7.1 Ordination procedure

As protocol 3, protocol 4 performs PCoA on a distance matrix calculated with Extended Gower coefficient of dissimilarity, combining Euclidean distances on transformed compositional data (50%) and RRD on ranked petrographic data (50%). In this case, we are not filtering out the shipwreck samples, but we do exclude the true outliers (IND, observations with no group assigned) so they don’t pollute visualization.

prot4_Shipwreck_2d <- apply_ordination(# no true outliers
                                       cleanAmphorae[!isTrueIND,],
                                       "4", # select protocol 4
                                       exception_columns = excep_cols,
                                       variable_tags = varCode,
                                       coda_override = chemVars16,
                                       coda_transformation_method = "CLR")

prot4_Shipwreck_3d <- apply_ordination(# no true outliers
                                       cleanAmphorae[!isTrueIND,],
                                       "4", # select protocol 4
                                       exception_columns = excep_cols,
                                       variable_tags = varCode,
                                       coda_override = chemVars16,
                                       coda_transformation_method = "CLR",
                                       dimensions = 3)