Establishing a Methodology for Determining Handedness in Lithic Materials As a Proxy for Cognitive Evolution

Home , Eraillure, Experimental archaeology, Lithic analysis

EXPERIMENTAL ARCHAEOLOGY AND HOMINID EVOLUTION:

ESTABLISHING A METHODOLOGY FOR DETERMINING HANDEDNESS IN

LITHIC MATERIALS AS A PROXY FOR COGNITIVE EVOLUTION

Lana Ruck

A Thesis Submitted to the Faculty of

The Dorothy F. Schmidt College of Arts and Letters

In Partial Fulfillment of the Requirements for the Degree of

Master of Arts

Florida Atlantic University

Boca Raton, FL

December 2014

ACKNOWLEDGEMENTS

I would like to thank my thesis committee members, Dr. Douglas Broadfield, Dr.

Clifford Brown, and Dr. Kate Detwiler, for their constant support and help with developing this project, as well as the head of the Department of Anthropology, Dr.

Michael Harris, for his insights. This project would not have been possible without the help of my volunteer flintknappers: Ralph Conrad, Mike Cook, Scott Hartsel, Ed Moser, and Owen Sims, and my raw materials suppliers: Curtis Smith and Elliot Collins. I would also like to thank Miki Matrullo and Katherine Sloate for cataloging my handaxes and flakes and aiding me in creating a blind study. Special thanks to Justin Colón and Dr.

Clifford Brown for assessing a random sample of my flakes, adding objectivity to this study. Finally, I would like to thank Dr. Natalie Uomini for her constant help and support of my project.

iv ABSTRACT

Author: Lana Ruck

Title: Experimental Archaeology and Hominid Evolution: Establishing a Methodology for Determining Handedness in Lithic Materials as a Proxy for Cognitive Evolution

Institution: Florida Atlantic University

Thesis Advisor: Dr. Douglas Broadfield

Degree: Master of Arts

Year: 2014

Human handedness is likely related to brain lateralization and major cognitive innovations in human evolution. Identifying handedness in the archaeological record is, therefore, an important step in understanding our cognitive evolution. This thesis reports on experiments in identifying knapper handedness in lithic debitage. I conducted a blind study on flakes (n=631) from Acheulean handaxes replicated by right- and left-handed flintknappers. Several flake characteristics significantly indicated handedness, with a binary logistic regression correctly predicting handedness for 71.7% of the flakes.

However, other characteristics were not associated with handedness. This is a result of personal knapping styles, as additional analyses show that individual knappers associate with some attributes better than handedness does. Continued work on these methodologies will enable analysis of Paleolithic assemblages in the future, with the

v ultimate goal of tracking population-level hominid handedness rates through time and using them as a proxy for cognitive evolution and language acquisition.

vi DEDICATION

To my closest friends and family, who now know more about handedness than they ever cared to. Thanks for putting up with me every day.

EXPERIMENTAL ARCHAEOLOGY AND HOMINID EVOLUTION:

ESTABLISHING A METHODOLOGY FOR DETERMINING HANDEDNESS IN

LITHIC MATERIALS AS A PROXY FOR COGNITIVE EVOLUTION

LIST OF TABLES ...... x LIST OF FIGURES ...... xii INTRODUCTION ...... 1 History of “Handedness” ...... 1 Brain Complexity: Hemispheric Specialization and Lateralization ...... 3 Issues in Studying Human Asymmetries ...... 4 Lithic Analysis and the Value of Paleolithic Technology ...... 6 Experimental Archaeology ...... 7 PREVIOUS RESEARCH ...... 10 Living Humans: Asymmetries in the Body and Brain ...... 10 Non-human Primates: Is Handedness an Autapomorphy? ...... 15 The Hominid Fossil Record: Direct and Indirect Evidence for Handedness ...... 18 METHODS ...... 30 Creation and Collection of Materials ...... 30 Preliminary Lithic Cataloging and Analysis ...... 32 Identification of Handedness-Indicative Features—Toth and Rugg and Mullane ...... 35 Identification of Handedness-Indicative Features—Bargalló and Mosquera ...... 38 Reconstruction of Assemblages and Statistical Methods ...... 41 RESULTS ...... 45 Assessment of Toth’s and Rugg and Mullane’s Methods ...... 45 Assessment of Bargalló and Mosquera’s Methods—non-statistical measures...... 48 Binary Logistic Regression—which traits predict handedness? ...... 52 Additional Analysis—Effects of Knapping Style on Flake Debitage ...... 74

viii Inter-observer Comparisons—Assessment of Bargalló and Mosquera’s method’s reliability ...... 78 CONCLUSIONS...... 85 Synthesis of Results for the Toth, Rugg and Mullane, and Bargalló and Mosquera methodologies ...... 85 Additional Considerations and Recommendations ...... 89 Implications of this Study: Applying Experimental Data to Fossil Assemblages ...... 94 Cognitive Evolution and Language Acquisition ...... 96 APPENDICES ...... 97 A—Subset Handedness Inferences by Handaxe ...... 98 B—Overall Frequency Data for Technical Features ...... 100 C—Binary Logistic Regression Case Processing Summaries ...... 103 D—Technical Features Frequency Data by Handedness ...... 104 E—Technical Features Frequency Data by Knapper ...... 108 F—Overall Frequency Data for Technical Features: Observer B ...... 115 G—Overall Frequency Data for Technical Features: Observer C ...... 118 H—IRB Approval ...... 121 REFERENCES ...... 122

LIST OF TABLES

Table 1: Summary of classification sets and their associated types ...... 38 Table 2: Meta-assemblage breakdown by handaxe and knapper ...... 42 Table 3: Summary of the Toth (1985) cortex-based method ...... 46 Table 4: Summary of the Rugg and Mullane (2001) cone of percussion method ...... 48 Table 5: Evaluation of overall handedness inferences by handaxe ...... 49 Table 6: Binary logistic regression values for the cone of percussion subset...... 56 Table 7: Measures of significance for the cone of percussion regression ...... 56 Table 8: Summary of predictive correctness for the cone of percussion regression subset...... 58 Table 9: Binary logistic regression values for the eraillure scar subset...... 59 Table 10: Measures of significance for the eraillure scar regression ...... 59 Table 11: Summary of predictive correctness for the eraillure scar regression subset...... 59 Table 12: Binary logistic regression values for the platform subset...... 60 Table 13: Measures of significance for the platform regression...... 60 Table 14: Summary of predictive correctness for the platform regression subset ...... 61 Table 15: Binary logistic regression values for the cortex subset...... 62 Table 16: Measures of significance for the cortex regression ...... 62 Table 17: Summary of predictive correctness for the cortex regression subset ...... 63 Table 18: Binary logistic regression values for the fracture location subset...... 64 Table 19: Measures of significance for the fracture location regression ...... 64 Table 20: Summary of predictive correctness for the fracture location regression subset...... 64 Table 21: Binary logistic regression values for the full data set...... 65 Table 22: Measures of significance for the full regression ...... 66 Table 23: Summary of predictive correctness for the full regression ...... 67 Table 24: Binary logistic regression values for the final step (17) of the backwards regression...... 68 Table 25: Measures of significance for steps 1 and 17 of the backwards regression ...... 68 Table 26: Summary of predictive correctness for steps 0, 1, and 17 of the backwards regression ...... 69 Table 27: Binary logistic regression values for the final unsaturated model...... 70 Table 28: Measures of significance for the final regression ...... 70 Table 29: Summary of predictive correctness for the final regression ...... 70 Table 30: Chi-squared values for the cone of percussion subset by handedness (left) and knapper (right) ...... 75 Table 31: Chi-squared values for eraillure scars and the platform subset by handedness (left) and knapper (right) ...... 76

x Table 32: Chi-squared values for cortex and fracture locations by handedness (left) and knapper (right) ...... 77 Table 33: Summary of Observer B's overall handedness inferences by handaxe ...... 79 Table 34: Summary of Observer C's overall handedness inferences by handaxe ...... 79 Table 35: Fleiss' Kappa values for each characteristic using all three observers ...... 81 Table 36: Weighted Cohen's Kappa values for each characteristic using pairs of observers ...... 83

LIST OF FIGURES

Figure 1: Handaxes created by knappers ...... 34 Figure 2: Assortment of labelled flakes, ventral aspect ...... 34 Figure 3: Ventral (left) and dorsal (right) faces of a right-oriented cortical flake ...... 35 Figure 4: Ventral (left) and dorsal (right) faces of a left-oriented cortical flake ...... 35 Figure 5: Ventral face of flake with a left-skewed cone of percussion ...... 36 Figure 6: Ventral face of flake with a right-skewed cone of percussion ...... 37 Figure 7: Ventral face of flake with a centered cone of percussion ...... 37 Figure 8: Graphic description of flake characteristics and their associated handedness inferences, modified from Bargalló and Mosquera (2013)...... 40 Figure 9: Ventral views of two right-handed flakes showing opposite skew in characteristics ...... 51 Figure 10: Dorsal view of two right-handed flakes showing variable cortex location. Flake 870, left; Flake 648, right...... 51 Figure 11: Example of flake characteristics, associated matrices, and beta values...... 72

xii

INTRODUCTION

Unique traits of Homo sapiens, such as bipedal locomotion, large and complex brains, culture, and language are often cited—by professionals and the public—as what make us human. One often-ignored aspect of human uniqueness, however, is handedness, although recent research has begun to highlight its importance in our evolution. Between

85 and 90% of living humans are right-handed, and extreme, population-level hand dominance is considered by many to be a unique part of what makes us us (Annett 1985;

Corballis 1983). In order to determine if this is true, researchers must investigate the benefits, uniqueness, and evolutionary origins of handedness in our species. As is common in paleobiology, particularly regarding the human fossil record

(paleoanthropology), this task can quickly become complicated. Despite this, several multidisciplinary approaches to understanding handedness in Homo sapiens, other non- human primates, and fossil hominids have been developed. Combining our current knowledge of handedness in extant humans, its homologous manifestations in non-human primates, and data from fossil evidence may enable us to progress in our understanding of what it means to be human.

History of “Handedness”

Right- vs. left-hand dichotomies are present in the historic record as early as four- thousand years ago, and human societies have described and disputed the nature of handedness in literature, philosophy, and science (Corballis 1991; Hewes 1993). While

1 the majority of the historical record of handedness is speculative and even superstitious, discussions in the light of recent scientific paradigms have concluded that handedness is a legitimate topic of study. Perhaps the first scientific inquiries into handedness involved psychological experiments on apraxic and aphasic patients in the late nineteenth century

(Springer and Deutsch 1981). Researchers like Marc Dax, Paul Broca, and Carl Wernicke studied patients with brain damage, and linked specific areas of damage to symptoms involving manual motor impairment, loss of speech, or decreased language comprehension (Corballis 1983; Geschwind and Galaburda 1987; Springer and Deutsch

1981).

In the past century, we have gained a better understanding of the relationship between handedness and the human brain. In the mid-1900s, removal of the corpus callosum (which allows communication between the brain’s left and right hemispheres) and lesions in other parts of the brain (e.g., parietal cortex of either hemisphere) were relatively common medical procedures. These so-called “split-brain studies” were performed on individuals with epilepsy or other previous brain damage, and their results definitively linked the motor function of our hands to specific areas in the brain

(Coolidge and Wynn 2009; Corballis 1983; Geschwind and Galaburda 1987; Kochetkova

1978; Springer and Deutsch 1981). While many split-brain studies focused on language comprehension and speech production, and changes in manual motor performance were noted as secondary manifestations of brain damage, these studies incited specific research on manual motor characteristics, including natural asymmetries. Handedness has now become a topic of interest in many fields, including developmental psychology, genetics, linguistics, paleoneurology, primatology, and others.

2 Brain Complexity: Hemispheric Specialization and Lateralization

Large brains are not unique to the human lineage, and most primates have relatively large brains for their body size (Aiello and Dunbar 1993; Cantalupo and

Hopkins 2008; Corballis 2007; Hopkins and Rilling 2000; Rilling 2006). What makes us unique is that our brain is three times bigger, relatively speaking, than any other extant primate (Falk 1980; Holloway et al. 2003; Rilling 2006, 2008). One very important aspect of brain growth in our ancestors is hemispheric specialization. As hominid brains got larger, connectivity within brain halves became more important than connectivity between halves, and each half of the brain likely became more and more isolated from the other (Cantalupo and Hopkins, 2008; Corballis 2009; Hopkins and Rilling 2000; Stout and Chaminade 2012). Lateralization of the brain occurs when, due to hemispheric specialization/isolation, one side of the brain becomes functionally dominant to the other for specific tasks (Cantalupo and Hopkins, 2008; Falk 1987; Hopkins and Rilling 2000;

Vallortigara and Rogers 2005). Early evidence of hemispheric specialization in apraxic and aphasic patients proved that although the human brain can sometimes recover from hemispheric damage, certain tasks are typically performed by either the left- or right- hemisphere (Geschwind and Galaburda 1987; Springer and Deutsch 1985).

Increases in technology, such as functional magnetic resonance imaging (fMRI), computed tomography (CT/CAT), positron emission tomography (PET) scans, and functional transcranial Doppler ultrasonography (fTCD), have allowed researchers to monitor neural blood-flow during fine motor movements, speaking, listening, and other tasks in living humans (Balzeau, Glissen, and Grimaud-Hervé 2011; Bishop, Watt, and

Papadatou-Pastou 2009; Hecht et al. 2014; Rilling 2008; Stout, Toth, and Schick 2006;

3 Stout et al. 2008; Uomini and Meyer 2013). By understanding which areas of the brain respond (or “activate”) during various tasks, we can understand which parts of the brain associate to which behaviors. These studies have supported early split-brain research and confirmed a contralateral link between our hands and their associated brain hemisphere

(the left hemisphere controls most right-handed actions, and the right hemisphere controls most left-handed actions) (Amunts et al. 1996; Corballis 1991; MacNeilage 2005; Stout and Chaminade 2012). The contralateral sensorimotor system is present in several animal species, including primates, hominids, and modern Homo sapiens (Cantalupo et al. 2008;

Fagot and Vauclair 1991; McGrew and Marchant 1997; Phillips and Hopkins 2007;

Phillips and Thompson 2012). Today, much of the human brain has been extensively mapped, and the two halves of the brain are generally well-understood by neurobiologists, psychologists, and anthropologists, among others. Additionally, manual motor lateralization has been repeatedly linked extensively and nearly exclusively to neural areas responsible for human language, making handedness a viable proxy for understanding aspects of brain lateralization and language acquisition in our species and our ancestors (Corballis 2003; Hecht et al. 2014; Marshack 1976; Noble and Davidson

1996; Ruck 2014; Stout, Toth, and Schick 2000; Stout et al. 2008).

Issues in Studying Human Asymmetries

Functional and morphological asymmetries in living humans’ bodies and brains are much less of a mystery today than in the past, but debates about the classification, inheritance, and evolution of handedness are common. Few evolutionary models exist for the development of human handedness, and models involving brain lateralization and

4 language acquisition are even more speculative in nature (Annett 1985, 1998; Bryden et al. 1997; Calvin 1993; Corballis 2002; McGrew and Marchant 1997; Provins 1997;

Reynolds 1993). One leading theory regarding handedness inheritance is that of Marian

Annett: the “right-shift theory” is a simplified genetic-based model for handedness inheritance, claiming that distributions of handedness in extant human populations fit a simple RS+/RS- allele pattern (where RS+ represents a right-shift in hand dominance, and RS- reflects equality between two hands) (Annett 1985). Criticisms of Annett’s theory stem from inherent epigenetic issues, such as: classifying handedness as naturally dichotomous (when in fact it is a spectrum) and identifying individual genes responsible for “phenotypic” handedness (Crow 2009; Crow et al. 2009; McManus1985). Regardless of these issues, few alternate explanations of the biological bases of handedness and its link to hemispheric asymmetry in the brain exist, and the “right-shift” concept is extremely useful when studying humans, non-human primates, and hominids.

Extending current knowledge on human handedness and brain lateralization backward into the fossil record has been extremely difficult due to an inherent issue in paleoanthropology: preservation. We do not understand language acquisition in hominids, for example, because words do not fossilize. This is the case for many of the characteristics that distinguish Homo sapiens from others, such as: specific genetic changes, cognitive complexity, cultural innovation, etc. Despite decades of field work in

Africa, Asia, the Near East, and Europe, paleoanthropologists often have less primary evidence in the form of fossils than many other fields, and thus must develop methodologies that allow them to glean as much information as possible from small

5 sample sizes (Cashmore, Uomini, and Chapelain 2008; Holloway 2010; Toth and Schick

1986; Wynn 2002).

