Skip to main content
Top
Published in:

Open Access 02-11-2024 | Special Issue Paper

Unveiling the automation—wage inequality nexus within and across regions

Authors: Roberta Capello, Simona Ciappei, Camilla Lenzi

Published in: The Annals of Regional Science | Issue 4/2024

Activate our intelligent search to find suitable subject content or patents.

search-config
loading …

Abstract

Since the1800s, automation technologies have been interpreted as a source of displacement effects, largely conceptualised and empirically proved in a vast literature. This paper claims that, despite their non-manufacturing nature, metropolitan regions are not exempted by the negative effects of automation on wage inequalities across workers’ groups. The paper empirically proves this statement by analysing the effects on jobs and wage differentials among groups of workers associated with the diffusion of robot technologies in Italian NUTS3 regions in the period 2012–2019. Results show that automation technologies in the form of robotisation do displace jobs, harming particularly low-skilled workers in non-metropolitan manufacturing regions, where inter-group wage inequalities increase. However, through the creation of high-skilled jobs, also cities experience a rise of inter-group workers inequalities. These results call for appropriate policies to cope with the changing occupational profiles requested by the labour market.
Notes

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

1 Introduction

The rapid and impressive technological advances of the last decade in multiple fields ranging from artificial intelligence to robotics have brought to fore once more the debate on the relationship between technology and society in general and the harmful consequences of technology for jobs in particular (Brynjolfsson and McAfee 2017; Frey and Osborne 2017; Economist 2023; Lancaster 2023). A recurrent conclusion made by many scholars and commentators is that the new technological landscape has created unfavourable conditions for balanced growth and socio-spatial resilience, amplifying a chief paradox of the present time. The co-occurrence of powerful technology (and accelerating technological change in the view of many) with stagnating median wages and decoupling of economic and employment growth (McAfee and Brynjolfsson 2017) have generated a generalised sentiment of techno-pessimism if not of automation anxiety (Autor 2015; Akst 2013) and of a diffused discontent (McCann 2020; Rodríguez-Pose 2018; Caselli et al. 2021a, b).
There is rich evidence of the rise of inter-personal inequalities in the past 40 years, showing that a small percentage of individuals and communities did enjoy improved economic prosperity, while the vast majority did not benefit from the rise of aggregate wealth (Feldman et al. 2021). This unbalance has been particularly well documented in the case of the USA starting from the work of Piketty and Saez (2003) and confirmed also for the last years (Kemeny et al. 2022; Alvaredo et al. 2018; Chancel et al. 2022). Importantly, a novel spatial dimension has characterised the rise of income and spatial inequalities, marking a clear divide between a small group of big, wealthy and high-income (elite) metropolitan regions and the remaining ones (Kemeny and Storper 2020). Similar conclusions have been drawn for the UK case; in particular, the disproportionate concentration of high-skilled workers was found as the main source of regional labour market disparities, measured in terms of wage differences (Overman and Xu 2022). More generally, these income and regional disparities are evident all over OECD countries, where remote regions and those far from cities continue to lag behind metropolitan regions in terms of GDP per capita levels and growth’ (OECD 2022, p. 17).
The implications of technological transformations deepening inequalities among workers (and social) groups and across different territorial contexts (particularly metropolitan vs non-metropolitan ones) remain therefore a compelling topic to be analysed, with important consequences for policy actions (Autor et al. 2022). In fact, with the expansion of the range of tasks performed by technologies, the deterioration of labour markets conditions can aggravate further, in terms of overall job losses  (i.e. the number of jobs being cut is likely to be greater than that of those being created) but primarily of a widening of inequalities across workers' groups, as a consequence of the selective and unbalanced displacement, reinstatement and productivity effects of the new technologies on different occupational categories (Acemoglu and Restrepo 2019, 2022; Autor et al. 2022).1
Existing literature has extensively investigated the consequences of enhanced robotisation for the compression of the labour share, stressing the labour-saving nature of these technologies, especially in the case of the most routine and manual jobs in manufacturing sectors (Autor 2022; Acemoglu and Restrepo 2020, 2022; Dauth et al. 2021). However, the consequences in the labour market of the increasing automation of the manufacturing environment could be more complex than what documented so far (Acemoglu and Restrepo 2022). In fact, the effects of robots’ adoption can go beyond the manufacturing sector, through two main channels. First, the operation of composition effects can re-allocate low-skilled workers from the manufacturing to the service sector due to their displacement in the former with relatively limited consequences on inequalities (Acemoglu and Restrepo 2022; Capello et al. 2022a). Second, automation of the most routine and manual jobs can come together with the creation (i.e. reinstatement) of new complementary high-skilled jobs in the services sector, as documented in the literature on manufacturing servitisation (Dauth et al 2021; De Propris and Storai 2019), leading to an increase in the wage bill share of high-skilled workers and a worsening of inter-group wage inequalities. The complementary effects between service and manufacturing sectors for low-skilled workers are expected to take place primarily in non-metropolitan regions, whose economies are mainly manufacturing production-based ones, with a net final effect of job displacement induced by automation. The service-based economy nature of metropolitan regions, instead, would open to reinstatement effects of automation in favour of high-skilled workers, making also cities unexpectedly exposed to the negative distributive outcomes generated by automation (Capello and Lenzi 2023).2 This situation may be exacerbated by the increase in low-skilled, low-paid workers displaced in non-metropolitan regions that search for a job in low-skilled service activities in nearby metropolitan regions.
Quite disappointingly, these aspects have been relatively neglected so far in the literature. In particular, the analysis of the labour market and inequality outcomes deriving from the adoption of the new automation technologies in different territorial contexts (metropolitan vs non-metropolitan) has remained largely unexplored, even if some evidence is emerging very recently (Capello and Lenzi 2023).3
The present paper aims at filling these gaps by presenting a conceptualisation and prima facie evidence on the displacement, reinstatement, and productivity effects of automation in metropolitan (cities) vs non-metropolitan region, so to highlight the role of space in the automation-inequality nexus.
More specifically on the conceptual grounds, by analysing the impact of robots’ adoption, both in terms of employment and wage inequalities across sectors, occupational groups and types of regions, this paper aims not only at confirming automation displacement effects, especially for low-skilled manufacturing workers (as largely emphasised in the literature), but primarily at emphasising its reinstatement and productivity effects, and their spatial dimension, i.e. whether these effects differ according to the territorial context where robot adoption takes place (service-based metropolitan regions vs manufacturing-based non-metropolitan ones) and are subject to spatial spillover mechanisms. In particular, metropolitan regions can experience a widening of wage differentials by becoming the locus of high- and low-skilled job reinstatement and productivity effects, thus amplifying wage differentials within cities. Moreover, the specialisation in manufacturing of the non-metropolitan regions and the specialisation in services of the metropolitan ones generate expectations on different effects of automation on wage inequalities between cities.
On the empirical grounds, the paper examines the inequality effects of automation, measured in terms of robot penetration rate as common practice in the literature, in the context of Italian cities (i.e. NUTS3 regions). Italy represents an ideal empirical case for multiple reasons. First, Italy is the second largest robot market in Europe after Germany (International Federation of Robotics – IFR 2021) and the second manufacturing country in Europe in terms of employment and value added (EUROSTAT 2021). Moreover, in all parts of the country from North to South, Italy presents metropolitan areas as much as industrial and rural ones, making the Italian territory highly heterogeneous. Importantly, technological intensity and propensity to introduce new technologies are marked by an enduring divide between macro-areas (i.e. North, Centre and South). Finally, even if cities are predominantly service-led economies (Baumol 1967; Camagni et al. 2023), some of the Italian ones still present a strong industrial vocation and tradition, supporting further the selection of Italy as testbed for the analysis conducted in this paper.
The rest of the paper is organised as follows. The next section sets out the conceptual framework and the hypotheses to be tested empirically. The measurement and empirical approach are presented in Sect. 3, while the results are discussed in Sect. 4. The last section concludes the paper with some final policy considerations.

