I'm unconvinced by their method to assess whether temperatures truly decreased. Why wouldn't a methodological error not also apply to the veterans dataset? E.g. all thermometers of an era reading slightly hot (rather than them all being less accurate in any direction)