Due to this, research on extant humans and non-human primates has become a necessary component of paleoanthropology, and these studies are sometimes more useful than the fossil remains themselves. According to N. Toth and K. Schick, leading researchers on hominid tool use:

[t]he paleoanthropologist's dream is to discover some relict population of

protohumans . . . who have remained virtually unchanged in some 2 million years

of hominid evolution. . . The possibility of such a discovery is, alas, nonexistent

. . . [so] we are restricted to . . . the imperfect analogues of modern phenomena in

our attempts to gain insights into how early hominids lived (Toth and Schick

1986: 1).

A multi-faceted analytical approach often incites skepticism and debates, but currently comprises the only viable option for studying the most significant aspects of the human character, and is particularly important in the study of handedness.

Lithic Analysis and the Value of Paleolithic Technology

Preservation will always be an issue for paleoanthropologists, but some aspects of human evolution do preserve relatively well, including stone tools. Lithic technology dates from around 2.5 million years ago (mya), and is directly associated with several hominid species, including all later species of the genus Homo (Ambrose 2001; Bordes

1968; Toth and Schick 1986). The Paleolithic period spans the Plio—Pleistocene transition, and dates from around 2.5 mya to 10,000 years ago (10 kya); it represents over

6 95% of the history of technology, and has three main phases: Lower, Middle, and Upper

(Bordes 1968). The Lower Paleolithic (2.5 mya-500 kya) includes the earliest stone tools: the Oldowan industry, comprised of simple flaked cobble/pebble tools, and the

Acheulean, characterized by more intensive manufacture, bifacial flaking, and somewhat more varied end-products (Bordes 1968; Hopkinson and White 2005; Noll and Petraglia

2003). The Middle and Upper Paleolithic periods (300 kya-10 kya) represent an explosion in tool forms, manufacture methods, and extreme cultural variation of lithic forms. They include a multitude of industries, such as (but not limited to): Mousterian,

Aterian, Gravettian, and Solutrean traditions (Ambrose 2001; Bordes 1968).

Lithic analysis is a significant aspect of paleoanthropology and archaeology, and over a century of research has been devoted to establishing industry-specific descriptive and analytical methods (Andrefsky 1998). The main goal of lithic analysis was creating typologies and sequences of stone tools for much of the profession’s history. However, in the 1960s, many researchers shifted away from description and chronology towards understanding the process of flintknapping and what it meant for the peoples they studied

(Pelegrin 2009; Whittaker 2007). While much is currently known about the manufacture and use of Paleolithic stone tools, paleoanthropologists are just beginning to develop approaches relevant for studying handedness and brain lateralization via stone tools.

Experimental Archaeology

A recent trend in lithic analysis that is particularly relevant to this thesis involves the experimental re-creation of Paleolithic tools (see Amick and Mauldin 1989; Ferguson

2010; Keeley 1980; Toth and Schick 2009). Experimental archaeology has existed for

7 several decades (Andrefsky 1998), and has enabled us to understand how previous peoples—including our hominid ancestors—relied upon material culture and even how they lived (Davidson and McGrew 2005; Moore 2012; Nowell and Davidson 2010; Toth and Schick 2009; Wynn 2002). A key aspect of lithic-based experimental archaeology is flintknapping. Flintknapping is the process by which stone tools are made, and is a hobby for many professional archaeologists and members of the public (Whittaker 2007). In the initial stages of stone tool manufacture, knappers often hold raw material (such as chert, quartzite, obsidian, or basalt) in their non-dominant hand and hit it with another material

(either a hammerstone or billet) rapidly with their dominant hand; this process is called direct percussion, and is the only necessary step for manufacturing many Paleolithic tools

(Andrefsky 1998; Semenov 1964; Whittaker 2007). Since these materials have predictable fracture mechanics (see Cotterell and Kamminga 2000), knappers can remove flakes successively until they have made their desired product or exhausted their core.

Removed flakes can either be used as tools themselves, be modified to form more complex tools, or get discarded, and there is evidence of each in the fossil record

(Cotterell and Kamminga 1979; Semenov 1964; Toth and Schick 1986). Later stages of flintknapping (present in some Mode II and Mode III Paleolithic tools) include additional techniques of flake removal, such as pressure flaking and other modifications (e.g. heat treating and hafting) (Bordes 1968; Semenov 1964; Whittaker 2007).

Flintknapping experiments have historically been useful for determining manufacture methodologies likely used by hominids, but combined with neural scanning technologies, the re-creation of Paleolithic tools has changed the nature of lithic-based experimental archaeology, adding a cognitive aspect (see de Beaune, Coolidge, and

8 Wynn 2009; Gibson and Ingold 1993; Nowell and Davidson 2010; Roux and Bril 2005).

This cognitive shift in lithic analysis, combined with new understandings of handedness and brain lateralization, has led to several studies attempting to establish handedness in the hominid fossil record in the 1980s and 1990s, with a recent revival in interest in the twenty-first century (for a review, see Uomini 2001, 2006, 2009). Studies on the evolution of handedness are limited in number, and only a few of them reflect experimental approaches to inferring handedness from lithic evidence (Bargalló and

Mosquera 2013; Patterson and Sollberger 1986; Pobiner 1999; Rugg and Mullane 2001;

Toth 1985; Uomini 2001, 2006). These seven publications are speculative in nature and often debated, but they represent innovative approaches in paleoanthropology and form the basis of this thesis.

9 PREVIOUS RESEARCH

Because higher cognition and language are seen by many to be uniquely “human” traits, research on these topics is almost immeasurable; publications span from implausible, erroneous speculations to dense, intellectual debates. Literature specifically regarding the link between handedness, brain lateralization, and language is slightly more manageable, and provides the necessary context for this thesis. However, the scope of the proposed research specifically involves stone tools, and the works reviewed below discuss manual and neural lateralization as it relates to tool use and manufacture in extant humans, non-human primates, and finally, extinct hominids.

Living Humans: Asymmetries in the Body and Brain

There are several classification schemes and methodologies for determining handedness of individuals, but the study of handedness in living humans has become increasingly complex in recent decades. Historically, handedness studies have gone from simple right- vs. left-hand dichotomies to reflecting the natural spectrum of hand preference and performance. Simple questionnaires, such as the Edinburgh Handedness

Inventory and The Waterloo Handedness Questionnaire are often cited as over-reporting strict hand preference, while actualistic and ethological approaches are considered as having more accurate representations of the spectrum of manual motor lateralization

(Cashmore, Uomini, and Chapelain 2009; McGrew and Marchant 1996).

10 Other recent directions in research discard the dichotomization of handedness overall, instead focusing on the differential roles of each hand for bimanual actions

(Guiard 1987). Despite debates regarding the classification of handedness, one widely accepted scheme, introduced by W. C. McGrew and L. F. Marchant (1997) has four main categories:

1) Hand Preference: individual bias towards one hand for a single task.

2) Hand Specialization: individual bias towards one hand across multiple tasks.

3) Task Specialization: a group of individuals biased to the same hand for single tasks.

4) Handedness: a group of individuals biased towards the same hand for multiple tasks.

In the same publication, McGrew and Marchant (1997) propose varying levels of manual laterality as a framework for future research: Level 1, where all individuals have no hand preference whatsoever; Level 2, where most subjects show some hand preference, but there is incomplete lateralization of the overall population; Level 3, where all subjects show complete right- or left-hand specialization, but the side remains randomly distributed (indicative of a point in time before the right shift occurred); Level 4, where most individuals are significantly lateralized to one side of the distribution, but are not exclusively hand-dominant for all tasks; and Level 5, where all individuals are completely lateralized in one direction (McGrew and Marchant 1997). According to these schemes, many agree that Homo sapiens exhibit Level 4 right-handedness in all extant populations.

11 Recent quantitative research has shown that skeletal markers of handedness can often be found in living humans, and even historic and pre-historic human remains (Shaw

2011; Ubelaker and Zarenko 2012). Skeletal remains can indicate and preserve many bilateral asymmetries—including handedness—via life-long differential use of limbs (e.g. lifting heavier loads with one arm vs. the other; writing solely with one hand, etc.).

Publications on handedness estimations via the scapulae, clavicles, humeri/radii/ulnae, and various hand bones are extensive (see Steele 2000 for a review). Steele also discusses the study of cranial bilateral asymmetries, including brain-scans and cranial landmarks.

Studying handedness via cranial asymmetries can be methodologically problematic, but theoretically it is a valid approach. This is due to another well-established population- level asymmetry in the brain: the left-lateralized expansion of the planum temporale and other areas associated with language. The planum temporale is a region of the brain within the Sylvian fissure in the parietal cortex of both brain hemispheres, just posterior the motor cortex in the temporal region. It is the most asymmetrical area in the brain, sometimes up to five times larger on the left half of the brain—and at least a little larger on the left side in 65% of living humans (Uomini 2001, 2006). One important area of the left-cerebral planum temporale is Wernicke’s area, which is used for language comprehension in Homo sapiens. Over a century of research, including the earliest split- brain studies, has linked Wernicke’s area to another left-lateralized cortical area for language production: Broca’s area, a part of the inferior frontal gyrus, is located just anterior to the motor cortex in the frontal lobe (Corballis 2007, 2009; Geschwind and

Galaburda 1987; Springer and Deutsch 1981).

12 Aside from their accepted roles in the production and comprehension of human language, Broca’s and Wernicke’s areas have a unique relationship with the primary sensory and motor areas of the brain: in the left hemisphere, Broca’s area is anterior to the primary motor cortex, while Wernicke’s is posterior to the primary sensory cortex

(Corballis 2007, 2009; Steele and Uomini 2009; Stout and Chaminade 2012; Stout et al.

2008). As mentioned before, humans have a contralaterally linked somatosensory system, implying that the areas responsible for sensory input and motor control of the right hand are theoretically bounded by the areas responsible for language comprehension and production in humans. While roughly 96% of right-handed individuals have the predicted left-hemispheric location of Broca’s and Wernicke’s areas, only 40% of left-handers have

Broca’s and Wernicke’s areas in their right cortical hemisphere (Corballis 1991).

Certainly, a link between handedness, cerebral lateralization, and language exists in our species, but little is known about the mechanisms involved in the evolution of these traits

(Ruck 2014).

Research on tool manufacture and use in humans has elucidated some of the mysteries related to handedness and brain lateralization, and many researchers have noticed a stronger right-hand preference for tasks where fine motor precision or rapid movements are required—such as writing, throwing, sewing—versus left-hand recruitment for slow, supportive, or generalized tasks—including holding paper in place while writing, fabric while sewing, etc. (Guiard 1987; Steele and Uomini 2009; Uomini

2009). In terms of flintknapping, this bimanual system (right-hand precision, left-hand support) is essential to successful tool manufacture (Uomini 2009). In order to further understand this, N. Uomini (2009) conducted an ethological study on handedness where

13 writing, nut-cracking, and lithic refitting strategies were studied at a festival on unknowing subjects. Interestingly, each hand had distinct, predictable roles for each task: the hand holding the hammer for nut cracking matched the writing hand in all cases, but lithic refitting had the most ambidextrous role differentiation, and writing hand did not predict which hand held supportive material vs. which hand placed the refits (Uomini

2009).

In a PET scan study, D. Stout, N. Toth, K. D. Schick, and T. Chaminade (2008) observed brain activation centers of both novice and expert flintknappers producing Early

Stone Age (ESA) tools, including Oldowan (novices and experts) and Acheulean (experts only) types; they noted more bilateral neural activation in experts, suggesting that left- hand support recruitment and bimanual coordination was more important to experts than novices (Stout et al. 2008). In a study conducted by N. Geribàs, M. Mosquera, and J. P.

Vergès (2010), novice and “expert” (meaning any experience with flintknapping, even solely theoretical) knappers were asked to make Acheulean handaxes out of brick; all knappers used their right hand for knapping, and only one was left-handed for writing.

Both groups were successful at creating a handaxe, but video analysis of the two groups showed that novices used percussion more often than experts (more rapid, focused, right- hand use), and experts used more turns (more slow, supportive, left-hand use) (Geribàs,

Mosquera, and Vergès 2010).

Just as the left hemisphere is associated with language functions and right- handedness, the right hemisphere is associated with visuospatial information processing, perception, sequencing (of time and events, etc.), and left-handedness (Corballis 1989,

2007; Geschwind and Galaburda 1987). The left-hand support roles for tool manufacture

14 and increased turning/planning in experts vs. novices contrast to right-hand action mechanisms, like hard hammer percussion and precision; these traits also correspond directly to lateral asymmetries in the brain (Stout 2005; Stout et al. 2008). Informed by

Guiard’s (1987) bimanual or complementary role differentiation (CRD) model, and

McGrew and Marchant’s (1997) classification schemes, these studies on tool production have suggested that flintknapping has a complex relationship with handedness and brain lateralization, and that perhaps the adaptation of stone tool production is either a product of, or a precursor to, current manifestations of handedness and language in Homo sapiens.

Non-human Primates: Is Handedness an Autapomorphy?

Understanding the biological bases of behavior in Homo sapiens has historically provoked several studies on non-human primates, particularly with the great apes, and the same is true specifically regarding brain lateralization, language, tool use, and handedness. One important conclusion of non-human primate studies is that often, there is evidence for human-like morphological and functional asymmetries to some degree in our primate relatives. It is true that human brains are much more complex than any other extant primate, but many of our hemispheric asymmetries, once thought to be unique to our species, actually have homologs in other primates. Although sample sizes are currently small and it seems that individual variation is much greater in apes, some species show morphological asymmetries between hemispheres in the primary motor cortex, planum temporale, inferior frontal gyrus, and primary visual cortex (Balzeau,

Glissen, and Grimaud-Hervé 2011; Cantalupo et al. 2008, 2009; Gannon et al. 1998;

15 Holloway, Broadfield, and Yuan 2003; Hopkins and Rilling 2000; Sherwood et al. 2003).

As discussed before, these areas correspond directly to language, handedness, and visuospatial planning in humans, and thus it is likely that selective pressures caused the intensification of these already existing asymmetries in the neural substrates of our non- human primate, and eventually, hominid ancestors (Hopkins, Russell, and Cantalupo

2007).

Due to the link between our hands, brains, and words, non-human primate manual specialization is important to researchers interested in brain evolution, and studies on primate handedness are far more common than those on cerebral hemispheric lateralization. W. D. Hopkins, who has conducted over 20 studies on chimpanzee handedness at Yerkes Primate Research Center and even more on other species, has argued in several publications that many non-human primates exhibit population-level handedness, similar to humans. He conducted a meta-analysis in 2011 and concluded that gorillas, chimpanzees, and bonobos have all exhibited population-level right-handedness at some point, whereas orangutans have exhibited left-handedness, likely due to unique ecological factors influencing posturing and locomotion (such as feeding) (Hopkins et al.

2011).

Another meta-analysis, conducted by W. C. McGrew and L. F. Marchant (1997), suggested that the individual variability noted in primate hemispheric specialization is reflected in their behavior, including handedness. In their analyses, McGrew and

Marchant acknowledged that there are several issues in non-human primate handedness studies—most importantly, theoretical and methodological inconsistencies between studies—but they concluded overall that non-human primates do not show population-

16 level handedness in any way (McGrew and Marchant 1997). While many studies confirmed low-level (1 or 2), randomly distributed hand preferences in primate populations across species, McGrew and Marchant believe that only chimpanzees (Pan troglodytes) show reliable evidence of level 3 specialization, meaning most individuals were highly lateralized, but randomly between left- and right-hands. This perhaps indicates that the last common ancestor (LCA) between chimps and hominids had not experienced selective pressures for the right-shift (Annett 1985; McGrew and Marchant

1997).

J. Fagot and J. Vauclair (1991) have advocated differential analyses of hand preferences in non-human primates based on tasks; they argue that symmetrical handedness distributions likely exist for “low-level” activities (i.e. reaching), and that increasing task complexity, or “high-level” tasks, likely produce skewed distributions in either direction (Fagot and Vauclair 1991). This view provides a possible intermediary between Hopkins’s and McGrew and Marchant’s meta-analyses, and also parallels evolutionary implications for the selective pressures that lead to human handedness today. As discussed above, tool manufacture is an inherently bimanual activity that is heavily lateralized in humans, and may be naturally more lateralized to right-hand action/left-hand support roles (Hecht et al. 2014; Stout et al. 2008; Steele and Uomini

2009). Studies on non-human primate tool use and manufacture demonstrate the importance of low- vs. high-level tasks on primate handedness as well.

Although very few studies conducted on non-human primates explicitly utilize bimanual role-differentiation theories, they have implicitly adopted Guiard’s (1985) and

Fagot and Vauclair’s (1991) frameworks. Tool-use studies in primates, especially in

17 gorillas, chimps, and bonobos are perhaps more common than any others, and often confirm lateral biases congruent with level 3, task-specialized, manual recruitment.

