Do robots boost productivity? A quantitative meta-study

This meta-study analyzes the productivity effects of industrial robots. More than 1800 estimates from 81 primary studies are collected. There is strong evidence that the empirical literature on the productivity effect of robots suffers from a substantial positive publication bias. This finding is observed across all measures of productivity used in the primary literature and is robust to several modern meta-analytic estimators. Beyond publication bias, there is only limited evidence for a productivity-increasing effect of robots, which so far have exerted at best a marginal boost. My analysis of the drivers of heterogeneity among the findings of primary studies points to adjustment costs at low intensities of robot use as well as diminishing returns at more advanced levels of robotization. My findings are robust to addressing model uncertainty through Bayesian model averaging. Finally, several explanatory factors for the emergence of a productivity paradox in the context of robotics are discussed.