Reduction in compositional error is incorrectly calculated

[mikemc]

Calculating the error with adist as in

error <- pred %>%
    group_by(Mixture_type, Bias_type) %>%
    summarize(
        RAdist = sqrt(mean(adist(Observed, Predicted))),
        Adist2 = mean(adist(Observed, Predicted)^2),
(...)
        )

is incorrect, as the Aitchison distance must be calculated within samples and then averaged. This is currently done twice in the Brooks2015 analysis, though the results are not currently used or referenced in the manuscript.