Research

My research asks how structural economic change reshapes democratic politics — and how institutions decide who wins and loses in the process. It proceeds on three connected fronts: the political economy of artificial intelligence, education and skill formation, and the politics of wealth and housing. Many projects sit at their intersections; each paper is listed once, under its centre of gravity.

AI and the Future of Work

Technological change runs through everything I do. I first worked on measurement: how to capture what workers actually do on the job, and why automation polarises employment in some countries but upgrades it in others. Generative AI has sharpened these questions — it reaches into cognitive work, spreads faster than institutions adapt, and is experienced first-hand by millions — and shifted my focus from labour-market outcomes to political ones. My current work asks what AI actually does to productivity and inequality, how the experience of working with it reshapes what people want from the state, and how firms weigh short-term automation gains against training the next generation of workers. Beyond my own papers, I invest in building this emerging field: I convene panels at major conferences (APSA, EPSA, SASE), organised a two-day Politics of AI workshop at Nuffield College in November 2025, and am editing a special issue on the topic in Comparative Political Studies.

Publications

Matthias Haslberger, Jane Gingrich, Jasmine Bhatia. 2025. Rage Against the Machine? Generative AI Exposure, Subjective Risk, and Policy Preferences. Journal of European Public Policy.
Replication Data

Abstract How does novel technology change public policy demands? Scholars interested in the effect of automation on policy preferences have commonly argued that exposure to automation technology increases subjective risk, which in turn predicts demand for insurance. Generative AI potentially challenges this dynamic. Based on a pre-registered online experiment with a sample of 1,041 UK working-age adults we show that direct exposure to generative AI in realistic work tasks does not increase subjective risk but does strengthen support for activating social policy. To understand this constellation of attitudes, we argue that exposure to technology may activate sociotropic preferences to support individuals who might be negatively affected by AI. Text analysis shows cautious optimism and thoughtful engagement with the implications of AI for work and social policy. Our findings suggest that the current uncertainty over the relative winners and losers from AI opens a window of opportunity to expand activating social policies.

Matthias Haslberger. 2022. Rethinking the measurement of occupational task content, The Economic and Labour Relations Review 33(1):178–199.
Replication Data

Abstract Which tasks workers perform on their jobs is critical for how technological change plays out in the labour market. This crucial insight sparked a large literature on routine-biased technological change which argues that routine occupations with a high share of repetitive and codifiable tasks are at risk of being automated. This paper makes the case for rethinking how we operationalise occupational task content. Based on survey data from 27 European countries between 2000 and 2015, I construct novel measures of routine task intensity and task complexity at the ISCO-88 2-digit level. Comparing them to existing operationalisations, I show that the proposed indices lead to improvements in several critical areas. The task dimensions have a straightforward theoretical interpretation as they capture the essence of the routine-bias and skill-bias arguments and are operationalised to better align theory and measurement. Furthermore, my indices create new opportunities for research by allowing researchers to analyse within-occupation change and country-differences in occupational task content. My paper can therefore contribute to a more sociologically informed understanding of technological change. The indices will benefit both sociologists and labour economists in investigating the nature of recent employment trends in Europe and formulating policies to deal with these challenges.

Matthias Haslberger. 2021. Routine-biased technological change does not always lead to polarisation: Evidence from 10 OECD countries, 1995–2013, Research in Social Stratification and Mobility 74 (August 2021).

Abstract This article deals with a central paradox in the occupational polarisation literature: most scholars accept that technological change is biased against routine-intensive occupations, but in many countries, we do not see the pattern of occupational polarisation that the theory usually predicts. I argue and show empirically using a dataset of 10 OECD countries between 1995 and 2013 that technological change is both routine-biased and skill-biased, but that the result of routine-biased technological change may be occupational upgrading rather than polarisation. This is due to differences in occupational routine-wage hierarchies: only where routine occupations cluster around the middle of the wage distribution are we likely to see polarisation. Where routine occupations are concentrated near the bottom of the wage hierarchy, upgrading occupational change is the norm. Based on research on the US, the former has been widely assumed, but it does not hold true in all countries. Overall, this article shows that much previous work on routine-biased technological change and polarisation was built on premises that do not travel well. This underscores the importance of comparative research for building and testing robust general theories.