2 Automation and inter-group workers inequalities: testable hypotheses

The impact of automation technologies on labour markets has been subject to an increasing number of studies in the last decade, highlighting a general displacement effect, particularly harmful for low-skilled, routine and manual workers in manufacturing sectors, and a widespread sentiment of techno-pessimism (Akst 2013; Frey and Osborne 2017; Autor 2022; Acemoglu and Restrepo 2022).4
However, empirical results present some heterogeneity largely related to the various combination between the reinstatement, displacement, productivity, and composition effects in different types of firms, sectors, periods of time and territorial contexts (Dauth et al. 2021; Caselli et al. 2021a, b; Dottori 2021; Nannelli et al., 2023). The framework set out by Acemoglu and Restrepo offers a useful starting point to reconcile these apparently scattered empirical findings and to conceptualise the heterogeneous effects of automation technologies across occupations, sectors, and territorial contexts as much as their implications for the widening of inequalities across workers groups and places (Acemoglu and Restrepo 2022).
Since automation technologies hit predominantly manufacturing production activities, in non-metropolitan regions the following effects can be expected for low-skilled manufacturing workers, whose manual and routine tasks are easily codifiable, replicable, and thus automatable (Autor et al. 2003; Acemoglu and Restrepo 2022):
  • a displacement effect of existing (automated) tasks and jobs in the manufacturing sector;
  • a productivity effect (and consequently a rise of wages) of the non-automated ones that are not displaced;
  • a reallocation of displaced low-skilled manufacturing workers into low-skilled service jobs (i.e. ripple or composition effect in the definition of Acemoglu and Restrepo 2022). This effect can partly compensate the labour-saving effects of automation technologies, leading to an expansion of employment not necessarily paralleled by a productivity effect (Capello et al. 2022a).
As long as the magnitude of the displacement effects is greater than the one of the productivity and composition effects, the wage bill share of low-skilled workers in non-metropolitan regions is likely to shrink.
At the same time, the same automation technology is also expected to influence high-skilled employment and wages by:
  • creating new tasks in high-skilled jobs and, therefore, impacting labour demand through a productivity effect (i.e. an increase in high-skilled wages and wage bill share) since labour has a comparative advantage in the newly introduced tasks with respect to technology
  • generating a reinstatement effect because of the expansion of the task set.
These effects are likely to take place especially in manufacturing-related services, as documented in the literature on manufacturing servitisation (Dauth et al 2021; De Propris and Storai 2019), even if limited in terms of magnitude (Vendrell-Herrero and Bustinza 2020).
Summing up these considerations on the two categories of workers, the combination of positive reinstatement and productivity effects for the high-skilled workers group, even if modest, and negative and sizeable displacement effects for the low-skilled workers group (marginally compensated by ripple or composition effects) is expected to lead to a worsening of existing inequalities in non-metropolitan regions (Hypothesis 1).
Importantly, the novelty of the paper is to claim that the reinstatement and productivity effects of automation do not remain bounded in manufacturing regions where it takes place, but can spread to neighbouring metropolitan ones, that are unexpectedly affected by automation technology adoption through spatial spillover mechanisms. In metropolitan areas, two effects can be expected to be at work:
  • a partial reallocation of low-skilled workers from non-metropolitan regions in low-skilled service sectors with very low wages (Acemoglu and Restrepo 2022);
  • an increase in demand for new high-skilled workers in advanced business services necessary for the automation process to be technically and operationally managed, mainly localised in metropolitan regions (De Propris and Storai 2019).
Therefore, in metropolitan regions automation is expected to lead to a polarisation of employment and of wages inequalities (Hypothesis 2).
Whether these hypotheses are right or not is a matter of empirics, presented in the following sections.

3 Data and methods

The empirical test of the framework outlined in the previous section requires a clear definition and measurement of three key concepts: displacement/reinstatement and productivity effects, diffusion of automation technologies, and city.
First, displacement, reinstatement, and productivity effects have been operationalised by examining the effects (i.e. expansion or contraction) of automation on the employment share and the wage bill share of sector–occupation-specific groups of workers, following an approach widely used in the literature (Acemoglu and Restrepo 2022). Specifically, analysing how automation impacts the employment share helps identifying the existence of displacement (contraction) or reinstatement (expansion) effects for specific groups of workers. Complementing this information with that on the impact of automation on the wage bill share of the same groups of workers (rather than on the unit wage by categories), while controlling for the respective employment share in the regressions, provides indications about the existence of productivity effects.
The workers’ groups considered in this paper are combinations of occupational groups and sectors (Table 1), identified with the goal of capturing the opposite extremes of the occupational hierarchy and work-related income distribution, namely low-skilled and high-skilled occupations, within specific sectors. Within the manufacturing sector, high-skilled occupations are identified as those related to managerial and professional jobs, whereas low-skilled occupations are identified as those related to the operation of plants and machine (see for a similar approach Acemoglu and Restrepo 2022). Within services, high-skilled occupations are identified as those related to managerial and professional jobs in those knowledge intensive business services that potentially might support the digital transition of manufacturing firms as knowledge providers (ICT, professional, scientific, and technical activities). Low-skilled occupations, instead, are the elementary ones in all consumers and business-oriented services complementary to knowledge intensive business services.
Table 1
Identification of high- and low-skilled occupations in manufacturing and service sectors
Category
Occupation groups
Sectors
Low-skilled employees in manufacturing
Class 7 of Italian Occupation Classification (Plant and machine operators and assemblers)
Section C of the statistical classification of economic activities (NACE)
High-skilled employees in manufacturing
Classes 1 (Legislators, officials, and managers) and 2 (Professionals) of Italian Occupation Classification
Section C of the statistical classification of economic activities (NACE)
Low-skilled employees in services
Elementary occupations—Class 8 of Italian Occupation Classification
Sections G–H–I–K–L–N–R–S–T of the statistical classification of economic activities (NACE)
High-skilled employees in services
Classes 1 (Legislators, Officials and Managers) and 2 (Professionals) of Italian Occupation Classification
Sections J–M of the statistical classification of economic activities (NACE
Data on employment and wages were derived from the Italian Labor Force Survey (RFL) conducted by the Italian National statistical office (ISTAT), which is the main source of statistical information on the Italian labour market. Specifically, microdata were aggregated by combinations of occupational groups and sectors and by NUTS3 region, for the period 2012–2019, covering the time frame from the aftermath of the debt crisis to the pre-COVID crisis.
The second operationalisation issue regards the main independent variable, namely the diffusion of automation technologies, measured as the number of industrial robots. Starting from national data provided by the International Federation of Robotics (IFR) for the period 2004–2017, three weights have been applied to apportion the industrial robots stock at NUTS3 level: (i) the number of manufacturing workers at the NUTS3 level over the national manufacturing employment, (ii) the number of blue-collar workers (ISCO8 occupational group) over the national blue-collar employment, and (iii) the share of households with broadband, all calculated in relation to the country. The selection of these weights aligns with existing literature (Dauth et al 2021) and is based on the expectation that robot adoption is more likely to occur in regions with (i) a higher concentration of industrial activities and (ii) equipped with better digital infrastructure.
Lastly, the identification of cities follows an administrative approach based on NUTS3 regions.5 Cities have been defined according to the average population level in the period of interest (2012–2019). Different thresholds were set and tested, respectively, the top 5%, 10%, 15%, 20%, and 25% most populous NUTS3 regions.6
Estimations have been carried out using a generalised least squares (GLS) random effects model with robust standard errors clustered at NUTS2 level.7 The presence of a time-invariant variable (metropolitan region) has made the use of a panel-fixed effects model unfeasible. Furthermore, the results of the Breusch and Pagan LM test for random effects (reported at the bottom the estimation tables in Appendix) suggest that a random effects model should be preferred to a pooled OLS model. In details, the estimated equations are as follows:
$$\begin{aligned} {\text{Employment share}}_{{o-j,i,t}} = & ~ + {\text{Robot density}}_{{i,{\text{ }}t - 2}} + ~{\text{Metropolitan region}}_{{\text{i}}} + ~{\text{Metropolitan region}}_{i} \\ & \quad \times {\text{ Robot density }} + _{{i,t - 1}} + {\text{Time FE}} + {\text{Regional FE}} + {\text{ }}\varepsilon _{{o,j,i,t}} \\ \end{aligned}$$
(1)
$$\begin{aligned} {\text{Wage bill share}}_{{o-j,i,t}} = & ~ + {\text{Robot density}}_{{i,{\text{ }}t - 2}} + ~{\text{Metropolitan region}}_{i} + ~{\text{Metropolitan region}}_{i} \\ & \quad \times {\text{ Robot density }} + _{{o,j,I,t - 2}} + _{{i,t - 1}} ~~ + {\text{Time FE}} + {\text{Regional FE}} + {\text{ }}\varepsilon _{{o,j,i,t}} \\ \end{aligned}$$
(2)
(Eq. 2).
where subscript o is the occupational group, j represents sector, i is the NUTS3 region, and t indicates the year.
Importantly, the interaction between the metropolitan region dummy variable and the robot density one allows calculating the marginal effects of robot diffusion in metropolitan vs non-metropolitan regions. Moreover, a matrix of control variables (X) was included in line with existing literature (Author et al. 2020, Acemoglu and Restrepo 2020; Dauth et al. 2021), accounting for:
  • The share of tertiary-educated employment, controlling for the level of education of the population;
  • The share of workers whose age is below the median age, controlling for the opportunities of finding a job at younger  ages;
  • The share of female occupation, controlling for the different opportunities of finding a job being a female rather than a male;
  • The regional sectoral composition, controlling for employment share in agriculture and in public administration.
Regional fixed effects (NUTS2 level) and year fixed effects were included to control for unobserved heterogeneity at the regional level and temporal variations over the study period. Finally, Eq. (2) includes the employment share of the group of workers under examination to control for any mechanical correlation between employment and wage bill shares variations. The full list and description of the variables used in the models is presented in Table 2.
Table 2
Variables description
Variable
Description
Data source
Inequalities by sector and occupational group (dependent variables)
Share of employment for each occupational group and sectors. Time span: 2012–2019
Share of wage bill for each combination of occupational group and sectors. Time span: 2012–2019
ISTAT—RFL microdata
Robot density (average)
Ratio between the stock of robots and the number of employees in manufacturing computed over a five-year period. Time span: t-6—t-2
International Robot Federation (IFR)
Metropolitan region
Dummy = 1 if the NUTS3 region is classified as metropolitan
ISTAT data on resident population
Tertiary-educated employment (%)
Share of tertiary-educated people in t-1
ISTAT—RFL microdata
 < 49 aged employment (%)
Share of workers whose age is below the median age in t-1
ISTAT—RFL microdata
Female employment (%)
Share of female occupation in t-1
ISTAT—RFL microdata
Sectoral composition
Employment in public services (%) in t-1
Employment in agriculture (%) in t-1
ISTAT
Employment share
Share of employment by sector and occupation
ISTAT—RFL microdata
To mitigate endogeneity concerns and the potential risk of self-selection bias (i.e. regions that promote automation might simultaneously reduce their dependency on manufacturing), all independent variables were lagged. Admittedly, this solution, though still a standard approach in the literature, has some limitations and might not be entirely sufficient to exclude the presence of these problems (Reed 2015). Yet, it may remain sufficient in the present context where the focus is on the association between automation deepening and the variation of employment and wage bill shares of different occupational groups within and across regions.8
Finally, spatial spillover effects were considered. Specifically, Eqs. 1 and 2 were augmented to account for spillover effects from non-metropolitan regions to nearby metropolitan regions. This adjustment aimed to examine whether the diffusion of automation technologies in non-metropolitan regions affects employment in the services sector of nearby metropolitan areas (i.e. reinstatement and composition effects, respectively, of high-skilled and low-skilled workers). To do so, the robot density spatial lag was calculated using a symmetric spatial contiguity matrix between metropolitan and non-metropolitan regions. In the model, this spatial lag is interacted with the metropolitan region dummy variable, thereby capturing potential spillover effects from non-metropolitan to metropolitan areas. Additionally, the model accounted for potential spatial effects between contiguous non-metropolitan regions by including the robot density spatial lag derived from the symmetric spatial contiguity matrix between non-metropolitan regions. The robot density spatial lag which accounts for contiguity between metropolitan regions was omitted from the model, as the automation spatial spillovers hypothesis was tested solely for the top 5% metropolitan regions, none of which are contiguous.9

