Spatial distribution of artefacts and refit lines
The Rose diagram of Exp. 18 shows that about half of the pieces are located within a 75° interval to the southeast of the core (Fig. 2c), with the rest evenly distributed in all directions. The mean distance between products and the core is 343 mm, and the longest is 761 mm (Table 1). Analysis of the sequence (28 stages) shows that the majority of products landed to the northwest or south of the previous removal (Fig. 2e). More than 75% of the stages are less than 600 mm away from the preceding product (Fig. 6b). Refits have a mean distance of 385 mm (Table 1), and longer refit lines produce a largely elongated NW–SE trend (Fig. 2g).
Figure 3c shows that most artefacts in Exp. 40 occur in a 90° interval to the northeast of the core. More than 75% of core-product distances (Fig. 3b) are less than 400 mm, with outliers at 681 mm and 886 mm (Fig. 6a and Table 1). The knapping sequence (27 stages; Fig. 3d) shows a general trend of removals to move towards the east diagonally, and almost every other stage is to the North or South of its previous and next stage, forming a N–S trend (Fig. 3e). This is corroborated by the refit orientation circular histogram (Fig. 3g), which also indicates an additional trend from NE to SW. Many of the refits are less than 400 mm from one another (Fig. 6c), thus following the pattern of core–product distances.
In the case of Exp. 54, most pieces are within a 120° interval east from the core (Fig. 4c). Core–flake distances range between 72 and 1162 mm (Fig. 6a). There are 21 stages in the sequence, which shows an E–W trend (Fig. 4e). Most distances between successive stages are less than 600 mm, although there is a larger amount of variation in the lower end of the distribution than in freehand experiments. The Rose diagram of refits (Fig. 4g) shows a strong unimodal pattern along the E–W axis. Distances between conjoining pieces present a normal distribution, and are all ≤ 970 mm, with a mean of 475 mm (Table 1; Fig. 6c).
As shown in Fig. 5c, most Exp. 56 artefacts are clustered in a Rose diagram to the NE of the core, in a narrow range of 75°. This reduction sequence contains 18 stages, where successive products are distributed randomly with respect to the previous detachment (Fig. 5d), with no pattern except a singular trend towards 240°–255°. The unimodal and symmetrical NE to SW distribution (Fig. 5g) may be linked to clustering near the anvil. On the other hand, one-to-one refit distances (mean = 467 mm; see Table 1) show a wider variation than the other experiments (Fig. 6c).
ANN results show that the average distance in all experiments is clearly lower than the average distance in a hypothetical random distribution, indicating that materials follow a clustered distribution (likelihood higher than 99%; Table 2). This is supported by Global Moran’s statistics and the Getis–Ord method, indicating a dominance of high-concentration clusters. Ripley’s K Function shows that clustering is maintained up to a distance of 1384–1396 mm, beyond which a dispersed pattern dominates (Table 2). Maximum clustering is reached at a distance of 380 mm for Exp. 18, and 464 mm for the rest of experiments.
Figure 7 shows kernel maps of density and dispersion patterns of artefact frequency for each experiment. Exp. 18 has the lowest mean density, and Exp. 54 the highest mean density (Table 3). Exp. 40 has the highest maximum density (0.08881), which determines the highest range of density data, with a standard deviation of 0.00891. Exp. 18 contains the lowest maximum density (0.05862) and the lower standard deviation (0.00548; Table 3). This indicates that Exp. 54 and Exp. 56 share similar density patterns, whereas Exp. 18 and Exp. 40 have a higher variability and more extreme values. Dispersion of artefact frequency shows the highest concentration in front of the knapper for Exp. 40, and also in front, but slightly displaced to the right, for Exp. 18. Highest concentration areas can be clearly distinguished in the density maps (Fig. 7a), and are also detected as hot spot clusters by Gi* statistics with 95% level of confidence (Fig. 7b).
Both Exp. 18 and Exp. 40 also show an area of low artefact concentration that forms a distal arc from right to the left of the knapper. In Exp. 18, this low concentration area shows a scattered pattern, whereas in Exp. 40 the low concentration area has a ring shape and is separated from the highest concentration area by a discontinuous strip with no material. Areas of low concentration do not constitute statistically significant cold spots, according to the Gi* method. The highest concentration areas in Exp. 54 and Exp. 56 are also defined by density maps and Gi* hotspots, and are strongly patterned to the right of the knapper.
