Sage Journals: Discover world-class research

Abstract

This article explores the effects of increasing robot adoption on workforce composition, wages and wage inequality in the manufacturing sector. Using longitudinal data from the German Institute for Employment Research (IAB), industrial robot sales data and survey data from the IAB Establishment Panel, we examine the impact of robots on total employment and wage structures at the establishment level from 2008 to 2017. We find that while robotisation contributes to overall employment and wage growth, its effects vary across worker groups. High- and middle-skilled workers benefit more from employment and wage increases, whereas low-skilled and routine-intensive workers experience fewer gains. In contrast to skill-biased and task-biased technological change theories (SBTC and TBTC), robots do not significantly increase wage inequality within establishments. Instead, firms mitigate inequality, suggesting that organisational policies play a key role in shaping distributional outcomes. Works councils also influence wage dynamics, benefiting middle-skilled more than low-skilled workers.

Keywords

employment establishments industrial robots inequality technological change

Introduction

The integration of robots into industrial manufacturing has increased significantly in recent decades, transforming production processes across multiple sectors. This technological advancement has attracted significant scholarly attention, particularly in industries with high rates of robot adoption, where evidence points to a shift towards higher-skilled workers. However, relatively little attention has been given to the internal dynamics within firms, especially how these technologies reshape workforce composition and affect wages and wage inequality. This oversight is important because firms, as meso-level actors, can buffer external labour market trends, mediating or even decoupling from broader macro-economic processes (Thompson, 2013).

This article seeks to address this gap by examining how increasing robot adoption affects workforce composition, wages and wage inequality within establishments. We argue that market-based theories, such as skill-biased and task-biased technological change (SBTC and TBTC), only partially capture the internal dynamics at play. While robots do indeed shift workforce composition, their effects on firms’ internal wage structures are more nuanced, with firms adopting strategies to limit internal wage inequalities. These efforts are shaped by organisational imperatives, such as workforce cohesion, motivation and the avoidance of reputational damage, which are not fully accounted for by external labour market trends (Leete, 2000).

Our theoretical framework is grounded in the concept that robots substitute certain types of labour, leading to unequal opportunities for different worker groups to adapt to technological change (Boyd and Holton, 2017; Damelang and Otto, 2024; Parolin, 2020). While some studies have demonstrated labour displacement in specific regions and industries (Acemoglu and Restrepo, 2020; Dauth et al., 2021), the catastrophic job losses predicted by earlier automation research (Frey and Osborne, 2017) have largely failed to materialise. Instead, most research indicates a net positive effect of robot adoption on employment and wages (Klenert et al., 2023), although the distribution of these gains remains uneven across different worker groups (de Vries et al., 2020; Hötte et al., 2023). In particular, lower-skilled workers tend to benefit less from these productivity gains, raising concerns about widening inequality.

While much of the literature has focused on labour market outcomes, the role of establishments in shaping the consequences of robotisation remains underexplored. We argue that firms are crucial to understanding the social dynamics of robot adoption, as they control key decisions on technology investments, hiring, redundancies and wage policies (Baron and Bielby, 1980). Longitudinal studies at the firm level are therefore necessary to capture the intra-organisational effects of robots, which macro-level analyses may obscure (King et al., 2017; Müller, 2024).

Economically, robots may reshape employment composition by increasing demand for workers in complementary jobs while reducing the need for those in substitutable roles. This shift could lead to wage divergence and rising inequality, which is consistent with SBTC and TBTC theories (Acemoglu, 2002; Acemoglu and Autor, 2011; Katz and Murphy, 1992). However, firms do not always respond directly to market forces. Organisational research indicates that companies often decouple internal wage policies from external labour market trends because of concerns about fairness norms and reputation (Leete, 2000). Unequal wage structures can demotivate employees, harm morale and damage a firm’s reputation as a fair employer (Van Dierendonck and Jacobs, 2012). As a result, firms function as ‘inequality regimes’, actively working to contain internal wage disparities, even in the face of technological advancements such as robotisation.

This article contributes to the literature by analysing workforce composition, wage structures and intra-establishment wage inequalities in tandem. We move beyond traditional SBTC and TBTC frameworks by considering how collective organisational dynamics, such as the presence of works councils, moderate the impact of robotisation. Works councils, as worker representation bodies, play a critical role in wage bargaining and workforce protection, potentially limiting the extent of inequality within robotised firms (Doellgast, 2009; Streeck, 1997). Our empirical analysis focuses on German manufacturing establishments, drawing on longitudinal employer–employee data from the German Institute for Employment Research (IAB), supplemented with industrial robot sales data from the International Federation of Robotics (IFR).

By employing establishment-level fixed-effects (FE) models over the 2008–2017 period, this article sheds light on the nuanced role of robot adoption in shaping firm-level wage inequality and employment composition. Our findings not only contribute to the debate on automation and labour markets but also highlight the importance of firm-level institutional mechanisms in mitigating inequality.

Robots and firm-level outcomes: What do we know?

Compared with macro-level analyses, relatively few studies have examined the impact of robots at the firm level, and the existing research offers partly contradictory conclusions. While most studies suggest that the net effect of robot adoption on firm employment and average wages is positive (Acemoglu et al., 2020, 2023; Barth et al., 2020; Domini et al., 2020, 2021; Genz et al., 2021; Koch et al., 2021), others conclude that automation events slow down employment growth (Bessen et al., 2020) and lead to a higher separation rate of incumbent workers within firms (Bessen et al., 2025). Only a few studies provide deeper insight into inequality between subgroups of workers. These studies indicate that low-skilled (Acemoglu et al., 2023), middle-skilled (Dixon et al., 2021), or blue-collar and production workers (Acemoglu et al., 2023; Humlum, 2022) might be negatively affected by declining employment due to robot adoption. One central mechanism seems to be that while robots increase overall employment, they adjust the workforce composition by reducing the employment share of production workers (Acemoglu et al., 2020).

The impact of technology-induced productivity gains and changes in workforce composition on wage inequality remains somewhat unclear. Some studies show that robot adoption can contribute to rising wage inequality between skill groups (Barth et al., 2020; Bessen et al., 2025) and occupations (Acemoglu et al., 2023; Humlum, 2022). However, other studies find no significant impact of robots on wage inequality (Domini et al., 2020, 2021). These mixed results suggest that purely market-driven approaches such as SBTC and TBTC are insufficient to explain the effect of robots on wage inequality.

From an empirical perspective, most firm-level studies have compared robot-adopting and nonadopting firms to identify differences in employment outcomes and productivity (Acemoglu et al., 2023; Barth et al., 2020; Genz et al., 2021; Humlum, 2022; Koch et al., 2021). As Bekhtiar et al. (2024) reported, this comparison can lead to an overestimation of the impact of robots on productivity and wages, as differences between robot-adopters and nonadopters are partly influenced by sample composition. An alternative approach is to examine not only the first implementation but also the impact of quantitative increases in robot investment in firms that are already using robots.