Several studies on common chimpanzees (Pan troglodytes) and capuchin monkeys

(Cebus apella) have consistently linked hand preference for tool use with cerebellar and cortical asymmetries in the brain (Byrne 2005; Cantalupo et al. 2008; Hopkins et al.

2007; Lonsdorf and Hopkins 2005; Phillips and Hopkins 2007; Phillips and Thomson

2012; Westergaard and Suomi 1994). Little work has been done exclusively on stone tool use in non-primates, as most studies cover a range of activities, including a variety of feeding mechanisms (Byrne 2005; Matsuzawa 2001). Additionally, handedness in each of the groups studied by Hopkins, Phillips, Westergaard, and others seems to be randomly distributed between right and left (see meta-analyses).

The overall consensus from non-human primate studies is that to some extent, handedness and brain lateralization are homologous in all primates, but there is no evidence of a non-human primate source of the right-shift in handedness. Manual motor preference as a result of tool manufacture culminates in human right-hand preference, and this phenomenon is likely an autapomorphy that makes Homo sapiens, and our hominid ancestors, unique. More studies on non-human primate use of stone tools (and perhaps even manufacture, in the case of individuals who have been taught) are required in order for us to understand the role of lithic technology in our evolution.

The Hominid Fossil Record: Direct and Indirect Evidence for Handedness

The paucity of the hominid fossil record is a primary obstacle in studying handedness and cognitive evolution in the hominid lineage, but this has not stopped

18 researchers from trying. There are a number of methods for studying hominid handedness, both directly and indirectly, and the same is true for brain evolution

(Cashmore 2009; Uomini 2009). There are several studies on bilateral asymmetries

(including handedness) in living humans and skeletal remains using cranial and postcranial measures, and the same methods have been used on hominid remains (see

Lazenby 2001; Steele 2000; Steele and Uomini 2005, 2009 for a review). Studies on fossilized post-cranial elements include those on Neanderthal remains from various locations in Europe and the Nariokotome Boy (H. erectus/ergaster), but limitations in preservation of upper limbs (especially those which are validated as paired to one individual) have prevented much work. Despite this, in all current cases, right-side measurements are larger than their left-side counterparts, suggesting, at the very least, asymmetrical biomechanical loading in Homo neanderthalensis and perhaps in H. erectus

(Liguria 1997; Trinkaus et al. 1994; Vandermeersch and Trinkaus 1995).

Direct assessment of cranial features, especially dentition, in Neanderthals has also provided indications of population-level right-handedness in Middle- and Upper-

Pleistocene hominids. It is widely accepted that Neanderthals used their teeth for non- masticatory purposes, often as a “third hand” when manipulating objects, particularly tools and food (Bax and Ungar 1999; Lozano et.al. 2009l Volpato et al. 2012). Living humans also use their teeth as tools in varying frequencies, and ethnographic research on populations with heavy non-masticatory dental use have shown micro-striations on buccal surfaces of incisors and canines, among other features (Lozano et al. 2009). These features have been shown to accurately reflect the handedness of individuals as well via directionality of striations (Bermúdez de Castro, Bromage, and Jalvo 1988; Fox and

19 Frayer 1997). Research on Neanderthal teeth—specifically using buccal striation directionality—from a multitude of sites (Atapuerca, Krapina, La Quina, Regourdou 1,

Shanidar, Sima de los Huesos, etc.) has confirmed that Neanderthal populations across geographic localities were predominantly right-handed (Bax and Ungar 1999; Bermúdez de Castro, Bromage, and Jalvo 1988; Fox and Frayer 1997; Lozano et al. 2009; Volpato et al. 2012).

In addition to skeletal and cranial remains, symbolic evidence from Upper

Paleolithic Homo has strengthened claims of hominid handedness, with analyses on cave paintings and carved-bone items confirming a right-hand bias (Faurie and Raymond

2004; Marshack 1976). These forms of evidence have led to a general consensus that by the time of the Neanderthals, the hominid lineage shows clear evidence of the right-shift in handedness, and that later populations of genus Homo may have had comparable-to- modern levels of right-handers, at around 90% (Uomini 2011).

Other evidence useful to paleoanthropologists interested in hominid handedness, brain lateralization and language are endocasts (internal casts of the brain case, which potentially mirror neural tissues). Over 200 hominid brain endocasts exist, and Ralph L.

Holloway and others have dedicated much of their lives to the study of hominid cognitive changes via endocast analysis. Holloway’s extensive studies on hominid endocasts have helped to establish a timeline of events in the evolution of our ancestors’ brains, and confirm human-like patterns of cranial asymmetry in many Pleistocene individuals.

Asymmetries associated with handedness, tool use, and language in Homo sapiens appear as early as 2.0 mya, evidenced by endocast characteristics including: nonallometric size increase, reorganization of the frontal lobe (expansion of Broca’s area), and left-

20 occipital/right-frontal petalias (counterclockwise torque) (Holloway 2008). According to this evidence, hominid neural circuitry was changing much earlier than population-level right handedness shows up definitively in the fossil record.

The plethora of data confirming right-handedness in later Homo is useful, but

Holloway’s work, combined with knowledge on tool-use and cerebral asymmetries, suggests we can find earlier evidence of a shift towards rightward manual motor lateralization—and perhaps other asymmetries associated with language—via alternate analyses. In order to push our record of hominid handedness back in time, we must turn to evidence other than hominid skeletal remains, which are currently too limited to provide conclusive evidence. Some studies have sought to assess hominid handedness via cutmarks on faunal bones, and include microscopic analysis of butchered animal remains from various hominid sites and experimental re-creation of hominid prey-butchery

(Bromage and Boyde 1984; Pickering and Hensley-Marschand 2008). Despite the promising nature of such an approach, it has not successfully produced any new information on the handedness of Plio-Pleistocene hominids, perhaps due to lack of control in experimental contexts (Pickering and Hensley-Marschand 2008).

The final approach, from which this thesis is derived, is the experimental determination of handedness from lithic evidence. The earliest speculation about determining handedness from lithic materials was by S. A. Semenov, in his experimental studies on various Paleolithic tools. Semenov (1964) was particularly interested in the biomechanics of stone tool production, and thus had an inherent interest in handedness.

There are several inferences to the right-handedness of Paleolithic flintknappers throughout his work (Semenov 1964). Other than Semenov’s work, lithic-based

21 approaches to hominid handedness were non-existent until N. Toth, an American paleoanthropologist and flintknapper, published “Archaeological Evidence for

Preferential Right-handedness in the Lower and Middle Pleistocene, and Its Possible

Implications” in 1985. In the work, Toth (1985) argued that the orientation of cortex on successively removed lithic flakes can be used to infer handedness from hominid tool assemblages; Toth applied his method, which was informed by experimental re-creation of simple core-scrapers, to assemblages from Koobi Fora, Kenya, dated to 1.9 and 1.5 mya, and found similar right-to-left flake ratios. He concluded that this was initial evidence of high rates of right-handed knappers in early hominid populations, even including Homo habilis and H. erectus (Toth 1985).

This publication inspired much excitement about the possibility of early hominids being behaviorally closer to us than previously thought; however, debates regarding the integrity of Toth’s (1985) method ensued. Two critiques of this study note that Toth’s main assumption (of clockwise rotation of the core in the left hand of right-handed knappers and counterclockwise core rotation for opposite knappers) is not necessarily true of all knappers (Patterson and Sollberger 1986; Pobiner 1999). Several attempts have been made to reassess Toth’s technique, and Pobiner’s (1999) study showed variable success in designating handedness to a single knapper for multiple events, based on manufacture changes between sessions. Pobiner exclusively used novice knappers, though, so variability in flakes between knapping events may reflect the subjects’ lack of experience (Geribàs et al. 2010; Pobiner 1999).

Outside of Toth’s study (and the reviews it triggered), no attempts were made to identify handedness in lithic materials for decades. However, in a (1986) monograph on

22 La Cotte de Saint Brelade, a Paleolithic site in Jersey, J. M. Cornford published evidence of handedness in lithic resharpening techniques (Cornford 1986). There is extensive literature on differential retouch of lithic materials, particularly of hafted tools, and evidence of “prehensility” in Upper Paleolithic tools reinforces investigations of handedness and lithic remains. However, much like skeletal studies, the assemblages that could be studied regarding prehensility, retouch, and hafting are often later in time, when established evidence of right-handedness is not strongly debated.

In 2001, G. Rugg and M. Mullane published a new method of study for inferring hominid handedness via stone tools, based on the skew of the cone of percussion on lithic flakes. Rugg and Mullane (2001) experimentally show that an overall rightward skew in the cone of percussion for flakes of a single knapping event successfully predict a right- handed knapper, and the reverse for a left-handed one. They had four right-handed knappers and four left-handed knappers of mixed gender and expertise (including complete novices) remove flakes from nodules via direct percussion. No formal tools were made by knappers, as the study focused on a debitage characteristic (the cone of percussion). A total of 299 flakes were collected, but only 44 showed clear right- or left- skew in the cone of percussion. Rugg and Mullane correctly identified 75% of these flakes, but they never assessed the handedness of an actual lithic assemblage, meaning no concrete information on hominid handedness was gained from their approach.

Additionally, the low number of flakes that was assignable is discouraging. Despite the methodological and theoretical issues, Rugg and Mullane conclude that the two methods

(theirs and Toth’s) should be used in conjunction with each other, and that:

23 [t]he [statistical] chances of the Toth method and the cone of percussion method

both independently happening to reach the same conclusions about an assemblage

would be significantly smaller [than using only one method]” (Rugg and Mullane

2001:258).

In their conclusions, Rugg and Mullane note that they will not be making further attempts to assess handedness via lithic remains, and other than unpublished work from N. Uomini

(2001 and 2006), no additional studies followed.

Uomini’s unpublished studies (M.Sc., 2001, and Ph.D., 2006) on handedness and

Paleolithic materials reflect, without a doubt, the most significant reviews of both theory and practice in assessing hominid handedness. In her 2001 thesis, Uomini attempted to reproduce Toth’s (1985) and Rugg and Mullane’s (2001) studies, and applied their techniques to Paleolithic assemblages from British Lower Paleolithic sites, Swanscombe and Purfleet. Uomini notes that “[b]oth the [Toth] and the [Rugg and Mullane] method were quite easy to learn and apply to large numbers of archaeological specimens.

However, the reliability of the methods proved disastrous” (Uomini 2001:69). Uomini applied each method on 664 flakes from both sites, and none of her resultant ratios paralleled those from other publications (Uomini 2001:69). In fact, she concluded that the flakes from both sites seem to be distributed to left- and right-handedness randomly (as is seen in primates), implying that no right shift was present in hominids at the sites (dated between 400 and 200 kya), or that the methods developed by Toth and Rugg and Mullane were ineffective at determining handedness (Uomini 2001).

Despite this, Uomini continued research on lithic indicators of handedness, and her dissertation (2006) was on inferring handedness from two more British Lower

24 Paleolithic sites—Boxgrove and High Lodge. In addition to the Toth (1985) and Rugg and Mullane (2001) approaches, Uomini used a third, new approach, derived from

Cornford’s (1986) work on tranchet flakes, which are unique, final flakes removed from stone tools in order to create one straight cutting edge (Uomini 2006). Uomini also used comparative data from 12 living knappers in her study, who were part of a week-long flint workshop at the Lejre Historical-Archaeological Research Centre in Denmark. All subjects had experience knapping, although it was variable by subject. In the experimental part of her study, Uomini did an extensive qualitative analysis of the flintknapping process for each individual via interviews and video recordings. She simply asked knappers to produce “. . . what they felt capable of. . .” and “[o]nly the most experienced knappers attempted to make tranchet flakes on bifaces” (Uomini 2006:78).

First, Uomini (2006) assessed the accuracy of her three methods (Toth’s, Rugg and Mullane’s, and Uomini’s new tranchet method) by assessing materials of known handedness from the experimental knapping sessions, which included single-platform cores, Oldowan flakes, and tranchet flakes. Uomini had variable success using Toth’s cortex model (e.g. one subject’s handedness was almost perfectly predicted, while another’s was almost completely misjudged). She was unable to statistically evaluate

Rugg and Mullane’s cone of percussion (CoP) method, but found an even distribution of right- and left-skewed cones in most cases. Additionally, the tranchet flake method produced an even distribution of right-strike and left-strike characteristics (Uomini 2006).

In an attempt to explain the weaknesses of each method, Uomini assessed the video data from each knapper, and found that flintknapping was highly stylistic and individualized.

Knappers each had their own techniques for rotating and percussing cores, which were

25 often incompatible with the assumptions of the handedness models. Unfortunately, these experimental results imply that all three approaches are unreliable in predicting knapper handedness.

Because the assumptions and validity of each model were refuted by the experimental approach, Uomini (2006) did not attempt the Toth (1985) or Rugg and

Mullane (2001) methods on the Lower Paleolithic assemblages. She looked instead at lateralized resharpening and holding constraints of tranchet scars of the bifaces from the sites, inspired by Cornford’s work at La Cotte (Uomini 2006; Cornford 1986). The High

Lodge site had 20 handaxes, and the Boxgrove site was particularly important because it preserved a right-handed knapping scatter in situ, 8 additional scatters, remains of Homo heidelbergensis teeth (with right-skewed buccal striations), 406 handaxes (with 306 showing tranchet negatives) and over 100 usable flakes (Uomini 2006). Her analysis on these materials relied upon angle measurements and presence of hackles, and often used refit pieces in comparison to their negative scars. Using this new method, she proved that tranchet flakes can significantly indicate laterality, but they are relatively rare in the paleoarchaeological record. Despite the unsuccessful attempts in her works, Uomini still advocates experimental approaches to studying hominid handedness, and suggests future research involving Scanning Electron Microscopy (SEM) and other new approaches in favor of replicating previous studies (Uomini 2006).

The final publication relevant to this thesis is “Can hand laterality be identified through lithic technology?” by A. Bargalló and M. Mosquera (2013). Much like

Uomini’s works, Bargalló and Mosquera assess Toth’s and Rugg and Mullane’s original methodologies, but they also introduce five new potential indicators of handedness on

26 lithic remains. Updates to Toth’s (1985) method include designation of cortex into quadrants instead of simple left- vs. right-orientation. Updates to Rugg and Mullane’s

(2001) method include the addition of the “extraction axis,” which is seen as an extension of the cone of percussion into the distal end of the flake. Perhaps inspired by Uomini’s work (2001 and 2006), Bargalló and Mosquera also added hackles and eraillure scars to their methodology, but these were not used within the context of tranchet flakes exclusively, because normal flakes demonstrate them as well. Other additional technical features identified by Bargalló and Mosquera include: structure and inclination of the striking platform, location of impact point on platform, as well as locations of fractures on broken flakes, including cortical flakes and non-cortical flakes (Bargalló and M.

Mosquera 2013:8).

In their publication, Bargalló and Mosquera (2013) did not apply their techniques to actual assemblages, and instead focused solely on experimental verification of knapper handedness. However, due to lack of expert knappers, they had to use novices, and thus they only required subjects to produce flakes (Bargalló and Mosquera 2013). Their first goal was to determine that the technical features they listed were regularly produced by all flintknappers, and thus were not subject to individual knapping styles, or even handedness. For this assessment, Mann-Whitney U tests were used to compare production rates of each feature by left- and right-handed groups. Only a few features showed statistically significant differences between handedness groups, as every knapper produced each technical feature in varying frequencies. This suggests that Bargalló and

Mosquera’s features are inherent to lithic manufacture and viable for handedness

27 analyses, but they used simple frequency data for this analysis, making Mann-Whitney U testing technically inappropriate.

According to Bargalló and Mosquera (2013), Toth’s (1985) and Rugg and

Mullane’s (2001) methods were unsuccessful at determining knapper handedness, both independently and in conjunction with each other, although they used the Mann-Whitney

U values as evidence. In order to better associate their own suite of technical features to knapper handedness, Bargalló and Mosquera conducted a correspondence analysis on flake characteristics by knapper, and hoped to find clear separation of data by handedness. Their analysis revealed that no single trait on lithic flakes is accurate in predicting handedness, but that a combination of traits as a whole can predict knapper handedness relatively well. They note that, in general, left-sewed characteristics indicate a left-handed knapper, whereas right-skewed characteristics indicate a right-handed one.

Several flakes analyzed showed a mixture of these features, however, and Bargalló and

Mosquera state that “single flakes cannot be ascribed with certainty to a right- or left- handed knapper” as well as entire assemblages can. Bargalló and Mosquera conclude by urging further studies, particularly ones with expert knappers, and this publication will likely generate replicative studies and debates as Toth’s 1985 publication did (Bargalló and Mosquera 2013).

