Fig 1.
Data patterns derived from 8730 datable entries of 178 eastern Mediterranean sites from the ICRATES database. Left: percentage of sites with each pottery ware; note the disproportionate dominance of the earliest ware (ESA—teal), and its decrease with the introduction of the western-produced ITS (light green). We will later refer to this metric as Pattern A. Right: percentage of sites with a certain number of different wares. Note the dominance of sites with no or only one ware. We will refer to this metric as Pattern B. The two patterns capture two different aspects of the changes we are studying. Pattern A shows how each different tableware spreads to settlements throughout the time period under study, whilst pattern B shows how the diversity of tablewares changed from one settlement to another and over time. To ensure that the models of social learning we explore capture both aspects of these changes we test their ability to reproduce both patterns A and B at the same time.
Table 1.
Typological, chronological references and possible region of production for major tablewares studied in this paper.
Fig 2.
Geographical distribution of the 178 eastern Mediterranean sites from the ICRATES database. Each point represents one archaeological site. Presumed production regions of ESA, ESB, ESC, ESD are labeled. Map created using data from Natural Earth (http://www.naturalearthdata.com/).
Table 2.
Model parameters.
Table 3.
Prior distributions for parameters to be inferred by the ABC.
Fig 3.
Impact of the number of bins on the data pattern.
The number of bins used to describe the data increases from left (2 bins) to right (200 bins).
Table 4.
Value of epsilon for all steps of the ABCPMC.
Fig 4.
ABC accepted simulation ratio.
Evolution of the ratio between the number of accepted simulation runs (ie simulations generating output s where δ(s, d) < ϵstep) and the total number of simulations needed to accept 500 simulations at each step of the ABC algorithm and for the three models. The first step has been removed as it represents an ϵ big enough to accept any simulation run, which leads to a ratio of 1 for all models.
Table 5.
Summary of the Bayes factors between all models.
Fig 5.
Posterior distributions for the three models.
Posterior distributions of parameters drawn using the 500 accepted simulations from the last ABC step (ϵ = 0.0099).
Fig 6.
Posterior distribution and high density regions for the model of independent learning.
Marginal posterior distributions of the independent learning Model’s parameters. The boxplots at the bottom of each graph shows the 75% HDR (darker green) and the 95% HDR (lighter green). The vertical line indicates the mode of the distribution.