Fig. 4

Empirical distributions of the difference between predicted and observed fPARi values for the five statistical methods, and the associated mean squared error (MSE). Note that in the log-transformed linear model, predictions had to be re-transformed to the observation level. MSE are defined as the mean squared difference between predicted and observed fPARi. Log-transformed log-transformed linear model, Bayes (beta) Bayesian model assuming a beta distribution of the data, MLE (beta) maximum likelihood estimation assuming a beta distribution of the data, MLE (normal) maximum likelihood estimation assuming a normal distribution of the data, LSE least squares estimation, techniques