In this article, we focus specifically on the case of Germany, which presents a particularly interesting context for analysing organisational and institutional determinants. Germany leads Europe in industrial robot density, especially in the automotive and machinery sectors, which account for a large share of the country’s robot usage (IFR Statistical Department, 2017). Consistent with international findings, research shows that robot adoption in Germany positively impacts productivity and overall employment while shifting demand towards more skilled workers (Dauth et al., 2021). However, some studies emphasise the uneven effects across industries and worker groups, with low-skilled workers and those in routine-intensive occupations (RIOs) facing greater risks of displacement (Dengler and Matthes, 2018). What makes Germany particularly noteworthy is its strong vocational training system, which provides robust support to middle-skilled workers, and its system of works councils, which offer institutionalised collective representation at the establishment level. Moreover, we observe works councils in many firms serving as the employees’ collective voice. How these structures mitigate the effects of robots within firms remains an open question.

Effects of robots on within-establishment workforce composition and wages

The theoretical discussion about the impact of new technologies on the workforce is largely dominated by two market-based approaches: the concept of skill-biased inequality, which argues that workers’ ability to adapt to new technologies depends on their skill set and educational attainment (Acemoglu, 1999, 2002; Goldin and Katz, 2008), and the task-biased inequality concept, which states that the routine intensity of occupations is a key factor in determining the likelihood of technical substitution (Acemoglu and Autor, 2011). Both approaches suggest that workers whose productivity is not enhanced by robotic production will be less in demand, leading to lower wages and higher unemployment risks. However, these approaches, which focus primarily on the broader labour market, overlook the fact that establishments can decouple from labour market trends to some extent (Bewley, 1999; Frank, 1984).

We argue that firms will specifically mitigate the effects of robots on wages, independent of workforce composition changes. Additionally, collective strategies, such as works councils representing workers within establishments, may help mitigate the impact of robots, particularly for lower-skilled workers.

Labour demand and wage levels

Two factors must be considered to predict how an increase in robots will affect overall employment in firms: productivity gains and labour substitution. Robots significantly boost productivity in manufacturing (Graetz and Michaels, 2018) and can increase revenue for robot-firms using them (Koch et al., 2021), potentially driving higher labour demand and firm growth. However, robots may also replace human labour (Frey and Osborne, 2017), which could reduce employment in these establishments. Since both mechanisms can occur simultaneously, which effect prevails remains an empirical question. However, new technologies generally create strong demand for workers with relevant skills and often lead to significant productivity gains, which is consistent with the majority of existing studies (Hötte et al., 2023). Therefore, we expect increased robot usage to have a positive, rather than negative, impact on total employment within establishments:

H1 : As the share of robots increases, total employment in the establishment increases.

In addition to their positive impact on employment, we expect robots to increase average wages. Robots enhance productivity, leading to higher firm output (Humlum, 2022; Koch et al., 2021). In a competitive system, this gain at the firm level should also translate into higher wages for workers (Barth et al., 2020). Consequently, we hypothesise the following:

H2 : As the share of robots increases, the average wages of all workers within the establishment increases.

Workforce composition

Despite the anticipated growth in net employment, employment opportunities are likely to be unevenly distributed across different worker groups. Establishments are expected to adjust their workforce composition to meet the changing needs brought about by increased robot use through mechanisms such as hiring, dismissals, or retirements. As a result, the employment share of more productive workers should increase, whereas the share of less productive workers more exposed to technological change should decrease.

The SBTC approach assumes that technological investments increase job requirements, increasing demand for skills and human capital (Acemoglu, 2002; Katz and Murphy, 1992). The canonical model often compares higher-skilled workers (with a college degree) and lower-skilled workers (Acemoglu and Autor, 2011). This model suggests that high-skilled workers adapt more quickly to new technologies, creating a productivity advantage that leads to greater inequality. However, this model may not fully capture the complexities of the German vocational training system (Deissinger, 2015). Dividing non-college workers into middle-skilled (vocational degree) and low-skilled (no degree) workers provides a more nuanced understanding.

In summary, SBTC argues that demand for high-skilled workers will increase, whereas low-skilled workers and, to some extent, middle-skilled workers are more likely to be replaced by robots. This leads to the following hypothesis:

H3a: As the share of robots increases, the share of lower-skilled compared with high-skilled workers within the establishment decreases.

In contrast, the TBTC concept emphasises differences in task groups as the main factor driving inequality (see also Gil-Hernández et al., 2024). Robots, AI and other technologies can substitute for some tasks but are less capable of automating complex tasks such as interactive or analytical tasks (Acemoglu and Autor, 2011; Autor et al., 2003). Routine tasks, particularly manual and cognitive tasks, are the most vulnerable to automation (Spitz-Oener, 2006). Occupations vary in their degree of automation potential (Josten and Lordan, 2019), and workers in RIOs are the most at risk. Dengler and Matthes (2018) identified occupations with more than 70% routine tasks as particularly vulnerable. Thus, as organisations adopt more automation, the demand for workers in RIOs should decline relative to that of non-RIO workers. TBTC leads to the following hypothesis:

H3b : As the share of robots increases, the share of RIOs compared with all other workers within the establishment decreases.

Wage inequality

We expect robots to positively impact overall employment and wages while also changing the composition of the workforce within establishments. This raises the question of whether these two mechanisms affect wage inequality and whether firms distribute the gains from robots equally or unequally among their employees.

The SBTC framework assumes that higher-skilled workers, in particular, adapt more effectively to new technology, leading to increased productivity for this group. In contrast, lower-skilled workers retained by the establishment may be reassigned to other low- or medium-skilled tasks that are not yet automated, resulting in little to no productivity gains for this group. Moreover, the decreasing demand for lower-skilled workers is expected to result in lower wages. Consequently, SBTC predicts the following hypothesis:

H4a : Wage increases from robot adoption are (relatively) lower for lower-skilled than for high-skilled workers.

Similarly, within the task-focused framework of TBTC, it is assumed that workers engaged in routine tasks benefit less from productivity gains because of increased robot use. This leads to the following TBTC-based hypothesis:

H4b : Wage increases from robot adoption are (relatively) lower for RIOs than for workers in non-RIOs.

In both cases, we would expect an increase in wage inequality within establishments, driven by lower productivity and reduced demand for certain groups of workers. However, when organisational factors are considered, traditional market mechanisms may not always accurately reflect the reality within firms. As the literature on wage fairness and employer reputation shows, firms are often concerned with internal wage structures, adhering to principles of fairness and equity in pay (Alves and Rossi, 1978). When workers compare their wages, establishments have an interest in preventing wage disparities between different worker groups to avoid dissatisfaction (Mohrenweiser and Pfeifer, 2023). To achieve this, wages can be decoupled from individual productivity and reallocated from more productive workers to less productive workers (Frank, 1984).

This does not mean that labour and product market conditions do not influence personnel policies, but firms tend to balance market forces with internal wage equity, thus mitigating the effects of external forces. Firms can be considered ‘inequality regimes’ where management can influence the level of inequality within the firm. From this perspective, increasing robot investments may not necessarily lead to rising wage inequality within establishments, as pure market mechanisms might suggest. We propose the following alternative hypothesis:

H5 : Contrary to H4, we do not expect effects of increased robot use on intra-firm wage inequality.

