Industrial robots have only modest links to job quality across Europe once controls are included, but they reduce physical risks and appear to narrow the gender gap in autonomy while women still face higher work intensity.

Main Finding

Conditional on rich individual controls and country–industry fixed effects, industrial robot exposure in Europe is only weakly associated with overall job-quality outcomes. However, effects differ by job-quality dimension and gender: higher robotisation is linked to lower physical-risk exposure for both sexes, a persistent female disadvantage in work intensity (which may widen with robot density), and a narrowing female–male gap in work autonomy as robot exposure rises.

Key Points

• Data and scope: Micro–macro linkage using EWCTS 2021 (individual-level working-conditions survey, ~70k respondents across 36 European countries) matched to IFR 2021 industrial-robot stock by country × industry.
• Job-quality outcomes: three binary indicators from EWCTS harmonised recodes — high work intensity, exposure to physical risks, and high autonomy.
• Robot measure: country–industry robot exposure (operational robot stock scaled by employment); author models both an extensive margin (any exposure: binary) and an intensive margin (log of positive exposure).
• Econometric approach: weighted logistic regressions with individual/job controls (age, education, contract type, full-time, public sector, tenure, workplace size, occupation), country and industry fixed effects, and standard errors clustered at country–industry level. Gender heterogeneity captured via interactions between gender and both robot-exposure components. Fairlie nonlinear decompositions quantify contributors to gender gaps.
• Descriptives: women slightly higher on work intensity (0.68 vs 0.65) and physical risks (0.29 vs 0.27); men higher on autonomy (0.51 vs 0.47). Large cross-country heterogeneity in robot exposure (e.g., Germany ~62.8 robots/10k workers; Serbia/Greece ~2).
• Main patterns from predicted probabilities:
  • Physical risks: decline with robot exposure for both sexes; profiles largely similar.
  • Autonomy: at low robot exposure women have lower predicted autonomy than men, but women’s autonomy increases with robot density while men’s remains flat → gap narrows.
  • Work intensity: women consistently show higher predicted intensity across exposure levels; men’s intensity tends to decline slightly with robotisation while women’s is stable or increases modestly → gap persists or widens.
• Decomposition: robot density explains little of absolute gender gaps in work intensity and autonomy; occupation, work arrangements, and macro-level factors (country/industry) explain substantially larger shares.

Data & Methods

• Data sources:
  • EWCTS 2021: ~70,000 respondents from 36 European countries; final calibrated sampling weights used.
  • IFR (World Robotics) 2021: operational stock of industrial robots by country × industry; employment from Eurostat (NACE Rev.2) used to compute exposure.
• Outcome construction: binary indicators for (1) high work intensity, (2) exposure to physical risks, (3) high autonomy — using EWCTS harmonised job-quality recodes.
• Robot exposure specification:
  • robots_excs = 1 if exposure_cs > 0 (extensive margin), 0 otherwise;
  • robots_exp_logcs = log(exposure_cs) if exposure_cs > 0, 0 otherwise (intensive margin).
• Baseline model (weighted logit): Pr(Y_ics = 1) = Λ(α + β1 robots_excs + β2 robots_exp_logcs + β3 Female_i + β4 (robots_excs × Female_i) + β5 (robots_exp_logcs × Female_i) + X_i'γ + μ_c + μ_s)
  • Λ = logistic CDF; X_i includes demographics, work arrangements, occupation, establishment size, etc.; μ_c and μ_s are country and industry fixed effects.
  • Standard errors clustered at country–industry level.
• Robustness/limitations acknowledged by author: cross-sectional design (causality limited), exposure measured at aggregated country–industry level (potential measurement error; unobserved within-industry heterogeneity), single year snapshot (EWCTS 2021 / IFR 2021).

Implications for AI Economics

• Broaden focus beyond employment and wages: robotisation (and by extension task‑automation technologies including AI) reshapes job quality in multi-dimensional ways — physical safety, autonomy, and workload — with gender-differentiated patterns that are not captured by employment/wage summaries alone.
• Automation can improve worker safety uniformly (lower physical risks), supporting policies that promote automation of hazardous tasks. This is a clear social benefit of (robot/AI) adoption.
• Distributional and gender effects are heterogeneous:
  • Women do not uniformly gain from robotisation in terms of workload; the persistent (and possibly widening) female disadvantage in work intensity suggests automation may recompose tasks in ways that leave women with high-intensity responsibilities (or increase monitoring/throughput pressures).
  • The narrowing autonomy gap implies automation can reallocate task content or elevate the scope of tasks where women gain relative autonomy — but gains are context-dependent.
• Policy implications:
  • Active labour and workplace policies should pair automation adoption with task redesign, workload management, and gender-sensitive implementation to avoid increasing unpaid/undervalued intensification for women.
  • Training and reskilling should be targeted not only by occupation/skill but by gendered task profiles to ensure equitable access to autonomy-enhancing opportunities.
  • Monitoring and evaluation of automation projects should track job-quality indicators (not only productivity/wages), disaggregated by gender.
• For AI economics research:
  • Need for longitudinal, firm- and task-level data to identify causal pathways (who performs which tasks pre- and post-adoption) and to separate technology effects from sorting and compositional changes.
  • Important to study complementarities between robots and AI (perception/prediction systems) — combined systems may amplify or alter the gendered patterns observed for industrial robots alone.
  • Decomposition evidence here indicates that occupations, work arrangements, and macro context matter more than robot density per se for explaining gender gaps → models of automation impacts should explicitly incorporate occupational task structure, institutional context, and gendered labour-market sorting.
• Overall: automation/AI policy should be designed with a gender lens, addressing both the potential benefits (safety, autonomy gains in some contexts) and risks (work intensification and unequal distribution of gains).