Probability Modeling
Bayesian inference models utilize multivariate probability distributions to represent uncertainty regarding the proportions of multiple source categories in a mixed sample. Practitioners employ dirichlet prior distributions to establish the baseline probabilities of geographic origins before incorporating the measured geochemical data of flax fibers. This statistical choice ensures that the estimated proportions of regional sources always sum to one.
The mathematical framework accommodates varying levels of prior confidence depending on historical trade records and agricultural output figures.
Statistical Formulation
The distribution uses a vector of positive real numbers called concentration parameters to control the shape and density of the prior probability density function. In textile provenance models, dirichlet prior distributions represent the expected distribution of soil elements or isotope ratios across different agricultural regions. When these parameters are set to equal values, the prior assumes no initial bias toward any specific geographic origin.
If historical shipping manifests indicate that a particular region supplied most of the flax, the corresponding parameters are scaled higher to reflect this prior knowledge. This mathematical flexibility allows the model to update its predictions as new analytical measurements become available, providing a stable foundation for the posterior simulation.
Allocation Inference
Mixing models combine these prior assumptions with empirical measurements to compute the posterior distribution of the origins. Through this integration, the model assigns a credible interval to the geographic source of the tested linen fabric. This method provides a mathematical way to handle the natural variations in soil chemistry.
Accurate source allocation prevents fraud in the supply chain by flagging discrepancies.