Working Papers and Work in Progress

Matthias Haslberger, Jane Gingrich, Jasmine Bhatia. No Great Equalizer: Experimental Evidence on Productivity Effects of Generative AI Use in the UK Labor Market
(Invited to Resubmit at Research Policy)
SSRN Working Paper

Abstract An emerging consensus holds that generative artificial intelligence (AI) equalizes workers’ performance within tasks, reducing productivity differences across workers. Existing research has largely studied productivity within single occupational groups and task structures. Whether this equalizing pattern generalizes to the labor market at large remains unclear. Observed performance equalization within groups of workers is compatible with both increasing and decreasing inequality between groups. To distinguish these outcomes, we conducted a large pre-registered online experiment with a sample of the UK working age population which randomly assigned participants to treatments that encouraged or discouraged the use of ChatGPT and then asked them to complete a set of realistic work tasks. We find that ChatGPT use increased productivity in all tasks, with greater benefits observed in more complex and less ambiguous tasks. However, compression effects between tasks were limited. Moreover, ChatGPT use did not affect productivity differentials between gender, age, educational or occupational groups.

Striking the Balance? Firms’ Training Strategies in Response to Generative AI

Abstract Practical learning from experts is key to the mastery of almost any job. It constitutes the backbone of collective skill formation systems and is equally crucial in systems emphasising on-the-job learning. Yet, generative AI threatens the complementary relationship between master and apprentice, senior and junior worker: tasks performed by entry-level workers with limited experience tend to be particularly exposed to substitution by AI. If these positions are replaced, skill development is impeded and overall workforce skill levels are likely to suffer in the medium- to long-term. This creates a dilemma for profit-maximising firms: automation may boost short-term productivity at the expense of long-term productivity growth. This paper studies how firms in Germany and the US navigate this challenge. Building on the Varieties of Capitalism literature, I argue that differences such as higher employee turnover and a greater emphasis on shareholder value incentivise American firms to prioritise AI adoption, while German firms maintain a stronger focus on training younger workers even if it means foregoing short-term gains. I test this argument empirically using data on job openings in sectors that are exposed to or sheltered from AI. This allows me to trace differential rates of change between the groups of occupations and across countries in a) the balance between junior and senior vacancies and b) the task composition of junior roles. Empirical support for my argument highlights a hitherto underappreciated potential long-term consequence of generative AI that may undermine its productivity-enhancing effects.

Artificial Intelligence, Voice at Work, and Politics

Abstract Artificial intelligence (AI) is reshaping work and politics in myriad ways. The emerging literature on the politics of AI focuses largely on the direct effects of exposure to AI on expectations, political preferences, and behavior. I argue that this risks ignoring a key site of political preference formation: the workplace (Kitschelt and Rehm 2014). To what extent individuals have a say in how their work is organized and how new technologies are implemented is likely to shape their political reaction to the proliferation of AI. This chapter first summarizes the current literature on AI and politics before theorizing how voice at work shapes individuals’ political reactions to AI exposure and testing the hypothesized associations using original survey data from the United States and the United Kingdom.

Who Goes for Growth? Personality and Policy Priorities in the AI Economy
(with Jasmine Bhatia, Patrick Emmenegger, and Jane Gingrich)

Abstract Artificial intelligence is likely to generate substantial economic gains while distributing them unevenly. How do citizens navigate the resulting tension between growth and equality? We examine whether policy priorities in the AI economy depend on individuals’ likely exposure to AI and their personality. Using a representative US survey of 5,000 working-age adults, we combine Big Five personality measures and a behavioral measure of risk orientation with a highly personalized information treatment identifying respondents as likely winners or losers from AI. Contrary to preregistered expectations, we find largely sociotropic responses: likely winners shift toward prioritizing equality, while likely losers shift toward growth. Among losers, this shift is concentrated among respondents who are surprised by and find the treatment credible. Personality strongly structures baseline policy priorities, but generally does not moderate responses to personalized AI risk. Behavioral risk orientation is the main exception. The findings highlight a tension between stable psychological predispositions and dynamic responses to technological change.