This effect can be further amplified by labour institutions such as unions or works councils, which help ensure that wages are not distributed too unevenly between different groups (Kristal and Cohen, 2017; Parolin, 2020). We focus specifically on the effects of works councils, which represent workers in an establishment. However, the formation of a works council is not automatic; it requires an election by the workforce. In establishments with a works council, the effect of robots on wages may be more moderated than in those without worker representation. Works councils may act as a counterbalance to management, advocating for worker rights, fair wages and job security, thereby reducing the class-based inequalities that typically emerge with technological change.

From a class-based perspective, works councils may limit the extent to which higher-skilled workers disproportionately benefit from robot adoption. Instead of widening wage gaps, works councils may encourage wage policies that ensure a more equitable distribution of productivity gains across different skill levels, helping to mitigate wage inequality between classes within the establishment. Several studies have highlighted the role of works councils in preventing wage polarisation. For instance, Streeck (1997) argued that collective bargaining and worker representation can impose limits on employers’ ability to introduce wage-differentiation strategies that benefit only the highest-skilled workers. Similarly, Doellgast (2009) showed that works councils help secure wage increases for lower-skilled workers, preventing the emergence of a two-tier wage system. Therefore, in establishments with works councils, we might expect lower wage polarisation due to robots, as these councils protect lower-skilled workers from the negative effects of automation, such as job displacement and wage stagnation. This leads to our final hypothesis:

H6 : In establishments with works councils, increased robot use leads to higher wages and fewer staff cuts for lower-skilled workers.

Data and estimation strategy

Data and sample

For our empirical analysis, we aim to identify the impact of increasing robot usage on different establishment-level outcomes. Specifically, we analyse (1) total employment, (2) employment shares, (3) average wages across different worker groups, and (4) within-establishment wage inequality between these groups. We use longitudinal employer–employee data provided by the IAB, supplemented with information on the operational stock of industrial robots at the industry level and data on the task structure of various occupations. Below, we describe the datasets and sample used in our empirical analysis.

We analyse the effects of robots at the establishment level, defined as the physical location where a firm’s business activities occur. Our main data source for measuring changes in workforce composition, wages and wage inequality within establishments is the Linked-Employer-Employee Data (LIAB) from the IAB. The LIAB combines process-generated person-level data collected by the German Federal Employment Agency with establishment-level data from the IAB Establishment Panel (IAB-EP), a large-scale German establishment survey (Schmidtlein et al., 2019). The key advantage of the LIAB is that it includes information on wages, employment status and educational attainment for all workers liable to social security who are employed in establishments that participated in the IAB-EP between 2008 and 2017 (Schmidtlein et al., 2019). This structure allows us to track changes in employment, workforce composition and wage structures over a 10-year period. Additionally, the survey data enrich the administrative data with background information.

We measure changes in robot usage via well-established data (Acemoglu and Restrepo, 2020; Dauth et al., 2021; Graetz and Michaels, 2018) from the World Robotics Survey (WRS) conducted by the IFR. The WRS is an annual survey of robot suppliers, covering approximately 90% of the global robot market (Dauth et al., 2021; IFR Statistical Department, 2017). It provides industry- and country-specific information on annual robot sales and the operational stock of industrial robots. In this study, we focus on the operational stock of robots in different manufacturing subindustries as a key measure for increasing robot usage. The focus on manufacturing has the advantage that most manufacturing firms have already adopted robots, which reduces the endogeneity bias (Bekhtiar et al., 2024). To avoid overestimating the share of robots in high-employment industries, we weight the operational stock by 1000 workers in the corresponding industry via employment data from the Federal Statistical Office. Additionally, we harmonise the industry classification codes between the WRS and LIAB data (cf. Damelang and Otto, 2024; Dauth et al., 2017), giving us annual robot shares for 53 manufacturing subindustries.

An important distinction in our study is between RIO workers and non-RIO workers. To calculate the routine intensity of occupations, we use the BERUFENET expert database, which provides an overview of the occupational tasks in nearly all German occupations. Dengler and Matthes (2018) classified tasks into five main dimensions on the basis of expert ratings: analytic, interactive, manual non-routine, and manual and cognitive routine tasks. Following this approach, we define all occupations in which more than 70% of tasks are potentially substitutable (manual and cognitive) as RIOs. This specification reflects the idea that workers in occupations with a high share of replaceable tasks have more difficulty adapting to increased automation.

In our analytical sample, we exclude nonmanufacturing establishments and those with fewer than 10 full-time workers, as employment outcomes in these establishments are highly susceptible to outliers. Our final annual panel dataset covers the period from 2008 to 2017, including an average of 2060 establishments per year in the manufacturing sector, representing approximately 621,500 workers.

Measures

As a core measure of changes in robot usage, we rely on the operational stock of industrial robots per 1000 workers in an industry. This industry-specific measure allows us to quantify changes in robot shares across all 53 manufacturing subindustries in our sample. However, the initial measure from the WRS provides no information on variation between establishments within the same subindustry. It does not account for the fact that some establishments invest more heavily in robots, whereas others do not use robots at all. For our analysis of within-establishment inequality, we need an approximation of the distribution of new robot units across establishments within each subindustry. We address this by creating an establishment-level weight via survey-based information from the IAB-EP.

The weight consists of two parts: the first part accounts for whether an establishment uses industrial robots. The second part weights the operational stock of robots on the basis of the total investment made by each establishment compared with the median investment of other establishments in the manufacturing sector.

For the first part, we use the IAB-EP data on whether an establishment uses ‘program-controlled means of production (e.g. industrial robots)’ to identify establishments that do not use robots. Unfortunately, this variable is only available in the 2017 wave of the IAB-EP. To solve this problem, we apply a multiple imputation (MI) procedure (e.g. Johnson and Young, 2011), estimating the probability of robot usage on the basis of the other variables in the IAB-EP that are highly correlated with robot usage in 2017 and available across all panel waves (supplemental material Table S1). After imputation, approximately 67% of the establishments in the full sample were classified as ‘robot-using’ (Table S2).

For the second part, we used annual information from the IAB-EP on total investments in euros. Since robot investments are cost intensive, we assume that establishments with higher overall investment levels are more likely to increase their robot share than establishments that invest less. To capture medium-term investment trends, we calculated the total investments per 1000 workers over the last three years. Each establishment’s investments were then compared with the median investments of all manufacturing establishments in the same period. The weight takes values between 0 and 1 if an establishment’s investments are below the median, and values greater than 1 if its investments exceed the median. Thus, when allocating the robot share in subindustries to the establishment level, we underweight the share for establishments with below-average investments and overweight it for those with above-average investments. In summary, the establishment weight w_est can be described as follows:

$\begin{array}{l} w_{e s t} = r o b o t u s a g e (0 / 1) * \frac{o v e r a l l i n v e s t m e n t s e s t a b l i s h m e n t / 1000 w o r k e r s}{m e d i a n i n v e s t m e n t s a l l e s t a b l i s h m e n t s / 1000 w o r k e r s} \end{array}$

In summary, our final robot investment proxy consists of the share of industrial robot units in each subindustry per 1000 workers, multiplied by the establishment weight, w_est. To account for the exponential distribution of the measure (Figure S1), we log-transform the robot measure. Approximately 38% of the establishments either do not use robots or have not made any recent investments.

