This site is undergoing review. Some annotations were human-generated, some AI-generated — all are being verified.

Robots at Work

Graetz, Michaels

2015Review of Economic Statistics (REStat)124 citations

Observational labor marketCausalTheoretical model

Automation / RobotsAugmentation vs. substitution

Summary

Graetz and Michaels use panel data on industrial robot adoption across 17 countries and 14 industries from 1993-2007, employing instrumental variables based on task replaceability and reaching/handling intensity, to estimate robots' causal impact on productivity, employment, and wages

Main Finding

Increased robot use from 1993-2007 contributed approximately 0.36 percentage points to annual labor productivity growth (15% of total growth), raised total factor productivity and lowered output prices, with no significant effect on total employment but reduced low-skilled workers' employment share

Primary Datasets

IFR Industrial Robots

Other

IFR industrial robot data; 17 countries panel (1993-2007)

Secondary Datasets

INE sectoral GDP deflators

Key Methods: Panel data analysis with country and industry fixed effects; two instrumental variables strategies (replaceability index and reaching & handling index) to address endogeneity; OLS and 2SLS estimation with two-way clustering
Sample Period: 1993-2007
Geographic Coverage: International
Sample Size: 238 country-industry pairs (17 countries × 14 industries) over 1993-2007
Level of Analysis: Industry, Country
Occupation Classification: None
Industry Classification: ISIC