Cognitive Traits and Expectations of AI’s Effects
(with Jasmine Bhatia, Patrick Emmenegger, and Jane Gingrich)

Abstract Who does the public expect to win and lose from generative AI? Using a survey of 5,000 employed Americans, we show that expectations are sharply stratified: the rich and highly educated are expected to benefit, while the poor, less educated, and older workers are expected to lose. Group members sometimes diverge strikingly from outside observers — young people rate their own generation's prospects far more pessimistically than others do, while high-income respondents are more optimistic about the rich. Because reactions to technological change are fundamentally reactions to uncertainty, we theorize that personality traits — as markers of risk orientation, optimism, and soft skills — shape these expectations alongside demographics. They do: extraverted and agreeable respondents expect groups across society to fare better, while neurotic and open respondents expect worse, holding demographics constant. Expectations about AI's distributive consequences are strongly personality-laden.

The New Politics of Artificial Intelligence
(with Thomas Kurer, Aina Gallego, and Nicole Wu)

Abstract In this introductory paper for the *Comparative Political Studies* special issue, we argue that the proliferation of AI creates a new politics that differs fundamentally from the routine-biased technological change (RBTC) paradigm of recent decades. We identify three main shifts: 1) AI creates new winners and losers, who are endowed with different power resources; 2) the political demands of these groups so far do not align with the populist response of the losers from RBTC; 3) the governance of AI is highly politically salient at an early stage before dramatic labour market effects have materialised, potentially opening a window of opportunity for proactive policy reform. We highlight the substantive implications of this paradigm shift, as well as the new challenges and opportunities this creates for political science research.

Obtaining Validated and Nuanced Occupational Information in Online Surveys

Abstract Online surveys often collect only coarse or imprecise occupational information from respondents, limiting the scope of possible analyses. I introduce a novel AI-assisted protocol that allows researchers to collect detailed information on occupational titles and core tasks and dynamically validate it directly in the survey. I benchmark the approach against existing procedures and show that for trivial cost, researchers can obtain much more nuanced occupational information.

Education and Skill Formation

Education systems are where societies decide how to prepare people for economic change. Much of my work centres on Europe’s most distinctive skill institution — dual vocational education and training — and documents a paradox: VET demonstrably delivers, compressing wage inequality and dampening the political anxieties that automation otherwise fuels, yet families across Europe increasingly treat it as a second-best option. I study both sides of this tension: how skill formation shapes workers’ life chances and what they demand from the welfare state, and why VET’s attractiveness keeps eroding even where it performs best. A complementary historical strand examines how partisan politics built modern secondary education in the first place.

Publications

Matthias Haslberger, Patrick Emmenegger, Niccolo Durazzi. 2026. The Missing Link: Technological Change, Dual VET, and Social Policy Preferences. British Journal of Political Science.
Replication Data

Abstract How does technological change affect social policy preferences? We advance this lively debate by focusing on the role of dual vocational education and training (VET). Existing literature would lead us to expect that dual VET increases demand for compensatory social policy and magnifies the effect of automation risk on such demands. In contrast, we contend that dual VET weakens demand for compensatory social policy through three non-mutually exclusive mechanisms that we refer to as (i) material self-interest; (ii) workplace socialization; and (iii) skill certification. We further hypothesize that dual VET mitigates the effect of automation risk on social policy preferences. Analyzing cross-national individual data from the European Social Survey and national-level data on education systems, we find strong evidence for our argument. The paper advances the debate on social policy preferences in the age of automation and sheds new light on the relationship between skill specificity and social policy preferences.

Patrick Emmenegger, Matthias Haslberger, Anna Wilson. 2026. Caught in a Downward Spiral? The Relative Attractiveness Deficit of Vocational Education and Training. Sociology of Education.

