ABSTRACT

While there are studies that show a positive or negative impact of robots on wages, a meta-analysis of 2,586 estimates from 52 studies in this paper finds that when one looks at the literature as a whole, there is no clear evidence of a sizable impact of robots on wages.

KEYWORDS:

Additionally, can also be used to test if publication bias exists by checking the smoothness of the distribution around points where statistical significance is determined (Gerber and Malhotra Citation2008). The two vertical red lines show the t-statistic values of 1.96 and −1.96, the threshold associated with a 5% significance level. There is a clear spike in the frequency of observations just beyond the threshold, suggesting there could be publication bias.

shows the ordinary (OLS) and weighted (WLS) least squares analysis and estimates that are corrected for publication bias using the FAT-PET-PEESE approach. Weighted estimates give each study equal weighting.Footnote1

Table 1. OLS, FE, RE & UWLS results.

For interpretation, the values in have been converted back to PCC values. This allows us to use the threshold values by Doucouliagos (Citation2011) who define a small effect as a PCC above 0.07, and medium effect as above 0.17, and a large effect as above 0.33. The overall takeaway from these results is that there is little evidence for a statistically or economically significant overall relationship between robots and wages. None of the estimated ‘effects beyond bias’, statistically significant or not, are near the threshold for even a small effect, indicating that on average, there is little correlation between robots and wages across the literature. The unweighted OLS PET results produced a statistically significant (at the 5% level), but small, positive estimate; however, this significance and positive sign is removed once the PEESE estimation, which corrects for publication bias, is used.

In , there is little evidence of publication bias from the FAT-PET-PEESE tests. All ‘publication bias’ coefficients in the PET analysis fail to reach statistical significance. Other estimators used in Logchies (Citation2024) do detect some publication bias, however which makes it hard to drawn firm conclusions regarding the level of publication bias.

While the average effect across studies is insignificant, there is a wide range of estimates across the literature, which we analyse using a meta-regression technique called Bayesian Model Averaging. The results are shown in .Footnote2

Table 2. BMA results.

The PIP shows the estimated likelihood that the variable is included in the true specification and therefore should be controlled for. Some of the variables that have high PIP, and therefore potentially significant impacts on the estimates found in the literature, include the standard error (to capture publication bias), the way the wage definition is defined, the level of data being used, controlling for specific variables or restricting the dataset in some way, such as a specific industry. These all have a PIP above 0.5, and adding these variables to an OLS regression gives an adjusted R-squared of 0.785, so 78.5% of the variation in Fisher’s Z values can be explained by these variables. However, there are also some variables that one may think would have an impact but are insignificant in determining the relationship between robots and wages, such as worker skill level and whether fixed effects were used in an estimate’s regression.

IV. Conclusion

In this paper, we present the results of a meta-analysis of the current literature measuring the impact robots have on wages. The main finding from our analysis is that, across the literature we find on average no sizable effect of robots on wages. Almost all regression models fail to achieve statistically significant results. The models that do achieve statistical significance, are all far below the thresholds for even a small effect (Doucouliagos Citation2011).

The results presented here are in line with two other recent working papers (Schneider, Jurkat, and Klump (Citation2023) and Dario, Alessandro, and Jelena (Citation2024)) that also do a meta-analysis of robots on wages like the one presented here.Footnote3 Taken together these 3 papers provide robust evidence that while robots can be bad for wages in specific contexts, the general statement that ‘robots are bad for wages’ is unlikely to be correct.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 Each estimate is weighted by the inverse of the number of estimates for each study. Additional estimates that do not correct for publication bias or are based on different estimators, including WAAP (Ioannidis, Stanley, and Doucouliagos Citation2017), TOP10 (Stanley, Jarrell, and Doucouliagos Citation2010), Stem (Furukawa Citation2019), endogenous kink method (Bom and Rachinger Citation2019), MAIVE (Irsova et al. Citation2023), and RTMA and MAN (Mathur Citation2024) can be found in Logchies (Citation2024).

2 We only include variables with a PIP above 0.5. Full results can be found in Logchies (Citation2024).

3 These 3 papers were written independently from each other, and we only became aware of the 2 other papers after we had analysed and written up our analysis.

References

• Bom, P. R. D., and H. Rachinger. 2019. “A Kinked Meta-Regression Model for Publication Bias Correction.” Research Synthesis Methods 10 (4): 497–514. https://doi.org/10.1002/jrsm.1352.
• Claburn, T. 2022. “Industrial Robots in China Push People Out of Jobs, Slash Wages.” The Register. December 6. https://www.theregister.com/2022/12/06/robots_china_employment/(open in a new window).
• Dario, G., P. Alessandro, and R. Jelena. 2024. “Will Robot Replace Workers? Assessing the Impact of Robots on Employment and Wages with Meta-Analysis.” GLO Discussion Paper No. 1395. https://hdl.handle.net/10419/282318(open in a new window).
• Doucouliagos, C. 2011. How Large is Large? Preliminary and Relative Guidelines for Interpreting Partial Correlations in Economics. Deakin University. https://hdl.handle.net/10536/DRO/DU:30106880(open in a new window).
• Furukawa, C. 2019. Publication Bias Under Aggregation Frictions: Theory, Evidence, and a New Correction Method. SSRN. https://doi.org/10.2139/ssrn.3362053.
• Gerber, A., and N. Malhotra. 2008. “Do Statistical Reporting Standards Affect What is Published? Publication Bias in Two Leading Political Science Journals.” Quarterly Journal of Political Science 3 (3): 313–326. https://doi.org/10.1561/100.00008024.
• Harthorne, M. 2017. “Here’s Proof Robots are Taking Jobs and Cutting Wages.” Fox News. March 29. https://www.foxnews.com/tech/heres-proof-robots-are-taking-jobs-and-cutting-wages(open in a new window).
• Ioannidis, J. P. A., T. D. Stanley, and H. Doucouliagos. 2017. “The Power of Bias in Economics Research.” The Economic Journal 127 (605): F236–F265. https://doi.org/10.1111/ecoj.12461.
• Irsova, Z., P. R. D. Bom, T. Havranek, and H. Rachinger. 2023. Spurious Precision in Meta-Analysis of Observational Research. Prague: Charles University. https://meta-analysis.cz/maive/(open in a new window).
• Logchies, T. 2024. Do Robots Decrease humans’ Wages? Harvard Dataverse. https://doi.org/10.7910/DVN/TLCPRS.
• Mathur, M. B. 2024. “P-Hacking in Meta-Analyses: A Formalization and New Meta-Analytic Methods.” Research Synthesis Methods 15 (3): 483–499. https://doi.org/10.1002/jrsm.1701.
• Schneider, F., A. Jurkat, and R. Klump. 2023. Robots and Wages: A Meta-Analysis. Kiel, Hamburg: ZBW - Leibniz Information Centre for Economics. https://doi.org/10.2139/ssrn.4666798.
• Stanley, T. D., S. B. Jarrell, and H. Doucouliagos. 2010. “Could it Be Better to Discard 90% of the Data? A Statistical Paradox.” American Statistician 64 (1): 70–77. https://doi.org/10.1198/tast.2009.08205.