necessarily distributing equally among workers. Labor market institutions, however imperfect, play their default redistributive role.
The studies consulted do not permit concluding that the mechanism is exhausted in OECD countries. In Germany, robots initially created skilled jobs by restructuring production toward the high end. But this complementarity has limits: beyond a certain stock of automated capital, productivity gains accumulate on the capital side rather than the labor side, unless specific institutions—strong collective bargaining, redistributive taxation, financed continuous training—perform active redistribution.
The study does not establish that the effect of technology on employment is conditioned by political choices.
There exists a competing reading, notably carried by Schumpeterian growth economists like Philippe Aghion: creative destruction is a condition of progress, and markets eventually adjust if given time and means. From this perspective, the polarization observed in OECD countries is temporary: new sectors emerge, absorb displaced workers, and technological unemployment remains structurally limited. The WEF, in its Future of Jobs 2025 report, projects 92 million jobs displaced by 2030 and 170 million new jobs by 2030; estimates of 85 million displacements and 97 million creations by 2025 came from the 2020 report.
The problem is that this logic assumes an institutional adjustment capacity that emerging countries have not yet built. The Schumpeterian transition takes time. It requires training systems, safety nets, human capital markets that redirect workers. This threshold dynamics and institutional adjustment is not documented. The panel does not allow comparison of the width of an adoption window with prior anticipations.
The ten years that matter
This dynamic has measurable temporality. The 2019 OECD report on automation determinants already documented disparities in penetration between economies. The IFR observes long-term growth in Asia, while installations in the Americas and Brazil evolve more cyclically and heterogeneously.
Over the full sample and the 2005-2021 period, the estimated association between new installations and unemployment is negative. No critical level of robot density is provided in the article.
Adoption rates and political choices influence automation trajectories in emerging countries. That is precisely where the issue lies. The gap between declared adoption and actual use of automation already observed in European SMEs suggests the transition is slower in fact than in installation statistics. Emerging countries may benefit from additional delay.
But that delay is not guaranteed.
What is guaranteed, however, is the asymmetry of risks. OECD countries traversed the destructive phase with social protection systems, high per capita incomes, and labor markets capable of absorbing part of the shocks. Emerging countries have none of these shock absorbers. A rapid transition to automation without institutional preparation risks producing significant social costs.
Two trajectories, a single starting point
The 2024 panel fits into an analysis of possible trajectories for automation in emerging countries.
In the first, low robot penetration economies maintain an institutional architecture that orients automation toward sectors where the machine complements human labor rather than replacing it. In this case, they traverse the favorable phase documented by the study, improve overall productivity, expand their middle class, and gradually build the redistributive institutions necessary to cushion the next phase. This scenario assumes industrial policy capable of evaluating the sectoral effect of each wave of automation.
In the second, automation advances according to scenarios of origin and complementarity of robots, under the effect of international competitive pressure and the macroeconomic effects that result, while local institutions face constraints likely to make adaptation difficult, despite potentially longer delay than advanced economies had. Displaced workers may encounter high retraining barriers: training systems remain limited, social protection nets are often less comprehensive and generous, and new sectors may emerge without necessarily absorbing all displaced workers. Polarization may also affect certain emerging economies, but its causes, timing, and intensity do not necessarily reproduce those of wealthy countries.
This second scenario deserves to be named clearly. It is not inevitable. But dependency on global value chains that emerging countries already experience in other sectors shows that external pressure can catch local institutional capacities off guard. Automation follows the same logic.
What would make the tipping point visible
The concrete track is not to slow automation in emerging countries. Slowing technical progress in economies that need productivity gains to escape poverty would be counterproductive. The track is to have tools to anticipate the tipping point before it produces irreversible effects.
Two signals are identifiable. The first is the evolution of the robot stock per 10,000 workers in the major emerging economies. This signal exists: the International Federation of Robotics publishes this data annually. What is missing is a sectoral interpretation framework that permits early warning.
The second signal is the unemployment rate in manufacturing industries of non-OECD countries with strong robot growth. If this rate begins to increase while the rest of the economy remains stable, the tipping point may be near. Measuring this signal annually, by country and by sector, would allow anticipating and adapting policies before inversion becomes irreversible.
Regional observatories of robot penetration in non-OECD countries, calibrated on these two indicators, would fulfill a function that current international institutions do not ensure. The WEF publishes global projections. The OECD tracks its members. The International Federation of Robotics aggregates stocks. No one produces the early warning by emerging country that would allow concerned governments to act before the tipping point.
Conditioning incentives for robot investment in these countries on criteria for net effect on sectoral employment, evaluated annually, would be a next step. This idea is difficult to implement: it assumes statistical capacity and political will that many of these governments do not yet have. But it is consistent with what the panel data establish: a measurable association between robot installations and unemployment in the observed sample.
The policy decision that data make possible
The study improves the quality of the argument available for deciding. Until now, the debate over technology and employment in emerging countries was divided between technological enthusiasm, which promised jobs without condition, and catastrophism, which rejected automation as a universal threat. Neither of these positions permitted action in an operational manner.
The panel of 63 countries provides an aggregate estimate over 2005-2021, with breakdowns by sex, age, and education level. The study estimates a statistical association in its sample, without establishing general predictability.
Governments of emerging countries must build the necessary institutions to cushion a rapid transition to automation. Acemoglu has posed this question at the macro scale for years. The study brings comparable data at the global economy scale.
That is this research’s contribution: to document heterogeneity in associations with unemployment according to worker characteristics, and to enable debate over measurement tools and adaptation policies.
Sources
- ScienceDirect – Econometric study on robot density and unemployment, panel 63 countries, 2005-2021 (2026)
- Daron Acemoglu – Nobel Lecture: Institutions, Technology, and Prosperity, American Economic Review (2025)
- World Economic Forum – Future of Jobs Report 2025 (weforum.org)
- OECD – Determinants and Impact of Automation, 2019
- International Federation of Robotics (IFR) – World Robotics Report, annual robot stock data
