Case Study 8: Spatial Statistics and Multi-Proxy Methodologies: Lessons from Flixton Island 2, North Yorkshire

Introduction

Flixton Island 2 is one of the lesser known sites from the Vale of Pickering, Yorkshire, that was excavated as part of the ERC and Historic England-funded ‘POSTGLACIAL Project’, under the direction of Nicky Milner (University of York) and Barry Taylor (Chester) from 2012 to 2014.

Post-excavation analyses of the Late Upper Palaeolithic and Early Mesolithic lithic assemblage and associated soil samples from Flixton were undertaken and the datasets visualised and statistically analysed as part of an ERC/AHRC-funded PhD by the author. A site monograph is in preparation.

Background

While Star Carr was on the shores of what would have been palaeo-Lake Flixton (now peaty farmland), Flixton Island itself was a small bipartite island in the lake, not far to the east. Rather than finding the rest of a small scatter (as originally anticipated based on previous work by local archaeologist John Moore in the 1940s and the Vale of Pickering Research Trust and Chantal Conneller in the 1990s and 2000s), a surprising palimpsest of over 21,000 finds of Late Upper Palaeolithic and Mesolithic material was retrieved during the POSTGLACIAL project by a team largely made up of dedicated volunteers and students.

In a bid to understand the deposition of material on the ‘dryland’ area of the island in particular, where trench depth was a matter of tens of centimetres and there was no clear stratigraphy to differentiate occupations or use events, a programme of geochemical sampling was designed by Lisa-Marie Shillito and Helen Williams to complement the retrieval of the dense deposit of lithic artefacts. Organic preservation was very poor and there were no clear archaeological features that might have suggested a reason for the deposition.

Lithic and geochemical sediment analyses were then undertaken as PhD research by the author. Spatial visualisation and statistical analyses were applied to integrate the datasets generated to form a multi-proxy study (Rowley 2017). The aim was to maximise the information retrieved from the two main forms of evidence from the site: knapped stone and sediment.

Methodologies

Detailed recording and high-resolution sampling were both key to generating viable datasets for this approach. Every find larger than a thumbnail from an archaeological (non-topsoil) layer was assigned a unique find number and 3D-located by total station. This was most efficient when finds were geolocated in batches periodically, having been tagged in the ground as an interim step by excavators, employing a dedicated team of two people while excavation could largely continue around them.

Smaller debitage was collected and bagged by horizontal 1m² grid square and context. Topsoil finds and those retrieved from the sieving (100% of sediments) were also collected this way. This provided a phenomenal resource for spatial analysis of the lithic artefacts, only hindered by the fact that negative numbering was used for grid squares for a time, which generated errors in the dataset and meant some information had to be discarded.

Recording orientation, akin to applications at Boxgrove and Seamer C, would have added a further layer of information that could have been advantageously utilised. Naturally, the higher the resolution of the spatial data, the more intimate the level of spatial patterning that can be investigated.

All the lithic material was subject to visual assessment. Similar to Star Carr, and many other sites, this involved recording typological categories, suggested raw materials, and signs of pre- and post-depositional modifications or damage, in a relational database. Every row represented a unique find, including sieved and spall pieces. This allowed the database of artefact attributes to be easily merged with the 3D-location data so that it could be integrated into a GIS, visualised and statistically analysed. Pieces bagged by grid square and context were generally visualised using representational symbology overlaid on an estimated grid layout.

This work would be further complemented by a comprehensive refitting study akin to Conneller’s work at Star Carr, and although it was not feasible within the time constraints of the PhD research, a York undergraduate student dissertation yielded promising preliminary results (Nash 2017).

Archaeological sediment samples were taken for geochemical laboratory analyses or analysed in the field at high resolution horizontally across the in-situ occupation layers. At Star Carr, there were clear features that could also be sampled.

