May 17, 2019, 11:30–12:30
Understanding how social information is used in human populations is one of the challenges in cultural evolution. Fine-grained individual-level data, detailing who learns from whom, would be most suited to answer this question empirically but this kind of data is difficult to obtain, especially in pre-modern contexts. Therefore inference procedures have often been based on population-level data in form of frequency distributions of a number of different variants of a cultural trait at a certain point in time or of time-series that describe the dynamics of the frequency change of cultural variants, often comprising sparse samples from the whole population. This data situation leads to a classical inverse problem inviting caution in linking population-level frequency patterns to individual-level learning processes. Using generative inference frameworks we demonstrate that there exist theoretical limits to the accuracy of inferring processes of social learning from aggregated data highlighting the problem of equifinality, especially in situations of sparse data. In other words, we should not expect a one-to-one mapping between populationlevel statistics and underlying learning processes and the consistency between any one specific learning hypothesis and empirical data should be interpreted in this context. Crucially we show the importance of rare variants for inferential questions. The presence, or absence, of rare variants as well as the spread behaviour of innovations carry a stronger signature about underlying processes than the dynamic of highfrequency variants. On the example of the choice of baby names, we illustrate that the consistency between empirical data, summarized by the so-called progeny distribution, and hypotheses about social learning processes depends entirely on the completeness of the data set considered. Analyses based on only the most popular variants, as it is often the case in studies of cultural evolution, can provide misleading evidence for underlying processes of social learning.
Anne Kandler (Max Planck Institute), “Inferring processes of social learning from cultural frequency data”, IAST General Seminar, Toulouse: IAST, May 17, 2019, 11:30–12:30.