Authors

Alex Mes

University of Cambridge

Abstract

Why do people move? Understanding the different underlying processes behind human expansion can have implications in elucidating the nuances in interactions between migrant populations and incumbent communities; between different subsistence strategies; and between humans and their environments. Taking one of the fastest and largest spatial range migrations in human history, the Bantu expansion, this paper looks at reconstructing movement dynamics. In just a few thousand years, Bantu languages proliferated over more than 9 million square kilometers to become the largest language group in Africa, spoken by over 350 million people today. Major economic and cultural changes took place across sub-Saharan Africa during this time, with the dispersal associated with a ‘bantu cultural package’ consisting – wholly, or in part, at various times – of a more sedentary lifestyle, thick-walled pottery, iron metallurgy, cattle-keeping and crop cultivation.

Behind all large-scale human dispersals, such as the spread of the Neolithic in Europe, lies a body of literature seeking to categorise this movement as demic migration vs. acculturation. The same debate surrounding the Bantu expansion sees supporters predominately for demic diffusion, cultural diffusion, or combinations thereof. In addition, there is support for the hypothesis that in the later stages of the Bantu spread, language and farming dispersed simultaneously through a demic expansion. Framing dispersal debates in this way is not always helpful. It rigorously restricts to only two ways of movement: the migration of the social group as a self-contained entity, or the borrowing of ideas by local populations without any population admixture. The reality in any human migration is far more complex: the units of migration (individuals, families, communities), as well as the localised tempo and direction of dispersal shifted frequently. Agency for social action needs to be afforded to both the incumbents in a given area and the migrants – both having an equal ability to be influenced by the other to adopt appropriate cultural traits [Robb and Miracle].

It is the localised migration dynamics that this paper aims to examine through the archaeological record: understanding the eastern stream of the Bantu expansion as a series of regional and complex movements, focusing on variations in the tempo of dispersal and arrival times of sub-Saharan eastern Early Iron Age within smaller geographic areas and the potential ecological drivers determining observed differences in the dispersal process. This is done in a Bayesian framework: calculating movement rates using quantile regression and regional arrival times using hierarchical phase models and intrinsic conditional autoregressive (ICAR) models.

Regression analyses have a long history in archaeology: beginning with the pioneering work of Edmonson in the 1960s who used the earliest occurrences of pottery, copper and maize to determine the rate of diffusion of neolithic conditions globally. The idea was built on by Ammerman and Cavalli-Sforza in 1971, who used a linear regression on 14C dates to determine the rate of the neolithic expansion in Europe. Subsequently, mathematical models have been proposed to analyse archaeological transitions and classify the driving mode behind spreads. The calculated rate from regression analyses on radiocarbon dates is compared to predicted rates determined from either a demic diffusion model or cultural diffusion model, from which the calculated spread is classified as being driven by one of these mechanisms. From this point, there have been several refinements to mathematically model archaeological transitions. However, there are major limitations to these types of analyses: (i) the analysis is dependent on the spatial scale, how one selects local sample sizes and defines a nominated origin (from which the distance and therefrom speed is calculated), (ii) the method uses mean calibrated dates and does not take measurement uncertainty into account, (iii) typically the use of linear regression yields single dispersal rates, which dramatically generalises movement dynamics. The method can be adapted but it remains difficult to determine if the variations in calculated dispersal rates are genuine and not caused by measurement uncertainty, calibration error, or sample size.

In the first part of this paper, I calculate the dispersal rate of the eastern sub-Saharan EIA using a Bayesian quantile regression, which mitigates some of the limitations facing ordinary linear regression. Focusing on the 90th and 99th percentiles is used to specifically examine the distribution of the earliest arrival dates. Using MCMC sampling, a Gaussian process adaption which allows for localised variation in the dispersal rate is explored. Computational intensity, and how this can restrict approaches to large archaeological datasets is briefly discussed.

The second part of this paper focuses on hierarchical phase modeling. While perhaps not having as long-established a history as regression-based analyses in archaeology, Bayesian phase model analyses of radiocarbon dates have been used in several different studies: typically in stratigraphic contexts to accurately and precisely chronologically delineate cultural phases, estimate the arrival of cultural traits or crops, or to model migration arrival times. A core advantage of Bayesian phase models is that they are able to take measurement uncertainty into account when determining estimated arrival times. Limitations include: (i) the calculations are sensitive to how the spatial regions are defined, and (ii) uneven sampling leads to some sites contributing multiple dates and the interdependence of these dates biases the arrival time estimation.

The arrival times of the eastern EIA in geographic regions are examined using a Bayesian phase model, where a hierarchical structure is introduced to account for the bias introduced by sample interdependence. This framework is then built on by exploring ICAR models where spatial autocorrelation between neighbouring regions is taken into consideration. There are, however, other factors which influence how regions are related to other another, and the model can be adapted to include ecological measures, such as elevation – quantifying to what extent these potential drivers explain the observed differences in the dispersal process.

The results from these two approaches are presented in order to build a more detailed description of the localised dynamics of the EIA expansion. Throughout, biases which influence the data – such as uneven spatial and temporal sampling density – are acknowledged and measurement uncertainty (the error on the calibration curve and the sample’s 14C age error) is accounted for in all models.

References

Robb, John, and Preston Miracle. “Beyond migration versus acculturation: new models for the spread of agriculture”. In Proceedings-British Academy, 144:99. Oxford University Press Inc., 2007.