4 Results

Estimation of Eqs. (1) and (2) offers support to the hypotheses advanced in Sect. 2. Specifically, Figs. 1 and 2, together with Table 3, display the displacement/reinstatement effect (i.e. decrease/increase in a specific occupation and sector employment share, columns 1–4) and the productivity effect (i.e. increase in a specific occupation and sector wage bill share, keeping constant the respective employment share, columns 5–6) in regions characterised by different settlement structures, identified according to different thresholds of urbanisation intensity as detailed in Sect. 3. It is important to underline that the effects on the wage bill share presented in Figs. 1 and 2 can be interpreted as a change in the unit wage of each workers’ category, and therefore as a proxy for the productivity effect, since the estimates of the effects of automation on the wage bill share control for the employment share in the same workers’ category. In this way, the effects on the wage bill share can be associated with changes in unit wages rather than to the number of workers in that particular workers’ category (Table 4).
Table 3
Reinstatement, displacement, and productivity effects of automation by occupations and sectors in metropolitan vs non-metropolitan regions
 
Reinstatement/displacement effect (employment share)
Productivity effect (wage bill share)
 
High-skilled occupations in manufacturing
High-skilled occupations in services
Low-skilled occupations in manufacturing
Low-skilled occupations in services
High-skilled occupations in manufacturing
High-skilled occupations in services
Low-skilled occupations in  manufacturing
Low-skilled occupations in  services
Top 5% metropolitan regions
− 0.0006
0.0035***
− 0.0051***
0.0014
− 0.0009
0.0051***
− 0.0060***
0.0011
[0.0006]
[0.0008]
[0.0019]
[0.0014]
[0.0009]
[0.0012]
[0.0019]
[0.0011]
Non-metropolitan regions
− 0.0009***
0.0001
− 0.0029**
0.0022***
− 0.0015***
0.0001
− 0.0035**
0.0013
[0.0002]
[0.0003]
[0.0013]
[0.0008]
[0.0003]
[0.0003]
[0.0015]
[0.0008]
Top 10% metropolitan regions
− 0.0010***
0.0009
− 0.0020
0.0037***
− 0.0017***
0.0013
− 0.0027
0.0021***
[0.0004]
[0.0013]
[0.0024]
[0.0010]
[0.0006]
[0.0020]
[0.0026]
[0.0008]
Non-metropolitan regions
− 0.0008***
0.0003
− 0.0031***
0.0021**
− 0.0014***
0.0003
− 0.0038***
0.0013
[0.0002]
[0.0003]
[0.0012]
[0.0008]
[0.0003]
[0.0004]
[0.0014]
[0.0009]
Top 15% metropolitan regions
− 0.0012***
0.0011
− 0.0026
0.0041***
− 0.0020***
0.0016
− 0.0033
0.0024***
[0.0003]
[0.0013]
[0.0023]
[0.0010]
[0.0005]
[0.0019]
[0.0024]
[0.0008]
Non-metropolitan regions
− 0.0008***
0.0003
− 0.0031**
0.0022**
− 0.0014***
0.0003
− 0.0038***
0.0013
[0.0002]
[0.0003]
[0.0012]
[0.0009]
[0.0003]
[0.0005]
[0.0014]
[0.0009]
Top 20% metropolitan regions
− 0.0014***
0.0011
− 0.0020
0.0042***
− 0.0023***
0.0015
− 0.0028
0.0023***
[0.0003]
[0.0012]
[0.0020]
[0.0010]
[0.0005]
[0.0018]
[0.0022]
[0.0006]
Non-metropolitan regions
− 0.0008***
0.0003
− 0.0032 **
0.0021***
− 0.0014***
0.0003
− 0.0038***
0.0013
[0.0002]
[0.0003]
[0.0013]
[0.0008]
[0.0003]
[0.0004]
[0.0014]
[0.0009]
Top 25% metropolitan regions
− 0.0012***
0.0005
− 0.0022
0.0034***
− 0.0021***
0.0006
− 0.0030
0.0020***
[0.0002]
[0.0008]
[0.0016]
[0.0007]
[0.0004]
[0.0012]
[0.0020]
0.0006
Non-metropolitan regions
− 0.0008***
0.0003
− 0.0033**
0.0018**
− 0.0013***
0.0003
− 0.0039***
0.0011
[0.0002]
[0.0004]
[0.0013]
[0.0009]
[0.0003]
[0.0005]
[0.0014]
[0.0009]
Standard errors clustered at NUTS2 level in brackets. *** p < 0.01, ** p < 0.05, * p < 0.1
Table 4
Spatial reinstatement and productivity effects of automation in metropolitan vs non-metropolitan regions (marginal effects)
 