These approaches to assessing handedness in lithic materials all have merits and weaknesses, but all authors maintain the claim that knapper handedness is difficult, but possible, to detect in lithic debitage. Toth’s original method (1985) is likely too simplified for later manufacture techniques, where single-axis rotation is insufficient for completing the tool, but it still may be viable for the earliest lithic technologies.

28 Likewise, Rugg and Mullane’s (2001) assumptions do not account for an expert knapper’s ability to manipulate the direction of percussion in non-traditional ways, but they may hold true for the earliest stone tool makers. Both Uomini (2006) and Bargalló and Mosquera (2013) suggest that well-associated assemblages are better for analysis than single flakes, with the ideal being preserved knapping scatters with multiple refits.

However, considering the paucity of the paleoarchaeological record, especially in the

Lower Paleolithic, a method that works on lone flakes would be far superior. All of these issues must be addressed by additional research before new data can be gained on the evolution of hominid handedness.

29 METHODS

The purpose of this thesis is to test three of the previously established methods for determining handedness of lithic materials. These methods include:

1) Toth’s (1985) cortex model

2) Rugg and Mullane’s (2001) cone of percussion model

3) Bargalló and Mosquera’s (2013) technical features model

Under the assumptions of these previous works, right- and left-handers create predictable, differentiable characteristics on lithic materials they produce, but more work is needed to confirm or falsify the applicability and replicability of these methodologies. The ultimate goal of this thesis is to delineate clearer, more effective approaches for the experimental determination of handedness in lithic materials, so that data gained from these approaches can eventually be used on fossil remains if they are appropriate.

Creation and Collection of Materials

Because this is an experimental approach, all lithic materials were produced by modern-day expert flintknappers. Two right-handed and two left-handed expert knappers

(classified as those who are comfortable with, and used to, producing stone tools via flintknapping) were asked to produce three Acheulean handaxes each, in different flintknapping sessions. In order to avoid ascribing meaning to individualized lithic characteristics, more than one knapper of each hand preference was needed, and all knappers selected had at least 10 years of experience working with lithic materials.

30 Acheulean handaxes were chosen because they are restricted to the Lower Paleolithic period, but they reflect a major cultural transition in our hominid ancestors, and have more complex manufacture methods than those involved in Oldowan tool manufacture

(Hopkinson and White 2005; Schick and Clark 2003; Stout, Toth, and Schick 2006).

Additionally, they do not require pressure flaking, which is difficult to study in terms of handedness determination, and outside the scope of previous studies. There is also a general consensus that the change from Oldowan to Acheulean industries coincides with a significant cognitive shift in hominid species, which has been supported by neural studies on stone-tool production (Geribàs, Mosquera, and Vergès 2010; Moore 2012;

Stout et al. 2012; Uomini 2013).

Materials were provided to each subject in order to maintain control over external sources of error or noise (i.e. differing conchoidal fracture patterns, hardness of rocks used, amount of initial reduction required, etc.). Although consistency is difficult when dealing with lithic raw materials, the same source materials were used in an attempt to establish as much experimental control as possible. Each knapper received three Edwards

Plateau chert nodules and one hammerstone, all from a single source. Each flintknapper created their handaxes while sitting, using direct percussion and nothing else. They were directed to create handaxes between 9 and 12 cm in maximum dimension, instead of producing a set number of flakes or completing a set number of blows, in order to maintain a realistic mixed meta-assemblage. All materials produced in the creation of the tools (including the finished product, flakes, hammerstones, and shatter) were collected by the knappers and returned to Florida Atlantic University for analysis. Production and collection of materials was ultimately determined by each knapper, but they were urged

31 to lay down a tarp to prevent loss of debitage in each session. The products of each knapping session were kept separated by handaxe (one, two, and three) and by knapper

(one through four) in order to facilitate cataloging.

The original intent was to get 12 handaxes made by four knappers, but one right- handed knapper was unable to complete one handaxe due to a low-quality nodule, and one left-handed knapper was unable to complete any handaxes for unknown reasons.

Because of these issues, two additional left-handed knappers were recruited at a knap-in in northern Florida, and each made one Acheulean handaxe, also using Edwards Plateau chert that I provided. One knapper made the handaxe on-site using only a copper billet, and the other later sent the handaxe and debitage to me, like the originally recruited knappers. The same cataloging and analytical methods were used on these materials.

Preliminary Lithic Cataloging and Analysis

Ultimately, 10 handaxe “assemblages” were made for this thesis, five by right- handers and five by left-handers (see Figure 1). Upon arrival at Florida Atlantic

University, all flake debitage for each handaxe was sorted and labeled by two undergraduate anthropology students. The students cataloged one handaxe at a time, using the following procedure:

1) Randomly assign the assemblage a letter (A through J),

2) Label the associated handaxe with the assemblage letter,

3) Separate flakes over two (2) cm in length/width using a 2 x 2 cm geological sieve, and

4) Put smaller flakes/shatter in a bag labeled by assemblage (A-J).

32 Afterwards, each flake over two (2) cm in maximum dimension was then labelled with a random number between 1 and 1,000 (because no more than 1,000 flakes were expected).

All labeling was done with a Faber-Castell Pitt (India ink) pen and clear gloss, as is standard in the field (see Figure 2). A Microsoft Excel spreadsheet was used to randomize the numbers (1-1,000) for each flake. In the Excel spreadsheet, the students also recorded each flake’s associated handaxe (A through J), the flintknapper it came from, and finally, the “true handedness” of the flake (i.e., the handedness of the knapper).

After each assemblage was coded, all flakes were mixed together in a single “meta- assemblage.” This process was necessary because it created a blind analysis by removing the possibility that I would know the true handedness of the flakes and subconsciously take that information into account during my analysis. I analyzed the coded meta- assemblage using Toth’s (1985), Rugg and Mullane’s (2001), and Bargalló and

Mosquera’s (2013) techniques. Data recording and analysis were facilitated using a

Microsoft Access database, and then exported to Microsoft Excel and eventually SPSS for analysis.

Figure 1: Handaxes created by knappers

Figure 2: Assortment of labelled flakes, ventral aspect

34 Identification of Handedness-Indicative Features—Toth and Rugg and Mullane

The first handedness analysis conducted was based on the cortex model introduced by Toth in 1985, so only cortical flakes were analyzed. This analysis was done in the following manner: flakes were oriented with the platforms superior to the terminations and dorsal (cortical) surface facing the viewer. Using judgment by eye, cortex was assigned right-oriented or left-oriented based on predominance of cortical surface area. Figures 3 and 4 show the ventral and dorsal faces of flakes showing a right- cortical orientation and a left-cortical orientation, respectively.

Figure 3: Ventral (left) and dorsal (right) faces of a right-oriented cortical flake

Figure 4: Ventral (left) and dorsal (right) faces of a left-oriented cortical flake 35 The next analysis replicated Rugg and Mullane’s cone of percussion method from

2001. Only flakes with identifiable platforms and observable cones of percussion were used for this method. As stated in the original publication, the use of a protractor in determining the angle (or skew) of the cone of percussion was “not usually possible, as the striking platform was not usually exactly straight” (Rugg and Mullane 2001:253). My original intent was to still attempt the protractor method, but it was, indeed, not feasible for this assemblage. Thus, cones and ridges were both simply categorized by inspection as left-skewed, right-skewed, or centered (see Figures 5, 6, and 7).

Figure 5: Ventral face of flake with a left-skewed cone of percussion

Figure 6: Ventral face of flake with a right-skewed cone of percussion

Figure 7: Ventral face of flake with a centered cone of percussion

37 Identification of Handedness-Indicative Features—Bargalló and Mosquera

The last method is derived from Bargalló and Mosquera’s recent (2013) publication, which is also a review of Toth’s (1985) and Rugg and Mullane’s (2001) methodologies. This analysis is the central component of this thesis, and was completed on all cataloged flakes. Data were recorded using a Microsoft Access database with the following 15 fields, shown in Table 1.

Table 1: Summary of classification sets and their associated types Characteristic Observation Options Cone of Percussion Left, Center, Right, None Hackles Left, Distal (Center), Right, None Ripples Left, Distal (Center), Right, None Extraction Axis Left, Distal (Center), Right, None Ridge Angle Negative (Left), Center, Positive (Right), None Platform Type Linear, Punctiform Platform Inclination Left, Center, Right, Sinuous, None Impact Point Left, Center, Right, None Cortex (in Quadrants) Any combination of A, B, C, and D Fracture Locations Any combination of a-g Eraillure Scar Locations Left, Center, Right, None Overall Handedness Inference Left, Right, Indeterminate

These features were noted on all flakes collected, and flakes with no identifiable trace of a particular feature (e.g. distal flakes with no preserved platform) were labeled 999— which is coded by SPSS as “missing” data, and thus not included in statistical analyses

(Field 2005). Items coded “None” for various traits were included in analyses, so they represent characteristics that were present, but for some reason un-classifiable, often due to ambiguity (not clearly left, right, or center, but still present).

Each feature had its own specific methodology. See Figure 8, as well as Bargalló and Mosquera (2001:8) for an additional description of these methods. For most flakes, the following procedures were used:

38 1) Identify dorsal/ventral and proximal/distal orientation of flake; orient the flake with the platform upwards.

2) The following characteristics, when present, all required a judgment of their orientation as either Left, Distal (or Center), or Right: cone of percussion, hackles, ripples, extraction axis, ridge angle, and eraillure scar. Those observations were made with the ventral face placed upward toward the viewer.

3) For the following characteristics, the extraction axis of the flake was vertically oriented with the ventral face toward the viewer, and the platform was observed for: type, inclination, and impact point location.

4) Fracture locations were designated by looking at the ventral face of each flake.

5) Finally, the dorsal face was assessed for cortical flakes only.

For every flake, five independent handedness deductions were made, one for each of the following groups of characteristics: Cone of percussion, which included joint consideration of the cone of percussion, hackles, ripples, extraction axis, and ridge; platform, which included joint consideration of the platform type, inclination, and impact point; cortex; fractures; and eraillure scars.

Based on these five subsets (which is equivalent to considering all the observations made on a single flake), an overall handedness inference was made for every flake: “right,” “left,” or an “indeterminate” option for ambiguous flakes, or cases where some observations indicated right-handedness while others indicated left- handedness on the same flake.

Figure 8: Graphic description of flake characteristics and their associated handedness inferences, modified from Bargalló and Mosquera (2013). I. General flake characteristics on ventral face (left) and dorsal, cortical face (right). II. Cortex classifications on the dorsal surface for left-handers (cortex AC, left), and right-handers (cortex BD, right). III. Cone of Percussion (CoP) characteristics on the ventral face for unidentifiable handedness (left), left-handers (center), and right-handers (right). The top row shows the actual cone of percussion and the ridge at the proximal end of the flake; the middle row shows distal ripples and an arrow generalizing the extraction axis; the bottom row shows proximal eraillure scars—which were recorded as binary data—and distal hackles (note that these have opposite orientation to the other CoP characteristics). IV. Platform characteristics, including impact point and platform orientation for left- and right-handers (top right), as well as these characteristics, and platform type, which did not indicate handedness. V. Fracture locations (A-G) on the ventral face, reorganized by their handedness associations, with unassociated location C (left), left-handed locations (center), and right-handed locations (right).

40 Bargalló and Mosquera (2013) found that some of the previously mentioned characteristics did not show clear separation between right-handed and left-handed knappers, such as fracture location, impact point on the platform, and cortex. In order to correlate these ambiguous features with handedness, I conducted an informal pilot study using flakes produced by two novice female knappers—one right-handed (myself) and one left-handed (a fellow graduate student). In this pilot study, the handedness of the flakes was known to me, so I identified handedness trends for as many features as I could with my sample. Based on these data, right-handed designations were usually made for flakes with a right-extraction axis (including Cone of Percussion [CoP], impact point, and fractures D and G—See Figure 8) and left-oriented eraillure scars, while left-handed designations were usually made for left-extraction axes (with fractures at E and F). As

Bargalló and Mosquera found, however, many of the flakes in this pilot study, as well as many of the flakes from the main study, also often exhibited a mosaic of these traits.

Reconstruction of Assemblages and Statistical Methods

After all flakes were analyzed, the original coding catalog, which contained the handaxe/assemblage label, associated flintknapper, and “true handedness” for each flake was given to me. I used these data to regroup my individual flakes by handaxe and finally by knapper, reconstructing the initial assemblages sent to me by the knappers themselves.

Table 2 describes each labeled handaxe by knapper, handedness, and number of flakes over 2 x 2cm in dimension, which were included in the following analysis.

41 Table 2: Meta-assemblage breakdown by handaxe and knapper Handaxe Knapper Knapper Number of % of Meta- A SH HandednessRight Complete Flakes 62> assemblage9.8 B MC Right 2 cm 16 2.5 C SH Right 89 14.1 D EM Left 100 15.8 E MC Right 64 10.1 F MC Right 11 1.7 G OS Left 62 9.8 H EM Left 57 9.0 I EM Left 92 14.6 J RC Left 78 12.4

For this thesis, I used the Statistical Package for the Social Sciences (SPSS) software for all analyses. I first replicated the methods of Toth (1985) and Rugg and

Mullane (2001). As mentioned before, Bargalló and Mosquera (2013) conducted Mann-

Whitney U tests and a correspondence analysis in order to determine which features explain right- vs. left-handed variability the best. Instead of this method, I used binomial logistic regression on my data. I chose binary logistic regression because of the nature of the data collected and the structure of the hypothesis I wanted to address. Most other statistical tests would have been invalid for those purposes, which is a largely unaddressed issue in the previous studies.

Logistic regression was the most appropriate technique because it is designed to predict a binary dependent variable. Logistic regression can take many data types for independent variables, including binary and nominal-scale data, which compose the entirety of the data collected (Harrell 2001:217). Like Bargalló and Mosquera’s (2013) correspondence analysis, binary logistic regression can also indicate which attributes contribute most to identifying knapper handedness. Correspondence analysis, though, is mainly an exploratory technique, whereas logistic regression is also predictive. Logistic

42 regression is also non-parametric, which was important because of the characteristics of the data sets.

An important aspect of binary logistic regression is that it results in an equation, which represents the predictive model created from data analysis. If an effective regression model is found for determining handedness in my experimental lithic remains, the resultant equation can be used to determine handedness probability values for flakes of unknown handedness—including archaeological ones—in the future. No previous research has provided a similar product, and an effective regression equation could eliminate the difficulties associated with analyzing single flakes. A working equation would also provide a more quantitative approach for assessing flakes, where a calculated p-value reflects the probability of handedness, instead of a simple judgment by an observer.

Many weaknesses present in previous studies can be ascribed to observational biases (see Rugg and Mullane 2001: 254-255, for a discussion on observer objectivity). so I also wanted to examine the reliability and validity of the methods proposed for inferring handedness. Therefore, after I completed all analyses on each flake in the meta- assemblage, I took a stratified random sample of flakes that was also assessed by one other anthropology graduate student and one anthropology faculty member—both of whom are skilled in lithic analysis—using the same methods. These two observers only used Bargalló and Mosquera’s (2013) approach, because Toth’s (1985) and Rugg and

Mullane’s (2001) methods are implicit in their work. The graduate student and the professor each re-analyzed a sample of 50 flakes, 25 of which overlapped, so all three observers (including myself) had analyzed them. After the independent analyses were

43 completed, cross-comparisons were made for each observer using Fleiss’ Kappa and weighted Cohen’s Kappa, which are standard tests for inter-rater agreement of qualitative data (Ben-David 2008; Fleiss et al. 1969; Nichols et al. 2011; Warrens 2011). The motivation for this aspect of the methodology is that many of the characteristics described by Bargalló and Mosquera (2013) appear ambiguous or subjective in nature, and no previous attempts to address inter-observer error in determining handedness from lithic debitage have yet been made.

44 RESULTS

In total, 631 flakes were analyzed as part of this thesis. It is important to note that only 242 (38.4%) flakes came from right-handed knappers, while 389 (61.6%) came from left-handers simply because the latter produced on average more flakes per handaxe.

Ideally, the experimental assemblage would have been closer to 50% right-handed and

50% left-handed, but knappers were not instructed to produce a specific number of flakes, so the experimental distribution reflects natural variation that is undoubtedly common in actual assemblages. For my assessment of Bargalló and Mosquera’s (2013) methods, the binary logistic regression cut-off value was not the standard 0.5, which would have been appropriate for a 50-50 distribution, but 0.6, to reflect the relative predominance of left-handed flakes in the meta-assemblage (Neter et al. 1996; Sarkwar and Midi 2010; van der Heijden 2012). Methods for Toth’s (1985) evaluation, Rugg and

Mullane’s (2001) evaluation, and inter-observer comparisons were not changed.