We measure changes in total employment by analysing changes in establishment size over time. The variable ‘establishment size’ includes all workers liable to the social security system. To test our hypotheses related to the SBTC and TBTC concepts, we also analyse changes in the total employment of low-skilled, middle-skilled and high-skilled workers, as well as the number of workers in the RIO and non-RIO groups.

An increase in the net employment of a group does not necessarily mean that the group is more strongly represented within the establishment. Therefore, we measure changes in workforce composition by analysing the employment share of low-skilled, middle-skilled and high-skilled workers, along with the share of RIO workers compared with non-RIO workers.

We measure average wages in an establishment by considering the inflation-adjusted daily wages of all full-time workers. Additionally, we analyse group-specific wages for our five groups of interest: low-skilled, middle-skilled and high-skilled workers, and workers in RIOs and non-RIOs. By comparing group-specific wage changes, we test the extent to which different types of workers benefit from increasing robot usage.

Finally, we measure relative within-establishment wage inequality between our groups of interest. We calculate the skill wage gap by dividing the difference between the average wages of high-skilled workers and the average wages of low- or middle-skilled workers by the average wages of high-skilled workers. Similarly, we calculate the task wage gap by dividing the difference between the average wages of non-RIO workers and RIO workers by the average wages of non-RIO workers. Both measures of relative wage inequality indicate the percentage by which the less vulnerable group earns more than the potentially more vulnerable group.

In all the models, we account for wage censoring, which affects high incomes above a certain threshold in the administrative data. We follow other studies (King et al., 2017) and use a standardised method of wage imputation (Schmucker et al., 2018).

Control variables

In addition to the increase in robot investment, other potential factors at the establishment and industry levels can influence wage inequality and explain changes in organisational composition. Thus, we control for additional factors at the establishment and industry levels. First, we account for the influence of labour institutions by including works council presence and collective wage agreements from IAB-EP data. Second, to capture changes in the industry’s economic conditions, we use annual revenue data from a German Federal Office of Statistics survey (2020), ensuring that our measured effects reflect robot usage rather than broader industry trends. Third, except in overall employment models, we control for changes in workforce composition, including establishment size, mean age and shares of women and foreign workers

Methodological approach

The empirical analysis proceeds as follows. First, we estimate the impact of changes in the robot share on total employment and on employment within different subgroups, including high-, middle- and low-skilled workers, as well as RIO workers and non-RIO workers. In these models, we use a reduced set of covariates, excluding workforce composition controls, to show the unadjusted impact of robots on employment. As we have used a MI approach to identify establishments that do not use robots, we estimate a range of 20 potential effects for different possible outcomes of the MI for all FE models and calculate the average coefficients across these models. Consequently, the uncertainty from the MI is also taken into account when calculating the standard errors (Wäsche, 2024).

Next, we analyse how an increase in our robot investment proxy variable affects the employment shares of all skill and task groups. In these models, we include workforce composition controls, as changes in employment shares may be related to other structural changes within the establishment that are not caused by robots.

In the third step, we analyse the mean inflation-adjusted wages of all workers, along with the mean wages of high-, middle- and low-skilled workers, as well as the relative skill wage gaps between them. Similarly, we examine the wages of RIO workers and non-RIO workers and the relative task wage gap between the two groups. In the models analysing the impact of robots on the wages of different subgroups, the number of observations in the FE models is lower than that in the full sample (N = 16,086). This reduction is due to the exclusion of establishments without information on the average wage for the respective subgroup, as they do not employ at least one person from that group.

Our central explanatory variable is the log-transformed share of industrial robots per 1000 workers in an industry, weighted by the establishment weight (the ‘robot investment proxy’). We analyse the impact of a percentage increase in our robot investment proxy variable on the dependent variables. Since we are interested in absolute changes in workforce composition and wages, we primarily use models with a linear-log specification. For total employment, we switch to log-log models to interpret log-transformed dependent variables, as establishment sizes in our sample vary considerably.

One potential issue with our approach is that the effects of new robots may take time to materialise. Here, our medium- and long-term perspective is particularly valuable, as we include the lagged effect of our robot investment proxy in all of our FE models. Comparing the impact of robots in period t with that in period t – 1 allows us to address the issue of timing and to understand whether establishments adjust their employment structure and wages in the year of the investment (direct or preceding effects) or rather in the following year (time-lagged effects).

Findings

Descriptive findings

We begin with a brief descriptive overview of our data. In Table 1, we compare the employment and wage structure of establishments with different levels of robot usage from 2008 to 2017. For this purpose, we divide establishments into three groups: first, those that do not use robots; second, less robotised establishments that use robots but fall in the lower half of our robot investment proxy; and third, highly robotised establishments with values above the median of our robot investment proxy. As shown in Table 1, highly robotised establishments are, on average, substantially larger (568 workers) than those with lower (282 workers) or no (106 workers) robot utilisation. However, aside from these differences in employment size, the establishments are very comparable in terms of workforce composition. We find only minor differences in the employment shares of low-skilled and workers in RIOs across establishments. Both potentially vulnerable groups are slightly more represented in highly robotised establishments than in those with few or no robots, suggesting that a significant proportion of these workers are still employed in these establishments.

Table 1.

Descriptives establishments.

	Non-robot using	Less robotised	Highly robotised	All establishments
Establishment characteristics
Establishment size	106.52	282.18	568.34	302.82
Weighted robots / 1000 workers	0	5.13	42.46	14.58
Works councils	35.6%	55.5%	59.8%	49.2%
Collective agreement	34.4%	44.7%	50.1%	42.4%
Employment shares
High-skilled	11.5%	12.8%	10.6%	11.6%
Middle-skilled	78.1%	76.5%	77.8%	77.5%
Low-skilled	9.0%	9.7%	10.6%	9.7%
Non-RIO	64.7%	63.0%	60.6%	62.9%
RIO	35.3%	37.0%	39.4%	37.1%
Daily wages (€)
High-skilled	125.33	141.90	146.00	137.34
Middle-skilled	79.93	88.95	90.19	85.91
Low-skilled	72.31	78.41	79.62	76.85
Non-RIO	92.38	104.47	105.69	100.26
RIO	77.742	85.201	83.726	82.085
Observations	N = 7820	N = 6491	N = 6286	N = 20,597

Note: RIO: routine-intensive occupation.

