Modern methods for old data: An overview of some robust methods for outliers detection with applications in osteology

•Some non-robust methods for outliers detection are still widely used in archaeology.•Robust methods for univariate and multivariate outliers detection are presented.•The concept of cellwise outliers seems promising in osteology.•All R script files written for this review are available online.•The manuscript, entirely written in Org mode, is reproducible via a Docker image. Whereas outlier detection is routinely performed in archaeological sciences and may have a substantial impact on subsequent discussion and interpretations, modern and robust methods are rarely employed in our disciplinary field. The detection of univariate outliers mainly relies on the well-known rule of “sample mean plus or minus two standard deviations”, whose the lack of robustness is illustrated in this article. Furthermore, specific and efficient methods for multivariate outliers seem to be very little known and rarely used through the literature published in the Journal of Archaeological Science: Reports. To fill this gap, this article aims to present and summarize some robust methods well suited to the data usually gathered in archaeological and anthropological sciences, for both univariate and multivariate outliers. Robust methods for correlation and linear regression, whose results remain correct even in presence of strong outliers, are also illustrated. Methodological guidelines are discussed, in the light of applications in osteology. All the results (figures and tables) presented in this article can be fully reproduced with the companion R code available online, thus providing to the researchers some examples of templates for outliers detection.