This past few months I’ve been prompted into working on some scarred or culturally modified tree data that I recorded near Weipa during a series cultural heritage consultancy projects between about 2003 and 2007. The reason for looking at this again was that I was fortunate enough to be hosting/supervising Masters student Emily Shepard from Portland State University who was out here on an EAPSI scholarship to work on this material with me. It was a great chance to blow the dust off some good data collected under trying circumstances during many months in the field. It’s not often you get a chance like that.
The project we’ve been working on has involved looking at ‘sugarbag’ scarred trees. These are trees scarred by Aboriginal people cutting holes (or apertures) to access honey and wax from the nests of various species of Australian native stingless bee. Alun Salt wrote a great post about some of my work on CMTs here last year and it’s well worth a read. The question Emily and I have been looking at this past few months involves using the data I collected to identify trends and patterns that give us some insight into the intensity of wild honey collection. Emily has worked through and made sense of the original data, re-analysed photos and completed most of the statistical analyses. I turned my attention to spatial statistics, a mildly terrifying method, but one that I think more archaeologists should employ.
Spatial statistics are simply tools in a Geographic Information System (GIS) that use statistics to “cut through the map display and get right at the patterns and relationships in the data” (Mitchell 2009:2). They do require reasonable familiarity with using GIS software, as well as access to decent software that can perform the analysis. I found it quite challenging to begin with, partly as I’ve had no formal training in statistics or GIS, but if you need to identify patterns in the way archaeological data are distributed then it’s well worth the investment of time. There are a bunch of more simple tools archaeologists can use to find patterns in their data, such as proximity analysis, and these give good insights on simple questions such as ‘what is the relationship between site location and distance to water’. Resulting data can be quickly and easily exported to conventional statistical software. But GIS can do a lot more than make maps and summarise basic patterns such as this.
Cluster analysis is something that I’ve been interested in for some time, in part because my Doctoral research involved looking at clusters of midden sites and trying to make sense of them. With the scarred tree data, we were interested in discovering whether we could find clusters of similar variables in our dataset of >1500 sugarbag scars. We did.
We looked at the frequency of scars across our study area. Figure 1 shows aggregated number of scars within 500 metre raster cells. This is a great means of visualising datasets in a relatively simple manner and helped us to identify general areas of high frequency scarring. However, it doesn’t provide a clear indication of whether there are finer or more localised trends within this dataset, or whether the things we think are ‘clusters’ meet tests for statistical significance.
We then used two local statistical measures to further explore whether there are any specific clusters of high scar frequencies. We used Anselin’s Local Moran statistic and the very nicely named Local Getis-Ord Statistic (or Gi*). I won’t go into details of how these work, but see this guide for a start if you’re interested. Figure 2 shows the resulting data. What we were looking for particularly were areas where both techniques pointed to a a number of cluster points in relatively close proximity to each other. You can see a few of these in this image.
I suspect the results are probably not that exciting to look at without any more detailed context, but the approach has enabled us to identify clusters in the data that weren’t noted from visual inspection alone. Given some success here, we decided that it was worth exploring clustering of other variables and the one that we had most success with was identifying clusters of larger scar aperture area, shown in Figure 3.
The result indicated the high frequency scarring locations broadly correlated with large aperture sizes and that there were even more subtle trends we needed to think about. I won’t go into what we think the results mean, partly because we haven’t completed our paper yet, but these methods provide a useful insight that can be used alongside other standard statistical tests that archaeologists often use.
Our dataset is not perfect: it’s uneven and there are major gaps which have limited our ability to take these analyses any further. Despite that, I think these tests are still worth exploring for archaeological spatial data. I’m especially fascinated by the potential of these kinds of tests for picking out clusters in more evenly distributed data, such as looking for clusters of particular artefact types or sizes within large surface scatters.
I’ve picked out a few books and articles below that I found really quite useful and that are worth reading if you’re interested in exploring this material in more detail. There is surprisingly little written about spatial analysis and spatial statistics in archaeology, which I find baffling given our love affair with conventional statistics.
Some useful sources