Time-dependent taphonomic site loss leads to spatial averaging: implications for archaeological cultures

Archaeologists often define large regional cultural areas based on material culture similarities between sites. However, the archaeological record is inherently incomplete due to taphonomic and discovery biases, affecting the search for cultural areas. While the impact of incompleteness on temporal trends has been studied, its effect on spatial patterns due to time-dependent site loss remains under-researched. Cultural classification, fundamental to archaeology, involves defining bounded, homogeneous cultural groupings with distinct material cultures. This practice has been critiqued for its difficulty in drawing discrete boundaries in material culture variation and directly mapping these onto cultural or ethnic groups. Despite these critiques, cultural classification persists, often employing informal comparisons or quantitative methods to explore geographic distance and similarity. Long-distance similarities in material culture are typically explained by migration, cultural transmission, or convergent evolution. However, the vast scale of some prehistoric cultural regions (e.g., Aurignacian, Gravettian, Acheulian) raises questions about whether material culture similarities at such scales truly reflect sociocultural processes. This study hypothesizes that the incomplete nature of the archaeological record leads to spatial averaging, overestimating cultural regions. To test this, it uses a spatially autocorrelated, culturally independent dataset: soil classifications. Soil types provide a spatially complete dataset with temporal stability and spatial autocorrelation, unaffected by cultural processes, allowing the study of spatial subsampling effects without invoking cultural explanations.

The paper reviews existing literature on the impact of taphonomic and discovery biases on archaeological interpretations, particularly concerning temporal trends and the definition of cultural areas. It highlights previous studies demonstrating the distortion of behavioral signals due to time-dependent factors, referencing works by Surovell and Brantingham (2007), Surovell et al. (2009), Contreras and Meadows (2014), Miller-Atkins and Premo (2018), and Perreault (2018, 2019). The authors critically assess the challenges of cultural classification, citing Barth (1981), Reynolds and Riede (2019), Shennan (1994), and Lucy (2005) on the limitations of mapping material culture variation onto discrete cultural boundaries. The use of quantitative methods to explore geographic distance and similarity in defining cultural structure is examined, mentioning Renfrew (1977), Kimes et al. (1982), Shennan et al. (2015), and Lycett (2019). The concept of Tobler's First Law of Geography and its relevance to both cultural and non-human geographic patterns is discussed, referring to Nekola and White (1999) and others' work on distance decay of similarity in ecological communities. The paper also explores existing explanations for long-distance material culture similarities, including migration, cultural transmission, and convergent evolution, citing Crema et al. (2014), Ross and Atkinson (2016), and O'Brien et al. (2018). Finally, the authors acknowledge previous work on time averaging’s impact on the spatial spread of cultural signatures (Miller-Atkins and Premo, 2018), setting the stage for their investigation into the effects of spatial averaging.

The study used the European Soil Database v2.0, converting soil polygons to point data using ArcGIS. Two datasets were analyzed: one encompassing all of Europe (~25,000 points) and another containing only German points (~2,500 points). Thirteen soil attributes (Table 1) were selected to approximate informal archaeological comparisons based on trait presence/absence and artifact proportions. The methodology involved repeatedly constructing decay curves of similarity from a random point to other points. Similarity between points was calculated as a percentage of matching categorical attributes. A generalized additive model (GAM) was used to fit a smoothed function to the data, and the first local minimum in similarity was identified. This process was repeated for progressively smaller proportions of points (10% intervals from 90% to 10%, 1% intervals from 10% to 1%), reflecting potential site loss in archaeological contexts. To model data loss due to taphonomy, an exponential function was fit to radiometric dates for European Paleolithic sites (Vermeersch, 2019), calculating the proportion of points remaining at 10 ka BP, 50 ka BP, and 100 ka BP. The methodology was repeated 100 times for each proportion to capture variation. For each subset, the distance corresponding to the first minimum in similarity was recorded. A negative power function was fit to the absolute distance to the first minimum in similarity by percentage of total points via nonlinear least squares regression for the European dataset, while a log-linear model was used for the German dataset. Mann-Whitney U-tests compared distances to the first minimum at different modeled ages of deposition (10 ka, 50 ka, 100 ka). Finally, the study examined how the location of the starting point affected the region of similarity.

The analysis revealed a significant negative relationship between the percentage of points used and the distance to the first minimum in similarity, both on a continental scale (Europe) and a regional scale (Germany). For the European dataset, a negative power function best described the relationship, indicating that as fewer points are considered, the distance over which things are similar increases. This relationship was also confirmed for the German dataset, although a log-linear model provided a better fit. The analysis of "aged" samples, based on modeled site survival rates, showed significant differences in the distance to the first minimum between the 10 ka and 100 ka samples, and between the 10 ka and 50 ka samples. The difference between the 50 ka and 100 ka samples was not significant, suggesting a point beyond which the relationship becomes negligible. The R-squared values in the regressions were small, reflecting the variation in similarity at given distances due to different reference points. Despite this variation, the negative relationships were statistically significant, highlighting the impact of the proportion of points on the size of the similarity region. Furthermore, the geographic location of the initial comparison point significantly impacted the size of the identified region of similarity. The area of similarity at the lowest proportion of points could vary substantially depending on the location of the initial comparison point.

The findings demonstrate that spatial averaging, resulting from incomplete datasets, leads to overestimation of similarity regions. This effect is inversely proportional to the amount of data remaining, potentially affecting older periods and regions with significant landscape changes. The study highlights the impact of the starting point of comparison on the size of defined similarity regions, emphasizing the influence of chance and research history on the definition of cultural areas. These results are comparable to those observed with time averaging in previous studies, indicating that both temporal and spatial averaging contribute to biases in archaeological interpretations. The paper concludes that the size of material culture similarity regions does not accurately reflect population or cultural dynamics due to incomplete data, suggesting the need for a more nuanced interpretation of the archaeological record.

This research demonstrates that spatial averaging, a consequence of incomplete archaeological datasets, significantly impacts the identification of similarity regions and the interpretation of cultural patterns. The amount of spatial averaging is inversely related to the data available, leading to overestimation of similarity in older periods and regions with high landscape change. The location of the initial comparison point also influences the size of defined similarity regions, introducing additional bias. These findings highlight the need for careful consideration of data completeness and the limitations of using similarity region size to infer population and cultural dynamics in archaeological research. Future research should investigate the combined impact of temporal and spatial averaging and refine methodologies to account for these biases.

The study's reliance on soil data, a proxy for archaeological assemblages, may limit the direct applicability of the findings to archaeological contexts. The relatively small R-squared values in the regression models highlight the high variability in similarity patterns, underscoring the complexities of interpreting spatial relationships. While the study uses random subsampling to simulate data loss, it does not directly account for the non-random nature of actual archaeological site preservation and discovery.