Using 2438 estimates collected over 32 studies, I conduct a meta analysis to investigate the relationship between robotics and employment. Using both fixed and random effects weighted least squares regressions, I find little evidence of a clear relationship between changes in robot use and employment, nor do I find evidence of publication bias within this literature. Using Bayesian model averaging and backwards stepwise regressions I explore factors that may influence the size of the robot-employment relationship. I find evidence that variables for the level at which analysis is conducted, as well as source of data used are very likely to feature in the true robot-employment specification, but none of these meet the minimum threshold for being categorized even as a small effect size. I also find evidence that controls for population size, gender shares and ethnicity shares are likely to be present in the true robot-employment specification, and are each found to have a small effect size. Specifications controlling for population size or ethnicity shares tend to have a smaller estimated effect of the impact of robots on employment, while specifications controlling for age shares tend to give larger estimates of the robot-employment effect. Finally, I find that there exists a pronounced interest within the literature for analysis of already developed nations, but only few papers on developing countries, which limits the generalisability of our findings. One further needs to recognise that there exists some significant flaws in the IFR dataset, which is the data source for the majority of analyses estimating the robot- employment effect size. Such flaws threaten the validity of the results of both empirical studies using IFR data and dependent studies, such as this meta analysis.