21. Calibrating Bone Durability

21.1.1 Do Skeletal Elements Differ in Their Amounts of Bone Tissue per Unit Volume?

A casual observer handling a disarticulated skeleton laid out on a tabletop intuitively grasps that not all skeletal parts contain the same amount of bone tissue in relation to their overall size. The sternum and sacrum are very light, being composed nearly entirely of cancellous tissue with a thin veneer of cortical bone , while carpals and tarsals are much heavier due to their much denser compact bone tissue. As was the case with bone fracture patterns, properties inherent to the elements themselves, rather than the modifying actors , can strongly influence element survival in the face of destructive processes, and thereby structure element frequencies . Human selectivity may produce the initial element frequencies in an assemblage, but actualistic research in paleontology and zooarchaeology has shown that subsequent mechanical or chemical forces can modify these original frequencies according to different elements’ intrinsic properties, including their surface area in relation to their volume, as it did with weathering and the amount of bone tissue per unit volume.

21.1.2 Can this Variability be Measured with a Degree of Replicability?

As part of his palaeoanthropological research on differential skeletal element survival (Chap. 2), Brain (see 1981 for an overview) measured modern goat elements’ specific gravity, using the time-honored water-displacement method. He noted that the differences in modern element survival rates when subjected to carnivore gnawing or ungulate trampling correlated well with this measure. Behrensmeyer (1975) also recorded observations on the specific gravities of skeletal elements from modern taxa analogous to those found in fossil deposits in the Lake Turkana basin. Binford and Bertram (1977) attempted to extend Brain ’s use of specific gravity to their study of skeletal element destruction by Alaskan working dogs and wolves, but found the method wanting as a readily replicable process. Thus, by the early 1980s, paleontologists and zooarchaeologists were increasingly concerned with how best to distinguish selective transport of body segments from differential destruction by various agents, especially carnivores . Some sought more readily replicated methods for assessing the bone tissue density of various skeletal elements .

Lyman (1984) observed that mammalian bone tissue has a single specific gravity and physical density, but that individual skeletal elements , with their differing combinations of compact and cancellous tissue, as well as their internal voids for marrow , comprise variable volumes of such tissue, with different resulting “volume densities.” Lyman (1984: Table 1) reviewed previous specific gravity studies, observing a wide variation in values produced for similar elements. He argued that variations in the bone tissue composition of skeletal elements , and hence their survival potentials, seriously complicate inferences about human behavior from archaeofaunal element frequencies and Binford ’s utility indices.

Lyman (1984) introduced a new level of replicability in estimating bone tissue density with the application of biomedical scanning equipment, specifically single-beam photon densitometry, to measure bone tissue in skeletal elements . At that time, photon absorptiometry was widely applied in measuring the bone tissue in patients at risk of osteoporosis, based on a representative element, usually the distal radius and ulna of the forearm. The single photon (“X-ray”) beam moves over a plate on which the scanned tissue rests. A detector on the plate registers the beam, with the tissue of the scanned tissue partially impeding the beam.

Because the photo beam sensor registers a two-dimensional scan line across the element, the volume of the bone tissue in a three-dimensional element must be estimated with a formula that converts the photon-derived reading into a three-dimensional figure. Lyman originally estimated bone volume at each scan site as the product of the beam scan’s width (1/8 in.), the maximum bone length, and the maximum bone thickness. Later, he used an average bone thickness (Lyman et al. 1992). The method treats all skeletal elements as cross-sectionally rectangular at the scan site and assumes that the element is a solid, lacking medullary or inter-trabecular spaces. These assumptions may not be warranted, as will be seen below. Nonetheless, Lyman ’s scan-site approach has guided later research and development on this topic.

Lyman initially called the resultant measure of bone tissue in three dimensions “bulk density,” which he defined as, “the ratio of the weight of a volume of a substance to the volume of that substance including the pore space volume” (1985: 226). Thus, the statistic should reflect the degree to which an element is tightly packed with bone tissue. By this criterion, vertebrae have low bulk density, and carpal and tarsal bones have very high bulk density. Later, Lyman (e.g. 1992) changed to the term “volume density” (VD ) as the more accurate name for the same index.

