Probabilistic graphical model for the hierarchical modelling setup. The white nodes outside the grey plane represent the global regression (hyperparameters) of the model. The circular nodes inside the grey plane are the latent variables (underlying true values) in the model, each corresponding to a shaded observation node. The observations are conditioned on their uncertainties, shown by the smaller solid nodes. If no metallicity observation exists, we draw an estimate from a prior as indicated by the dashed circle.