An evaluation of the robustness of network based diffusion analysis.
Speaker: Glenna Evans (University of St.Andrews)
Here we examine the robustness of a recently developed method for studying social transmission of behaviour in groups of animals: network based diffusion analysis (NBDA). We fit an NBDA model to diffusion data derived from observations of foraging bouts in starlings (Sturnus vulgaris) given knowledge of their patterns of association, as well as to simulated data, and derive estimates of key parameters in the model. By deploying known values of key variables in simulated data we assess the robustness of the NBDA method. We employ a reversible jump Markov chain Monte Carlo (RJMCMC) algorithm to discriminate between models, and thereby determine whether the null model (with no social component) or the full model (containing both an asocial and social component) is most likely. We also show proof of principle by accurately reconstructing the original parameter values of the simulated data. Our analysis extends the current use of NBDA models to incorporate random effects and facilitate model discrimination. Furthermore, we show that NBDA models can be used to analyse diffusions and derive association patterns from common behaviours that do not involve learning. This methodology is likely to be particularly useful to deal with datasets that include many covariates and that can be fitted with a correspondingly large number of competing models.