High-skilled occupations in services
Low-skilled occupations in services
 
Reinstatement/displacement effect (employment share)
Productivity effect (wage bill share)
Reinstatement/displacement effect (employment share)
Productivity effect (wage bill share)
Top 5% metropolitan regions
0.0024***
0.0032***
− 0.0013
− 0.0012
[0.0003]
[0.0004]
[0.0011]
[0.0007]
Non-metropolitan regions
0.0001
0.0001
− 0.0002
− 0.0001
[0.0001]
[0.0002]
[0.0007]
[0.0004]
Standard errors clustered at NUTS2 level in brackets. *** p < 0.01, ** p < 0.05, * p < 0.1
Full estimates are reported in Appendix (Tables 5, 6, 7, 8, and 9), while marginal effects are in Table 3.
The joint reading of Fig. 1 and Table 3 highlights a significant displacement effects of low-skilled manufacturing workers, which is paralleled by a moderately greater reduction of their wage bill share, suggesting that this group of workers experiences a contraction in terms of the share of jobs and wages, compared with the other workers’ categories. On the other hand, low-skilled services workers experience a likely composition (and possibly a reinstatement) effect, although in this case it is not paralleled by a productivity effect. Importantly, and unexpectedly, also high-skilled manufacturing workers are subject to a displacement effect without experiencing a compensatory productivity or composition effect.
Put shortly, in non-metropolitan areas:
  • None of the workers groups enjoy a productivity effect (i.e. there are no wage increases);
  • At best, there seems to be a composition effect, involving the (partial) reallocation of low-skilled manufacturing workers into low-skilled services jobs;
  • The displacement of high-skilled manufacturing workers is not associated with a productivity effect, and it is neither counterbalanced by a composition nor a reinstatement effect into services (Fig. 1).
The absence of productivity effects for all occupational groups together with a generalised displacement effect in the manufacturing sector suggests a reduction of average wages; this negative outcome is particularly severe for low-skilled manufacturing workers, which thus experience a deterioration of their conditions with respect to the other occupational groups, warning about potentially enlarging inequalities, a result consistent with Capello and Lenzi (2023) and Dottori (2021).10
The case of metropolitan regions presents two interesting specificities with respect to non-metropolitan regions (Table 3 and Fig. 2). First, the top 5% cities experience an expansion of the share of high-skilled service workers, possibly due to a combination of reinstatement but also composition effects in response of the shrinking of the share of high-skilled workers in manufacturing taking place in all cities, regardless of the population threshold used to define metropolitan regions. This result may suggest potential spillover effects from second-tier cities to the top 5% ones. Interestingly, this result is consistent with Hypothesis 2 but proves to be also highly selective in space. In fact, this effect vanishes when relaxing the definition of metropolitan regions to include the top 10% most populous regions (or more) (Table 3), which have on average a lower specialisation in services than large ones. Second, the share of low-skilled service workers enlarges in response to automation, with the unexpected exclusion of the top 5% metropolitan regions suggesting the possibility of reverse spillovers from the top cities to the second-tier cities. This result confirms the widening of inequalities and the polarisation of employment within and across regions, as posited in Hypothesis 2; in relative terms, the share of top occupations expands in the top 5% (i.e. first rank) cities, while the share of elementary occupations expands in large but second-rank ones. This interpretation is also reinforced by the results on the productivity effect. While the magnitude of the productivity effect is larger than the reinstatement/composition effect for high-skilled service workers in the top 5% cities, the opposite occurs for low-skilled service workers in the other cities (Fig. 2 and Table 3). Put shortly, a larger fraction of highly skilled workers enjoys an even larger wage bill share in the top 5% cities, while a smaller fraction of low-skilled workers accrues a proportionally smaller wage bill share in large but second-rank cities, thus enlarging the occupational and spatial divides and the separation between metropolitan elites and enlarging and impoverishing urban communities.
More interestingly, Fig. 3 (together with Table 9 in Appendix) shows the effects of robot diffusion in non-metropolitan regions on contiguous metropolitan ones and indicates that the top 5% cities enjoy an expansion of the share of high-skilled service workers in response to automation deepening in contiguous non-metropolitan regions. A possible interpretation for this result is that sectoral and spatial complementarities enable high-skilled workers to relocate across sectors (i.e. from manufacturing to services) and possibly across regions (i.e. from non-metropolitan to top metropolitan ones). Importantly, this spatial and sectoral reallocation does not exclude the reinstatement of new high-skilled services jobs, complementary to metropolitan as much as non-metropolitan manufacturing activities. However, these positive sectoral and spatial spillover effects do not apply to low-skilled workers, which seem neutral to automation processes in nearby regions.

5 Conclusions

The paper analysed conceptually and empirically the effects of automation on workers' group inequalities, taking into consideration reallocation effects between sectors and regions. The traditional displacement effects of low-skilled workers emerged once again in non-metropolitan regions.
However, the originality of the paper was the interpretation of the effects of automation on inter-group inequalities beyond non-metropolitan areas. In fact, through sectoral employment reallocation effects from manufacturing to services, the worsening of workers' group inequalities was expected to spread to metropolitan regions. In particular, the empirical analysis has pointed out that the largest cities are the place where high-skilled jobs opportunities related to automation are created.
The city is therefore not exempted from the impacts of automation. It is in the largest cities where talented people have the highest probability to increase their share of job opportunities and their wage bill in the service sector through automation. A contraction of the share of jobs instead characterises high-skilled manufacturing workers in all settings, while complementary low-skilled workers in the service sector enjoy an expansion of their job share. Except for high-skilled workers in services, none of the occupational groups considered in the analysis enjoy a productivity effect, suggesting a deterioration of average income, with the ultimate effect of a worsening of workers' group inequalities not only within regions but especially between top cities and the remaining regions.
In conclusion, while one cannot ignore the benefits stemming from automation, it is wise not to disregard the spatial dimension of workers’ group inequalities that such technologies are able toan produce. They represent an urgent issue requiring timely policy reply and intervention, like appropriate policies to cope with the changing occupational skills requested by the labour market. This is true for manufacturing regions, but also for regions that are not so clearly exposed to automation technology risks. Also in these regions, anticipatory policy interventions could be appropriate to avoid a widening of disparities in the future.
Tackling the differentiated effects of automation may include a broad menu of actions, which may also bring the risks of mitigating one source of inequalities (e.g. across regions) at the expense of the other (e.g. across occupational groups). The development of a broad and comprehensive policy framework and a far-reaching set of initiatives in multiple domains ranging from education to social, innovation, and industrial policies represents therefore an ambitious as much as a necessary strategy to fight multidimensional inequalities. In this respect, the EU has undertaken an unprecedented financial effort through the launch and coordination of the National Recovery and Resilience Plans, the NextGenerationEU program, and European Structural Investment Funds (Santos and Conte 2024), which represents a unique opportunity, not to be missed, to address the growing inequalities in the EU and enhance equitable labour markets throughout the Union.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://​creativecommons.​org/​licenses/​by/​4.​0/​.

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendix

Appendix

Tables 5, 6, 7, 8 and 9
Table 5
Automation impact on employment and wage bill share—high-skilled occupations in manufacturing
 
Dependent variable: employment share in high-skilled occupations manufacturing
Dependent variable: wage bill share in high-skilled occupations manufacturing
1
2
3
4
5
1
2
3
4
5
Robot density
− 0.0009***
− 0.0008***
− 0.0008***
− 0.0008***
− 0.0008***
− 0.0015***
− 0.0014***
− 0.0014***
− 0.0013***
− 0.0013***
[0.0002]
[0.0002]
[0.0002]
[0.0002]
[0.0002]
[0.0003]
[0.0003]
[0.0003]
[0.0003]
[0.0003]
Top 5% regions
− 0.0008
    
− 0.0026
    
[0.0060]
    
[0.0092]
    
Top 5% regions × robot density
0.0003
    
0.0006
    
[0.0005]
    
[0.0007]
    
Top 10% regions
 
0.0021
    
0.0026
   
 
[0.0036]
    
[0.0058]
   
Top 10% regions × robot density
 
− 0.0002
    
− 0.0002
   
 
[0.0003]
    
[0.0005]
   
Top 15% regions
  
0.0049***
    
0.0076***
  
  
[0.0018]
    
[0.0029]
  
Top 15% regions × robot density
  
− 0.0005**
    
− 0.0007**
  
  
[0.0002]
    
[0.0003]
  
Top 20% regions
   
0.0060***
    
0.0095***
 
   
[0.0017]
    
[0.0028]
 
Top 20% regions × robot density
   
− 0.0006***
    
− 0.0009***
 
   
[0.0002]
    
[0.0003]
 
Top 25% regions
    
0.0053***
    
0.0086***
    
[0.0014]
    
[0.0023]
Top 25% regions × robot density
    
− 0.0005***
    
− 0.0008***
    
[0.0002]
    