Lyman (1984) scanned multiple sites (Fig. 21.1) on axial and appendicular skeletal elements of 13 deer (Odocoileus virginianus and O. hemionus), plus a sheep (Ovis aries) and pronghorn antelope Antilocapra americana. He targeted scans to locations of Binford ’s MGUI values, such as the proximal and distal femur, adding long bone shafts, for which Binford had not assigned MGUI . He also made multiple scans for elements treated in aggregate by Binford , for example vertebral regions, ribs, carpals, and tarsals, but later averaged these values to compare with the MGUI . For any element to which Binford had given distinct (e.g. proximal and distal) MGUI values, he retained distinct values (Table 21.1).

Table 21.1

Summary of Lyman's (1985) scan sites for deer skeletal elements, with Volume Density (VD) values, and Lam et al.'s (1998) caribou CTScan Bone Mineral Density (BMD) values, according to their BMD1 and BMD2 calculations (see text for details)

Element	N	Lyman (1985) deer VD	Lam et al. (1998) Caribou CTScan
			BMD1	BMD2
Cranium (petrosal)	1	N/S		1.29
Mandible	2
DN1		0.55	0.65
DN2		0.57	0.75	1.05
DN3		0.55	0.63	1.07
DN4		0.57	0.67	1.06
DN5		0.57	0.56	1.05
DN6		0.31	0.66	0.84
DN7		0.36	0.98	1.01
DN8		0.61	0.99
Atlas	1
AT1		0.13	0.47
AT2		0.15	0.42
AT3		0.26	0.49
Axis	1
AX1		0.16	0.62
AX2		0.10	0.42
AX3		0.16	0.42
Cervical	5
CE1		0.19	0.45
CE2		0.15	0.43
Thoracic	13
TH1		0.24	0.38
TH2		0.27	0.53
Lumbar	6
LU1		0.29	0.49
LU2		0.30	0.45
LU3		0.29	0.51
Sacrum	1
SC1		0.19	0.37
SC2		0.16	0.40
Innominate	2
IL1		0.20	0.43
IL2		0.49	0.70	1.02
AC1		0.27	0.64
PU1		0.46	0.58
PU2		0.24	0.54
IS1		0.41	0.67	0.94
IS2		0.16	0.30
Rib	26
RI1		0.26	0.47
RI2		0.25	0.49
RI3		0.40	0.62	0.96
RI4		0.24	0.65	0.94
RI5		0.14	0.40	0.40
Sternum	6
ST1		0.22	N/S
Scapula	2
SP1		0.36	0.66	1.01
SP2		0.49	0.73	1.04
SP3		0.23	0.73
SP4		0.34	0.69	1.01
SP5		0.28	0.48
P Humerus	2
HU1		0.24	0.26
HU2		0.25	0.31	0.44
Sh Humerus
HU3		0.53	0.61	1.12
D Humerus
HU4		0.63	0.62	1.08
HU5		0.39	0.48
P Radius	2
RA1		0.42	0.53
RA2		0.62	0.57	1.08
Sh Radius
RA3		0.68	0.73	1.09
D Radius	2
RA4		0.38	0.38	0.97
RA5		0.43	0.49
P Ulna	2
UL1		0.30	0.49
UL2		0.45	68	0.84
D Ulna	2
UL3		0.44	NS	NS
P Metacarpal	2
MC1		0.56	0.63	0.92
MC2		0.69	0.69	1.08
Sh Metacarpal
MC3		0.72	0.79	1.10
D Metacarpal	2
MC4		0.58	0.59	1.01
MC5		0.49	0.48
MC6		51	0.68
P Femur	2
FE1		0.41	0.39
FE2		0.36	0.35	. 52
FE3		0.33	0.35	0.74
Sh Femur
FE4		0.57	0.57	1.15
D Femur
FE5		0.37	0.40	0.61
FE6		0.28	0.32
FE7		N/S	0.30
Patella	2
PA1		0.31	0.57
P Tibia	2
TI1		0.30	0.35
TI2		0.32	0.44	1.01
Sh Tibia
TI3		0.74	0.71	1.13
D Tibia	2
TI4		0.51	0.53	1.12
TI5		0.50	0.39	0.73
Fibular	2	0.52	0.68
Naviculocuboid	2
NC1		0.39	0.56
NC2		0.33	0.62
NC3		0.62	0.55
Astragalus	2
AS1		0.47	0.68
AS2		0.59	0.70
AS3		0.61	0.63
Calcaneus	2
CA1		0.41	0.52
CA2		0.64	0.80	0.94
CA3		0.57	0.66
CA4		0.33	0.73
P Metatarsal
MR1		0.55	0.58	0.90
MR2		0.65	0.57	1.10
Sh Metatarsal
MR3		0.74	0.65	1.08
D Metatarsal	2
MR4		0.57	0.54	1.08
MR5		0.46	0.41
MR6		0.50	0.59
Phalanx 1	8		0.56
P11		0.36	0.48
P12		0.42	0.56	0.92
P13		0.57	0.71
Phalanx 2	8
P21		0.28	0.49	0.61
P22		0.25	0.64	0.72
P23		0.35	NS
Phalanx 3	8
P31		0.25	0.48
Carpals	10
Scaphoid	2	0.98	0.70
Lunate	2	0.83	0.67
Cuneiform	2	0.72	0.71
Magnum	2	0.74	0.69
Unciform	2	0.78	0.72

