Using Generalized Estimating Equations to Evaluate Activity in Human Skeletal Remains.
by
Sara K. Becker
University of North Carolina, Chapel Hill

Skeletal measures of activity can be used to reconstruct workload levels, repetitive motions, and mobility in past human societies, but working with multiple data points recorded from an individual’s skeleton can come with some scalar problems. For example, if activity measurement is reduced to an overall present or absent count per individual, the reduction in sample size may result in a loss of very specific pathology data, as well as be insufficient to address research questions. However, if activity indicators are calculated on a per data point basis, one individual with multiple positive scores may skew statistical results when looking for patterns of activity, or could be a violation of the independence of data required for many statistical tests. In this paper, these issues were solved by using generalized estimating equations (GEE). GEE was used to model the multiple recorded data points on each joint surface for osteoarthritis and each muscle attachment point for musculoskeletal markers, while keeping data linked with each individual specimen. These data were then tested for significant chronological and spatial differences using the chi-square statistic. Overall, the results correlate well with archaeological artifactual data, helping understand the work and activity that laborers performed in prehistory. Additionally, GEE is flexible enough to accommodate variables that are not normally distributed, small sample sizes, and most importantly, randomly missing or unobservable variables (Ballinger 2004; Ghislatta and Spini 2004; Liang and Scott 1986; Rochon 1998), all of which are very important when studying skeletal samples that vary in actual bones recovered from archaeological sites.

Date received: July 22, 2012