[0.0003]
Tertiary-educated employment (%)
0.0068
0.0066
0.0072
0.0070
0.0067
0.0096
0.0092
0.0103
0.0099
0.0094
[0.0197]
[0.0197]
[0.0197]
[0.0198]
[0.0199]
[0.0304]
[0.0304]
[0.0305]
[0.0305]
[0.0307]
 < 49 aged employment (%)
0.0069
0.0068
0.0059
0.0053
0.0058
− 0.0031
− 0.0034
-0.0048
− 0.0057
− 0.0049
[0.0086]
[0.0086]
[0.0084]
[0.0085]
[0.0085]
[0.0122]
[0.0121]
[0.0118]
[0.0117]
[0.0118]
Female occupation (%)
0.0253
0.0264
0.0259
0.0261
0.0257
0.0431*
0.0448*
0.0439*
0.0440*
0.0436*
[0.0186]
[0.0185]
[0.0187]
[0.0187]
[0.0187]
[0.0261]
[0.0264]
[0.0266]
[0.0263]
[0.0265]
Public services employment (%)
0.0082
0.0075
0.0081
0.0120
0.0125
0.0007
− 0.0004
0.0011
0.0079
0.0087
[0.0211]
[0.0209]
[0.0179]
[0.0172]
[0.0173]
[0.0333]
[0.0332]
[0.0286]
[0.0271]
[0.0272]
Agriculture employment (%)
− 0.0292
− 0.0331*
− 0.0302*
− 0.0297*
− 0.0280*
− 0.0456
− 0.0517*
-0.0459*
− 0.0449
− 0.0429
[0.0186]
[0.0193]
[0.0165]
[0.0174]
[0.0168]
[0.0295]
[0.0312]
[0.0268]
[0.0279]
[0.0269]
High-skilled manufacturing employment (%)
     
− 0.0003
− 0.0003
-0.0003
− 0.0004
− 0.0004
     
[0.0015]
[0.0015]
[0.0015]
[0.0015]
[0.0015]
Constant
− 0.0029
− 0.0032
− 0.0032
− 0.0035
− 0.0038
0.0096
0.0091
0.0091
0.0085
0.0080
[0.0179]
[0.0176]
[0.0174]
[0.0174]
[0.0173]
[0.0229]
[0.0225]
[0.0222]
[0.0221]
[0.0220]
Observations
804
804
804
804
804
804
804
804
804
804
R-squared overall
0.4734
0.4728
0.4818
0.4836
0.4855
0.4719
0.4711
0.4801
0.4823
0.4849
Breusch and Pagan Test for RE (Chi2)
222.75***
223.85***
205.99***
204.07***
199.68***
213.47***
214.71***
197.01***
193.89***
187.65***
Standard errors clustered at NUTS2 level in brackets. Constant, year, and NUTS2 dummy included. *** p < 0.01, ** p < 0.05, * p < 0.1
Table 6
Automation impact on employment and wage bill share—high-skilled occupations in services
 
Dependent variable: employment share in high-skilled occupations in services
Dependent variable: wage bill share in high-skilled occupations in  services
1
2
3
4
5
1
2
3
4
5
Robot density
0.0001
0.0003
0.0003
0.0003
0.0003
0.0001
0.0003
0.0003
0.0003
0.0003
[0.0003]
[0.0003]
[0.0003]
[0.0003]
[0.0004]
[0.0004]
[0.0004]
[0.0005]
[0.0004]
[0.0005]
Top 5% regions
− 0.0137**
    
− 0.0206**
    
[0.0062]
    
[0.0091]
    
Top 5% regions × robot density
0.0033***
    
0.0050***
    
[0.0008]
    
[0.0012]
    
Top 10% regions
 
0.0025
    
0.0030
   
 
[0.0074]
    
[0.0112]
   
Top 10% regions × robot density
 
0.0006
    
0.0010
   
 
[0.0012]
    
[0.0018]
   
Top 15% regions
  
− 0.0007
    
− 0.0012
  
  
[0.0071]
    
[0.0106]
  
Top 15% regions × robot density
  
0.0009
    
0.0013
  
  
[0.0012]
    
[0.0018]
  
Top 20% regions
   
0.0005
    
0.0006
 
   
[0.0064]
    
[0.0098]
 
Top 20% regions × robot density
   
0.0008
    
0.0012
 
   
[0.0011]
    
[0.0016]
 
Top 25% regions
    
0.0036
    
0.0052
    
[0.0041]
    
[0.0062]
Top 25% regions × robot density
    
0.0002
    
0.0003
    
[0.0007]
    
[0.0010]
Tertiary-educated employment (%)
− 0.0026
− 0.0037
− 0.0026
− 0.0034
− 0.0038
− 0.0032
− 0.0050
− 0.0034
− 0.0045
− 0.0050
[0.0174]
[0.0170]
[0.0171]
[0.0171]
[0.0175]
[0.0241]
[0.0237]
[0.0237]
[0.0239]
[0.0243]
 < 49 aged employment (%)
− 0.0211*
− 0.0216*
− 0.0231**
− 0.0220*
− 0.0219*
− 0.0205
− 0.0209
− 0.0231
− 0.0216
− 0.0214
[0.0119]
[0.0120]
[0.0117]
[0.0118]
[0.0114]
[0.0187]
[0.0185]
[0.0180]
[0.0179]
[0.0174]
Female occupation (%)
0.0220
0.0230
0.0221
0.0193
0.0216
0.0321
0.0332
0.0324
0.0283
0.0316
[0.0192]
[0.0188]
[0.0190]
[0.0182]
[0.0195]
[0.0242]
[0.0228]
[0.0230]
[0.0218]
[0.0236]
Public services employment (%)
0.0100
0.0087
0.0024
0.0131
0.0122
0.0229
0.0213
0.0125
0.0281
0.0272
[0.0206]
[0.0210]
[0.0207]
[0.0204]
[0.0206]
[0.0291]
[0.0302]
[0.0291]
[0.0288]
[0.0289]
Agriculture employment (%)
− 0.0519***
− 0.0555***
− 0.0553***
− 0.0447**
− 0.0564***
− 0.0582*
− 0.0628**
− 0.0628**
− 0.0465*
− 0.0627**
[0.0201]
[0.0207]
[0.0206]
[0.0187]
[0.0191]
[0.0302]
[0.0309]
[0.0306]
[0.0281]
[0.0276]
High-skilled services employment (%)
     
− 0.0013
− 0.0012
− 0.0012
− 0.0012
− 0.0012
     
[0.0023]
[0.0022]
[0.0022]
[0.0022]
[0.0022]
Constant
0.0155
0.0154
0.0176
0.0163
0.0158
0.0135
0.0135
0.0164
0.0144
0.0136
[0.0184]
[0.0187]
[0.0181]
[0.0183]
[0.0187]
[0.0261]
[0.0263]
[0.0254]
[0.0254]
[0.0260]
Observations
830
830
830
830
830
830
830
830
830
830
R-squared overall
0.431
0.377
0.3647
0.3805
0.3688
0.4406
0.3743
0.3636
0.3808
0.3690
Breusch and Pagan Test for RE (Chi2)
543.76***
633.55***
653.79***
654.76***
669.80***
568.31***
683.85***
696.88***
698.13***
713.81***
Standard errors clustered at NUTS2 level in brackets. Constant, year, and NUTS2 dummy included. *** p < 0.01, ** p < 0.05, * p < 0.1
Table 7
Automation impact on employment and wage bill share—low-skilled occupations in manufacturing
 
Dependent variable: employment share in high-skilled occupations in manufacturing
Dependent variable: wage bill share in high-skilled occupations in manufacturing
1
2
3
4
5
1
2
3
4
5
Robot density
− 0.0029**
− 0.0031***
− 0.0031**
− 0.0032**
− 0.0033**
− 0.0037***
− 0.0040***
− 0.0040***
− 0.0041***
− 0.0042***
[0.0013]
[0.0012]
[0.0012]
[0.0013]
[0.0013]
[0.0012]
[0.0012]
[0.0012]
[0.0012]
[0.0013]
Top 5% regions
− 0.0020
    
0.0005
    
[0.0159]
    
[0.0158]
    
Top 5% regions × robot density
− 0.0022
    
− 0.0026
    
[0.0023]
    
[0.0023]
    
Top 10% regions
 
− 0.0245*
    
− 0.0255
   
 
[0.0135]
    
[0.0159]
   
Top 10% regions × robot density
 
0.0012
    
0.0012
   
 
[0.0015]
    
[0.0016]
   
Top 15% regions
  
− 0.0145
    
− 0.0167
  
  
[0.0122]
    
[0.0135]
  
Top 15% regions × robot density
  
0.0006
    
0.0007
  
  
[0.0014]
    
[0.0015]
  
Top 20% regions
   
− 0.0203*
    
− 0.0233**
 
   
[0.0109]
    
[0.0117]
 
Top 20% regions × robot density
   
0.0012
    
0.0013
 
   
[0.0012]
    
[0.0013]
 
Top 25% regions
    
− 0.0170*
    
− 0.0200**
    
[0.0089]
    
[0.0094]
Top 25% regions × robot density
    
0.0011
    
0.0013
    
[0.0009]
    
