Saturday, January 16, 2010

Archaeological Patterning across Southern Wyoming: Part 2

In a previous post, I visually examined the distributions of some artifact, feature, and site types across southern Wyoming along and East-West gradient. This inspection produced evidence for a transition between the Plains and Wyoming Basin cultural patterns in the vicinity of Rawlins. In this post, I examine the same data using a range of statistical techniques and come to the same conclusion - but with cooler charts! You will need to read the previous post first in order for this one to make sense.

Visual inspection of east/west distribution graphs, discussed above, identified eight variables as potential differentiators between the eastern and western areas of southern Wyoming. These are stone cairns and alignments, ceramics, FCR and thermal features, ground stone, housepits, rock art, stone circles, and steatite artifacts. To assess the statistical significance of these visually-identified differences, a chi-squared analysis was conducted. The visual inspection of the graphs suggested that the boundary between the eastern and western parts of the study area should be placed somewhere between R80W and R100W. For most artifact and feature types, the change was observed between R80W and R90W. For the purposes of the chi-squared analysis, therefore, Ranges were placed into two groups: a western group, located in R85W and further westward, and an eastern group, located east of R85W.

Stone CirclesCeramicsGround StoneFCR ThermalCairns AlignmentsRock ArtHousepitsSteatiteTotals

Table 1: Selected feature and artifact type counts organized by east/west. East/West line placed between Ranges 84W and 85W

Table 1 contains the counts of the potentially significant feature and artifact types organized into eastern and western groups. Pearson’s chi-squared test shows the difference between these two groups to be highly significant (χ2=2083.691, df=7, p<2.2x10-16).
Figure 1: Adjusted chi-squared residuals, artifact and feature types by East vs. West. East/West line placed between Ranges 84W and 85W. From left to right: cairns/alignments (C/A), ceramics (Cer), FCR/thermal features (FCR), ground stone (GS), housepits (HP), rock art (RA), stone circles (SC), and steatite (ST).
Further information can be gleaned from an inspection of the adjusted chi-squared residuals (Figure 1). These residuals provide a measure of the importance of specific variables in distinguishing between the two groups. Values above +2 and below -2 indicate significant deviations from expected values in the contingency table, providing a convenient threshold of statistical significance. When plotted, the residuals clearly show that only five of the eight variables meet this threshold of statistical significance. These are cairns and alignments, ceramics, FCR and thermal features, ground stone, and stone circles. Of these, ground stone is the weakest differentiator and stone circles the strongest. In general terms, then, the eastern group has significantly more sites with cairns/alignments, ceramics, and stone circles, and significantly fewer sites with FCR and thermal features than does the western group. This is not to say that there is no difference between the two areas in terms of rock art, housepits, or steatite artifacts, merely that given the current sample these variables do not significantly differ between the two groups. All three of these variables involve very low site counts (Table 1).


In order to assess the degree to which the five variables identified as significant in the chi-squared analysis distinguish between the eastern and western portions of the study area, a multidimensional scaling (MDS) analysis was conducted. This technique assess the n-dimensional distance between Ranges based on the values of the five variables identified above, and attempts to approximate that relationship in a reduced number of dimensions, such that the data can be visualized on a graph. Only Ranges with more than 50 prehistoric occupations were included in the analysis. The analysis was conducted using the cmdscale function of the R statistical software package.
Figure 2: Multidimensional scaling of site assemblages, grouped by Range.
The results (Figure 2) indicate a clear separation of eastern and western areas in the sample. All of the Ranges west of R89W are located in a tight cluster in the lower right corner of the graph, while all Ranges east of R80W are clearly separated from the western cluster. The tight clustering of the western samples reflects large sample sizes and low variance, while the wider spread of eastern samples is probably produced by lower sample sizes. Interestingly, samples from R80-90W are located between the two other groups in a tight linear cluster, suggesting that this area is a kind of transition zone, consisting of a mix of the attributes that identify the western and eastern archaeological patterns. This could be produced either by a true cultural blend, or, perhaps more likely, by a cyclical shifting of the boundary between these two patterns with changes in paleoenvironmental conditions.
Figure 3: Multidimensional scaling of site assemblages, grouped by Range, labeled E/W. Ranges 80-90W excluded from analysis.
Figure 3 represents the results of a second MDS analysis, this one excluding Ranges in the transition zone (R80-90W). In this figure, Ranges are labeled with a single letter indicating membership in the western or eastern group. The separation of the groups in this graph is complete, with all of the western samples located in a tight cluster in the lower right corner. When the intermediate transition zone is excluded from the analysis, therefore, the groups are completely separated by this technique, with no overlap whatsoever.
Figure 4: Principal Components Analysis biplot of site assemblages, grouped by Range, labeled E/W. Ranges 80-90W excluded from analysis.
Figure 4 displays a biplot of the results of a principal components analysis (PCA) of Ranges using the same five variables that were employed in the MDS analysis. The analysis was conducted using the princomp function of the R statistical software package. The results are the same as the MDS analysis, but the superposition of variables on the biplot provides additional information on what differentiates these clusters. The eastern and western groups are completely separated on Principal Component 1, which is comprised primarily of the cairns/alignments, stone circles, and FCR/thermal variables. Ceramics and ground stone, though statistically significant, are the primary constituents of Principal Component 2, and are only weak differentiators of the two groups.


The results of this analysis suggest that archaeological patterns in southern Wyoming are strongly separated into distinct eastern and western patterns. The western pattern is characterized by a high frequency of FCR and thermal features, and the eastern pattern by high frequencies of stone circles, cairns, and alignments. The western pattern is predominant from R91W westward, and the eastern pattern from R79W eastward. The area in R80-90W appears to constitute a transitional zone, where elements of both patterns are combined. This could represent a cultural gradient between the two areas, the Plains to the east and the Wyoming Basin to the west. However, it seems more likely that this transition is produced by the east/west movement of the Plains/Wyoming Basin ecotone with changes in critical paleoenvironmental parameters such as temperature and precipitation. However, and adequate evaluation of this hypothesis would require analysis of the east/west pattern distinction through time, which is impossible with the available data.

Another significant factor that should be addressed but is currently intractable is the extent to which these western and eastern patterns reflect surface visibility and therefore differential site discovery and site recording in the southeastern and southwestern portions of the state. It is possible that high surface visibility in the Wyoming Basin leads to a higher discovery rate for ephemeral sites containing FCR and thermal features (and not containing stone features). It is not possible to evaluate this hypothesis at present. However, the fact that there is no significant difference between the two areas in terms of lithic artifact frequencies would suggest that differential site discovery is not the dominant factor structuring archaeological patterning at this large scale. A formal test of the influence of surface visibility is needed, though, before this possibility can be entirely discarded.