In terms of inflation-adjusted average wages, we observe substantial differences between establishments that use robots and those that do not. As expected, robot utilisation appears to be positively correlated with the economic success of the establishment, leading to higher wage levels for all employees. Additionally, for most groups of workers, average wages are slightly higher in highly robotised establishments than in less robotised establishments. Notably, however, RIO workers are the only group with lower average wages in highly robotised establishments than in less robotised establishments (approximately €1.5 less per day). This descriptive finding offers an initial indication that not all worker groups may benefit equally from a greater share of robots.

Finally, it is noteworthy that in highly robotised establishments, both the skill and task wage gaps are already greater than those in less robotised and non-robot-using establishments (Figure 1). High-skilled workers in these establishments earn 34% more than middle-skilled workers do and 45.4% more than low-skilled workers do, which is 5.7 to 4.9 percentage points more than in non-robot-using establishments and 1.5 to 0.9 percentage points more than in less robotised establishments. Similarly, the wage gap between non-RIO and RIO workers in highly robotised establishments is 3.3 percentage points greater than that in non-robot-using establishments and 1.1 percentage points greater than that in less robotised establishments. All the differences are statistically significant (5% level). These findings suggest that relative wage inequality between groups that are more and less vulnerable to technological change is already greater in establishments that have made greater investments in robots in the past. However, the extent to which these differences are due to selection effects (e.g. more productive firms investing in robots) or the increasing use of robots remains an open question. To address this question, we rely on FE models, which are presented in the next section.

Figure 1.

Skill and task wage gaps by degree of robot utilisation.

Multivariate findings

The multivariate analysis first examines the impact of an increasing robot share on changes in total employment and the employment shares of each skill and task group within establishments (Figure 2). The initial FE model indicates a positive relationship between our robot measure and overall employment within establishments, which is consistent with H1. A 1% increase in the robot share during period t and preperiod t – 1, as measured by our proxy variable, leads to a significant increase in total employment (‘all workers’) of slightly less than 0.01%. Given an average growth of 25% in the robot share between 2008 and 2017 in manufacturing, an establishment with 1000 workers and average investments has grown by approximately two workers over the course of 10 years. Accordingly, robots have a relatively small but sustainable impact on employment growth at the establishment level.

Figure 2.

Fixed-effect models: Impact of the robot investment proxy in periods t and t − 1 on employment.

The subsequent FE models further analyse how robots affect the total employment of high-skilled, middle-skilled and low-skilled workers within establishments. The results indicate that the employment of high-skilled and middle-skilled workers increases as robot utilisation increases. In contrast, low-skilled workers do not seem to benefit from robots in terms of employment growth. This difference in absolute growth rates across skill groups is an initial indicator that increasing robot usage may lead to changes in workforce composition. Similarly, we compare the impact of robots on employment within establishments in our two task groups: non-RIO workers and RIO workers. Like the skill models, we find significant differences between these groups. With a rising share of robots, employment for non-RIO workers increases significantly. However, when robot investments in the same year and t – 1 are considered, the hiring of workers in the potentially vulnerable RIO group does not increase significantly. In summary, the positive effects of robots on overall employment seem to outweigh the negative effects. However, low-skilled workers and RIO workers do not benefit from these employment gains in establishments with increased robot deployment.

To investigate how these employment changes affect the relative workforce composition, we now analyse shifts in employment shares across skill and task groups. In terms of skill groups, robots appear to contribute to an increasing share of high-skilled workers within establishments, whereas the share of low-skilled workers declines. Robot investments made in the pretreatment period t – 1 seem to have a significant effect on the growth of high-skilled worker shares and the decline of low-skilled worker shares, suggesting that robot deployment takes time to increase demand for skilled labour. However, the proportion of middle-skilled workers remains stable, which is consistent with our expectations of H3a. Furthermore, these findings suggest that the shifts in employment predicted by the SBTC concept occur mainly between high-skilled and low-skilled workers, with minimal impact on middle-skilled workers with vocational degrees.

In terms of task groups, the coefficients indicate a trend: as the robot share increases, the proportion of RIO workers decreases relative to non-RIO workers. This trend aligns with our expectations from H3b, suggesting a decline in the relative demand for workers in occupations with more than 70% substitutable tasks. However, owing to the large heterogeneity within the RIO group and the conservative significance test used, the coefficients do not differ significantly from zero.

Subsequently, we analyse the impact of robots on average wages within establishments and on relative wage inequality between different subgroups (Figure 3). The first FE model shows that an increasing robot share has a positive effect on the inflation-adjusted daily wages of all workers in an establishment, which is statistically significant at the 0.1% level. The wage gains from robots beyond inflation adjustments are generally modest, particularly if we compare them with previous studies that compare robot adopters with nonadopters (Acemoglu et al., 2023; Humlum, 2022). To some extent, this can be explained by our research design, as we mainly examine the effect of additional robots in already robotised establishments. Another possible explanation for these relatively modest wage gains is that additional robots may generate delayed spillover effects that manifest in wage gains in subsequent periods. Empirical evidence for this assumption can be found when examining the coefficient of robot investments in the preperiod t – 1. Here, we observe that robot deployment at t – 1 has an additional effect on average wages. Overall, the findings support H2, indicating that robots increase average wages in establishments, although the wage gains are small.

Figure 3.

Fixed-effect regressions: Impact of the robot investment proxy in periods t and t − 1 on wages.

We continue analysing the mean wages of the different subgroups. The FE models indicate that the inflation-adjusted wages of all skill groups – high-, middle- and low-skilled workers – increase with a greater share of robots. However, contrary to H4a, we find no significant difference in average wages; all skill groups seem to benefit equally from an increase in robots within the establishment. Accordingly, there is no evidence that only high-skilled workers experience a wage premium as the share of robots increases. When comparing routine and non-routine jobs, we observe a wage premium for the latter when considering the effects of robot investments in the preperiod t – 1. However, there are no significant wage differences between the two groups when comparing the estimates, which contradicts H4b. Additionally, as shown in the lower part of Figure 3, in terms of relative wage inequality, we find no statistically significant effects of robots on either the skill wage gap between high-skilled and middle-skilled workers or high-skilled and low-skilled workers or the task wage gap. This absence of an increase in wage inequality supports our assumption in H5 that the pure market forces proposed by the SBTC and TBTC economic approaches only partially capture the real dynamics within establishments. Instead, the finding aligns with the idea that firms seek to limit internal wage inequality by reallocating wages across skill and task levels. Particularly for robot-using firms, which already start with greater inequality, allocating the additional gains solely to high-skilled workers would be challenging. Therefore, contrary to demand-drive explanations, even less-qualified workers profit to some extent from the overall productivity gains.

However, the extent to which collective bargaining mechanisms contribute to this finding remains an open question. This is addressed in H6, which posits that in establishments with works councils, increased robot use would lead to higher wages and fewer staff cuts for lower-skilled workers than in establishments without works councils. However, this hypothesis is only partially supported, as the results present a more nuanced picture.

First, no significant difference was found between establishments with and without works councils regarding the effect of robots on total employment (Figure 4). In establishments without works councils, the increase in employment due to robots is significantly different from zero, whereas in those with works councils, no such significant change is observed. The effect is significantly different between establishments with and without works councils (0.0290, p > t = 0.034), indicating that works councils do not appear to influence overall employment levels in response to increased robot use.