Assessment of Toth’s and Rugg and Mullane’s Methods

The following handedness inferences for each handaxe were made based on whether a majority of the flakes for that handaxe were designated right- or left-skewed.

For Toth’s (1985) method, this relied upon the distribution of cortex on cortical flakes, where predominance of cortex was classified as left or right. Table 3 shows the frequencies of left- and right-classified cortical flakes by handaxe; 159 flakes were not included in this analysis because they were fragments or shatter that could not be oriented

45 properly, and 251 flakes were either fully cortical or non-cortical, and therefore did not aid in determining handedness. Thus, only about 35% of the flakes collected were viable for analysis using this method.

Table 3: Summary of the Toth (1985) cortex-based method Knapper Flake Handedness Handaxe Percent Correct? Handedness Frequency Inference Right Left Right Left A Right 5 12 29.41 70.59 Left No B Right 5 3 62.50 37.50 Right Yes C Right 13 10 56.52 43.48 Right Yes D Left 6 12 33.33 66.67 Left Yes E Right 4 9 30.77 69.23 Left No F Right 4 1 80.00 20.00 Right Yes G Left 7 5 58.33 41.67 Right No H Left 6 4 60.00 40.00 Right No I Left 8 10 44.44 55.56 Left Yes J Left 7 11 38.89 61.11 Left Yes

Using this method, six handaxe assemblages were correctly predicted, while four were not. Note that the binomial probability of randomly obtaining exactly 4 wrong and 6 right answers in 10 attempts is p = 0.205. Indeed, the cumulative binomial probability of getting 6 or more correct answers at random is ≈.377. So, obtaining 6 correct inferences could easily be pure luck. Furthermore, one might predict that if the method were working correctly, then assemblages with large numbers of flakes, that is, larger samples, would have been more likely to have been correctly identified. The opposite was actually the case: the two smallest assemblages, B and F, were correctly identified. The mean number of flakes in the correctly ( ) and incorrectly ( ) identified assemblages were similar. A Mann-Whitney U test does not find a significant difference between the numbers of flakes (p = .45). It is possible that the method works, but only weakly, so a much larger number of experiments are needed to discern the effect statistically.

46 Of the assemblages that were incorrectly predicted, two had relatively equal percentages of right- and left-cortical flakes, but all other assemblages had a clear predominance of either right- or left-cortex. In Toth’s original publication (1985), he interpreted two assemblages from Lower Paleolithic sites as having right-hand dominant knappers based on his own created assemblage of flakes, which roughly showed a 55:45 right-to-left distribution. However, in my recreation of his methods, assemblages with even stronger directional skew (e.g., handaxes E and G with an approximate 60:40 right- to-left distribution, or handaxe C with a 70:30 left-to-right skew) showed handedness opposite to the ratio, suggesting that the ratios themselves are unreliable predictors of handedness. It was hoped that Acheulean handaxe manufacture would be simple enough that Toth’s principle of core rotation would still apply, but it is likely that any bifacial manufacture technique requires violations of the uni-directional rotation he suggests, because the core must be flipped along all axes. Coupled with the other critiques of

Toth’s method, my reassessment the cortex method suggests that is inaccurate at predicting knapper handedness, and considering that it could only be applied to a minority of the flakes created, it is not sufficient as a stand-alone approach for assessing handedness in lithic debitage.

For Rugg and Mullane’s (2001) method, assessment was based only on right- or left-skew of the cone of percussion, as I, like they, found the protractor method to be unemployable. Only 49 of the 631 (7.7%) flakes analyzed had visible ridges, and many of these, as Rugg and Mullane stated, did not have platforms that facilitated the protractor method. Additionally, 255 flakes were fragments or shatter where the skew of the cone could not be identified, and 308 had centered cones of percussion, so only about 11% of

47 the flakes analyzed indicated knapper handedness via the cone of percussion. Table 4 shows the frequencies of left- and right-classified flakes by handaxe using Rugg and

Mullane’s method.

Table 4: Summary of the Rugg and Mullane (2001) cone of percussion method Knapper Flake Handedness Handaxe Percent Correct? Handedness Frequency Inference Right Left Right Left A Right 0 3 0.00 100.00 Left No B Right 2 2 50.00 50.00 Indeterminate No C Right 7 1 87.50 12.50 Right Yes D Left 5 8 38.46 61.54 Left Yes E Right 2 6 25.00 75.00 Left No F Right 0 1 0.00 100.00 Left No G Left 1 5 16.67 83.33 Left Yes H Left 3 1 75.00 25.00 Right No I Left 10 5 66.67 33.33 Right No J Left 2 4 33.33 66.67 Left Yes

Using this method, only four handaxe assemblages were predicted correctly, and the number of flakes that showed clear directional skew in the cone of percussion was far too small for this method to be beneficial on its own. The binomial probability of randomly obtaining exactly 6 wrong and 4 right answers in 10 attempts is p = 0.205, while the cumulative binomial probability of getting 4 or more correct answers at random is ≈.828. So, obtaining 4 correct inferences could be attributable to luck, and not even very good luck. However, as will be explained later, the modified cone of percussion method in Bargalló and Mosquera’s (2013) paper may be a more viable approach in terms of predicting knapper handedness.

Assessment of Bargalló and Mosquera’s Methods—non-statistical measures.

Before conducting any regressions on my flake classifications, I assessed how well my handedness predictions, informed by Bargalló and Mosquera’s original

48 publication (2013) and my pilot study on novice-flakes, fit the actual handedness of each flake, by handaxe. Unfortunately, a majority of flakes either had too many missing features, or a mixture of right- and left-skewed traits, resulting in a strong predominance of “Indeterminate” handedness classifications, labeled as IND (see Table 5: center columns). As shown in Appendix A, which lists handedness predictions by subset

(including cone of percussion, eraillure scars, platforms, cortex, and fractures), indeterminate classifications typically represent most of the sample, reaching over 90% of flakes in some cases. Additionally, percentages of flakes that were classified correctly by me as right- or left-handed do not differ from percentages predicted incorrectly. No major differences in predictive success exist between handaxe, handedness, or knapper, suggesting that simple observation via the Bargalló and Mosquera method is consistently no better than chance at predicting knapper handedness. Using the five subsets in a combined “overall handedness inference” had little impact on the accuracy of my judgments, but in some cases, it did reduce the number of indeterminate flakes, as shown in Table 5.

Table 5: Evaluation of overall handedness inferences by handaxe Knapper Hand N of N % N N Handed- % IND % Wrong -axe flakes Correct Correct IND Wrong ness A Right 62 8 12.90 39 62.90 15 24.19 B Right 16 8 50.00 2 12.50 6 37.50 C Right 89 17 19.10 60 67.42 12 13.48 D Left 100 21 21.00 59 59.00 20 20.00 E Right 64 12 18.75 32 50.00 20 31.25 F Right 11 3 27.27 5 45.45 3 27.27 G Left 62 10 16.13 48 77.42 4 6.45 H Left 57 9 15.79 41 71.93 7 12.28 I Left 92 21 22.83 48 52.17 23 25.00 J Left 78 25 32.05 40 51.28 13 16.67

49 As stated by Bargalló and Mosquera (2013), many lithic flakes showed a mosaic of traits, making flake-by-flake interpretations very difficult. They suggest that their method would be most effective on well-preserved knapping scatters, where all flakes can be analyzed with reference to each other. As mentioned before, however, preserved paleo-archaeological scatters are extremely rare in comparison to mixed assemblages (as seen, for example in taphonomic and refitting studies), so in my blind study, I had no intention of analyzing flakes in clusters, as I wanted to test the method on single flakes.

Figures 9 and 10 show ventral and dorsal views, respectively, of two flakes that came from the same handaxe (and same knapper). Flake 870 (left) exhibits many characteristics that suggest right-handedness, including left-skewed eraillure scar and hackles, right-skewed ripples, and a predominance of right-skewed cortex. I correctly classified this flake as right-handed. Flake 648, however, shows relatively ambiguous features, such as: centered ripples and non-skewed cortex (locations C and D, see Figure

8), as well as two features associated with left-handedness: right-skewed hackles and eraillure scar. I incorrectly inferred that this flake was left-handed.

Figure 9: Ventral views of two right-handed flakes showing opposite skew in characteristics

Figure 10: Dorsal view of two right-handed flakes showing variable cortex location. Flake 870, left; Flake 648, right.

51 As suggested by my overall percentages, and the above example, analyzing lone flakes by simple judgment using these techniques is an unreliable method for determining handedness, but work on sets of flakes may still be a viable avenue for future work. The method’s inaccuracy in informing my judgments could additionally be due to poor handedness associations for traits. For example, the association between platform inclination and handedness was relatively hard to make because it had shown no clear differentiation in either Bargalló and Mosquera’s (2013) or my own pilot study, so I rarely knew how to interpret it. Thus, additional statistical analysis may prove that it was my associations that were incorrect, even though the method actually works.

Binary Logistic Regression—which traits predict handedness?

Binary logistic regression measures how changes in an independent variable increase the log-odds of a binary outcome. Using the maximum likelihood method, each regression predicted how each variable change increased the likelihood of left- handedness, coded as the binary value of 1 (Hilbe 2009; Hosmer, Lemeshow, and

Sturdivant 2013; Menard 2002). Each regression model runs first without inputting data

(called Step 0) and thus predicts all valid cases as left-handed. Subsequent models include all viable cases for each selected independent variable, and report how each variable either increases or decreases the likelihood of left-handedness (indicated by the coefficient ß or beta statistic—henceforth simply B). This statistic represents the log-odds of each case resulting in a value of 1 (left-handedness) in the following equation:

= B0 + B1*x1 + B2*x2 + B3*x3 + B3*x3 . . . Bn*xn

52 Here, B0 represents the constant of the model calculated in Step 0, and x represents the value of each independent variable, which is a dummy-coded matrix assigned by SPSS in the case of nominal data, and either a zero (0) or one (1) in the case of binary data

(Cokluk 2010). The subscripts are an index indicating independent variables 1 to n and their associated beta coefficients. The symbol p in the expression to the left of the equal sign represents the probability that the dependent variable, y (handedness), equals 1 (left- handedness), while the part of the expression within the parentheses in which p appears is the odds derived from the probability. Hence, the left part of the equation is the log-odds.

Negative beta values indicate that the presence of a feature decreases the log-odds of a flake being from a left-handed knapper, and thus these features increase the overall probability of the flake being right-handed.

Each beta value is reported with its associated standard error, Wald statistic (a z- test based on a modified chi-squared distribution), degrees of freedom, and overall significance (based on of a modified two-tailed t-test). Finally, an exponentiated beta value is given, which represents the odds ratio for each independent variable (Field

2005). I originally intended to use the standard ɑ value of 0.05 to determine which variables were statistically significant, but so few fit this standard that a less-demanding ɑ value of 0.10 was adopted for all regressions. In addition to these, SPSS provides two measures intended to help you evaluate the regression model: two R-squared approximations are provided for each regression, as well as the Hosmer and Lemeshow test (Field 2005). In OLS and other types of regression, R-squared is a measure of how far each data point deviates from the best fit line, and these measures usually signify how much of the data’s variability is addressed by the model. Although the two R-squared

53 statistics function the same way for binary logistic regression, many researchers urge caution in their interpretation due to the nature of logistic regression (Harrell 2001; Hilbe

2009; Hosmer, Lemeshow, and Sturdivant 2013). The Hosmer and Lemeshow test is based off the chi-squared distribution, where the resultant p-values from the regression are grouped into deciles, and their observed and expected frequencies are compared

(Hosmer, Lemeshow, and Sturdivant 2013). Because a good regression should have different numbers of cases in each decile, in the case of the Homser and Lemeshow test, p-values above 0.05 suggest a properly specified model, because the observed and expected values are different. Many researchers also urge caution in interpreting the

Homser and Lemeshow p-value (Field 2005; Hilbe 2009; Tjur 2012).

A fully saturated regression model (i.e., one with all independent variables included) was run as a general evaluation of Bargalló and Mosquera’s (2013) classification system, but multiple, smaller “subset” regressions were also run for two main reasons. First, my own classifications of the meta-assemblage were so poor when based on only one category that I wanted to see if the regressions would also suffer in their predictive accuracy, and if the subset regressions would have different predictions than the saturated model. Second, Bargalló and Mosquera heavily stressed in their original article that many traits were not particularly good indicators of knapper handedness, and that “no single variable can be used to determine the laterality of the knapper, but rather the evidence of handedness lies in the combination of several variables” (Bargalló and Mosquera 2013: 1). This ultimately implies that collecting all data observations on each flake is necessary for the method to work, even though over

50% of the flakes they collected (663 out of 1,158, roughly 57%) showed only one

54 characteristic out of their original nine, and only one flake exhibited all of the characteristics. I ran regressions on the same five subsets of variables that I used to inform my original handedness classifications: cone of percussion, eraillure scars, platform, cortex, and fractures, in order to see if any of these subsets alone were as effective as the saturated model, indicating that they could be used instead of the entire methodology.

Finally, I ran a backwards regression from the fully saturated model, as well as the resultant unsaturated model, as an indicator of which dependent variables are most significant overall, and therefore contribute most to a correct prediction of flake handedness. This unsaturated model could also be used as a stand-alone data collection method for future analyses, especially where the full method is impractical or not applicable. As stated before, many of the flakes collected for this thesis did not exhibit several of the features that were supposed to be analyzed, and flake-fragments and shatter presented a large portion of the sample. (See Appendix C for an account of cases included in each model, as well as a percentage of missing cases [originally coded 999]).

If a satisfactory unsaturated, simpler model could be found (i.e., with high predictive power), it would be particularly useful for future work. Indeed, an unsaturated model focused on attributes that were usually present could, in theory, be more powerful than a saturated model because the latter loses many cases to missing data. A judiciously designed unsaturated model could retain more cases due to a lower rate of missing data and therefore produce more decisive results. All frequency data for the following regressions can be found in Appendix B.

55 The cone of percussion (CoP) subset included the following independent variables: cone of percussion, hackles, ripples, extraction axis, and ridge, which were all categorized as center, left, or right. In total, 411 flakes were included in this analysis

(65.1%), whereas 220 flakes (34.9%) were invalid for various reasons, such as missing data (e.g., no hackles), or an inability to properly orient the flake (see Appendix C). The

CoP regression model is shown in Table 6, and its various measures of significance in

Table 7.

Table 6: Binary logistic regression values for the cone of percussion subset. Results in bold-face type are statistically significant at α = 0.1. CoP Regression Model B S.E. Df Sig. Exp(B) CoP*Hack*Rip*ExtAx*Ridge 41 1.000 Cone of Percussion 3 .434 Cone of Percussion Center .211 .406 1 .604 1.235 Cone of Percussion Right .928 .694 1 .182 2.529 Cone of Percussion Left .685 .623 1 .271 1.984 Hackles 3 .925 Hackles Center .144 .376 1 .702 1.155 Hackles Right -.136 .421 1 .746 .873 Hackles Left -.143 .409 1 .727 .867 Ripples 3 .110 Ripples Center .709 .343 1 .038** 2.033 Ripples Right 0.177 0.6 1 0.768 1.193 Ripples Left 1.366 0.858 1 0.112 3.919 Extraction Axis 3 0.833 Extraction Axis Center -1.042 1.315 1 0.428 0.353 Extraction Axis Right -0.952 1.401 1 0.497 0.386 Extraction Axis Left -1.356 1.459 1 0.353 0.258 Ridge 3 0.078* No Ridge -1.231 0.546 1 0.024** 0.292 Ridge Center -0.539 0.502 1 0.282 0.583 Ridge Right -1.081 0.729 1 0.138 0.339 Constant 1.566 1.434 1 0.275 4.787 * P < 0.1 ** P < 0.05 *** P < 0.01

Table 7: Measures of significance for the cone of percussion regression Statistic p(df) Cox and Snell R Square .147 Nagelkerke R Square 0.198 Hosmer and Lemeshow Test 0.516(8)

56 As indicated by the model, only two variables significantly impacted the log-odds of a flake coming from a left-handed knapper: ripples and ridge. It is interesting to note that in both cases, it was centered ripples and ridges that contributed most to the model, instead of right-or left-skewed traits. Additionally, centered ripples increase the likelihood of a flake being left-handed, while a centered ridge seems to decrease these odds, which seems inconsistent with the assumptions that Bargalló and Mosquera operated under in their methodology (2013). According to the measures of significance for the model (two R-squared approximations and the Hosmer and Lemeshow chi- squared p-value), relatively little of the variability of the data is addressed by the CoP regression model. This is not unexpected, however, because it is likely that handedness contributes very little to the variability in flake morphology overall, with other contributors, such as material type, stage of removal, etc., having more significant impacts on flakes. As indicated by the Hosmer and Lemeshow test, this model is still significantly better at predicting handedness than Step 0, where no independent-variable regressors were used, so it is probably correctly specified (Hosmer, Lemeshow, and

Sturdivant 2013).

