A relatively high proportion of the morphologically identified specimens at Crvena Stijena (84.6%) and Le Piage (71.4%) could be assigned to a taxon through ZooMS analysis, Table 2), which is indicative of excellent collagen preservation. However, this proportion is considerably lower at Saint-Césaire (51.1%, Table 2). Despite the small dimensions of the finds, a large fraction of the indeterminate, non-piece plotted specimens from Crvena Stijena (44.7%) and Saint-Césaire (81.0%) could likewise be attributed to a taxon using ZooMS.
Patterns in the NISP sample
Because they provide the foundation of a large number of faunal interpretations, our comparisons begin with an analysis of the NISP sample (all layers are combined here to increase sample size, an analysis by layer is presented below). A chi-square test of independence shows no statistical difference in the taxonomic profiles produced by the two methods at all three sites (Crvena Stijena: χ2 = 14.0, p = .1709; Saint-Césaire: χ2 = 9.7, p = .2843; Le Piage: χ2 = 3.1, p = .6801, Fig. 1, data from Table S1), an indication that the counts are not affected by the method of identification. Out of 26 pairwise comparisons of taxa, only one shows a taxonomic difference larger than 7% (reindeer at Saint-Césaire, difference of 10.8% between the two methods, Table S1).
Because errors may cancel one another in comparisons of whole samples, an additional test consists in determining if the morphological and ZooMS identifications are consistent for the same specimen. In this case, we are ascertaining whether a given morphological identification—a Bos/Bison femur for instance—matches the taxon given by ZooMS for the same specimen. The percentage of agreement, that is, the proportion of matches relative to the sum of compared specimens, is moderate at Saint-Césaire (69.6%), whereas it is high at Crvena Stijena (91.8%) and Le Piage (93.2%, Table 3). A plot of the data suggests that errors of identification are more common when species are poorly represented, although the trend is not significant (r = 0.69, p = .0707, Fig. 2a, n > 4). The percentage of agreement in our dataset also seems inversely correlated with species diversity as measured by the reciprocal of Simpson’s Index, a pattern that is observed regardless of whether the identifications are based on the morphological or ZooMS counts (Fig. 2b).
Turning to patterns of skeletal representation, a chi-square test of independence and the adjusted standardized residuals shows a clear over-representation of ribs, and to a lesser extent, humeri and scapulae in the mismatch sample (χ2 = 31.2, p < .001, Fig. 3, categories for the test as in the histogram), the results showing a moderate effect size (Cramér’s V = 0.35). If ribs are excluded from the NISP sample, the percentage of agreement is substantially increased at Saint-Césaire (from 69.6 to 82.3%) where this body part is unusually abundant (63.0% in the sample examined here vs. 0% at Crvena Stijena and 0.9% at Le Piage). Conversely, long bones are less commonly represented in the mismatch sample.
Comparing the NISP sample with the indeterminate remains
While the two methods give comparable results for the NISP sample, what remains to be determined is whether results are also consistent when comparisons are performed with the indeterminate sample (specimens in this sample lack a morphological identification). At Crvena Stijena, comparing ZooMS identifications for morphologically indeterminate specimens with published NISP counts for the M5, M3 and M1 levels—three Middle Paleolithic occupations dated to Marine Isotope Stage 3—yield only small differences that are not statistically significant (M5: χ2 = 7.6, p = .3723; M3: χ2 = 4.3, p = .9773; M1: χ2 = 6.7, p = .8778, data from Table S4, Fig. 4). Unlike Crvena Stijena, the taxonomic abundances in the NISP sample at Saint-Césaire and Le Piage are significantly different from the ZooMS counts derived from the indeterminate sample (Saint-Césaire, US16: χ2 = 92.2, p < .0001; Le Piage, Early Aurignacian: χ2 = 110.1, p < .0001; Le Piage, Solutreo-Badegoulian: χ2 = 71.0, p < .0001, data from Table S5). The results for these comparisons show a moderate to strong effect size (Saint-Césaire, US16: Cramér’s V = 0.51; Le Piage, Early Aurignacian: Cramér’s V = 0.20; Le Piage, Solutreo-Badegoulian: Cramér’s V = 0.16). This lack of agreement means that the indeterminate sample contains an array of taxa that is different from that of the NISP samples, with large species such as horse, Bos/Bison and rhinoceros being systematically more common in the former samples (Fig. 5). These differences in taxonomic composition are confirmed by a chi-square comparison of the ZooMS counts for the NISP and indeterminate samples in the Early Aurignacian at Le Piage (χ2 = 19.9, p = .0013, Cramér’s V = 0.31, data from Table S6, note that the ZooMS counts by layer at the other sites are too small to allow direct comparisons between the NISP and indeterminate samples).
Although the indeterminate samples are smaller for the other units, the ZooMS analysis provides us with valuable qualitative information on species that are rare or otherwise undocumented in the selected samples. For instance, one hominin remain was identified by ZooMS in unit US15/16 at Saint-Césaire, an important finding given the scarcity of human remains for the early Upper Paleolithic (see Supplemental spreadsheet S1). Other rare species identified by ZooMS in the course of this study include one specimen attributed to Panthera (US15, Saint-Césaire).