Eastern Fennoscandia has a rich archaeological record. A good portion of the finds has been radiocarbon dated; an ongoing work that currently allows for spatial and spatio-temporal analysis of the data. Radiocarbon dating gives a relatively good estimate of the point in time when an archaeological artefact was deposited. Analyzing the distribution of finds on a geographical area for different time periods is likely to provide clues of differential human activity in the area. To assess whether the current data is sufficient for wide-ranging spatio-temporal analysis, we predisposed a section of this data for tests with a piece of statistical software suitable for the analysis.

In this paper, we present a Bayesian computing approach for spatio-temporal analysis of radiocarbon data from eastern Fennoscandia using the recently developed R-INLA software (http://www.r-inla.org). INLA stands for Integrated Nested Laplace Approximation, a statistical method that has found great use in approximating posterior marginals of non-Gaussian response variables frequently encountered in approximate Bayesian computing (Rue et al 2009). We employ a Poisson point process model to study the spread of archaeological depositions across the area of Finland and ceded Karelia. Point processes are a type of a random process and are well studied objects in probability theory (Daley & Vere-Jones 1988). In a Poisson point process the number of events in disjoint intervals are independent and have a Poisson distribution.

In our previous approach to Bayesian analysis with radiocarbon data from period 4000-3500 cal BCE (Kammonen et al. in press), we employed a spatial model using MCMC (Markov chain Monte Carlo) methods of the WinBUGS software (Lunn et al. 2000). To extend this approach into the spatio-temporal domain, we included radiocarbon dates from a longer period of time and phased in a spatio-temporal Poisson point process model in R-INLA. We used the “toolbox” recommended by the R-INLA developers to implement this modification (Illian et al. submitted). In the first phase we adapted our previous spatial model to R-INLA. A test run showed that results from the INLA-approach are congruent with our previous MCMC approach. In the second phase, we effectively replaced the spatial model with a Poisson point process model. The INLA-approach proved computationally more efficient. Moreover, introducing the new model is the first step in actual spatio-temporal analysis of archaeological data from eastern Fennoscandia.

Daley DJ & Vere-Jones D 1988 Springer

Illian et al submitted Annals of Applied Statistics

Kammonen et al in press CAA2011 proceedings

Lunn et al 2000 Statistics and Computing

Rue et al 2009 Journal of The Royal Statistical Society