Table 8 compares the Step 0 predictive correctness vs. the predictive correctness of the regression model. You can see that the model starts with all the flakes classified as the most common type, left-handed. After taking into account the information in the independent variables, it re-classifies them slightly better, correctly identifying some (58) of the right-handed flakes, but it also misclassifies some (32) of the left-handed flakes.

So, the correctly classified flakes consist of the 58right + 208left = 266 and 266/411 =

64.7%.

57 Table 8: Summary of predictive correctness for the cone of percussion regression subset Classification Table—CoP Step Handedness Percentage Correct Right Left Right 0 171 0.0 Left 0 240 100.0 0 Overall % Correct 58.4 Right 58 113 33.9 Left 32 208 86.7 1 Overall % Correct 64.7

Overall, the cone of percussion regression model only improved the predictive correctness by 6.3 percent (64.7-58.4). Combined with the low R-squared values and the beta value associations for centered ripples and ridges, this low predictive capability suggests that the characteristics within the CoP subset do not differentiate well between right- and left-handed knappers on their own. This is supported by my analysis of Rugg and Mullane’s (2001) cone of percussion method, as well as Bargalló and Mosquera’s

(2013) insignificant Kruskall-Wallis values for most CoP traits (Rugg and Mullane 2001;

Bargalló and Mosquera 2013:11).

Eraillure scars were analyzed separately because they were originally entered as binary data instead of categorical data, and in some cases, the locations of eraillure scars were not mutually exclusive (consider that some flakes can theoretically have eraillure scars in multiple locations on the ventral face). For this regression, all 631 cases were included, as no missing data occurred (see Appendix C). Tables 9 and 10 show the regression values and significance for the ERA model, respectively.

58 Table 9: Binary logistic regression values for the eraillure scar subset. Results in bold-face type are statistically significant at α = 0.1. ERA Regression Model B S.E. df Sig. Exp(B) Center -.193 .277 1 .485 .824 Right -.038 .263 1 .886 .963 Left -.447 .312 1 .152 .640 Constant .528 .094 1 .000 1.696 * P < 0.1 ** P < 0.05 *** P < 0.01

Table 10: Measures of significance for the eraillure scar regression Statistic p(df) Cox and Snell R Square .006 Nagelkerke R Square 0.008 Hosmer and Lemeshow Test 0.945(2)

According to this model, eraillure scar location is an insignificant indicator of knapper handedness, and as suggested by the eraillure scar frequency table in Appendix

B, eraillure scars are specific to each knapper, regardless of handedness. The R-squared values for this model are very low, but the Hosmer and Lemeshow goodness-of-fit test indicates that this model is overall quite significant compared to Step 0, and thus also correctly specified. Once again, it is interesting to note that all three beta values increase the likelihood of a flake being right-handed, suggesting that the assumed directional skew for knappers of differing handedness is incorrect. Table 11 shows the predictive correctness for Steps 0 and 1 of the model.

Table 11: Summary of predictive correctness for the eraillure scar regression subset Classification Table—ERA Step Handedness Percentage Correct Right Left Right 0 242 0.0 Left 0 389 100.0 0 Overall % Correct 61.6 Right 4 238 1.7 Left 4 385 99.0 1 Overall % Correct 61.6

59 In this model, the regression correctly classified 4 right-handed flakes, but also incorrectly classified 4 left-handed ones, resulting in no improvement in the classification. Bargalló and Mosquera (2013) included eraillure scars in their cone of percussion subset, but also found it to be insignificant in terms of knapper handedness

(Bargalló and Mosquera 2013:14).

The platform regression included platform type, platform inclination, and impact point, which had various subsets (see Appendix C). 450 flakes (71.3%) were included in this analysis, and 181 (28.7%) were not, typically because the platform was missing.

Tables 12 and 13 show the regression and its significance.

Table 12: Binary logistic regression values for the platform subset. Results in bold-face type are statistically significant at α = 0.1. Plat Regression Model B S.E. Wald df Sig. Exp(B) ImpPt * PlatI * PlatT 4.982 6 .546 Platform Type .090 .223 .163 1 .686 1.094 Platform Inclination 14.913 3 .002*** Platform Inclination Center -1.681 .559 9.034 1 .003*** .186 Platform Inclination Right -1.212 0.584 4.308 1 .038** .298 Platform Inclination Left -2.036 0.596 11.650 1 .001*** .131 Impact Point 0.063 2 .969 Impact Point Center .009 .303 .001 1 .976 1.009 Impact Point Right 0.073 0.359 0.041 1 .839 1.075 Constant 1.838 0.657 7.832 1 .005*** 6.282 * P < 0.1 ** P < 0.05 *** P < 0.01

Table 13: Measures of significance for the platform regression Statistic p(df) Cox and Snell R Square .051 Nagelkerke R Square 0.068 Hosmer and Lemeshow Test 1(6)

The platform regression model shows that platform inclination is quite significant in indicating knapper handedness, but once again, shows congruent, instead of opposite, beta values for center, left, and right inclinations, suggesting that directional skew is not a

60 good predictor, but perhaps that presence of a platform is in this data set. As stated before, platform inclination was particularly hard for me to interpret based on Bargalló and Mosquera’s (2013) work and my own pilot study, so its significance was unanticipated. Much like the ERA model, R-squared values suggest a poor-fitting model, but the Hosmer and Lemeshow p-value is extremely significant, likely because the model was so poor that the deciles required to perform the test were ensured to have differences between observed values and expected ones (Hosmer, Lemeshow, and Sturdivant 2013).

Table 14: Summary of predictive correctness for the platform regression subset Classification Table—Plat Step Handedness Percentage Correct Right Left Right 0 171 0.0 Left 0 239 100.0 0 Overall % Correct 58.3 Right Left Right 32 152 17.4 Left 29 237 89.1 1 Overall % Correct 59.8

Table 14 shows that despite the significance of platform inclination in indicating knapper handedness, the overall predictive correctness of the platform model increased by less than 2% because while it correctly classified 32 cases as right-handed, it also incorrectly classified 29 as left-handed, thus resulting in only a tiny net improvement.

Combined with its other weaknesses, platform characteristics alone are not ideal indicators of knapper handedness.

The cortex regression was based on every possible combination of the four cortical quadrants listed by Bargalló and Mosquera (2013), resulting in 15 categories, including fully cortical (all) and non-cortical (none) flakes. 526 (83.4%) flakes were included in this analysis, while 105 (16.6%) were not, because cortex quadrants could not

61 be oriented on the flake. Tables 15 and 16 show the values and significance of the CTX regression model.

Table 15: Binary logistic regression values for the cortex subset. Results in bold-face type are statistically significant at α = 0.1. CTX Regression Model B S.E. Wald df Sig. Exp(B) Cortex 21.643 15 .118 A -.352 .550 .410 1 .522 0.703 AB 0.446 0.676 0.435 1 0.509 1.562 AC -0.688 0.403 2.917 1 0.087* 0.502 AD 20.44 40192.97 2.59E-07 1 0.999 7.57E+08 ABC -0.757 1.422 0.283 1 0.594 0.468 ABD 20.44 28420.72 5.18E-07 1 0.999 7.57E+08 B -0.064 0.880 0.005 1 0.941 0.9375 BC 0.158 0.851 0.034 1 0.852 1.171 BD 0.340 1.165 0.085 1 0.769 1.40 BCD -1.21 0.400 9.234 1 0.002*** 0.296 C -1.34 0.579 5.394 1 0.020** 0.260 CD -0.59 0.329 3.218 1 0.072* 0.553 D -0.89 0.33 6.981 1 0.008*** 0.410 ABCD (all) -21.96 40192.37 2.99E-07 1 0.999 2.90E-10 None -0.636 0.292 4.746 1 0.029** 0.529 Constant 0.757 0.156 23.45 1 1.28E-06 2.133 * P < 0.1 ** P < 0.05 *** P < 0.01

Table 16: Measures of significance for the cortex regression Statistic p(df) Cox and Snell R Square .057 Nagelkerke R Square 0.076 Hosmer and Lemeshow Test 1(6)

For this model, cortex was overall insignificant for predicting knapper

handedness, but a few classes within this dependent variable were significant: C, D, AC,

CD, and BCD. Once again, this information is encouraging at first, as these categories manifest as clearly left- and right-skewed on actual flakes. However, once again, beta values for left-skewed predictors (C and AC) are in the same direction as right-skewed predictors (D and BCD), and one class (CD) should theoretically be insignificant (if Toth and Bargalló and Mosquera are correct). Like the other models, all significant predictors

62 have negative beta values. This model also has low R-squared values and a significant

Hosmer and Lemeshow p-value, likely due to the poor quality of the regression.

Table 17: Summary of predictive correctness for the cortex regression subset Classification Table—CTX Step Handedness Percentage Correct Right Left Right 0 195 0.0 Left 0 277 100.0 0 Overall % Correct 58.7 Right 53 142 27.2 Left 38 239 86.3 1 Overall % Correct 61.9

Table 17 shows the predictive correctness for the CTX model, which increases by only 3.2% after the regression. Much like Bargalló and Mosquera’s (2013) and my own evaluation of Toth’s (1985) method, it appears that overall, cortex is not a good indicator of knapper handedness, likely because the assumption about uni-directional rotation of the core is too simplified, especially for higher-order stone tools like Acheulean handaxes.

Like eraillure scar locations, fracture locations were not mutually exclusive (i.e., a flake can be fractured in more than one place, although the probability of multiple fractures is low), so these data were entered as binary data for each location. Thus, all

631 flakes were included in the fracture regression. Tables 18 and 19 show the regression and significance for the fracture model.

63 Table 18: Binary logistic regression values for the fracture location subset. Results in bold-face type are statistically significant at α = 0.1. Fractures Regression Model B S.E. df Sig. Exp(B) A Fracture .190 .725 1 .793 1.210 B Fracture .594 .483 1 .219 1.811 C Fracture .499 .373 1 .181 1.648 D Fracture -.376 .528 1 .476 .687 E Fracture -.344 .490 1 .483 .709 F Fracture .327 .381 1 .390 1.387 G Fracture -.038 .345 1 .913 .963 Constant .428 .091 1 .000 1.534 * P < 0.1 ** P < 0.05 *** P < 0.01

Table 19: Measures of significance for the fracture location regression Statistic p(df) Cox and Snell R Square .008 Nagelkerke R Square 0.011 Hosmer and Lemeshow Test 0.6(1)

For this model, no fracture locations were found to be significant, although there is some evidence of directional skew in the beta values. This model is overall insignificant at indicating knapper handedness, as were the other four, suggesting that these subsets of flake characteristics do not preserve evidence of right- and left- handedness very well. Table 20 shows that after the regression, not a single case was interpreted with a probability value of less than 0.6, and there was no change in the predictive correctness of this model. Likewise, Bargalló and Mosquera (2013) found fractures to be insignificant.

Table 20: Summary of predictive correctness for the fracture location regression subset Classification Table—Fractures Step Handedness Percentage Correct Right Left Right 0 242 0.0 Left 0 389 100.0 0 Overall % Correct 61.6 Right 0 242 0.0 Left 0 389 100.0 1 Overall % Correct 61.6

64 Despite the individual inefficacy of each regression subset, all data and independent variable categories were included in a saturated regression model. For this model, 403 (63.9%) flakes were viable for SPSS analysis, and 228 (36.1) were not. The intent of this regression was to determine if Bargalló and Mosquera (2013) were justified in claiming that a combination of all features was the best way to interpret handedness in lithic materials. The regression and model significance for all traits are show in Tables 21 and 22.

Table 21: Binary logistic regression values for the full data set. Results in bold-face type are statistically significant at α = 0.1. Saturated Regression Model B S.E. Wald df Sig. Exp(B) Cone of Percussion 3.890 3 .274 Cone of Percussion Center .641 .407 2.486 1 .115 1.899 Cone of Percussion Right .903 .633 2.034 1 .154 2.467 Cone of Percussion Left 1.001 .572 3.061 1 .080* 2.722 Hackles 1.100 3 .777 Hackles Center .100 .348 .082 1 .774 1.105 Hackles Right -.295 .374 0.620 1 .431 .745 Hackles Left .141 .381 .136 1 .712 1.151 Ripples 2.409 3 .492 Ripples Center .468 .330 2.013 1 .156 1.596 Ripples Right -.198 .612 .105 1 .746 0.820 Ripples Left .377 .674 0.312 1 .576 1.458 Extraction Axis 5.337 3 .149 Extraction Axis Center -2.149 1.336 2.589 1 .108 .117 Extraction Axis Right -1.257 1.394 .813 1 .367 .285 Extraction Axis Left -2.949 1.488 3.929 1 0.047** 0.052 Ridge 41.401 3 0 No Ridge -1.698 0.635 7.16 1 0.007*** 0.183 Ridge Center 0.268 0.636 0.178 1 0.673 1.308 Ridge Right -0.715 0.763 0.877 1 0.349 0.489 Eraillure Center 0.085 0.389 0.048 1 0.827 1.089 Eraillure Right 0.12 0.383 0.098 1 0.754 1.127 Eraillure Left -0.541 0.453 1.429 1 0.232 0.582 Platform Type -0.188 0.299 0.395 1 0.53 0.829 Platform Inclination 15.594 3 0.001 Platform Inclination Center -3.482 1.286 7.328 1 0.007*** 0.031 Platform Inclination Right -2.675 1.284 4.343 1 0.037** 0.069 Platform Inclination Left -3.99 1.308 9.305 1 0.002*** 0.019 Impact Point 0.714 2 0.7 Impact Point Center -0.102 0.414 0.061 1 0.805 0.903 Impact Point Right 0.187 0.472 0.156 1 0.693 1.205 65 Saturated Regression Model (Continued) B S.E. Wald df Sig. Exp(B) Cortex 9.428 15 0.854 A 0.953 0.928 1.053 1 0.305 2.593 AB 1.108 0.728 2.313 1 0.128 3.027 AC -0.322 0.543 0.352 1 0.553 0.724 AD -0.447 2.035 0.048 1 0.826 0.639 ABC -2.593 1.628 2.536 1 0.111 0.075 ABD -1.474 1.545 0.91 1 0.34 0.229 B 0.952 0.997 0.911 1 0.34 2.591 BC 0.257 0.919 0.078 1 0.78 1.293 BD 0.263 2.102 0.016 1 0.9 1.301 BCD 0.097 0.526 0.034 1 0.854 1.102 C -0.694 0.75 0.856 1 0.355 0.5 CD -0.137 0.45 0.092 1 0.761 0.872 D 0.092 0.48 0.037 1 0.848 1.097 ABCD (all) -22.026 21437.078 0 1 0.999 0 (Cortex) None -0.025 0.414 0.004 1 0.952 0.976 A Fracture 1.163 0.995 1.366 1 0.243 3.199 B Fracture -0.399 0.588 0.461 1 0.497 0.671 C Fracture -1.143 0.698 2.68 1 0.102 0.319 D Fracture -0.008 0.768 0 1 0.991 0.992 E Fracture 1.267 0.84 2.274 1 0.132 3.549 F Fracture 0.146 0.549 0.071 1 0.791 1.157 G Fracture -0.682 0.653 1.092 1 0.296 0.506 Constant 5.927 2.816 4.43 1 0.035** 375.139 * P < 0.1 ** P < 0.05 *** P < 0.01

Table 22: Measures of significance for the full regression Statistic p(df) Cox and Snell R Square .160 Nagelkerke R Square 0.216 Hosmer and Lemeshow Test 0.175(8)

In the saturated model, several traits were found as significant indicators of knapper handedness, including: left-skewed cone of percussion and extraction axis, centered ridge, and centered, left-, and right-skewed platform inclinations. These variables were also prominent in the subset regressions discussed previously. For every independent variable except cone of percussion, the beta value was negative, suggesting that presence of these traits is linked to right-handed knappers, which is congruent with

66 the subset regressions, and an important finding within the context of previous assumptions about directional skew. For this model, the R-squared values suggest that the model fits the data better than the subset regressions, although it is still not a good fit overall, likely an indication of the issue of knapper stylistic variability and other contributors to flake morphology. The Hosmer and Lemeshow test is still significant, but less so than the subset models, which likely means that a working regression model is still correctly specified.

