Bayesian Posterior Predictive Distributions for Assessing Soil Aggregation in Undisturbed Semiarid Grasslands

Traditionally, data on soil properties collected across experimental treatments are analyzed via analysis of variance comparisons of their mean values, with such comparisons usually requiring similar variances. This study focused on changing the analysis to center on the soil properties themselves instead of concentrating on parameters such as means. Such analysis on variables or observables can be accomplished by computation of Bayesian posterior predictive distributions, namely, distributions of observables conditional on the observed data. These distributions incorporate the mean and variance structure of the variables, a useful feature for including the variability inherent in natural systems. We built on a previous introduction to the use of Bayesian models for analysis of soil texture, aggregation, and organic matter storage to demonstrate the use of posterior predictive distributions for these variables of interest. Examining these distributions may provide more informative inferences when comparing soil properties between two treatments and these methods are easy to interpret.