[0.0010]
Tertiary-educated employment (%)
0.0307
0.0308
0.0296
0.0300
0.0307
0.0118
0.0119
0.0098
0.0102
0.0117
[0.0492]
[0.0491]
[0.0492]
[0.0492]
[0.0490]
[0.0486]
[0.0488]
[0.0486]
[0.0488]
[0.0484]
 < 49 aged employment (%)
0.1159**
0.1170**
0.1182**
0.1187**
0.1170**
0.1077***
0.1089***
0.1110***
0.1116***
0.1093***
[0.0467]
[0.0467]
[0.0464]
[0.0464]
[0.0466]
[0.0379]
[0.0378]
[0.0377]
[0.0377]
[0.0380]
Female Occupation (%)
− 0.0955**
− 0.0982**
− 0.0981**
− 0.0972**
− 0.0982**
− 0.0828*
− 0.0861*
− 0.0860*
− 0.0840*
− 0.0853*
[0.0475]
[0.0469]
[0.0473]
[0.0475]
[0.0476]
[0.0499]
[0.0489]
[0.0495]
[0.0493]
[0.0490]
Public services employment (%)
− 0.2790***
− 0.2781***
− 0.2684***
− 0.2820***
− 0.2786***
− 0.2682***
− 0.2688***
− 0.2559***
− 0.2757***
− 0.2725***
[0.0782]
[0.0773]
[0.0789]
[0.0788]
[0.0800]
[0.0763]
[0.0763]
[0.0763]
[0.0750]
[0.0771]
Agriculture employment (%)
0.0214
0.0241
0.0307
0.0209
0.0342
0.0433
0.0451
0.0507
0.0340
0.0515
[0.0575]
[0.0572]
[0.0569]
[0.0542]
[0.0518]
[0.0702]
[0.0708]
[0.0691]
[0.0669]
[0.0630]
Low-skilled manufacturing employment (%)
     
− 0.0448***
− 0.0449***
− 0.0450***
− 0.0452***
− 0.0449***
     
[0.0111]
[0.0110]
[0.0110]
[0.0110]
[0.0110]
Constant
0.1066**
0.1086***
0.1055**
0.1080***
0.1084***
0.1460***
0.1489***
0.1456***
0.1490***
0.1490***
[0.0418]
[0.0415]
[0.0413]
[0.0417]
[0.0415]
[0.0395]
[0.0395]
[0.0393]
[0.0400]
[0.0397]
Observations
830
830
830
830
830
830
830
830
830
830
R-squared overall
0.6780
0.6819
0.6768
0.6792
0.6768
0.6870
0.6906
0.6872
0.6906
0.6873
Breusch and Pagan test for RE (Chi2)
1379.62***
1368.23***
1368.06
1372.31***
1381.03***
1199.20***
1193.02***
1184.27***
1184.54***
1207.39***
Standard errors clustered at NUTS2 level in brackets. Constant, year, and NUTS2 dummy included. *** p < 0.01, ** p < 0.05, * p < 0.1
Table 8
Automation impact on employment and wage bill share—low-skilled occupations in services
 
Dependent variable: employment share in low-skilled occupations in services
Dependent variable: wage bill share in low-skilled occupations in services
1
2
3
4
5
1
2
3
4
5
Robot density
0.0022***
0.0021**
0.0022**
0.0021***
0.0018**
0.0014*
0.0013*
0.0013*
0.0013*
0.0011
[0.0008]
[0.0008]
[0.0009]
[0.0008]
[0.0009]
[0.0007]
[0.0008]
[0.0008]
[0.0008]
[0.0008]
Top 5% regions
0.0259**
    
0.0135
    
[0.0117]
    
[0.0100]
    
Top 5% regions × robot density
− 0.0008
    
− 0.0003
    
[0.0014]
    
[0.0014]
    
Top 10% regions
 
0.0059
    
0.0038
   
 
[0.0074]
    
[0.0053]
   
Top 10% regions × robot density
 
0.0016*
    
0.0009
   
 
[0.0008]
    
[0.0006]
   
Top 15% regions
  
0.0012
    
− 0.0003
  
  
[0.0075]
    
[0.0051]
  
Top 15% regions × robot density
  
0.0020**
    
0.0013**
  
  
[0.0009]
    
[0.0006]
  
Top 20% regions
   
0.0016
    
0.0011
 
   
[0.0074]
    
[0.0053]
 
Top 20% regions × robot density
   
0.0021***
    
0.0011*
 
   
[0.0008]
    
[0.0006]
 
Top 25% regions
    
0.0035
    
0.0014
    
[0.0065]
    
[0.0045]
Top 25% regions × robot density
    
0.0016*
    
0.0010*
    
[0.0008]
    
[0.0005]
Tertiary-educated employment (%)
− 0.0574
− 0.0550
− 0.0502
− 0.0519
− 0.0548
− 0.0512
− 0.0501
− 0.0473
− 0.0488
− 0.0502
[0.0444]
[0.0430]
[0.0428]
[0.0429]
[0.0425]
[0.0342]
[0.0333]
[0.0335]
[0.0335]
[0.0331]
 < 49 aged employment (%)
0.0446
0.0442
0.0382
0.0411
0.0434
0.0273
0.0276
0.0242
0.0262
0.0275
[0.0378]
[0.0387]
[0.0375]
[0.0374]
[0.0372]
[0.0282]
[0.0285]
[0.0275]
[0.0274]
[0.0273]
Female occupation (%)
0.0549
0.0510
0.0506
0.0414
0.0428
0.0406
0.0373
0.0368
0.0321
0.0327
[0.0424]
[0.0424]
[0.0416]
[0.0421]
[0.0413]
[0.0294]
[0.0294]
[0.0291]
[0.0301]
[0.0284]
Public services employment (%)
0.0829
0.0812
0.0587
0.0874
0.1015*
0.0548
0.0555
0.0418
0.0598
0.0673
[0.0646]
[0.0604]
[0.0600]
[0.0562]
[0.0585]
[0.0509]
[0.0479]
[0.0488]
[0.0473]
[0.0484]
Agriculture employment (%)
− 0.1314***
− 0.0950**
− 0.0954**
− 0.0653
− 0.0704
− 0.0651***
− 0.0428
− 0.0430
− 0.0309
− 0.0320
[0.0477]
[0.0391]
[0.0399]
[0.0409]
[0.0459]
[0.0234]
[0.0272]
[0.0264]
[0.0269]
[0.0320]
Low-skilled services employment (%)
     
− 0.0339***
− 0.0333***
− 0.0333***
− 0.0328***
− 0.0324***
     
[0.0105]
[0.0104]
[0.0103]
[0.0104]
[0.0104]
Constant
0.0052
0.0068
0.0126
0.0098
0.0087
0.0277
0.0280
0.0316
0.0293
0.0285
[0.0380]
[0.0368]
[0.0364]
[0.0366]
[0.0371]
[0.0304]
[0.0298]
[0.0290]
[0.0291]
[0.0289]
Observations
856
856
856
856
856
856
856
856
856
856
R-squared overall
0.3822
0.4128
0.4172
0.4311
0.4231
0.3455
0.371
0.3749
0.3743
0.3700
Breusch and Pagan test for RE (Chi2)
569.53***
496.36***
468.86***
451.04***
494.35***
488.78***
444.11***
425.43***
433.80***
460.21***
Standard errors clustered at NUTS2 level in brackets. Constant, year, and NUTS2 dummy included. *** p < 0.01, ** p < 0.05, * p < 0.1
Table 9
Spatial effects of automation on employment and wage bill shares
Dependent variable:
High-skilled occupations in services
Low-skilled occupations in services
Employment share
Wage bill share
Employment share
Wage bill share
Robot density
0.0001
0.0002
0.0022***
0.0014**
[0.0002]
[0.0003]
[0.0008]
[0.0007]
Robot density of contiguous regions (metropolitan and non-metropolitan regions)
0.0001
0.0001
− 0.0002
− 0.0001
[0.0001]
[0.0002]
[0.0007]
[0.0004]
Top 5% regions
− 0.0767***
− 0.1024***
0.0667*
0.0511*
[0.0088]
[0.0136]
[0.0366]
[0.0301]
Top 5% regions × Robot density of robot density of contiguous regions (metropolitan and non-metropolitan regions)
0.0023***
0.0031***
− 0.0010
− 0.0011*
[0.0003]
[0.0005]
[0.0008]
[0.0006]
Robot density of contiguous regions (non-metropolitan–non-metropolitan)
− 0.0000
0.0000
0.0000
− 0.0001
[0.0000]
[0.0001]
[0.0001]
[0.0001]
Tertiary-educated employment (%)
− 0.0037
− 0.0046
− 0.0567
− 0.0510
[0.0166]
[0.0231]
[0.0448]
[0.0345]
 < 49 aged employment (%)
− 0.0195
− 0.0185
0.0440
0.0261
[0.0122]
[0.0192]
[0.0381]
[0.0286]
Female occupation (%)
0.0194
0.0292
0.0579
0.0435
[0.0178]
[0.0219]
[0.0427]
[0.0294]
Public services employment (%)
0.0170
0.0310
0.0821
0.0565
[0.0201]
[0.0287]
[0.0628]
[0.0502]
Agriculture employment (%)
− 0.0583***
− 0.0692**
− 0.1292***
− 0.0669***
[0.0190]
[0.0288]
[0.0468]
[0.0242]
High-skilled services Employment (%)
 