Table 23: Summary of predictive correctness for the full regression Classification Table--Saturated Model Step Handedness Percentage Correct Right Left Right 0 167 0.0 Left 0 236 100.0 0 Overall % Correct 58.6 Right 95 72 56.9 Left 44 192 81.4 1 Overall % Correct 71.2

Table 23 shows that despite the relative inefficacy of the saturated model, its predictive correctness is much better than the subset models. This supports Bargalló and

Mosquera’s advocacy for holistic data collection and analysis. For the saturated model, predictive accuracy increased by 12.6%, and over 70% of the flakes were correctly classified as right- or left-handed, which is better than any previous attempt, including my own judgments.

The same saturated model was run in a backwards regression, which uses the changes in the -2LL (used to calculate the R-squared p-values), to determine which variables can be removed from the model without resulting in a significantly worse regression. This was done because a majority of predictors were insignificant in all models, and thus they may not be necessary information for future analyses. As it was for

67 the saturated model, 403 (63.9%) flakes were viable for SPSS analysis, and 228 (36.1)

were not. Tables 24 and 25 show the values and significance of the backwards regression

model for only the first and last step.

Table 24: Binary logistic regression values for the final Step (17) of the backwards regression. Results in bold-face type are statistically significant at α = 0.1 Backwards Regression Model—Final Step B S.E. Wald df Sig. Exp(B) Step 17 Ridge 50.326 3 .000 No Ridge -1.512 .552 7.498 1 .006*** .221 Ridge Center .178 .558 0.102 1 .750 1.195 Ridge Right -.152 .669 0.052 1 .820 .859 Platform Inclination 11.983 3 .007 Platform Inclination Center -2.724 1.046 6.778 1 .009*** .066 Platform Inclination Right -2.145 1.070 4.023 1 .045** .117 Platform Inclination Left -3.022 1.078 7.857 1 .005*** .049 C Fracture -1.155 .666 3.002 1 .083* .315 Constant 4.775 1.344 12.623 1 .000*** 118.561 * P < 0.1 ** P < 0.05 *** P < 0.01

Table 25: Measures of significance for steps 1 and 17 of the backwards regression Step Statistic p(df) 1 Cox and Snell R Square .160 Nagelkerke R Square 0.216 Hosmer and Lemeshow Test 0.175(8) 17 Cox and Snell R Square .117 Nagelkerke R Square 0.158 Hosmer and Lemeshow Test 0.495(8)

After 17 steps, the backwards regression model removed several characteristics,

keeping only the ridge, platform inclination, and fracture location C. Considering the

repeated significance of the ridge and platform inclination, these two features are not

surprising. Fracture location C, however, is the only fracture that has no directional skew,

and was thus never associated to knapper handedness, by both Bargalló and Mosquera

(2013) and myself; that it was found to significantly indicate right-handed knappers

further suggests that the assumptions about predictable directional skew in percussion

force are inadequate, at best. This reduction of the model leads to lower R-squared values 68 and a poorer fit of the data’s variability, but an increase in the Hosmer and Lemeshow p- value indicates a better specified model. These differences between Steps 1 and 17 suggest that a saturated model may be beneficial over an unsaturated one in some cases, but that an unsaturated model is still useable. This result corroborates the results found by

Bargalló and Mosquera, where a few variables accounted for a majority of the variability in the data, and a majority of them accounting for relatively little variability. Table 26 shows the predictive correctness for Steps 0, 1, and 17 of the backwards regression, which parallels the results from the saturated model overall.

Table 26: Summary of predictive correctness for Steps 0, 1, and 17 of the backwards regression Classification Table--Backwards Model Step Handedness Percentage Correct Right Left Right 0 167 0.0 Left 0 236 100.0 0 Overall % Correct 58.6 Right 90 62 59.3 Left 41 211 83.7 1 Overall % Correct 74.5 Right 97 55 63.8 Left 55 197 78.2 17 Overall % Correct 72.8

A final, unsaturated model was run using data from the backwards regression, in the hope that more cases would be valid for SPSS analysis. Overall, however, the removal of extraneous variables only resulted in 50 cases being added as viable, with 453

(71.8%) flakes included in the unsaturated model, and 170 (34.5%) still not viable for analysis. This model represents the largest number of cases analyzed, as well as the variables deemed most significant by previous regressions. Tables 27 and 28 show the unsaturated regression and its associated significance values.

69 Table 27: Binary logistic regression values for the final unsaturated model. Results in bold-face type are statistically significant at α = 0.1 Unsaturated Regression Model B S.E. Wald df Sig. Exp(B) Ridge 63.5709 3 0.0000 No Ridge -1.6282 0.5448 8.9327 1 0.0028*** 0.1963 Ridge Center 0.1280 0.5540 0.0534 1 0.8172 1.1366 Ridge Right -0.1699 0.6667 0.0649 1 0.7989 0.8438 Platform Inclination 13.3339 3 0.0040*** Platform Inclination Center -1.6880 0.5776 8.5401 1 0.0035*** 0.1849 Platform Inclination Right -1.1651 0.6105 3.6420 1 0.0563* 0.3119 Platform Inclination Left -1.9732 0.6303 9.8010 1 0.0017*** 0.1390 C Fracture -1.0055 0.5985 2.8221 1 0.0930* 0.3659 Constant 3.6163 0.9800 13.6154 1 0.0002*** 37.1983 * P < 0.1 ** P < 0.05 *** P < 0.01

Table 28: Measures of significance for the final regression Statistic p(df) Cox and Snell R Square .110 Nagelkerke R Square 0.148 Hosmer and Lemeshow Test 0.464(8)

This model is consistent with the previous two in its indication of significant dependent variable classes. Once again, it shows negative beta values for all significant predictors, with relatively low R-squared values, but a significant Hosmer and Lemeshow p-value.

Table 29: Summary of predictive correctness for the final regression Classification Table--Unsaturated Model Step Handedness Percentage Correct Right Left Right 0 192 0.0 Left 0 221 100.0 0 Overall % Correct 58.4 Right 124 62 66.7 Left 66 201 75.3 1 Overall % Correct 71.7

Like the full regression, the unsaturated model performs better than any of the subset models, as it improves overall predictions from 58.6% correct to 71.7% correct

70 (see Table 29). Despite the fact that it includes more cases than any other model, however, the unsaturated model, which includes only the platform inclination, ridge, and fracture location C, does not show improvements in overall predictive correctness when compared to the full regression, and removing insignificant independent variable predictors reduces the model’s fit for the data overall.

However, according to one additional measure of significance, the “coefficient of discrimination” (proposed by T. Tjur in 2012), it becomes clear that the unsaturated model can better discriminate between right-and left-handed flakes, likely due to the removal of extraneous predictor variables (Tjur 2012). The coefficient of discrimination is simply the difference between the average p-values for all cases predicted as group 1

(left-handed) and the average p-values for all cases predicted as group 0 (right-handed).

According to Tjur (2012), in a good regression, most values align at the extreme ends of the logistic curve (p close to 1 or 0), and few values lie in between. In this case, the coefficient of discrimination would be close to 1, indicating a good model. However, in a poor regression where dependent-variable predictions are not well informed by the data, p-values may cluster at either end of the spectrum, or in the middle, leading to low coefficients of discrimination. For the saturated model, this value is 0.17, indicating a lack of very high and very low p-values (or a lack of easily-distinguishable group associations). For the unsaturated model, this value is 0.41, indicating a clearer separation between left-associated p-values and right-associated ones.

According to the predictive correctness and the various measures of significance, the unsaturated regression model is valid as a stand-alone method for determining handedness in flakes. This model also provides a simple platform to demonstrate how

71 these regressions could be used on single flakes for which handedness is unknown. Recall that the formula for binary logistic regression is as follows:

= B0 + B1*x1 + B2*x2 + B3*x3 + B3*x3 . . . Bn*xn

In the case of the unsaturated regression, three B values will be used to calculate the p- value for each flake. B1 is determined by the dummy coded matrix for ridge (see Figure

11: top row), B2 is determined by the matrix for platform inclination (Figure 11: middle row), and B3 is determined by fracture location C (Figure 11: bottom row).

: 

Figure 11: Example of flake characteristics, associated matrices, and beta values.

As an example, flake 13 represents a left-handed flake that was predicted correctly, both by myself and by the unsaturated regression. It demonstrated a centered ridge, right platform inclination, and had no C fracture. The regression calculation for flake 13 follows:

= 3.616 + 0.128 – 1.165 – 1.055 = 1.574  = 0.82831

This value represents the probability of the flake falling into the binary category of 1, and is greater than the cutoff value of 0.6. Thus it is predicted as a left-handed flake. Flake 25 represents a correctly predicted right-handed flake that showed no ridge, a centered platform, and no C fracture, with the following regression calculation: 72

= 3.616 – 1.628 – 1.688 – 1.055 = -0.705  = 0.33061

This value is lower than the cutoff value of 0.6, and with the resultant group prediction of

0, or right-handed. Flake 16 was labeled as “Indeterminate” by myself, and it showed no skewed characteristics other than platform inclination, which was right-oriented, so its regression calculation is:

= 3.616 – 1.628 – 1.165 – 1.055 = -0.182  = 0.4549

The final p-value suggests that the flake is right-handed, when in fact it came from a left-handed knapper. In this case, both the regression and my judgment were inaccurate in determining knapper handedness. As shown by the predictive correctness of the unsaturated model, the binary logistic regression predicted knapper handedness incorrectly relatively often, and these cases often coincided with flakes that I myself deemed indeterminate.

The beta values and matrices herein presented could potentially be used on future data sets for which handedness is unknown as well, which was a major goal of this thesis.

Flakes could be cataloged using Bargalló and Mosquera’s (2013) original technique, and depending on which features were present in each flake, researchers could select the most appropriate regression equation for calculating handedness p-values. In future research, left-handedness would retain its binary value of one (1), but p-values could be interpreted with varying cut-off values, using 0.5 as a standard.

The addition of working regression equations for nominal data on flake characteristics is a large advance in determining handedness in lithic debitage, but issues remain. Even with the improvements in overall test performance in the holistic models, it is clear that all regressions explain very little of the variability in the data. Additionally,

73 the tendency for all variants of a dependent variable class (for example platform inclination: center, left, and right) to have congruent instead of opposite beta values, suggests that right- and left-skew in lithic traits do not correspond directly to flintknapper handedness, once again highlighting the weaknesses in the original theoretical premises of the reviewed works. It is important to note that each regression predicted left- handedness better than right-handedness, which could be due to the asymmetry of the meta-assemblage overall (61.6% of the flakes were produced by left-handers, whereas only 38.4% were produced by right-handers), but would need confirmation on a more equally distributed meta-assemblage. It is evident that binary logistic regression is a more appropriate form of data analysis, especially for Bargalló and Mosquera’s (2013) nominal classification scheme, but it is also clear that much more work needs to be done.

Additional Analysis—Effects of Knapping Style on Flake Debitage

As suggested by Uomini’s (2006) dissertation findings, it is likely that individualized knapping styles manifest in the flakes that knappers produce, ultimately leading to highly varied assemblages that are extremely difficult to evaluate.

Additionally, Bargalló and Mosquera (2013) stress the importance of looking at flakes in preserved clusters, instead of individually. Combined with Uomini’s research proving that each knapper has his or her own style, I decided to see if there were detectable differences in flake characteristics by knapper, instead of by handedness. For this analysis, a simple Pearson’s chi-squared test was run on each of Bargalló and Mosquera’s characteristics: cone of percussion, hackles, ripples, extraction axis, ridge, eraillure scars, platform type and inclination, impact point, cortex, and fractures. Chi-squared values

74 were calculated for each variable frequency by handedness first (frequency data for these tests can be found in Appendix D), as an analog to the binary logistic regressions. These tests, much like the various regressions, rarely resulted in significant differences between features and handedness. However, when chi-squared values are calculated by knapper, almost every characteristic shows significant differences in frequencies, even with an ɑ value of 0.05. These tests are based on frequency data in Appendices D and E, so there are varying degrees of freedom. The df is calculated for each table, where the column df

(n-1) and the row df (n-1) are multiplied. Thus, for each characteristic or row (hackles, cortex, fracture location A, etc.), the n does not change (e.g. the df for platform type is 1.

[linear or punctiform=2, 2-1=1), but the column n changes from 1 in the handedness chi- squared tests (left or right=2, 2-1=1) to 4 in the knapper chi-squared tests (5 knappers=5,

5-1=4). Comparative chi-squared values are shown for each feature in Tables 30-32.

Table 30: Chi-squared values for the cone of percussion subset by handedness (left) and knapper (right) Chi Squared Values by Handedness Chi Squared Values by Knapper Cone of Percussion Cone of Percussion Value df p-value Value df p-value Pearson Chi-Square 7.676 3 .053 Pearson Chi-Square 23.309 12 .025 Likelihood Ratio 7.595 3 .055 Likelihood Ratio 21.143 12 .048 Hackles Hackles Value df p-value Value df p-value Pearson Chi-Square 3.96 3 .266 Pearson Chi-Square 21.483 12 .044 Likelihood Ratio 3.947 3 .267 Likelihood Ratio 21.595 12 .042 Ripples Ripples Value df p-value Value df p-value Pearson Chi-Square 2.843 3 .416 Pearson Chi-Square 23.206 12 .026 Likelihood Ratio 2.818 3 .421 Likelihood Ratio 23.497 12 .024 Extraction axis Extraction axis Value df p-value Value df p-value Pearson Chi-Square 2.953 3 .399 Pearson Chi-Square 15.326 12 .224 Likelihood Ratio 2.908 3 .406 Likelihood Ratio 16.187 12 .183 Ridge Ridge Value df p-value Value df p-value Pearson Chi-Square 36.942a 3 .000 Pearson Chi-Square 104.437 12 .000 Likelihood Ratio 38.026 3 .000 Likelihood Ratio 116.814 12 .000

75 For the CoP cluster, ridge is the only characteristic that differs significantly by handedness and by knapper. This is not unexpected, as ridge was also found to be a significant predictor of knapper handedness in both the saturated and unsaturated regression models. However, when grouped by knapper, cone of percussion, hackles, and ripples also show significant differences.

Table 31: Chi-squared values for eraillure scars and the platform subset by handedness (left) and knapper (right) Chi Squared Values by Handedness Chi Squared Values by Knapper Center Eraillure Scar Center Eraillure Scar Value Df p-value Value df p-value Pearson Chi-Square 0.27 1 .603 Pearson Chi-Square 4.524 4 .340 Likelihood Ratio .272 1 .602 Likelihood Ratio 4.208 4 .379 Right Eraillure Scar Right Eraillure Scar Value df p-value Value df p-value Pearson Chi-Square 1.011 1 .315 Pearson Chi-Square 13.321 4 .010 Likelihood Ratio 1.040 1 .308 Likelihood Ratio 12.357 4 .015 Left Eraillure Scar Left Eraillure Scar Value df p-value Value df p-value Pearson Chi-Square 1.801 1 .180 Pearson Chi-Square 15 4 .005 Likelihood Ratio 1.793 1 .181 Likelihood Ratio 14.490 4 .006 Platform Type Platform Type Value df p-value Value df p-value Pearson Chi-Square 1.801 1 .180 Pearson Chi-Square 7.956 4 .093 Likelihood Ratio 1.793 1 .181 Likelihood Ratio 8.364 4 .079 Platform Inclination Platform Inclination Value df p-value Value df p-value Pearson Chi-Square 17.714 3 .001 Pearson Chi-Square 32.359 12 .001 Likelihood Ratio 19.524 3 .000 Likelihood Ratio 31.878 12 .001 Impact Point Impact Point Value df p-value Value df p-value Pearson Chi-Square 0.14 2 .932 Pearson Chi-Square 15.66 8 .048 Likelihood Ratio .141 2 .932 Likelihood Ratio 16.228 8 .039

For eraillure scars and platform characteristics, only platform inclination is found to differ significantly both by handedness and knapper. Once again, platform inclination was a valid indicator of knapper handedness in each regression that included it, so its chi- square values are expected. When grouped by knapper, right- and left-skewed eraillure scars, as well as impact point, are found to differ significantly.

