Summary | Introduction | Method of Analysis | Sample for analysis | Results of the functional analysis | Analysis of the functional results | Statistical analysis of use-wear data | Ethnographic data | Users | Change in edge angles | Relations between the four phases | Discussion

6.6 Statistical analysis of use-wear data, with R.S. Shiel

In order to examine the strength and validity of the use-wear method, the data from the use-wear analysis were examined so as to detect statistically significant patterns in design and use. First the data were checked for completeness. Artefacts were removed either if they did not have a complete ethnographic record (i.e. information missing on unused edge angle, name of user, amount of time used, actual use) or had proved impossible to examine microscopically on one or both sides due to their shape. The dataset was analysed using a range of univariate, bivariate and multivariate techniques. Before analysis began, histograms of each variable were examined to ensure that the data distribution was appropriate for the analysis planned. Where data were not normally distributed, non-parametric tests such as chi square were used. If this analysis demonstrated strong relationships, it was in some cases possible to combine variables to create new variables, which had a more-continuous and near-normal distribution.

As the use-wear data were recorded in three initial phases of investigation – morphological, macroscopic, microscopic – these data classes were analysed first. Once complete, the analysed use-wear results were compared with the ethnographic information, which constituted a fourth phase.

The 'morphological data' apply to the following attributes and measurements which are related to the morphology both of the artefacts and of their edges.

Table 6. Morphological characteristics of the tools
Variable MeanSD
length mm 53.11 20.89
width mm 39.06 17.29
thickness mm 12.75 5.88
Mass g 28.1 32.48
edge angle ° 41.95 15.21
edge length mm 26.9 13.69
edge thickness mm 8.29 4.63
Z Shape 2.29 1.34

There was a great deal of variation in all of the tools' morphological characteristics with SDs typically about half the mean except for the less variable edge angle and much more variable mass (Table 6). The morphological characteristics can be divided into four groups:

1. length, width, thickness and mass of the artefact;
2. edge angle, edge length, edge thickness (Figs 6, 7, 8, 9)
3. edge profile, number of used edges (Fig. 10)
4. artefact shape * (Fig. 11)

* The shape of an artefact was initially obtained through the ratio length/width as in Grace (1989) and were grouped in the following way; very long (>3); long (1.2–3); square (0.8–1.2); short (0.3–0.8); very short (<0.3).

Tools 8, 41 and 13 tools fell into Grace's profile groups 1–3 respectively and tools 5, 35, 14 and 8 were in his shape groups 1–4 respectively. The measured variables in the first group were strongly positively correlated (P<0.001) (Table 7), suggesting that all dimensions of tools increased concurrently.

In the second group, edge angle and edge length were significantly correlated (P<0.001) with edge thickness. Profile was not related to any other variable, and shape was only related to thickness (P<0.05); this relationship resulted from the presence of 5 unusually thin tools. In group 2 edge thickness was positively related to all the group 1 variables but edge length was only correlated to thickness and length, while edge angle was only correlated to thickness (P<0.001); this suggests that the overall size of a tool is important and is reflected in the usable edge. Twelve of the tools had two used edges, but there was no significant difference in size or shape of the tools with one or two used edges. There were too few tools with two used edges to relate to other variables using chi square.

Table 7. Correlation between measurements on stone tools (*, **, *** refer to significance at P<0.05, 0.01 and 0.001 respectively)
  Length Width Thick Eangle* Elength Ethick Shape
Width 0.537***      
Thick 0.594*** 0.652***     
eangle 0.229 0.008 0.472***    
elength 0.459*** 0.285* 0.446*** 0.133   
ethick 0.447*** 0.587*** 0.845*** 0.550*** 0.407***  
Z shape -0.203 0.233 -0.360** -0.489*** -0.111 -0.277* 
Mass 0.724*** 0.808*** 0.668*** 0.172 0.287* 0.604*** -0.002

The lack of a general relationship between shape (group 4) and variables in groups 1–3 indicated that the definition of 'shape' being used might not accurately reflect the shape of the artefacts. The dimensions of the tools were therefore rearranged into longest, intermediate and shortest for each tool to provide a measure of pure shape. The intermediate and shortest axes were standardised by dividing by the length of the longest side giving lengths coded I and S respectively, with I>S and both <1. The shape (Z) was then calculated as Z=I²/S. The logic behind this relationship is that as I and S <1 then if I is much smaller than 1 squaring it will create a very small number; if I is close to 1 then squaring it has little effect. This gives the utility of a range of values for Z from above 1, where Length (L) and I are >S, through 1 where L, I and S are similar, to <1 which is planar – a 'book' shaped tool with two axes much larger than the third, ~1 which is cubic – more like a brick in shape and all axes lengths similar, and <1 which is columnar with one long and two short axes. This system gives no indication of the size of the tool, only its shape. The revised definition of shape gave a negative correlation with edge angle (P<0.001) and with edge thickness (P<0.05) which suggests that not only size but also shape is a factor in determining usable artefacts. The equations below relate the edge angle and thickness to the tool shape Z.

Edge angle = 54.3 - 5.38Z  r=0.473***
Edge thickness = 10.4 - 0.92Z  r=0.266*

These suggest that edge angle and thickness decrease as the tool becomes more planar.

