Epistemic considerations about uncertainty and model selection in computational archaeology: A case study on exploratory modeling

Computational modeling in archaeology is approached from various perspectives (correlative, generalized behavior, and system-based) and can serve different goals (prediction, reconstruction, and exploration). Whatever the focus of any particular model, however, we have to deal with aspects of uncertainty, which are not easily tackled. The uncertainty relates to the very nature of archaeological datasets as well as to the model itself, and both types of uncertainty make it difficult to assess model performance and decide which (set of) models serve a given purpose best. In this chapter, we specifically discuss uncertainty and model selection in an exploratory context. For this purpose, we consider a case study on prehistoric hunter-gatherer landscape use in Flevoland (The Netherlands). We demonstrate strengths and weaknesses of the model for exploratory use and argue that, rather than data-driven statistical testing, robustness analysis is the main approach for dealing with the uncertainty. Although Bayesian statistical approaches could be useful to deal with model uncertainty, such analyses are frustrated by the notorious lack of unambiguous archaeological data. Robustness analysis and the development of theoretically underpinned selection tools, like “informativeness” (specificity) and “surprisingness,” can help to evaluate the models already at the front end of model building. In this way, we can identify families of models that seem to offer the best possibilities for hypothesis testing.