Notes

REStat; ~0.36pp annual productivity growth from robots; no significant total employment effect; Non-US paper; included for cross-national comparison [Claude classification]: IZA Discussion Paper No. 10540; authors use GMM-SYS methodology to address endogeneity and dynamics; first study to examine both R&D and embodied technological change (ETC - innovative machinery/equipment) simultaneously; R&D lagged 2 years, ETC lagged 1 year; descriptiveCrossTabulation=true because authors do not claim causal identification despite using IV-style techniques; heterogeneous effects by sector and firm size are central findings [Claude classification]: Published in Review of Economic Statistics. Robot density increased 150% across sample (0.58 to 1.48 robots per million hours). Quality-adjusted robot prices fell 80% from 1990-2005. Authors construct novel instruments based on task replaceability and reaching/handling intensity. Effects show diminishing marginal returns. Case study data from Universal Robots shows payback periods of 2-18 months (25-200% annual returns). Contribution comparable to steam technology (0.35pp, Crafts 2004) but over shorter period; lower than ICT contribution (0.6-1.0pp). No effect on labor share found. Replaceability instrument passes falsification tests on pre-adoption periods. [Claude classification]: Published in Review of Economic Statistics. Robot density increased 150% across sample (0.58 to 1.48 robots per million hours). Quality-adjusted robot prices fell 80% from 1990-2005. Authors construct novel instruments based on task replaceability and reaching/handling intensity. Effects show diminishing marginal returns. Case study data from Universal Robots shows payback periods of 2-18 months (25-200% annual returns). Contribution comparable to steam technology (0.35pp, Crafts 2004) but over shorter period; lower than ICT contribution (0.6-1.0pp). No effect on labor share found. Replaceability instrument passes falsification tests on pre-adoption periods. [Claude classification]: Published in Review of Economic Statistics. Robot density increased 150% across sample (0.58 to 1.48 robots per million hours). Quality-adjusted robot prices fell 80% from 1990-2005. Authors construct novel instruments based on task replaceability and reaching/handling intensity. Effects show diminishing marginal returns. Case study data from Universal Robots shows payback periods of 2-18 months (25-200% annual returns). Contribution comparable to steam technology (0.35pp, Crafts 2004) but over shorter period; lower than ICT contribution (0.6-1.0pp). No effect on labor share found. Replaceability instrument passes falsification tests on pre-adoption periods. [Claude classification]: Published in Review of Economic Statistics. Robot density increased 150% across sample (0.58 to 1.48 robots per million hours). Quality-adjusted robot prices fell 80% from 1990-2005. Authors construct novel instruments based on task replaceability and reaching/handling intensity. Effects show diminishing marginal returns. Case study data from Universal Robots shows payback periods of 2-18 months (25-200% annual returns). Contribution comparable to steam technology (0.35pp, Crafts 2004) but over shorter period; lower than ICT contribution (0.6-1.0pp). No effect on labor share found. Replaceability instrument passes falsification tests on pre-adoption periods. [Claude classification]: Published in Review of Economic Statistics. Robot density increased 150% across sample (0.58 to 1.48 robots per million hours). Quality-adjusted robot prices fell 80% from 1990-2005. Authors construct novel instruments based on task replaceability and reaching/handling intensity. Effects show diminishing marginal returns. Case study data from Universal Robots shows payback periods of 2-18 months (25-200% annual returns). Contribution comparable to steam technology (0.35pp, Crafts 2004) but over shorter period; lower than ICT contribution (0.6-1.0pp). No effect on labor share found. Replaceability instrument passes falsification tests on pre-adoption periods. [Claude classification]: Published in Review of Economic Statistics. Robot density increased 150% across sample (0.58 to 1.48 robots per million hours). Quality-adjusted robot prices fell 80% from 1990-2005. Authors construct novel instruments based on task replaceability and reaching/handling intensity. Effects show diminishing marginal returns. Case study data from Universal Robots shows payback periods of 2-18 months (25-200% annual returns). Contribution comparable to steam technology (0.35pp, Crafts 2004) but over shorter period; lower than ICT contribution (0.6-1.0pp). No effect on labor share found. Replaceability instrument passes falsification tests on pre-adoption periods. [Claude classification]: Published in Review of Economic Statistics. Robot density increased 150% across sample (0.58 to 1.48 robots per million hours). Quality-adjusted robot prices fell 80% from 1990-2005. Authors construct novel instruments based on task replaceability and reaching/handling intensity. Effects show diminishing marginal returns. Case study data from Universal Robots shows payback periods of 2-18 months (25-200% annual returns). Contribution comparable to steam technology (0.35pp, Crafts 2004) but over shorter period; lower than ICT contribution (0.6-1.0pp). No effect on labor share found. Replaceability instrument passes falsification tests on pre-adoption periods. [Claude classification]: Published in Review of Economic Statistics. Robot density increased 150% across sample (0.58 to 1.48 robots per million hours). Quality-adjusted robot prices fell 80% from 1990-2005. Authors construct novel instruments based on task replaceability and reaching/handling intensity. Effects show diminishing marginal returns. Case study data from Universal Robots shows payback periods of 2-18 months (25-200% annual returns). Contribution comparable to steam technology (0.35pp, Crafts 2004) but over shorter period; lower than ICT contribution (0.6-1.0pp). No effect on labor share found. Replaceability instrument passes falsification tests on pre-adoption periods. [Claude classification]: Published in Review of Economic Statistics. Robot density increased 150% across sample (0.58 to 1.48 robots per million hours). Quality-adjusted robot prices fell 80% from 1990-2005. Authors construct novel instruments based on task replaceability and reaching/handling intensity. Effects show diminishing marginal returns. Case study data from Universal Robots shows payback periods of 2-18 months (25-200% annual returns). Contribution comparable to steam technology (0.35pp, Crafts 2004) but over shorter period; lower than ICT contribution (0.6-1.0pp). No effect on labor share found. Replaceability instrument passes falsification tests on pre-adoption periods. [Claude classification]: Published in Review of Economic Statistics. Robot density increased 150% across sample (0.58 to 1.48 robots per million hours). Quality-adjusted robot prices fell 80% from 1990-2005. Authors construct novel instruments based on task replaceability and reaching/handling intensity. Effects show diminishing marginal returns. Case study data from Universal Robots shows payback periods of 2-18 months (25-200% annual returns). Contribution comparable to steam technology (0.35pp, Crafts 2004) but over shorter period; lower than ICT contribution (0.6-1.0pp). No effect on labor share found. Replaceability instrument passes falsification tests on pre-adoption periods.