Over the next decade, a variety of VD estimation studies, using either single-beam photon or Dual-energy X-ray Absorptiometry (DEXA) scans, were made with skeletal elements of other taxa. These included the marmot Marmota marmota (Lyman et al. 1992), the leporids: domestic rabbit Oryctolagus cuniculus, eastern cottontail Sylvilagus floridanus, arctic hare Lepus canadensis, black-tailed jackrabbit Lepus californicus (Pavao and Stahl 1999), the bison Bison bison (Kreutzer 1992), the South American camelids: guanaco Lama guancoe, vicuña Lama vicugna, llama Lama glama, and alpaca Lama pacos (Elkin 1995; Stahl 1999), domestic cattle Bos taurus, domestic sheep Ovis aries, farmed European wild boar Sus scrofa (Ioannidou 2003), the turkey Meleagris gallopavo (Dirrigl 2001), the flightless lesser rhea Pterocnemia pennata (Cruz and Elkin 2003), and the North American harbor seal Phoca vitulina, with partial skeletons of the harp seal Phoca groenlandica included (Chambers in Lyman 1994: Table 7.7).

These findings were applied to assessing bone density-dependent bone destruction in various archaeological cases using rank-order correlation coefficient analysis. Lyman himself (1984, 1985) assessed several archaeofaunas purported to reflect selective transport , such as the Gatecliff Shelter , using volume density figures, as well as extending arguments about the relationship between nutritional utility and bone volume density, which will be treated in a subsequent sections.

21.1.2.2 Most Accurate Estimator of Bone Mineral Density: X-Ray Computed Tomography

Over several years, Lam and colleagues extensively explored the sources of error in photon densitometry estimates outlined above and developed CTscan applications to zooarchaeology (Lam et al. 1998, 1999, 2003; Lam and Pearson 2004, 2005). Computed Tomography scanning is used in medicine to produce three-dimensional images of soft and hard internal tissues of the body. The technique resembles photon densitometry in that an X-ray beam passes through a body or body segment to a beam receptor, with calibrations for impedance of the beam. However, CTscans employ digital processing software to interpolate three-dimensional images from multiple, two-dimensional “slices,” or scans, taken around a single axis of rotation.

For estimating bone mineral density , CTscans can represent both the actual external shape of a scanned skeletal element and the shapes and sizes of any internal voids, thus effectively addressing the two problems of photon densitometry scans (Fig. 21.2). The attenuation or impedance of the X-ray beam is expressed in “CT numbers” or Hounsfield units (Lam et al. 1998: 561), reflecting the calcium mineral densities within the scanned elements. A clinical product that permits calibration of bone mineral density (BMD), a solution of dipotassium phosphate, is scanned along with the bone specimens as baselines for imaging attenuation (Lam et al. 1998: 562). The bone mineral density, BMD (gm/cm³) is derived from the three-dimensional imaging.