In summary, the morphological data suggest that the size and shape of an artefact are important characteristics of selection for use as there is a correlation between the shape and angle of the used edge and the overall size and shape of the pieces. The importance of shape in the selection process has been noted before (Brass 1998; White and Thomas 1972).

The second group, macroscopic data, covers all the macroscopic observations obtained at 10x magnification, related to edge damage. These included:

5. dorsal and ventral macro fractures (absent; <5 per 10mm; >5 per 10mm)
6. dorsal and ventral macro fracture type (flakes, snaps, steps) (Fig. 12)
7. edge rounding

The results from this phase were less clear than those from the other phases. The edge rounding did not vary, being repetitive in form, and was dropped from the analysis. Dorsal and ventral fracture frequency appeared to be related (Table 8).

Table 8. Number of tools with 1, 2 or 3 ventral or dorsal macrofractures
 Ventral fractures
Dorsal fractures 1 2 3
1 4 1 0
2 4 19 9
3 0 12 13

Although the most frequent category of each type of dorsal fracture frequency corresponded with the most frequent category of ventral macrofracture frequency, there were frequent exceptions; a significant relationship (P<0.05) could be only be found after amalgamating the absent and <5 per 10mm groups. Dorsal and ventral microfracture type were not significantly associated. When macrofracture frequencies are cross-tabulated with fracture types there are more tools with few fractures and snaps but more with many fractures and flakes, steps or combinations of these with snaps than would be expected. This was significant for the ventral surface and is shown in Table 9 (P<0.01). This was only found after infrequent combinations were amalgamated. With few significant relationships present, the contribution of this group of variables to analysis of function was not clear. As the only consistent relationship was between macrofracture frequency and macrofracture type, these were combined to create two new variables, dorsal and ventral macrofractures.

Table 9. Numbers of tools with snaps, flakes, steps or combinations of these on the ventral surface
Fracture frequency Snaps (sn) Flakes (F) Steps (st) F & sn F & st F & sn & st
<5 20 60510

The third group, microscopic data, covers the following microscopic observations (Grace 1989) at 200x magnification.

8. dorsal and ventral microfractures (absent; <5 per 10mm; >5 per 10mm)
9. dorsal and ventral microfracture type (flakes, snaps, steps, flakes and snaps) (Fig. 12)
10. dorsal and ventral micro rounding (absent, heavy, light)
11. dorsal and ventral microtopography (flat; undulating; ridged) (Figs 13, 14)
12. dorsal and ventral polish distribution (continuous; intermittent; absent) – dorsal and ventral polish distribution type (away from the edge; gapped; edge only even; edge only asymmetric; differential; absent) (Fig. 15)
13. dorsal and ventral polish invasiveness (edge only; <0.5D; >0.5D; absent) (Fig. 16)
14. dorsal and ventral micro linear features (parallel; perpendicular; angled; parallel and perpendicular; parallel and angled; perpendicular and angled; absent) (Fig. 17)
15. dorsal and ventral microstriations (parallel; perpendicular; angled; parallel and perpendicular; parallel and angled; perpendicular and angled; absent) (Fig. 18)
16. dorsal and ventral polish development. [A (individual pockets); A+ (large unconnected individual pockets); B (linked pockets); B+ (greater linkage between pockets); C (all over polish); D (linear)]. (Fig. 19).

The attributes on dorsal and ventral sides of the used edges tended to be well related to each other (Table 10). Where a valid chi square test was feasible (microfractures, micro rounding, polish distribution, polish distribution type, polish invasiveness, microfracture type) between the dorsal and ventral variables, a highly significant relationship (P<0.001) was found in all cases, but microfracture type had no significant relationship with other variables.

Table 10. Numbers of tools with different counts of dorsal and ventral microfractures
 Ventral microfractures
Dorsal microfractures2>=3

The dorsal and ventral values were therefore combined in order to halve the size of the dataset. Another important finding was that dorsal and ventral polish invasiveness and polish development were related (P<0.001). These dorsal/ventral attributes were therefore also combined to produce one new variable – polish invasiveness and development. This was well related to the variable 'polish distribution type' (P<0.01) for example, in that polish away from the edges was associated with it being in individual pockets.

PCA with these new composite variables showed that 'polish distribution type', 'polish invasiveness' and 'polish development' are related, and together explain 22.8% of total variance, but polish distribution is independent of these other polish variables. A further 15.6% of the variance is accounted for by 'linear features', 'striations' and 'number of fractures'. There was no clear grouping of other variables in the micro group.

The invasiveness of polish relates to the hardness of the material worked on. A hard, non-pliable material, such as bone, will produce polish only on the tip of the edge. Softer materials, such as meat or plant material, will produce a polish which is more invasive. The extent of polish development also relates to the material worked on and is investigated in the next section. The interrelationship of these two variables and their relation with 'polish distribution type' suggests that these three variables together are more useful than all of the other variables in determining probable use of an edge. It is noteworthy in this regard that polish distribution is related to microtopography, suggesting that of the three factors it may be the least useful as an indicator of use, relating more to the surface physical properties of the artefact. 'Fractures' and 'fracture types' did not show any significant relation to other variables.


© Internet Archaeology/Authors

Last updated: Wed Oct 8 2003