Abstract

In most archaeological and anthropological applications, time series data on frequency changes of different cultural traits (e.g. different types of pots found in the archaeological record) are the only available information about past cultural patterns. Researchers in these fields would benefit from being able to infer information about the underlying evolutionary processes that have produced those frequency changes. This poses a classical inverse problem: we aim to convert available information about changes in variant frequencies into information about the evolutionary processes that have caused those changes, but which cannot be observed directly. We discuss this issue in the context of social learning and develop a n-variant reaction- diffusion framework that specifies the consequences of different mixtures of learning strategies on the frequencies of cultural variants. We assume that changes in these frequencies are caused by the adoption decisions of individuals and distinguish between the two main adoption mechanisms, (i) social learning, in the form of both directly biased and frequency-dependent transmission and (ii) individual learning. Given this causal relationship between frequency change patterns of cultural variants and adoption decisions of the population we attempt (i) to gain theoretical insights into the question of which learning strategies lead to well-adapted populations in changing environments and (ii) to ‘reverse engineer’ conclusions about the learning strategies deployed in current or past population, given knowledge of how cultural variation has changed over space and time. Here the statistical concept of Approximate Bayesian Computation offers an efficient way of finding those parameter constellations by generating posterior distributions for each model parameter which indicate how likely different parameter values could have produced the observed frequency pattern. Naturally, the quality of the information that can be extracted from the observed change patterns depends on the adequacy of the assumed relationships between adoption decisions and cultural frequency change as well as the temporal and spatial resolution of the available frequency data. Importantly, we do not claim the existence of a unique relationship between observed sparse frequency data and underlying processes; to the contrary, we suspect that different processes can produce similar frequency pattern. However, our approach can help narrow down the range of possible processes that could have produces those pattern, and thus still be instructive in the face of uncertainty. Rather than identifying a single social process that explains the data, we focus on excluding processes that cannot have produced the observed changes in frequencies.

Reference

Anne Kandler (University College London), “Inferring learning strategies from frequency data”, IAST General Seminar, Toulouse: IAST, October 3, 2013, 15:30–16:30, room MS001.