Getis–Ord Gi* statistics (Fig. 7c, d) detected only hot spots or statistically significant concentrations of pieces with high length and weight values. For length, Exp. 18 and Exp. 40 show hot spots in front of the knapper which are very concentrated. Conversely, length patterns of Exp. 54 and Exp. 56 are defined by hot spots to the right of the knapper, and show a more dispersed pattern, suggesting a higher dispersion of large pieces. This dual pattern is less obvious in the weight variable. Exp. 18 and Exp. 40 again show concentrated hot spots in front of the knapper, and Exp. 54 presents dispersed hot spots towards the right of the knapper. However, weight hot spots in Exp. 56 show a low dispersion: they are located to the right of the knapper, but are concentrated. Hot spot descriptive statistics show low values for Exp. 18 (Table 4), indicating the presence of smaller pieces in this assemblage; for example, hot spots reaching a 95% confidence include pieces with a mean length of 98–110 mm for Exp. 40, Exp. 54 and Exp. 56, but only of 67 mm for Exp. 18 (Table 4). Similarly, weight hot spots in Exp. 18 have a mean weight of 33 g and of 41–47 g for the other three experiments. No cold spots or statistically significant concentration of pieces with low length or weight were detected in any experiment.
Circular dispersion of conjoining sets
Statistical tests of core-to-flake displacements show high statistical significance (p < 0.01) in all cases. This allows to reject the uniform distribution (Table 5) and indicates the presence of preferred orientations in the four experimental assemblages. Since omnibus tests reject uniformity against unimodal and multimodal distributions, and the Rayleigh test detects only unimodal orientations, all the circular core-to-flake distributions can be essentially considered as unimodal preferred orientations.
The circular distribution of mean direction in the core-to-flake displacements of Exp. 18 and Exp. 40 indicates some significant differences (Fig. 8). While azimuths in Exp. 18 are mainly concentrated between 80 and 150° (Table 5), with a mean direction of 108.5° and a modal direction between 145 and 150°, the vectors in Exp. 40 are located mainly in the first quadrant, with a mean direction of 49.9° and a mode of 50–55° (Table 5). On the other hand, Exp. 54 and Exp. 56 display similar mean directions (83.5° and 67.2°, respectively), which overlap with 95% of confidence interval (Fig. 8). Their modal directions are also very similar (90–95° for Exp. 54 and 80–85° for Exp. 56, Table 5 and Fig. 8), and the dispersion of data (indicated by R and K) is higher than in Exp. 18 and Exp. 40. The most concentrated data is found in Exp. 54 (R = 0.81, K = 2.99), while Exp.18 contains the highest dispersion (R = 0.44, K = 0.97).
Descriptive statistics also included weighting the displacement vectors with the distance covered by each piece from the core, and the artefact weight (Fig. 8a, Table 5). Results show that the mean direction does not vary significantly from the unfiltered data. Exp. 18 is where the mean changed the most significantly, varying close to 15° from weighted to unweighted data. On the other hand, the mean direction did not change significantly in Exp. 54 and Exp. 56, which show variations of only 2–3°. The modal direction is more variable, but with no similar pattern shared by all experiments (Table 5). Regarding dispersion, weighted statistics show an increase of the concentration in all experiments, excepting in the data weighted by the distance of Exp. 18, where the concentration is reduced slightly with respect to the unweighted statistics. In general, data weighted by artefact mass shows more concentrated parameters than data weighted by the distance, excepting in Exp. 56 (Table 5).
Clear differences in the axial data of refit lines exist between freehand and bipolar experiments. Statistical results of freehand knapping experiments do not allow rejecting the null hypothesis of uniformity with a > 95% confidence interval. Minimum p values are obtained for Exp. 18, where Rayleigh’s test and Rao’s Spacing test only reach p = 0.073 (92.7% confidence interval), and Exp. 40’s statistical significance is even lower (p > 0.15). Therefore, no solid preferred orientation can be proposed for freehand knapping refit lines. Conversely, bipolar knapping data (Exp. 54 and Exp. 56) show high statistical significance for the Rayleigh test (Z > 3.94; p < 0.019) and demonstrate evidence of departure from uniformity in Rao’s Spacing and Watson’s omnibus tests (0.025 < p < 0.005; Table 5). These results suggest that bipolar refit lines show strong unimodal preferred orientations.
Refit mean directions are different within the freehand experiments (Exp. 18 = 130°, Exp. 40 = 32°; Table 5), whereas bipolar refit lines show more consistent mean directions, located around the 90–270° axis (axis 92–272° for Exp. 54 and axis 76–256° for Exp. 56). A similar—although weaker—relationship was observed in the refit line mode, which is more consistent within bipolar experiments than within freehand experiments (Table 5). The mode percentage is higher in bipolar (8–10.7%) than in freehand knapping (7.8%) (Table 5). The concentration parameters of refit lines also indicate a very dispersed distribution for Exp. 40 (R = 0.11 and K = 0.22) and the highest concentration for Exp. 54 (R = 0.34 and K = 0.72), while Exp. 18 and Exp. 56 have similar intermediate values. However, when refit lines are weighted by the distance (Fig. 8b), concentration similarities between Exp. 18 and Exp. 56 disappear, although they are still positioned between the end values of Exp. 40 and Exp. 54.