Figure 4.

Employment models split into establishments with and without works councils.

When examining the workforce composition, we observe that in establishments with works councils, the employment share of high-skilled workers differs significantly from that of low-skilled. Specifically, the share of low-skilled workers decreases, whereas the share of high-skilled workers increases, with both effects being statistically significant (–0.0015, p > t = 0.010 and 0.0014, p > t = 0.006). The share of middle-skilled workers remains stable in both establishment types. This suggests that works councils may inadvertently support a shift towards a higher-skilled workforce.

In terms of wages, the data show that establishments with works councils experience significantly greater wage increases due to robot adoption than those without (Figure 5). High- and middle-skilled workers in these establishments benefit from significant wage increases, whereas the effect for low-skilled workers is not statistically significant. In contrast, in establishments without works councils, only moderate wage increases are observed, primarily for middle-skilled workers. This effect is significantly smaller than that of establishments with works councils (–0.3955, p > t = 0.004).

Figure 5.

Wage models split into establishments with and without works councils.

Regarding wage inequality (Figure 5), we do not observe any moderating effects of works councils on the impact of robots on wage inequality within establishments. This suggests that works councils do not play a role in mitigating the widening wage gap that may arise from technological advancements such as robotisation.

In summary, these results provide only partial support for H6. While works councils do not prevent the decline of low-skilled employment, they appear to facilitate a shift towards higher-skilled labour. This is likely because works councils in Germany tend to focus on representing middle-skilled workers, whose employment share remains constant and who benefit most from wage increases. However, works councils do not seem to moderate the effect of robots on wage inequality. Overall, works councils improve the situation for middle-skilled workers within establishments but not for low-skilled workers. The fact that the positive effects of robots are evident only for high- and middle-skilled workers partially supports the class-based approach to technological change. However, works councils, as institutions of collective action, do not prevent this effect.

Robustness checks

Figures S2 and S3 show that our findings are robust to other technological developments that may affect employment and wages such as investment in information and communication technology and, as a measure of increasing foreign trade, the share of business volume abroad. We also demonstrate that our findings remain robust when we exclude the highly robotised automotive industries and test the sensitivity of our results to establishment size. This was achieved by excluding small establishments with fewer than 100 workers (also Figures S2 and S3) and by including interaction effects between the establishment size and the robot coefficient (Figures S4 and S5). The new models indicate that the adverse effects of robots on low-skilled and RIO workers, in terms of employment and wages, are more prevalent in large establishments. Moreover, to examine whether changes in employment volume are driven by an increasing number of hires or a decreasing number of workers leaving the establishment, we conducted additional analyses focusing on worker flows. Figure S6 reveals a tendency for robots to increase inflow of workers, which appears to have a stronger explanatory effect than the slight decrease in the outflow of workers. Unfortunately, due to the relatively stable employment structure in Germany, where changes are minimal, we are unable to further analyse entries and exits separately for individual groups of workers, as this would quickly result in a problem of insufficient sample size. However, we can already see indications in the total employment models (Figure 2), as well as in the inflow and outflow models (Figure S6), that there do not appear to be substantial layoffs for any group of workers. This suggests that the selection bias related to employees who are laid off is likely to be relatively small compared with those who remain in the company. If redundancies have occurred, and thus a selection of workers has taken place, it likely happened during the initial stages of automation. To rule out biases from possible bad controls, we tested several model variants. In Tables S3 and S4, we excluded works councils and workforce composition controls to avoid distortions of timing or causal mechanisms of how robots operate through labour institutions. To address potential post-treatment bias, we estimated lagged models (Tables S5 and S6) using t − 1 values for robots and covariates. Tables S6 and S7 show interactions between robots and works councils only for establishments with stable council status (only comparing establishments with and without councils, but no change over time). Across all specifications, our main conclusions remain robust.

Discussion

Unlike many prior firm-level studies, this article systematically analyses the effects of increasing robot shares on workforce composition and wages within establishments. We focus on incremental changes in robot use within manufacturing establishments, going beyond the typical comparison of adopters versus nonadopters. This approach provides new insights into the mechanisms behind the effects of additional robots on labour within establishments. In line with previous studies (Acemoglu et al., 2023; Barth et al., 2020; Domini et al., 2020; Humlum, 2022), our results indicate that an increase in the share of industrial robots has predominantly positive net effects on overall employment and wages. As the share of robots increases, establishments experience employment growth and share some of the productivity gains with workers through higher average wages. These gains, however, appear to be smaller than the positive outcomes reported in earlier studies (Acemoglu et al., 2023; Humlum, 2022), which focused only on measuring differences between robot adopters and nonadopters rather than on the incremental changes within robot-adopting firms. Thus, our findings contribute to the literature by showing that, along with differences between adopters and nonadopters, subsequent automation stages in robot-adopting firms also exert a modest but long-lasting positive effect on employment and wages. This employment effect is not moderated by works councils.

However, while there are no substantial ‘losers’ in this process, we observe differences across qualification groups within establishments. In terms of employment, robots significantly increase the number of high-skilled, middle-skilled and non-routine workers. Low-skilled workers without vocational degrees and workers in potentially substitutable occupations experience no employment gains from increased robot use – but also no substantial decreases. Our findings suggest that concerns about widespread job displacement due to robots do not materialise at the establishment level (see also Dauth et al., 2021). Robots do not appear to significantly increase the outflow of workers, although there is a slight tendency for employment to decline among vulnerable skill and task groups. Nevertheless, robot investments lead to a gradual shift in workforce composition, with employment shares of high-skilled and non-routine workers increasing, whereas those of low-skilled and routine-intensive workers decline. Establishments are therefore gradually adapting their workforce structure to the changing conditions brought by increased robot use.

Regarding wage inequality across different skill groups, we find no evidence of increasing wage inequality as predicted by the SBTC or TBTC approaches. All skill groups appear to benefit from increased robot use, and the relative wage gap between high-, middle- and low-skilled workers remains stable. Coupled with a gradually changing workforce composition, this result aligns with findings that the rise in earnings inequality is driven primarily by between-workplace inequalities (Tomaskovic-Devey et al., 2020). Therefore, robots seem to contribute to this trend by increasing inequality between firms rather than within them.

The relatively stable skill and task wage gap may seem counterintuitive and inconsistent with SBTC and TBTC predictions. This could be due to firms’ interest in curbing internal wage inequality for reasons of motivation and reputation. Our results show that firms do not dismiss lower-skilled workers but also do not hire them, resulting in their declining share as higher-skilled groups expand. Nonetheless, those retained in the firm receive some share of the productivity gains from robot use, keeping wage inequality stable. This aligns somewhat with Parolin (2020), who concluded that labour institutions can protect workers against rising wage inequality, albeit at the cost of declining employment shares among workers susceptible to automation. However, works councils seem to protect middle-skilled workers more effectively than they do lower-skilled workers.