Lam and colleagues (1998) initially produced bone BMD estimates for elements of goat (Capra hircus), using the same scan sites illustrated by Lyman , and compared their BMD results with Lyman’s cervid VD estimates. Notwithstanding taxonomic difference among these ruminant species, the values were highly correlated. Because some debate existed over whether density indices are ordinal, they offered both linear regression analysis (R ² = 0.47, F = 27.80, P < 0·0001) and Pearson’s rho rank order correlation coefficient (r _s = 0.68, P < 0·0001) between BMD and VD . However, Lam et al. stressed that the correlation masks the fact that CT BMD values for some element portions diverged markedly and importantly from those of photon densitometry VD for the same sites. For example, the femoral mid-shaft is the highest-ranked in BMD in CTscans, whereas it ranks seventeenth in VD calculations, results also suggested by Elkin’s 1995 study. Significantly for studying density-mediated attrition , CTscan BMD values for long-bone epiphyses display strong and consistent differences from diaphyses of the same elements that were not displayed in VD values for the same scan sites (Lam et al. 1998: 564). Applying both indices to caprine bones in two Southwest Asian archaeofaunas for which based on other criteria, they expected substantial carnivore attrition , they found that the CTscan BMD values correlated more strongly with log %MAU than did non-shape-corrected photon densitometer VD values (Table 21.2). Lam and co-workers attributed this to the greater realism of CTscans in estimating the actual amount of bone tissue in an element.

Table 21.2

Lam et al.’s (1998: 567, Table 3) regression and Spearman’s rho statistics for the relationships between the log %MAU in the two archaeological assemblages, Kobeh Cave and ‘Ain Dara Mound and the bone mineral density values using CTscan-based BMD estimates and photo densitometer based Volume Density (VD) estimates

	Regression statistics			Spearman’s statistics
	R 2	F	P	r s	P
Kobeh log %MAU by CT	0.56	37.08	<0.0001	0.5	<0.0001
Kobeh log %MAU by VD	0.18	6.30	0.0181	0.40	0·0298
’Ain Dara log %MAU by CT	0.53	31.77	<0.0001	0.69	<0.0001
’Ain Dara log %MAU by VD	0.16	5.25	0.0296	0.40	0.0447

Lam et al. (1999) published BMD indices for reindeer/caribou Rangifer tarandus, a cervid, wildebeest Connochaetes taurinus, an African bovid, common zebra Equus quagga burchelli, and wild horse E. przewalskii. They used the same scan sites as Lyman ’s, and sometimes altered sites to adjust to osteological differences between ruminants and equids, and they added some elements such as the petrous temporal. The article details how two calculations were used to estimate BMD, depending upon whether the external cross-sectional outline at scan sites was sufficient for calculating the bone area (BMD₁) or whether internal cross-section of medullary space must be traced to take into account the size and shape of the void within the element (BMD₂).

Lam and colleagues (1999) compared BMD values among all taxa in their samples, plus Lyman’s deer and sheep and Elkin’s guanaco, the latter three being VD values (Table 21.1). They found that, especially among the ruminant artiodactyls, element BMD values were usually highly significantly correlated (p = <0.001). However, true to Kreutzer’s (1992) functional anatomical predictions, modest differences are found in such regions as the anterior cervicals, where differences in types of male-male competition probably have favored deposition of more or less bone mineral. The authors recommended that these extant BMD estimates are sufficient for application to most other taxa within the major zoological families represented, thereby sparing other researchers the time and expense of seeking to use CT equipment. Naturally, animals of divergent functional anatomy and osteology should be CTscanned. Lam et al. (2003) later argued that, in the absence of the opportunity to use technology to estimate bone mineral density in various elements, shape-corrected densitometer values can provide a reasonably reliable key to bone durability in bone without internal voids.

In their reviews of bone mineral density estimation methods, Lam and Pearson (2004, 2005) inserted an important methodological note: “Ironically, the increased technological sophistication of density studies have resulted in greater methodological variation than that observed among the original water displacement studies” (Lam and Pearson 2005: 102). They note that one of the main sources of the variability in VD estimates, besides those already outlined here, seems to be variation in different researchers’ scanning procedures, which they illustrate by divergent results in simple VD estimates of caprine elements in three studies (Lam and Pearson 2005: 102, Fig. 1). They note that many more researchers have attempted photon densitometer estimates than have done so with CTscanning and stress a need for explicit standardization of scanning procedures, which necessarily differ from the medical uses of single- or double-beam technology to scan bones in living tissues of patients. They thus call for new techniques and methods. As of the time of their publication, the authors were part of the only team to produce CTscan readings for a variety of skeletal elements and species, but they imply that a similar lack of uniformity in technique could emerge in using such technology.