Abstract Vocational education and training (VET) is hailed for easing skill shortages and fostering inclusion. However, little is known about the factors influencing the choice between VET and general education. We conducted a vignette experiment with over 11,500 respondents in seven European countries, asking them to assign fictitious 15-year-olds either to VET or general education based on achievement, motivation and sociodemographic profiles. Respondents consistently channelled students with low grades and motivation into VET. This bias weakened, but persisted, among respondents who view VET as offering favorable labor-market prospects. Boys, working-class youth and adolescents outside large cities were also steered towards VET, although achievement effects outweighed ascriptive ones. These patterns hold across countries and respondent subgroups, indicating that VET is widely perceived as the less desirable educational pathway. Our findings suggest that VET is caught in a downward spiral in which the relative unattractiveness of VET and academic drift reinforce each other.

Patrick Emmenegger, Matthias Haslberger. 2026. Yesterday’s Model for Tomorrow’s Economy? Dual VET and Wage Inequality in the Knowledge Economy, Journal of European Social Policy 36(2):185–198.
Dual VET Dataset

Abstract Dual vocational education and training (VET) is said to have positive economic effects. However, recent contributions suggest that the rise of the knowledge economy may undermine these positive effects because university-educated workers are better suited for the new knowledge-intensive jobs. This paper provides the first evidence on the relationship between dual VET and wage inequality in mature knowledge economies. Using a new dataset on 37 advanced economies from 1996 to 2020, we find that dual VET is associated with lower levels of wage inequality. This negative association is particularly strong in the lower half of the wage distribution, which suggests that academically weaker students are the main beneficiaries of dual VET. Using three different indicators of the knowledge economy, we find, contrary to the fears often espoused in the literature, no clear evidence that the knowledge economy erodes this negative association between dual VET and wage inequality.

Matthias Haslberger, Scherwin M. Bajka. 2026. Subjective Technology Risk and Education Preferences: VET as a Safe Haven or Dead End?, Regulation & Governance 20(2):621–634.

Abstract Education equips individuals with valuable skills to protect them against employment risks associated with the digital transition. As scholars debate whether vocational education and training (VET) or general education better insures against technology-induced employment risk, we ask how this type of risk, as perceived by individuals, shapes their education preferences. Our analyses, based on a survey of over 11,500 respondents across seven European countries, show that VET is regarded as a safe haven by those perceiving heightened risk. This relationship remains robust when controlling for various alternative explanations and is consistent across countries. Subgroup interactions indicate that men, high-income earners, respondents with tertiary education, and those politically on the right more strongly favor VET in response to subjective technology risk. Hence, our study suggests that VET's practical, job-oriented focus is perceived as better protection against the growing uncertainty over skill demands in the twin transition than general education.

Anja Giudici, Jane Gingrich, Tom Chevalier, Matthias Haslberger. 2023. Center-right Parties and Post-War Secondary Education, Comparative Politics 55(2):193–218.

Abstract The massification of secondary schooling constitutes the key educational project of the first post-war period. However, the resulting educational structures differed in terms of streaming and standardisation. Despite their historical opposition, center-right parties contributed to shaping these reforms. They opposed standardisation because their distributive strategy rested on support from elites and middle classes. However, their stance on streaming varied. Centre-right parties supported streaming when they were linked to teachers and private providers who opposed comprehensive reforms, but supported de-streaming where such groups aligned with the left. The analysis suggests that common partisan distributive aims can materialize as varied public service reforms, due their intersection with the productive environment. This paper shows these outcomes by tracing reforms shaped by center-right parties in Bavaria, France, and Italy.

Working Papers and Work in Progress

VET Systems and Socio-Economic Outcomes
(with Patrick Emmenegger)

Abstract This chapter examines the effects of vocational education and training (VET) systems on socio-economic outcomes. VET plays a prominent role in most European education systems, often enrolling more than half of a cohort at the upper-secondary level. The chapter reviews and synthesizes existing evidence on the effects of VET across five socio-economic outcomes, drawing on both micro-level and macro-level research. These outcomes are: (i) educational inequality and stratification, (ii) earnings inequality, (iii) employment levels and type of employment, (iv) (youth) unemployment rates, and (v) the potential of VET for social integration. Across these domains, we pay particular attention to distributional effects that cut across these five socio-economic outcomes, especially with regard to gender and migration background. We synthesize evidence from different social science disciplines, as social policy, political science, sociology, and economics all highlight different socio-economic outcomes of VET systems. Overall, we find that VET systems have important effects on socio-economic outcomes, although these effects vary markedly across institutional designs---most notably between fully school-based and dual VET systems.