Several methodological limitations must be considered when interpreting these findings. First, the robot measure from the IFR only includes information on robot shares at the industry level. While we attempted to mitigate this limitation by merging industry-level data with establishment-level information in our robot investment proxy, precision issues remain. This could result in over- or underestimating robot deployments within establishments. Additionally, our FE models may be biased by the tendency for more productive firms to invest more in robots, making it difficult to completely rule out endogeneity. More productive firms will always necessarily invest more in robots to remain productive, and it is hardly possible to separate the two effects. Although our focus solely on manufacturing industries reduces possible bias (Bekhtiar et al., 2024), the causal interpretation of these models is limited.

Second, due to data limitations, we were faced with a trade-off: either strengthen the temporal ordering in the causal inference models (e.g. by including lagged covariates), which would have significantly reduced the number of observations, or maximise the generalisability of the models across a larger number of establishments. We ultimately chose a middle-ground approach and showed through robustness tests with strict temporal sequencing and smaller sample sizes that our main conclusions remain unchanged.

Third, our findings are only partially transferable to industries outside manufacturing and to other technological advances. It is likely that technologies such as service robots or AI will affect different industries and worker groups. For example, service robots may impact lower- and middle-skilled workers in manual roles in sectors such as healthcare and hospitality, whereas AI poses a potential threat to higher-educated workers by automating complex tasks.

Finally, there are follow-up questions our research design could not address. For example, our data do not provide insight into the inequality between firm owners and the workforce (Kristal, 2013). Firm owners and shareholders may use digital technologies to extend their power over workers (Spencer, 2017), capturing a larger share of productivity gains.

Conclusion

In contrast to Barth et al. (2020) and Bessen et al. (2025), but in line with a large body of sociological literature (King et al., 2017; Kristal and Cohen, 2017), we have demonstrated that technological investments do not necessarily lead to further increases in wage inequality within establishments. Instead, a rising share of robots tends to have positive effects on employment and wages. However, establishments are gradually adjusting their workforce composition to meet the changing demands of new technologies. In the medium to long term, these shifts may disadvantage workers in shrinking groups because of their lower relevance for the establishment’s personnel policies. Works councils, which often primarily represent middle-skilled workers, do not seem to counteract this trend.

As a result, policies aimed at reducing labour market inequality caused by technological shocks should focus more on inequality between firms. The long-term changes in within-establishment workforce composition and the desire to limit intra-establishment wage inequality suggest that addressing labour market inequality through differences between firms and worker selectivity may be more effective. From an educational policy perspective, raising the qualification levels of the population is generally advisable to prevent the growth of low-skilled and routine jobs.

From a theoretical standpoint, we provide only partial evidence for the market-based theories SBTC and TBTC. While both approaches correctly predict that low-skilled workers and those in RIOs will be less in demand, with the share of middle-skilled and high-skilled workers increasing, the demand-based models fail to fully explain the wage effects observed in our data. Instead, approaches that incorporate fairness norms, employer reputation and collective bargaining offer greater insight into the impact of increased robot use on the internal wage structure of establishments. Moreover, when SBTC is considered, it becomes clear that comparing only two skill groups, as typically done in the classical canonical model, is insufficient in the German context. The impact of robots on middle-skilled workers with vocational degrees differs significantly from that on workers without vocational degrees. In contrast to the findings of Dixon et al. (2021) for Canada and Bessen et al. (2025) for the Netherlands, we found that middle-skilled workers in Germany show a degree of resilience to shifts in robot use and even experience positive outcomes in terms of employment and wages. Given the persistent shortage of skilled workers in the German labour market, establishments appear unable to lose well-trained, middle-skilled workers, despite increasing robotisation. By focusing on robot stocks, our article captures only one – but important – subset of recent technological changes in the world of work. Future studies should therefore investigate whether other technological advances, especially the use of AI, can effectively mitigate skilled labour shortages in German establishments and to what extent these technologies differ in their effects from the findings on robots presented in this article.

Supplemental Material

sj-docx-1-wes-10.1177_09500170251351260 – Supplemental material for Robotisation and Workforce Dynamics: Analysing Employment and Wage Effects within Manufacturing Establishments

Supplemental material, sj-docx-1-wes-10.1177_09500170251351260 for Robotisation and Workforce Dynamics: Analysing Employment and Wage Effects within Manufacturing Establishments by Michael Otto and Martin Abraham in Work, Employment and Society

Footnotes

Funding

The authors received no financial support for the research,authorship,and/or publication of this article.

ORCID iD

Martin Abraham

Supplemental material

Supplemental material for this article is available online.

Michael Otto finished his PhD in Economics and Social Sciences at Friedrich-Alexander University Erlangen-Nürnberg in 2024. His research focuses on the use of new technologies in the labour market with a special focus on automation technology. In particular,he investigates how these technologies affect social outcomes such as the risk of unemployment and wage inequality.

Martin Abraham is Professor of Sociology and Empirical Research Methods at the School of Business,Economics and Society,Friedrich-Alexander University Erlangen-Nürnberg,Germany. His research interests are labour markets,occupations,co-operation in economic exchange,work and family and social inequality.

References

Acemoglu

(1999) Changes in unemployment and wage inequality: an alternative theory and some evidence. American Economic Review 89(5): 1259–1278.

Acemoglu

(2002) Technical change, inequality, and the labor market. Journal of Economic Literature 40(1): 7–72.

Acemoglu

Autor

(2011) Skills, tasks and technologies: implications for employment and earnings. In: Card

Ashenfelter

(eds) Handbook of Labor Economics. San Diego, Amsterdam: Elsevier, 1043–1171.

Acemoglu

Restrepo

(2020) Robots and jobs: evidence from US labor markets. Journal of Political Economy 128(6): 2188–2244.

Acemoglu

Koster

HRA

Ozgen

(2023) Robots and workers: evidence from the Netherlands. Working Paper 31009. Cambridge, MA: National Bureau of Economic Research.

Acemoglu

Lelarge

Restrepo

(2020) Competing with robots: firm-level evidence from France. American Economic Review Papers and Proceedings 110: 383–388.

Alves

Rossi

(1978) Who should get what? Fairness judgments of the distribution of earnings. American Journal of Sociology 84(3): 541–564.

Autor

Levy

Murnane

(2003) The skill content of recent technological change: an empirical exploration. The Quarterly Journal of Economics 118(4): 1279–1333.

Baron

Bielby

(1980) Bringing the firms back in: stratification, segmentation, and the organization of work. American Sociological Review 45(5): 737–765.

10.

Barth

Roed

Schone

, et al (2020) How robots change within-firm wage inequality. IZA DP No. 13605, August. Bonn: Institute of Labor Economics.

11.

Bekhtiar

Bittschi

Sellner

(2024) Robots at work? Pitfalls of industry level data. Journal of Applied Econometrics 39(6): 1180–1189.

12.

Bessen

Goos

Salomons

, et al (2020) Firm-level automation: evidence from the Netherlands. AEA Papers and Proceedings 110(1): 389–393.

13.