21.1.3 Does a Negative Correlation Exist Between Nutritional Utility and Bone Durability?

Are the skeletal elements most likely to be selected for transport and processing because of their higher nutritional values also the most fragile in the skeleton and therefore, the most liable to be deleted by taphonomic processes? Lyman (1985, 1992) explored whether a negative relationship exists between utility indices and bone VD . If at least some high-utility bones have low volume density values, and some low-utility bones have high density values, he argued, this would present a problem in archaeological inference, because “inverse utility” curves could also result from in-place destruction of less durable elements. His reasoning was that elements or portions with considerable cancellous tissue were both reservoirs of nutrients (Chaps. 5 and 19) and more delicate than elements composed mainly of compact bone .

21.1.4 What Causes Destruction of More Fragile Elements or Portions of Elements?

Destructive processes affecting skeletal elements include consumption by carnivores , human culinary and consumption practices (marrow extraction, bone grease extraction, mashing and gnawing bone ends), human and animal traffic that crushes less dense bone, and post-depositional chemical and mechanical stresses , including the cumulative static loading of sedimentary layers. Whether and how one can tell which actor or other taphonomic process is involved in destruction of skeletal elements is a concern, and the lines of evidence outlined in Section 4 are relevant because actor/effector analysis can more closely specify the main process(es) that impacted an archaeofaunal sample. How to assess the dominant agent(s) of reduction in such assemblages is at present unclear, no “cookbook” approach exists for doing so. However, Chap. 17 provided examples of how various zooarchaeologists have approached this problem. Lyman (2008: 264–298) reviews the issues involved in tabulating such traces, which, like element frequencies , also become aggregate data , and he stressed (Lyman 1994: 335) that it is important to inspect existing specimens for percussion marks, carnivore action, and weathering , all of which are known to be related to processes that contribute to bone attrition , for hints at the processes and actors involved. The next section presents examples of how some zooarchaeologists have used bone mineral density and nutritional indices to explore the overlapping impacts of human selectivity and site formation.

21.1.5 How Can Knowledge of Bone Durability be Employed in Archaeofaunal Analysis?

This question can be rephrased as, is there a way to distinguish in situ destruction from selective transport ? Lyman (1985) suggested that, in addition to element frequencies , taphonomic and sedimentary evidence must be used to evaluate how heavily attritional processes may have affected an assemblage. For example, high frequencies of carnivore tooth marks or bone reduction patterns typical of carnivores on specimens should alert analysts to the possibility that such actors may have deleted less durable bones from the assemblage. In an application of his recommendations, Lyman (1985: 233) argued that Gatecliff Shelter ‘s (Thomas and Mayer 1983) inverse bulk utility pattern in the bighorn sheep assemblage could have been structured by forces other than human selectivity. Carnivore action is evident on specimens , and post-depositional rock falls affected the bone assemblage. He was careful to state that he has not “proved” that attrition rather than selective transport by humans is responsible for the curve. Rather, he urges circumspection in making such behavioral interpretations. Lyman noted that, in contrast, Speth ’s (1983) Garnsey archaeofauna shows very little evidence of carnivore action, fluvial transport, or weathering on the well-preserved bones. While skeletal parts of low utility were more abundant than those of high utility at the Garnsey Site , some high utility bones are present in very good states of preservation . Lyman argued that, in this case, the bone durability evidence supports Speth ’s interpretation of the assemblage as the product of human decisions based on relative utilities of elements.

