Detection of outliers with a Bayesian hierarchical model: application to the single-grain luminescence dating method.

The event model was proposed by Lanos and Philippe (2018) to combine measurements in the context of archaeological chronological dating. We extend this model to luminescence dating and define a new strategy to detect outliers from the hyperparameters of the event model. This procedure is applied to the combination of Gaussian measurements and luminescence age estimation. We illustrate through simulations that it is preferable, in terms of accuracy and precision, to exclude detected outliers rather than use the robust estimation method (e.g. the event model).