− 0.0014
  
 
[0.0023]
  
Low-skilled services Employment (%)
   
− 0.0342***
    
[0.0104]
Constant
0.0137
0.0108
0.0042
0.0290
[0.0179]
[0.0252]
[0.0389]
[0.0318]
Observations
830
830
830
830
R-squared overall
0.5050
0.5073
0.3809
0.3471
Breusch and Pagan test for RE (Chi2)
413.71***
435.39***
567.33***
490.23***
Standard errors clustered at NUTS2 level in brackets. Constant, year, and NUTS2 dummy included. *** p < 0.01, ** p < 0.05, * p < 0.1
Footnotes
1
The definitions and terminology adopted in this paper follow those introduced by Acemoglu and Restrepo (2019), as explained in Sect. 2.
 
2
The conceptual dichotomy contrasting metropolitan areas specialised in services to non-metropolitan areas specialised in manufacturing is well accepted in the literature since the seminal paper by Baumol (1967). Cities are conceptually associated with a service-led economy structure, while the spatial concentration of the manufacturing sector is predominant in non-metropolitan settings. This is particularly true in Italy, where the industrial vocation is typically associated with industrial districts, made of non-metropolitan, small- and medium-sized city areas, that contrasts the service-based economies of large cities. See for a discussion on this point Camagni et al. (2023) and Camagni (2024).
 
3
Some works have been developed by exploiting spatial data settings, but without comparing the effects of automation across different types of settlements (Fossen and Sorgner 2022; Autor and Dorn 2013; Dauth et al. 2021; Caselli et al. 2021a, b).
 
4
Empirical results are, however, subject to important nuances depending on the specific empirical setting considered. See among many others Acemoglu and Restrepo (2020) for the USA, De Vries et al. (2020) for the EU, Adachi et al. (2020) for Japan, Aghion et al. (2019) for France, Dauth et al. (2021) for Germany, Caselli et al. (2021a, b) for Italy.
 
5
(see Capello et al. (2022b) for details and justification of the appropriateness of this approach)
 
6
Roma, Milano, Napoli, Torino, Brescia, Bari (top 5%), Palermo, Bergamo, Salerno, Catania, Bologna (top 10%), Firenze, Padova, Verona, Caserta, Varese (top 15%), Treviso, Vicenza, Monza e Brianza, Venezia, Genova, Lecce (top 20%), Cosenza, Modena, Perugia, Messina, Foggia (top 25%).
 
7
Errors were clustered at NUTS2 level consistently with the regional fixed level (NUTS2), and it accounts for possible within NUTS2 regions correlation of the regressors and the errors.
 
8
An IV-2SLS approach would be ideally a preferable approach, but it would clearly complicate the analysis of the spatial spillovers, as explained in the following paragraph in the main text.
 
9
The empirical analysis focused on the spatial interactions of the top 5% cities only because these are the cities where the employment share and wage bill share of high-skilled jobs turn to expand in association with automation deepening, as shown in Sect. 4.
 