Grayson (1988) analyzed the archaeofauna from Last Supper Cave , Nevada, and found a pattern of element representation for Ovis canadensis very similar to that of Gatecliff Shelter . Given Lyman ’s discussion of differential bone durabilities, he explored these in relation to VD and to MGUI , using Kendall's tau rank order correlation analysis. He proposed that inverse utility curves caused by human selection should show a significant negative correlation between MGUI and relative frequencies of elements, expressed as %MAU, which in the the era he was writing, was commonly used – see Chap. 18 (Grayson 1988: 70). By contrast, he contended that, in situations where only dense bones have survived attrition , there should be a significant positive correlation between VD and the element frequencies . With the Last Supper Cave assemblage, Grayson found that the relationship between VD :%MAU was positive and very significant, while that between %MAU :MGUI was negative but not significant. Gatecliff Shelter produced a similar result, with VD :%MAU , P = 0.0001, and a negative correlation of MGUI :%MAU p = 0.05. Without asserting a systematic relationship between MGUI and VD , Grayson inferred that these assemblages were more heavily affected by attrition than by human selectivity. On the basis of bone surface modifications, he concluded that carnivores are the most likely causal agents. Grayson also evaluated Anavik, Binford ’s Nunamiut kill-butchery site: the site showed no significant relationship between element frequencies and VD , but did display a highly significant negative correlation between frequency of skeletal element and MGUI , supporting Binford ’s description of activities there.

Lyman (1991) built on Grayson ’s predictive model in evaluating whether and how 67 ethnoarchaeological, archaeological, and paleontological faunas conformed to Grayson ’s 1988 predictions, using Spearman’s rho rank order correlation coefficients. Lyman constructed a nine-cell matrix (1991: 130), with significance values for %MAU :MGUI on one axis and those for %MAU :VD on the other. Lyman ’s results indicate that the Anavik kill-butchery fauna falls where it “should,” as a kill-butchery site from which high-utility elements had been removed, with a significant negative correlation between %MAU and MGUI and no positive correlation of %MAU :VD . Of special note is the fact that, despite archaeological evidence that some archaeofaunas derive from kill-butchery episodes, no archaeological sample displayed a similar pattern to Anavik’s, although some showed negative correlations of both %MAU :MGUI and %MAU :VD . Lyman interprets this as reflecting attrition by post-depositional taphonomic processes operating on the original element frequencies produced by human selectivity.

Taking a very different approach to the study of bone attrition , Cleghorn and Marean ( 2004) argued that the evidence for carnivore, especially spotted hyena (Crocuta crocuta), bone destruction in Pleistocene Eurasian and African archaeofaunas is so pervasive that methodological coping tactics were required. Since actualistic research on hyenas’ bone-destroying capabilities in captivity and in the wild (Marean et al. 1992; Blumenschine 1988; Marean and Spencer 1991; Capaldo 1998) are demonstrated, Marean and Cleghorn recommended that zooarchaeologists in any region with such large, bone-consuming carnivores use only the highest survival elements and portions of elements to estimate MNE . For long bones, these would be diaphyseal fragments, rather than epiphyseal ends; for skulls, these would be teeth and petrous bones (Bar-Oz and Dayan 2007). See also Marean et al. (2004) for a discussion of ignoring diaphyseal fragments in analyses.

Stiner (2002) objected to this and other arguments for diaphysis -based estimates of abundance on several grounds, citing agreements between shaft-based and epiphysis -based MNE estimates in the Palaeolithic Mediterranean archaeofaunas with which she had worked. Stiner also noted that published captive hyena feeding experiments used sheep elements and body segments, which the hyenas readily consumed. She stated that it was an open question whether the much larger ungulates recovered from Pleistocene archaeological sites would have been so completely reduced by Crocuta.

More recently, Janzen and Cleghorn (2010) presented data from experiments in the same hyena colony used by Marean et al., using domestic cattle vertebrae, scapulae, innominates, and femora, some of the latter whole, others broken and with marrow removed. They reported that the extent of bone destruction by hyenas in the experiment appeared to depend upon the social rank of the individual given the elements. Nevertheless, they found that hyenas were capable of demolishing even complete cattle femurs. As with the smaller animals, the hyenas consumed vertebrae leaving only a few scraps. Femoral portions were often consumed nearly completely, but differences were apparent in the treatment of shafts, depending upon whether the hyenas were presented with whole versus marrow -extracted bones. These data suggest that, if hyenas can have access to raw skeletal parts discarded by humans, larger axial elements often will be consumed, as will at least the epiphyses of long bones. Thus, the experiments established the species’ capabilities, as have Lupo’s (1995) observations on spotted hyena destruction of uncooked large ungulate elements hunted by Hadza foragers, and Blumenschine ’s and others’ field-based experiments cited above.