Wealth and Housing

As wealth outgrows income as the foundation of economic security, housing has become a political fault line in advanced democracies. My work in this area — much of it with the ERC project WEALTHPOL — examines how asset ownership shapes democratic voice and fiscal politics. Homeowners feel heard by the political system in ways renters do not, especially later in life, and homeowners’ opposition to wealth taxation — amplified by renters’ uncertainty about their own interests — helps explain why wealth taxes remain politically stuck despite record inequality and mounting fiscal pressure. Ongoing projects extend this agenda to perceptions of inequality and preferences over the tax mix.

Publications

Matthias Haslberger, Mads Elkjær, Ben Ansell. 2026. Homeownership and Political Efficacy: How Housing Wealth Shapes Whether People Feel Heard. West European Politics.

Abstract Why do some citizens feel that political institutions are responsive to people like them, while others do not? Existing research highlights the role of education and income in shaping external political efficacy, but little is known about the role of wealth. We argue that homeownership serves as a key marker of economic and social success, shaping perceptions of political inclusion, particularly in later life. While younger renters may still aspire to homeownership, older renters are more likely to confront the realisation that they will never make it onto the property ladder, undermining their efficacy. We test this argument using original survey data from over 10,000 adults in the United Kingdom. Our findings show that homeownership is positively associated with external political efficacy, particularly among older individuals. These results underscore housing wealth as a critical driver of political inclusion and highlight its broader implications for democratic legitimacy in unequal societies.

Mads Elkjær, Ben Ansell, Laure Bokobza, Asli Cansunar, Matthias Haslberger, Jacob Nyrup. 2025. Why Is It So Hard to Counteract Wealth Inequality? Evidence from the United Kingdom, World Politics 77(3):515–561.
Replication Data

Abstract It has long been established that education and income affect people's political efficacy. Surprisingly, the role of wealth has been largely neglected in this literature. In this paper, we argue that housing wealth performs an insurance function and is thereby associated with higher internal and external political efficacy. Using data from the UKHLS and a representative survey including an experiment that was administered in England and Wales, we document a sizeable and statistically significant positive association of housing wealth and perceived wealth with efficacy. However, this relationship is less robust to sample attrition than between efficacy and education or income. We furthermore investigate whether informing respondents about house price inequality affects their efficacy. Our information treatments show no effect on external efficacy, while the effect on internal efficacy depends on the respondent correctly understanding the information: comprehenders show higher efficacy and non-comprehenders exhibit lower efficacy, compared to the control group. This suggests that views of government responsiveness (external efficacy) are not easily manipulated, while for people's view of their own understanding of politics (internal efficacy), comprehension matters more than content of the information treatment, in accordance with self-efficacy theory.

Working Papers and Work in Progress

Matthias Haslberger, Mads Elkjær, Ben Ansell. The Electoral Politics of Wealth Taxation: Housing, Information, and Public Opinion
(Under Review)

Abstract Despite rising wealth inequality and mounting fiscal pressures, governments have largely refrained from raising taxes on wealth. We argue that a key reason is opposition to wealth taxes among homeowners, which has contributed to a political environment where stronger taxation of wealth is electorally challenging to implement. This dynamic is exacerbated by information asymmetries, which prevent low-wealth renters from formulating preferences aligned with their material self-interest, effectively handing the political initiative to homeowners. Utilizing original survey data from Denmark, France, Germany, Ireland, Italy, the Netherlands, and Sweden, we find empirical support for the argument. Housing wealth increases the likelihood of stating a preference on wealth taxation, and homeowners and their children oppose more progressive taxation of inheritances, net wealth, and capital gains. These findings help explain why, despite pronounced inequality in asset ownership and severe budgetary pressures, wealth taxation remains underutilized in advanced democracies.

Perceptions of Inequality in Europe
(with Mads Elkjær)

Housing Wealth and Tax Mix Preferences
(with Mads Elkjær)

Other Writing

Neil Lee. 2024. Innovation for the masses: How to share the benefits of the high‐tech economy. Reviewed for Social Policy & Administration.