Recent technological advancements, particularly in artificial intelligence and robotics, have reignited debates about technological unemployment. The rise of technologies capable of performing both routine and non-routine tasks raises concerns about job displacement across various sectors, impacting both the quantity and quality of jobs available. This concern is further fueled by the observation that economic growth may decouple from employment growth, potentially leading to a widening of income inequality. While existing research has documented the displacement effects of automation technologies, measured primarily by robot adoption, the spatial heterogeneity of these effects, especially the contrast between urban and non-urban settings, remains largely unexplored. This paper addresses this gap by examining the impact of automation on wage inequalities in Italian cities, leveraging the country's diverse spatial settings and varying levels of technological intensity across different regions. The authors hypothesize that while automation leads to job displacement in both urban and non-urban areas, the presence of diverse sectors and a highly skilled workforce in cities might lead to a reorientation of occupations towards higher-skilled and better-paid roles. However, this could also exacerbate inequality by creating an 'elite' job market that excludes low-skilled workers.

The paper reviews existing literature on the impact of automation on employment and income distribution. Studies have shown a negative correlation between robot adoption and wage share, with varying nuances across countries and worker groups. However, most studies lack a detailed spatial analysis comparing urban and non-urban settings. Some studies have explored the role of ICT and robot adoption within urban environments, but a comprehensive comparison across different settlement types is lacking. The authors highlight the need to understand whether the negative effects of automation on low-skilled workers are universal or vary based on the labor market structure and skill supply in different areas.

The empirical analysis utilizes data from the Italian Labour Force Survey (RFL) for the period 2009-2019, aggregated at the NUTS3 level. The dependent variables are the employment-to-population ratio and the employment share by skill level (low, mid, and high). Robot adoption data is obtained from the International Federation of Robotics (IFR) and apportioned to the NUTS3 level using weights based on the NUTS3 employment share in manufacturing, blue-collar occupations, and households with broadband. The concept of 'city' is operationalized using different population size thresholds (top 25%, 20%, and 10% most populous NUTS3 regions). Two econometric models are employed. The first examines the impact of robot density and its growth rate on the employment-to-population ratio, controlling for regional economic structure, demographics, and labor force composition. This model includes interaction terms between automation variables and city dummies to compare effects across urban and non-urban areas. The second model analyzes the impact of automation on the employment share by skill level, using a pooled panel setting with controls similar to the first model. Both models use lagged independent variables to mitigate potential endogeneity issues. Random effects estimation is used in the first model, while pooled panel estimation is used for the second, with year and NUTS2 fixed effects and robust standard errors in both cases.

The analysis confirms the labor-saving nature of automation technologies, with robot density and its growth negatively affecting labor market participation. This effect is consistent across both urban and non-urban areas. However, urban areas exhibit higher labor market participation overall, reflecting agglomeration economies. The interaction terms between automation variables and city dummies are not statistically significant, indicating a similar magnitude of negative effects on employment participation in both urban and non-urban contexts. Regarding skill-level employment, the results show that automation consistently displaces low-skilled workers in both urban and non-urban areas. Mid-skilled employment remains largely unaffected. Interestingly, high-skilled employment expands in urban areas in response to automation but shows no significant change in non-urban areas. This suggests that in urban areas the displacement of low-skilled workers is not fully compensated by the creation of similar jobs, resulting in a contraction of the labor force.

The findings highlight that automation displaces low-skilled workers across all territorial settings, contradicting some previous studies that found positive effects at more disaggregated spatial scales. The authors suggest that negative effects at the aggregate (NUTS3) level outweigh any positive effects for individual firms. The differing occupational mixes in urban and non-urban areas contribute to different adjustment paths to automation. In non-urban areas, low-skilled worker displacement is not compensated by reinstatement, leading to labor market exit. In urban areas, high-skilled employment expands, suggesting a potential widening of income inequality due to the creation of an 'elite' job market. This finding underscores the role of cities as potential multipliers of income inequality resulting from automation.

The study provides strong evidence of the labor-displacing effects of automation across urban and non-urban areas in Italy. The findings reveal that while low-skilled workers are affected equally in both settings, the urban context exhibits a compositional shift towards higher-skilled jobs, potentially exacerbating inequality. Future research could explore the impact of technologies performing cognitive and non-routine tasks and the spatial spillover effects of automation-induced displacement on neighboring labor markets.

The study uses data aggregated at the NUTS3 level, which may mask variations within regions. The operationalization of 'city' relies on population thresholds, which might not fully capture the multifaceted nature of urban areas. Potential endogeneity issues, although mitigated by lagging independent variables, cannot be entirely ruled out. The study focuses primarily on the impact of robotization and might not fully capture the effects of other automation technologies.