However, I believe Stiner is correct to note that not all of Crocuta’s former ranges displayed the levels of carnivore packing and interspecific competition observed in some parts of East Africa (Chap. 12). Lower rates of bone consumption are hinted at in African areas where hyena populations are less dense (e.g. Lam 1992; Egeland et al. 2008; Kruuk 1972). If at all possible, the prudent zooarchaeologist will seek contemporaneous paleontological information on carnivore species diversity before choosing a counting strategy, and checking both diaphyseal and epiphyseal estimates may be a good way to proceed. This is especially relevant in light of Morin et al.’s (2016) blind tests results, in which estimates based on long bone diaphyses were notably poor estimators of original skeletal element abundances (Chap. 18). Some have contested Stiner ’s (2002) anatomical region profiling method, but her careful cross checking of epiphyseal versus shaft MNE determinations is an intelligent way to explore the extent of BMD -based attrition , as are Grayson ’s (1988) statistical explorations.

These matters may seem only of concern to paleoanthropologists, but any zooarchaeologist who believes that their assemblages were affected by some form of serious mechanical attrition and considers using only high-survival elements should consider how this in turn may introduce the potential for sample-size effects, and hence reduced statistical power, in their estimates. Faith and Gordon (2007) used simulation studies to assess whether analysts following Cleghorn and Marean ’s recommendation to use only high-survival elements risk bringing their sample sizes below levels that produce reliable results. They evaluated sample-size effects on commonly applied measures of association employed to discern the influence of selectivity, such as correlations of %MAU with SFUI (Chap. 20). Their goal was to explore whether decreasing sample sizes affected the rates of Type I errors (in which significant correlations exist in the sample, but they do not in the parent population ) or Type II errors (in which no correlation exists in the sample, but a significant one exists in the parent population ).

From each sample population, random draws of 250, 150, 100, and 50 elements were made, each 5000 times. The evenness (E) of the distribution of skeletal elements and Spearman’s rho correlation coefficients of SFUI to element frequencies were calculated for each draw.

For samples from the gourmet and unbiased populations, Type II errors increased as sample size decreased but remained relatively low. However, with bulk transport samples, where all but lowest utility elements are transported, Type II errors jumped from 1.9 to 48.7% as the sample MNE dropped from 250 to 50 (Faith and Gordon 2007: 876, Table 3). They conclude that bulk transport assemblages are especially sensitive to sample size effects . Recalling that the unconstrained parent population had no prior correlation between SFUI and element frequencies , Type I error rates are of interest: these occurred in 10–12% frequencies, regardless of sample size.

Applying the Shannon evenness index to each parent population and sample, the authors note that the bulk and unconstrained use patterns are quite similar, approaching or at 1.000, whereas the unbiased is “intermediate” in evenness (0.842), while the gourmet population, dominated by femora, has the greatest unevenness, at 0.369 (Faith and Gordon 2007: 876, Table 1). Reducing sample size differently affects evenness in samples drawn from the four populations. The gourmet strategy remains distinct from the others, regardless of sample size. At 250, the remaining three samples discriminate well from one another, but at 150 and 100 MNE , the evenness values of the bulk and unconstrained samples begin to overlap, while the unbiased sample remains distinct. At 50 elements, the bulk, constrained, and unbiased samples are indistinguishable. Faith and Gordon illustrate these points with three archaeofaunal analyses for which the data were available: Porc-Épic (Porcupine) Cave, Ethiopia, Die Kelders Cave , and Olduvai FLKN levels 1–2, showing possible sample size effects and possible transport strategies.

To sum up, when sample sizes drop below 150, correlation coefficients may not accurately reflect composition of the parent population . Samples from the relatively even, bulk, and unconstrained populations displayed greater Type II errors than did unbiased and gourmet samples. Bulk sample assemblages have the highest Type II error rates because a lack of correlation of SFUI with element frequencies is quite possible. Faith and Gordon argue that the Shannon evenness index can help analysts distinguish between transport strategies, although bulk and unconstrained transport strategies will produce similar evenness values. One should also recall that, except in mass kill situations, permutations of the bulk or unconstrained transport tactics were probably the most common accumulators of archaeofaunal deposits.