10
We wish to thank an anonymous reviewer for suggesting us this interpretation.
 
Literature
go back to reference Acemoglu D, Restrepo P (2019) Automation and new tasks: how technology displaces and reinstates labor. J Econ Perspect 33(2):3–30CrossRef Acemoglu D, Restrepo P (2019) Automation and new tasks: how technology displaces and reinstates labor. J Econ Perspect 33(2):3–30CrossRef
go back to reference Acemoglu D, Restrepo P (2020) Robots and jobs: evidence from US labor markets. J Polit Econ 128(6):2188–2244CrossRef Acemoglu D, Restrepo P (2020) Robots and jobs: evidence from US labor markets. J Polit Econ 128(6):2188–2244CrossRef
go back to reference Acemoglu D, Restrepo P (2022) Tasks, automation, and the rise in US wage inequality. Econometrica 90(5):1973–2016CrossRef Acemoglu D, Restrepo P (2022) Tasks, automation, and the rise in US wage inequality. Econometrica 90(5):1973–2016CrossRef
go back to reference Adachi D, Kawaguchi D, Saito YU (2020) Robots and Employment: Evidence from Japan, 1978–2017, RIETI Discussion Paper Series 20-E-051 Adachi D, Kawaguchi D, Saito YU (2020) Robots and Employment: Evidence from Japan, 1978–2017, RIETI Discussion Paper Series 20-E-051
go back to reference Aghion P, Antonin C, Bunel S (2019) Artificial intelligence, growth and employment: the role of policy. Economie Et Statistique / Economics and Statistics 510–512:149–164 Aghion P, Antonin C, Bunel S (2019) Artificial intelligence, growth and employment: the role of policy. Economie Et Statistique / Economics and Statistics 510–512:149–164
go back to reference Akst D (2013) Where have all jobs gone? Wilson Q 37(3):2–15 Akst D (2013) Where have all jobs gone? Wilson Q 37(3):2–15
go back to reference Alvaredo F, Chancel L, Piketty T, Saez E, Zucman G et al (2018) World Inequality Report 2018. World Inequality Lab Alvaredo F, Chancel L, Piketty T, Saez E, Zucman G et al (2018) World Inequality Report 2018. World Inequality Lab
go back to reference Autor D (2015) Why are there still so many jobs? The history and future of workplace automation. J Econ Perspect 29(3):3–30CrossRef Autor D (2015) Why are there still so many jobs? The history and future of workplace automation. J Econ Perspect 29(3):3–30CrossRef
go back to reference Autor DH, Dorn D (2013) The growth of low-skill service jobs and the polarization of the US labor market. Am Econ Rev 103(5):1553–1597CrossRef Autor DH, Dorn D (2013) The growth of low-skill service jobs and the polarization of the US labor market. Am Econ Rev 103(5):1553–1597CrossRef
go back to reference Autor DH, Levy F, Murnane RJ (2003) The skill content of recent technological change: an empirical exploration. Quart J Econ 118(4):1279–1333CrossRef Autor DH, Levy F, Murnane RJ (2003) The skill content of recent technological change: an empirical exploration. Quart J Econ 118(4):1279–1333CrossRef
go back to reference Autor DH, Dorn D, Katz LF, Patterson C, Van Reen J (2020) The fall of the labor share and the rise of superstar firms. Quart J Econ 135(2):645–709CrossRef Autor DH, Dorn D, Katz LF, Patterson C, Van Reen J (2020) The fall of the labor share and the rise of superstar firms. Quart J Econ 135(2):645–709CrossRef
go back to reference Autor D, Mindell D, Reynolds E (2022) The work of the future: building better jobs in an age of intelligent machines. MIT Press, CambridgeCrossRef Autor D, Mindell D, Reynolds E (2022) The work of the future: building better jobs in an age of intelligent machines. MIT Press, CambridgeCrossRef
go back to reference Autor DH (2022) The labor market impacts of technological change: from unbridled enthusiasm to qualified optimism to vast uncertainty, NBER Working Paper Series, Working Paper 30074 Autor DH (2022) The labor market impacts of technological change: from unbridled enthusiasm to qualified optimism to vast uncertainty, NBER Working Paper Series, Working Paper 30074
go back to reference Baumol W (1967) Macroeconomics of unbalanced growth: the anatomy of urban crisis. Am Econ Rev 57(3):415–426 Baumol W (1967) Macroeconomics of unbalanced growth: the anatomy of urban crisis. Am Econ Rev 57(3):415–426
go back to reference Camagni R (2024) Adam Smith: uncovering his legacy for regional science. Scienze Regionali 23(1):3–64 Camagni R (2024) Adam Smith: uncovering his legacy for regional science. Scienze Regionali 23(1):3–64
go back to reference Camagni R, Capello R, Lenzi C, Perucca G (2023) Urban crisis vs. urban success in the Era of 4.0 technologies: Baumol’s model revisited. Pap Reg Sci 102(3):589–612CrossRef Camagni R, Capello R, Lenzi C, Perucca G (2023) Urban crisis vs. urban success in the Era of 4.0 technologies: Baumol’s model revisited. Pap Reg Sci 102(3):589–612CrossRef
go back to reference Capello R, Lenzi C (2023) Automation and labour market inequalities: a comparison between cities and non-cities. NPJ Urban Sustain 3(1):56CrossRef Capello R, Lenzi C (2023) Automation and labour market inequalities: a comparison between cities and non-cities. NPJ Urban Sustain 3(1):56CrossRef
go back to reference Capello R, Lenzi C, Perucca G (2022a) The modern Solow paradox. In search for explanations. Struct Chang Econ Dyn 63:66–180CrossRef Capello R, Lenzi C, Perucca G (2022a) The modern Solow paradox. In search for explanations. Struct Chang Econ Dyn 63:66–180CrossRef
go back to reference Capello R, Caragliu A, Gerritse M (2022) Continuous vs. discrete urban ranks: explaining the evolution in the Italian urban hierarchy over five decades. Econ Geogr 98(5):438–463CrossRef Capello R, Caragliu A, Gerritse M (2022) Continuous vs. discrete urban ranks: explaining the evolution in the Italian urban hierarchy over five decades. Econ Geogr 98(5):438–463CrossRef
go back to reference Caselli M, Fracasso A, Traverso S (2021a) Globalization, robotization, and electoral outcomes: Evidence from spatial regressions for Italy. J Reg Sci 61(1):86–111CrossRef Caselli M, Fracasso A, Traverso S (2021a) Globalization, robotization, and electoral outcomes: Evidence from spatial regressions for Italy. J Reg Sci 61(1):86–111CrossRef
go back to reference Chancel L, Piketty T, Saez E, Zucman G et al (2022) World Inequality Report 2022. World Inequality Lab Chancel L, Piketty T, Saez E, Zucman G et al (2022) World Inequality Report 2022. World Inequality Lab
go back to reference Dauth W, Findeisen S, Suedekum J, Woessner N (2021) The adjustment of labor markets to robots. J Eur Econ Assoc 19(6):3104–3153CrossRef Dauth W, Findeisen S, Suedekum J, Woessner N (2021) The adjustment of labor markets to robots. J Eur Econ Assoc 19(6):3104–3153CrossRef
go back to reference De Propris L, Storai D (2019) Servitizing industrial regions. Reg Stud 53(3):388–397CrossRef De Propris L, Storai D (2019) Servitizing industrial regions. Reg Stud 53(3):388–397CrossRef
go back to reference De Vries GJ, Gentile E, Miroudot S, Wacker KM (2020) The rise of robots and the fall of routine jobs. Labour Econ 66(2):101885CrossRef De Vries GJ, Gentile E, Miroudot S, Wacker KM (2020) The rise of robots and the fall of routine jobs. Labour Econ 66(2):101885CrossRef
go back to reference Dottori D (2021) Robots and employment: evidence from Italy. Economia Politica 38:739–795CrossRef Dottori D (2021) Robots and employment: evidence from Italy. Economia Politica 38:739–795CrossRef
go back to reference Economist (2023), How to worry wisely about artificial intelligence, 20/4/2023. Last accessed 2 May 2023 Economist (2023), How to worry wisely about artificial intelligence, 20/4/2023. Last accessed 2 May 2023
go back to reference Feldman M, Guy F, Iammarino S (2021) Regional income disparities, monopoly and finance. Camb J Reg Econ Soc 14(1):25–49CrossRef Feldman M, Guy F, Iammarino S (2021) Regional income disparities, monopoly and finance. Camb J Reg Econ Soc 14(1):25–49CrossRef
go back to reference Fossen FM, Sorgner A (2022) New digital technologies and heterogeneous wage and employment dynamics in the United States: evidence from individual-level data. Technol Forecast Soc Chang 175:121381CrossRef Fossen FM, Sorgner A (2022) New digital technologies and heterogeneous wage and employment dynamics in the United States: evidence from individual-level data. Technol Forecast Soc Chang 175:121381CrossRef
go back to reference Frey CB, Osborne MA (2017) The future of employment: How susceptible are jobs to computerisation? Technol Forecast Soc Chang 114I:254–280CrossRef Frey CB, Osborne MA (2017) The future of employment: How susceptible are jobs to computerisation? Technol Forecast Soc Chang 114I:254–280CrossRef
go back to reference Kemeny T, Storper M (2020) The fall and rise of interregional inequality: explaining shifts from convergence to divergence. Scienze Regionali 19(2):175–198 Kemeny T, Storper M (2020) The fall and rise of interregional inequality: explaining shifts from convergence to divergence. Scienze Regionali 19(2):175–198
go back to reference Lanchester J (2023) “Putting the silicon in Silicon Valey” review of Chip war: the fight for the world’s most critical technology by Chris Miller (Simon and Schuster 2022), London Review of Books Lanchester J (2023) “Putting the silicon in Silicon Valey” review of Chip war: the fight for the world’s most critical technology by Chris Miller (Simon and Schuster 2022), London Review of Books
go back to reference McAfee A, Brynjolfsson E (eds) (2017) Machine, platform, crowd: harnessing our digital future. W W Norton & Co, London and New York McAfee A, Brynjolfsson E (eds) (2017) Machine, platform, crowd: harnessing our digital future. W W Norton & Co, London and New York
go back to reference McCann P (2020) Perceptions of regional inequality and the geography of discontent: insights from the UK. Reg Stud 54(2):256–267CrossRef McCann P (2020) Perceptions of regional inequality and the geography of discontent: insights from the UK. Reg Stud 54(2):256–267CrossRef
go back to reference OECD (2022) OECD Regions and Cities at a Glance 2022. Paris: OECD Publishing. https:// www.oecd -ilibrary.org/ urban -rural -and -regional -development/ oecd -regions-and -cities -at -a -glance -2022 14108660 -en (accessed June 2023) OECD (2022) OECD Regions and Cities at a Glance 2022. Paris: OECD Publishing. https:// www.oecd -ilibrary.org/ urban -rural -and -regional -development/ oecd -regions-and -cities -at -a -glance -2022 14108660 -en (accessed June 2023)
go back to reference Overman HG, Xu X (2022) Spatial disparities across labour markets. Oxf Open Econ 3:585CrossRef Overman HG, Xu X (2022) Spatial disparities across labour markets. Oxf Open Econ 3:585CrossRef
go back to reference Piketty T (2014) Capital in the twenty-first century. The Belknap Press of Harvard University Press, Cambridge, MACrossRef Piketty T (2014) Capital in the twenty-first century. The Belknap Press of Harvard University Press, Cambridge, MACrossRef
go back to reference Piketty T, Saez E (2003) Income inequality in the United States, 1913–1998. Quart J Econ 118(1):1–41CrossRef Piketty T, Saez E (2003) Income inequality in the United States, 1913–1998. Quart J Econ 118(1):1–41CrossRef
go back to reference Reed WR (2015) On the practice of lagging variables to avoid simultaneity. Oxf Bull Econ Stat 7(5):897–905CrossRef Reed WR (2015) On the practice of lagging variables to avoid simultaneity. Oxf Bull Econ Stat 7(5):897–905CrossRef
go back to reference Rodríguez-Pose A (2018) The revenge of the places that don’t matter (and what to do about it). Camb J Reg Econ Soc 11(1):189–209CrossRef Rodríguez-Pose A (2018) The revenge of the places that don’t matter (and what to do about it). Camb J Reg Econ Soc 11(1):189–209CrossRef
go back to reference Santos AM, Conte A (2024) Regional participation to research and innovation programmes under next generation EU: the Portuguese case. Pap Reg Sci 103(1):1–14CrossRef Santos AM, Conte A (2024) Regional participation to research and innovation programmes under next generation EU: the Portuguese case. Pap Reg Sci 103(1):1–14CrossRef
go back to reference Vendrell-Herrero F, Bustinza OF (2020) Servitization in Europe. In: De Propris L, Bailey D (eds) Industry 4.0 and regional transformations. Routledge, London, pp 24–41CrossRef Vendrell-Herrero F, Bustinza OF (2020) Servitization in Europe. In: De Propris L, Bailey D (eds) Industry 4.0 and regional transformations. Routledge, London, pp 24–41CrossRef
Metadata
Title
Unveiling the automation—wage inequality nexus within and across regions
Authors
Roberta Capello
Simona Ciappei
Camilla Lenzi
Publication date
02-11-2024
Publisher
Springer Berlin Heidelberg
Published in
The Annals of Regional Science / Issue 4/2024
Print ISSN: 0570-1864
Electronic ISSN: 1432-0592
DOI
https://doi.org/10.1007/s00168-024-01317-7