A range of geochemical methodologies was applied to these sediments to establish their attributes and composition which could be plotted visually as well as statistically analysed. Such a multi-proxy study is preferable to applying a single methodology because it allows better critical evaluation and greater confidence in the interpretation eventually proposed (see Shillito 2017). High-resolution, well-planned horizontal and/or vertical sampling is needed if datasets are to be integrated most effectively with high-resolution lithic artefact spatial data.

It was noted that archaeological sediment sampling was best complemented with a thorough strategy for both control sampling and running repeat readings, which allowed more robust and confident interpretation of the results. Results from applying a similar approach at Star Carr strongly suggested support for the proposed delineation of the ‘central structure’ and surrounding occupation zones (Rowley et al. 2018).

Spatial visualisation allows exploration of patterns in the datasets, but the human eye is excellent at identifying random patterning so statistical analysis establishes whether those patterns are likely to be genuinely significant. As with the geochemical methods, where multiple different statistical analyses were applied, the interpretative case could be better evaluated and more strongly argued. A little time invested in understanding the general mathematical principles behind the statistical analyses, which is much easier when there is a genuine dataset in mind than when learning purely theoretically, prevented misapplication of inappropriate statistical tests and built confidence in the arguments made.

Texts such as that by Field (2000) outlined the use of software packages like Statistical Program for Social Sciences (SPSS) statistics and Microsoft Excel to help with looking for trends and correlations in the dataset using such techniques as multiple variance and mass analyses. Such packages were found to be remarkably accessible whilst also helping to develop a better understanding of the statistical techniques themselves.

Online GIS training, tutorials and communities were a powerful bank of shared knowledge, and frequently more up-to-date than many textbooks, so these should always be explored, even if with caution. When visualising the data, accessibility, as well as general clarity, was emphasised (see below).

Results

While the geochemical results aligned well with the structural features at Star Carr (Rowley et al. 2018), the signals from Flixton immediately appeared to be more homogenised, which was unsurprising given the shallow stratigraphy and dense distribution of artefacts retrieved (Rowley 2017; Rowley et al. forthcoming).

However, applying the multi-proxy approach still allowed for areas of interest within the palimpsest of material to be suggested, where patterning in subtle variations was complementary. Fig 8.1 provides an example of the process of data visualisation and analysis for one of the POSTGLACIAL trenches at Flixton Island 2.

Discussion and conclusions

High-resolution recording and sampling takes time. The time investment needs to be evaluated according to the depth of information desired, but where that time can be dedicated a truly detailed, in-depth study of the life history of a particular site and its deposited assemblage can start to be recreated. This is not just an approach that should be taken at larger sites with clear patterning and structures, such as Star Carr and Stainton West, but indeed in some ways it is more important that it is applied at smaller, more ephemeral and enigmatic sites lacking good preservation or features because it can help elucidate how they were utilised.

Data visualisation should always be implemented with clarity and accessibility in mind.

Finally, carefully selected statistical evaluations should be used to test and strengthen any patterning proposed and multi-proxy methodologies can truly enhance understanding of the lithic assemblage when the output datasets are properly integrated.

Acknowledgements

With thanks to Chantal Conneller for her guidance on the lithics analysis and all the volunteers for their time and care during excavations.

References

Field, A, 2000 Discovering Statistics Using SPSS for Windows. Thousand Oaks, CA, Sage Publications Ltd

Nash, B, 2017 What does Lithic Refitting Suggest About Mesolithic Lifeways at Flixton Island 2? Unpublished BA Dissertation, University of York

Rowley, C, 2017 Activities at Flixton Island: integrating scientific approaches for the study of Early Mesolithic living at ephemeral sites. Unpublished PhD thesis, University of York

Rowley, C, French, C, and Milner, N, 2018 ‘Geochemistry of the Central and Western Structures’. In Milner, N, Conneller, C and Taylor, B (eds) Star Carr, Volume 2: studies in technology, subsistence and environment. York, White Rose Press, 161-74

Shillito, L-M, 2017 ‘Multivocality and multiproxy approaches to the use of space: lessons from 25 years of research at Çatalhöyük’. World Archaeology 49(2), 237-59

Back to Guidance | Previous Case Study | Next Case Study