Bessen

Goos

Salomons

, et al (2025) What happens to workers at firms that automate? The Review of Economics and Statistics 107(1): 125–141.

14.

Bewley

(1999) Why Wages Don’t Fall During a Recession. Cambridge, MA: Harvard University Press.

15.

Bielby

Kalleberg

(1981) The structure of occupational inequality. Quality and Quantity 15(1): 125–150.

16.

Boyd

Holton

(2017) Technology, innovation, employment and power: does robotics and artificial intelligence really mean social transformation? Journal of Sociology 54(3): 1–15.

17.

Damelang

Otto

(2024) Who is replaced by robots? Robotization and the risk of unemployment for different types of workers. Work and Occupations 51(2): 181–206.

18.

Dauth

Findeisen

Südekum

, et al (2017) German robots – the impact of industrial robots on workers. IAB-Discussion Paper 30, Nuremberg: Institute for Employment Research.

19.

Dauth

Findeisen

Südekum

, et al (2021) The adjustment of labor markets to robots. Journal of the European Economic Association 19(6): 3104–3153.

20.

Deissinger

(2015) The German dual vocational education and training system as ‘good practice’? Local Economy 30(5): 557–567.

21.

Dengler

Matthes

(2018) The impacts of digital transformation on the labour market: substitution potentials of occupations in Germany. Technological Forecasting and Social Change 137(1): 304–316.

22.

de Vries

Gentile

Miroudot

, et al (2020) The rise of robots and the fall of routine jobs. Labour Economics 66(1): 101885.

23.

Dixon

Hong

(2021) The robot revolution: managerial and employment consequences for firms. Management Science 67(9): 5586–5605.

24.

Doellgast

(2009) Still a coordinated model? Market liberalization and the transformation of employment relations in the German telecommunications industry. Industrial & Labor Relations Review 63(1): 3–23.

25.

Domini

Grazzi

Moschella

, et al (2020) For whom the bell tolls: the effects of automation on wage and gender inequality within firms. Available at: https://ssrn.com/abstract=3701517 (accessed 1 September 2024).

26.

Domini

Grazzi

Moschella

, et al (2021) Threats and opportunities in the digital era: automation spikes and employment dynamics. Research Policy 50(7): 104137.

27.

Frank

(1984) Are workers paid their marginal products? The American Economic Review 74(4): 549–571.

28.

Frey

Osborne

(2017) The future of employment: how susceptible are jobs to computerisation. Technological Forecasting and Social Change 114(1): 254–280.

29.

Genz

Gregory

Janser

, et al (2021) How do workers adjust when firms adopt new technologies? IZA DP No. 14626, August. Bonn: Institute of Labor Economics.

30.

Gil-Hernández

Vidal

Torrejón Perez

(2024) Technological change, tasks and class inequality in Europe. Work, Employment and Society 38(3): 826–851.

31.

Goldin

Katz

(2008) The Race Between Education and Technology. Cambridge, MA: Harvard University Press.

32.

Graetz

Michaels

(2018) Robots at work. Review of Economics and Statistics 100(5): 753–768.

33.

Hötte

Somers

Theodorakopoulos

(2023) Technology and jobs: a systematic literature review. Technological Forecasting and Social Change 196(1): 122750.

34.

Humlum

(2022) Robot adoption and labor market dynamics. Study Paper No. 175. Berlin: Rockwool Foundation Research Unit.

35.

IFR Statistical Department (2017) World robotics. Industrial Robots 2016: 1–482.

36.

Johnson

Young

(2011) Towards best practices in analyzing datasets with missing data: comparisons and recommendations. Journal of Marriage and Family 73(5): 926–945.

37.

Josten

Lordan

(2019) Robots at work: automatable and non-automatable jobs. IZA Discussion Papers, No. 12520. Bonn: Institute of Labor Economics.

38.

Katz

Murphy

(1992) Changes in relative wages. The Quarterly Journal of Economics 107(1): 35–78.

39.

King

Reichelt

Huffman

(2017) Computerization and wage inequality between and within German work establishments. Research in Social Stratification and Mobility 47(1): 67–77.

40.

Klenert

Fernández-Macías

Antón

J-I

(2023) Do robots really destroy jobs? Evidence from Europe. Economic and Industrial Democracy 44(1): 280–316.

41.

Koch

Manuylov

Smolka

(2021) Robots and firms. The Economic Journal 131(638): 2553–2584.

42.

Kristal

(2013) The capitalist machine: computerization, workers’ power, and the decline in labor’s share within U.S. industries. American Sociological Review 78(3): 361–389.

43.

Kristal

Cohen

(2017) The causes of rising wage inequality. The race between institutions and technology. Socio-Economic Review 15(1): 187–212.

44.

Leete

(2000) Wage equity and employee motivation in nonprofit and for-profit organizations. Journal of Economic Behavior & Organization 43(4): 423–446.

45.

Mohrenweiser

Pfeifer

(2023) Wage structures, fairness perceptions, and job satisfaction: evidence from linked employer-employee data. Journal of Happiness Studies 24(7): 2291–2308.

46.

Müller

(2024) Technological change, training, and within-firm wage inequality in Germany. European Sociological Review 40(3): 450–463.

47.

Parolin

(2020) Automation, occupational earnings trends, and the moderating role of organized labor. Social Forces 99(3): 1–26.

48.

Schmidtlein

Seth

Umkehrer

(2019) Linked employer-employee data from the IAB: LIAB cross-sectional model 2 (LIAB QM2) 1993–2017. FDZ-Datenreport 06/2019 EN. Nuremberg: The Research Data Centre of the Federal Employment Agency in the Institute for Employment Research.

49.

Schmucker

Ganzer

Stegmaier

, et al (2018) Establishment history panel 1975–2017. FDZ Datenreport 09/2018. Nuremberg: The Research Data Centre of the Federal Employment Agency in the Institute for Employment Research.

50.

Spencer

(2017) Work in and beyond the Second Machine Age: the politics of production and digital technologies. Work, Employment and Society 31(1): 142–152.

51.

Spitz-Oener

(2006) Technical change, job tasks, and rising educational demands: looking outside the wage structure. Journal of Labor Economics 24(2): 235–270.

52.

Streeck

(1997) German capitalism: does it exist? Can it survive? New Political Economy 2(2): 237–256.

53.

Thompson

(2013) Financialization and the workplace: extending and applying the disconnected capitalism thesis. Work, Employment and Society 27(3): 472–488.

54.

Tomaskovic-Devey

Rainey

Avent-Holt

, et al (2020) Rising between-workplace inequalities in high-income countries. Proceedings of the National Academy of Sciences of the United States of America 117(17): 9277–9283.

55.

Van Dierendonck

Jacobs

(2012) Survivors and victims, a meta-analytical review of fairness and organizational commitment after downsizing. British Journal of Management 23(1): 96–109.

56.

Wäsche

(2024) Quantifying uncertainty for data imputation methods. Uncertainty Quantification Dictionary. Available at: https://dictionary.helmholtz-uq.de/content/data_imputation.html (accessed 19 August 2024).

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

1.20 MB