Modelling caprine age-at-death profiles using the Gamma distribution

Citation metadata

Publisher: Elsevier B.V.
Document Type: Report
Length: 405 words

Document controls

Main content

Abstract :

Byline: Adrian Timpson [] (a,*), Rosalind E. Gillis (b,d), Katie Manning (c), Mark G. Thomas (a,e) Keywords GammaModel; Gamma distribution; Caprines; Age-at-death profiles; MCMC; Model comparison; Neolithic Highlights * Existing 'ideal' kill-off profiles are implausible models of husbandry practices during the early Neolithic. * The Gamma distribution provides a simpler model that fits data better than the current nine-age class models. * Summarising kill-off profiles with the Gamma distribution minimises compression and allows clearer graphical comparisons. * We calculate kill-off profile model likelihoods whilst accounting for both sampling error, and uncertain age classes. * R package 'GammaModel' developed to provide tools to perform these analyses on age-at-death count data. Abstract Age-at-death profiles constructed from archaeozoological data have been used for decades to infer the goals of prehistoric herd management strategies. Several 'ideal' profiles have been proposed as models for the optimal kill-off profiles that represent specific husbandry strategies, such as maximising milk or meat yields, which can then be compared to archaeological profiles. We evaluate the goodness of fit of ten caprine archaeological age-at-death profiles to five published idealised profiles, whilst properly accounting for sampling error and data where the age classes of observations are uncertain. We statistically reject all tested idealised profiles as plausible models to explain the data, and instead propose that a Gamma distribution provides a simpler and better general model to represent possible herd management strategies. Furthermore, we show that archaeological profiles can be summarised well using Gamma parameters, which allow multiple datasets (and models) to be easily compared and graphically represented together with minimal information loss, thus allowing clearer inferences to be drawn. Finally, we calculate likelihood distributions of the Gamma parameters, which provide confidence intervals that fully account for the uncertainties from small sample sizes and uncertain age classes. We have developed an R package 'GammaModel' to enable users to apply these tools to any age-at-death count data. Author Affiliation: (a) University College London, Gower St, London, WC1E 6BT, UK (b) Institut fur Ur- und Fruhgeschichte, Christian-Albrechts-Universitat, Johanna-Mestorf-Stra[sz]e 2-6, D - 24098, Kiel, UK (c) King's College London, Strand, London, WC2R 2LS, UK (d) CNRS -- Museum National d'Histoire Naturelle -- Sorbonne Universites, Archeozoologie, Archeobotanique: Societes, Pratiques et Environnement, (UMR 7209), CP56, 55 rue Buffon, F-75005, Paris, France (e) UCL Genetics Institute, University College London, Gower St, London, WC1E 6BT, United Kingdom * Corresponding author. Article History: Received 3 February 2018; Revised 31 July 2018; Accepted 29 August 2018

Source Citation

Source Citation   

Gale Document Number: GALE|A560111861