Application of a Bayesian CAR Prior to Analyzing Ancient Statistical Records of the Inca Empire

We use a Bayesian model to analyze ancient khipu, the knotted cords of yarn used by the Incas to record state statistics and preserve control over one of the largest empires in the Western Hemisphere. In order to explore the degree of Inca influence on the information storage and dissemination in the peripheral regions and any significant provincial differences, we developed a Bayesian statistical model that enables us to quantify the uncertainties among the unknown observations recorded in the Khipu Research Database (Urton and Brezine, 2005). We use the Bayesian conditional autoregressive (CAR) prior to incorporate spatial correlations among adjacent locations, allowing us to impute the locations for khipus with unknown locations. The results bolster our hypothesis of differences between the samples from the coastal regions associated with diverse cultures subordinate to the Incas and produced a consistent pattern along the coast. By utilizing such variables as types of knot, cord directionality, and colors in our multivariate model, we draw further implications of potential regional markers and distribution of power and control across the 15th and 16th-century Latin America.