76 Table 32: Chi-squared values for cortex and fracture locations by handedness (left) and knapper (right) Chi Squared Values by Handedness Chi Squared Values by Knapper Cortex Cortex Value df p-value Value df p-value Pearson Chi-Square 9.476 15 0.851 Pearson Chi-Square 62.104 60 0.401 Likelihood Ratio 9.909 15 0.825 Likelihood Ratio 68.211 60 0.218 A Fracture A Fracture Value df p-value Value df p-value Pearson Chi-Square 0.583 1 0.445 Pearson Chi-Square 10.747 4 0.03 Likelihood Ratio 0.567 1 0.451 Likelihood Ratio 10.198 4 0.037 B Fracture B Fracture Value df p-value Value df p-value Pearson Chi-Square 0.021 1 0.886 Pearson Chi-Square 2.214 4 0.696 Likelihood Ratio 0.021 1 0.886 Likelihood Ratio 2.444 4 0.655 C Fracture C Fracture Value df p-value Value df p-value Pearson Chi-Square 4.962 1 0.026 Pearson Chi-Square 9.281 4 0.054 Likelihood Ratio 5.577 1 0.018 Likelihood Ratio 8.51 4 0.075 D Fracture D Fracture Value df p-value Value df p-value Pearson Chi-Square 0.887 1 0.346 Pearson Chi-Square 5.128 4 0.274 Likelihood Ratio 0.933 1 0.334 Likelihood Ratio 5.212 4 0.266 E Fracture E Fracture Value df p-value Value df p-value Pearson Chi-Square 0.239 1 0.625 Pearson Chi-Square 2.735 4 0.603 Likelihood Ratio 0.234 1 0.628 Likelihood Ratio 3.571 4 0.467 F Fracture F Fracture Value df p-value Value df p-value Pearson Chi-Square 1.18 1 0.277 Pearson Chi-Square 10.848 4 0.028 Likelihood Ratio 1.23 1 0.267 Likelihood Ratio 15.953 4 0.003 G Fracture G Fracture Value df p-value Value df p-value Pearson Chi-Square 1.842 1 0.175 Pearson Chi-Square 5.785 4 0.216 Likelihood Ratio 1.942 1 0.163 Likelihood Ratio 4.824 4 0.306

Finally, cortex and fracture locations are the only characteristics that generally do not seem to differentiate by handedness or by knapper. As we saw in the regressions, fracture location C shows significant differences by handedness, but it does not show significant differences by knapper at ɑ=0.05. The chi-squared values do suggest that fracture locations A and F differ significantly by knapper, however.

Overall, these chi-squared tests show that knappers produce unique combinations of the technical features identified by Toth (1985), Rugg and Mullane (2001), and 77 Bargalló and Mosquera (2013), which opens up new avenues of research that are not necessarily related to identifying handedness. It has been repeatedly mentioned in previous studies that clusters of flakes are more viable for analysis than single flakes because they can be assessed within the context of each other. In many cases, however, clearly associated flakes are rare. Although this thesis has resulted in multiple useable regression equations for single flakes presenting any combination of traits, these models are not perfect and are relatively unreliable in some cases, and a working method for associating multiple flakes to a single hominid knapper may also be useful in assessing paleo-archaeological sites. Additionally, identifying a cluster of flakes from a single knapper at a site could be a preliminary step for assessing the handedness of their assemblage, using perhaps one of the regression equations above to calculate p-values for each flake in the cluster, and deriving a handedness inference from the results.

Inter-observer Comparisons—Assessment of Bargalló and Mosquera’s method’s reliability

The final evaluation I completed for this thesis was an assessment of the reliability of Bargalló and Mosquera’s (2013) methodology. As mentioned before, almost every flake characteristic used by Bargalló and Mosquera is categorical in nature, which first limits the available range of statistical assessment. An additional complication of using this system is that most characteristics are simply judged by eye, which introduces problems with replicability, particularly between observers. In order to test whether different observers classify flake characteristics in the same manner, two additional observers who specialize in lithic analysis assessed a stratified random sample of flakes.

78 All three observers assessed 25 flakes in common, and an additional 25 flakes were assessed by only two observers. Frequency data for each characteristic for each observer can be found in Appendices F and G.

Before testing inter-observer agreement, I assessed how well each observer identified overall knapper handedness for their sample of 50 flakes using Bargalló and

Mosquera’s (2013) methods (see Table 5 in my own assessment of Bargalló and

Mosquera’s methodology for comparison). Tables 33 and 34 show how well each observer classified their flakes as right- or left-handed by handaxe.

Table 33: Summary of Observer B's overall handedness inferences by handaxe Handedness Classifications: Observer B Knapper Hand- N of N % N % N % Handed- axe flakes Correct Correct IND IND Wrong Wrong ness A Right 1 0 0.00 0 0.00 1 100.00 B Right 0 0 0.00 0 0.00 0 0.00 C Right 2 1 50.00 0 0.00 1 50.00 D Left 6 2 33.33 2 33.33 2 33.33 E Right 8 1 12.50 3 37.50 4 50.00 F Right 4 3 75.00 0 0.00 1 25.00 G Left 8 2 25.00 3 37.50 3 37.50 H Left 3 2 66.67 1 33.33 0 0.00 I Left 7 4 57.14 2 28.57 1 14.29 J Left 10 5 50.00 2 20.00 3 30.00

Table 34: Summary of Observer C's overall handedness inferences by handaxe Handedness Classifications: Observer C Knapper Hand- N of N % N % N % Handed- axe flakes Correct Correct IND IND Wrong Wrong ness A Right 3 1 33.33 2 66.67 0 0.00 B Right 0 0 0.00 0 0.00 0 0.00 C Right 2 0 0.00 2 100.00 0 0.00 D Left 6 3 50.00 2 33.33 1 16.67 E Right 7 1 14.29 5 71.43 1 14.29 F Right 0 0 0.00 0 0.00 0 0.00 G Left 11 6 54.55 3 27.27 2 18.18 H Left 3 0 0.00 2 66.67 1 33.33 I Left 8 2 25.00 3 37.50 3 37.50 J Left 10 4 40.00 4 40.00 2 20.00

79 Both observers show results much like my own classifications, with

“Indeterminate” handedness inferences being quite common, and no significant differences between the percentage of correctly-classified vs. incorrectly-classified flakes, in many cases. As stated before, flakes often showed either a mixture of right- and left-associated characteristics, or a predominance of centered characteristics, often making handedness inferences rather difficult to make. Without the regression formulas, knapper handedness is extremely hard to identify in single flakes using Bargalló and

Mosquera’s (2013) classification scheme.

For the 25 flakes analyzed by all three observers, a Fleiss’s Kappa value was calculated to determine agreement, with values close to 0 indicating poor agreement and a value of 1 indicating full agreement between all three observers (Fleiss et al. 1969).

Table 35 shows Fleiss values and standard error (SE) for each characteristic, as well as each subset handedness inference, and finally the overall handedness inference.

80 Table 35: Fleiss' Kappa values for each characteristic using all three observers Fleiss Kappa Values: n=25 Characteristic Value SE Cone of Percussion 0.1232 0.0660 Hackles 0.1599 0.0659 Ripples 0.1587 0.0654 Extraction Axis 0.0966 0.0793 Ridge 0.0113 0.0806 Eraillure Center 0.0227 0.1155 Eraillure Right 0.2424 0.1155 Eraillure Left 0.3333 0.1155 Platform Type 0.2110 0.0885 Platform Inclination 0.0891 0.0671 Impact Point 0.2526 0.0766 Cortex 0.3777 0.0498 A Fracture 1.0000 0.0000 B Fracture 0.0135 0.1155 C Fracture 0.5273 0.1155 D Fracture 0.4565 0.1155 E Fracture 0.5273 0.1155 F Fracture 0.5210 0.1231 G Fracture 0.3608 0.1155 CoP Handedness Inference 0.0969 0.0831 ERA Handedness Inference 0.0919 0.0185 Platform Handedness Inference 0.0834 0.0775 Cortex Handedness Inference 0.3746 0.0865 Fracture Handedness Inference 0.3746 0.0865 Overall Handedness Inference 0.2118 0.0836

In general, Fleiss values suggest that inconsistency is indeed an issue when using

Bargalló and Mosquera’s (2013) methodology. Relatively few characteristics have values above 0.2, which reflects “partial agreement” between all three observers. Additionally, only 4 characteristics, all of which were fracture locations, have values above 0.5. The sample size for these tests was quite low at only 25 flakes out of 631 (less than 5%), and some standard error values (SE) are relatively large compared to their Fleiss values.

For the flakes that were only classified by two observers (n=50), a weighted

Cohen’s Kappa test was used to assess inter-observer differences in classifications.

Weighted Kappa values represent not only agreement between observers, but also

81 magnitude of disagreement (Nichols et al. 2011; Warrens 2011). For example, if one observer classified the ripples on a flake as right, and another classified them as centered, the disagreement would be less significant than if one classified them as right and the other classified them as left. Like Fleiss’ values, 1 represents full agreement and values close to 0 represent poor agreement. Table 36 shows the weighted Cohen’s Kappa values and associated standard error (SE) for each characteristic and handedness inference.

Comparisons between observer A and B, and A and C have a sample size of 50, whereas the comparison between observers B and C reflects the same 25 flakes that were assessed in the three-way Fleiss test above. By increasing the sample size for two pair-wise comparisons, I hoped to reduce the standard error, and a general pair-wise approach may indicate that one observer was more prone to disagreement with the other two.

82 Table 36: Weighted Cohen's Kappa values for each characteristic using pairs of observers Weighted Cohen's Kappa Values Observer Observer Observer Characteristic A-B SE A-C SE B-C (n=25) SE Cone of Percussion 0.1483 0.1058 0.1250 0.1270 0.2778 0.1546 Hackles 0.3143 0.1057 0.0989 0.1445 0.1294 0.1721 Ripples 0.2161 0.0972 0.0935 0.0972 0.2629 0.1709 Extraction Axis 0.0643 0.1114 0.0933 0.1113 0.2424 0.1751 Ridge 0.0329 0.0511 0.2444 0.1199 0.1596 0.0057 Eraillure Center 0.2599 0.1396 0.1379 0.2296 0.3902 0.2170 Eraillure Right 0.4595 0.2592 0.1814 0.1953 0.1804 0.0174 Eraillure Left 0.3961 0.2562 0.3386 0.2296 0.3067 0.2331 Platform Type 0.2529 0.1959 0.3094 0.1628 0.2021 0.2840 Platform Inclination 0.0721 0.1146 0.2770 0.1229 0.2097 0.0226 Impact Point 0.2385 0.1278 0.2574 0.1216 0.1983 0.0741 Cortex 0.3222 0.1417 0.4086 0.1219 0.5044 0.1290 A Fracture 1.0000 0.0000 0.707 0.0204 1.0000 0.0000 B Fracture 1.0000 0.0000 0.4944 0.0309 0.9798 0.0012 C Fracture 0.3450 0.2509 0.2105 0.2383 0.3595 0.3363 D Fracture 0.1071 0.2525 0.3056 0.2021 0.3595 0.3363 E Fracture 0.1935 0.2827 0.6575 0.3390 0.6269 0.2351 F Fracture 0.1583 0.2253 0.3363 0.2542 0.7500 0.1696 G Fracture 0.2372 0.2302 0.2857 0.2504 0.4118 0.2696 CoP Handedness Inference 0.0542 0.1483 0.2562 0.1123 0.1914 0.1333 ERA Handedness Inference 0.2392 0.1554 0.2540 0.1493 0.1635 0.1492 Platform Handedness Inference 0.0741 0.1328 0.1617 0.1353 0.1793 0.1497 Cortex Handedness Inference 0.1901 0.1054 0.3368 0.1362 0.4280 0.1697 Fracture Handedness Inference 0.2049 0.0983 0.4025 0.1387 0.6405 0.1538 Overall Handedness Inference 0.1192 0.1200 0.1302 0.1248 0.2325 0.1748

As indicated by the weighted Cohen’s Kappa values, no single observer seems to disagree with the other two more often across characteristics, and a sample size of 50

(versus 25) does not significantly impact SE values in most cases. It is interesting to note that observers B and C have more Kappa values above 0.5 than A and C or A and B, but this may be due to the sample size difference. Like the Fleiss tests, the weighted Cohen’s

Kappa values suggest that inter-observer subjectivity is a large issue with this methodology. In Rugg and Mullane’s (2001) study, they attempted to assess replicability by having the same observer look at each flake twice with a few days in between judgments.

83 They state that:

[t]here were no cases in which flakes that were judged assignable were judged to

have different orientation on both occasions, though there were cases in which a

flake was judged unassignable on one occasion and assignable on the other (Rugg

and Mullane 2001:255).

This suggests that even for a single observer, some issues in replicability exist for these classification systems. These issues are abundantly clear when classifications from multiple observers are compared, and considering the general inefficacy of the regressions based on the data from a single observer, alternative approaches for data collection should be investigated.

Overall, the additional observers found the classification scheme to be quite difficult to utilize, even though they were familiar with Bargalló and Mosquera’s (2013) methodology. In fact, the target sample size for this thesis, which was originally 50 flakes overlapped between the three observers and 100 flakes overlapped by each pair of observers, had to be reduced because of the time required for analysis. It seems that expertise in standard lithic analysis does not fully translate to assessing handedness in flakes, perhaps because the additional observers did not gain a clear understanding of the theoretical premise underlying these works on handedness from only reading Bargalló and Mosquera’s work. Future assessments of these methods should be completed by those who are familiar with all of the relevant literature, although my data suggest that the theoretical frameworks of the existing approaches for determining knapper handedness are not universally accurate.

84 CONCLUSIONS

Each of the methods reviewed as part of this thesis have various theoretical bases, motivations, and methodological approaches, but all of them have come to the same conclusions: definitive evidence of handedness is extremely difficult to find outside of studying living human beings, and lithic-based evidence of handedness is a complex issue. My results reflect the limitations present in the studies herein reviewed, and I have also introduced some new topics that need addressing. Still, several important contributions were made through this thesis, including:

1) An additional assessment of the applicability and replicability of original analytical methods (Toth 1985; Rugg and Mullane 2001).

2) A statistical confirmation of the viability and necessity of a combinatory methodology (Bargalló and Mosquera 2013).

3) An improved approach for turning data into actual handedness estimates via binary logistic regression.

Synthesis of Results for the Toth, Rugg and Mullane, and Bargalló and Mosquera methodologies

In the case of Toth’s cortex method (1985), less than half of the 2 x 2 cm flakes had cortex present, and all analyses of cortex location show that it is insignificant at indicating knapper handedness. Using judgment alone, I predicted the handedness of six out of 10 handaxes correctly, and binary logistic regression models repeatedly found

85 cortex location overall insignificant, although some individual locations did improve predictive correctness. As mentioned earlier, Toth’s original method (1985) requires a sequence of flakes rather than singular ones, and his basic assumption involves unidirectional rotation of the core in the knapper’s non-dominant hand. This reduction method could have been used in simple Oldowan manufacture, but decortication is an initial step in many biface manufacture techniques, and it relies upon multi-directional rotation of cores, where left- and right-skew of cortex is simply unrelated to directional core rotation. Additionally, cortex was one of the few independent variable classes that showed no significant differences between knappers as well and handedness, suggesting that modes of removal are likely not stylistic or personalized, but universal or intrinsic to biface manufacture. More generalized studies on biface production may help us to detect this formulaic rotation scheme, which would better help in assessing cortex location, but in order to use this for detecting handedness, refits would likely be necessary. In terms of inter-observer agreement, cortex showed relatively high agreement between raters, but overall this thesis suggests that cortex location should be abandoned in the context of studying handedness pending further work on decortication approaches for various

Paleolithic tool types.

My assessment of Rugg and Mullane’s (2001) methodology also proved disastrous in terms of my own judgments, which were only correct for four out of the ten handaxes. Using binary logistic regression, the cone of percussion subset was also found insignificant as an indicator of handedness. Interestingly, the presence and skew of a ridge on the cone of percussion was one of the only significant predictors of knapper handedness. This is an unfortunate finding, however, considering that only 49 flakes out

86 of the 631 (less than 10%) had right- or left-skewed ridges. Still, over 200 flakes showed a centered ridge, which was a significant indication of right-handedness in regressions, despite its theoretical neutrality in terms of handedness. When compared between knappers, it seems that two out of the five knappers were more likely to produce skewed ridges in general—although one was left-handed and the other was right-handed—and one knapper hardly produced ridges at all (see Appendix E). This implicates the role of personal knapping styles, and based on the levels of experience for these three knappers, I believe that the presence of a ridge is related to expertise, (much like hinge terminations, for example), although more work is needed to confirm this. The kappa values for ridge were quite low, suggesting issues with replicability, and the repeated failure of the protractor method, which could eliminate some subjectivity, does not seem like an alternate approach. Overall, the cone of percussion and ridge methods suggested by Rugg and Mullane should be used with caution. The authors noted that the chance of both

Toth’s (1985) method and their own being incorrect when combined should be very low, but in many cases, the combination of these methods would have led to contradictory or still-incorrect handedness assumptions (see Tables 3 and 4).

My review of Bargalló and Mosquera’s (2013) methodology introduces more promising avenues for further research than the CTX or CoP approaches, but this method showed mixed success overall. It is clear from my subset regressions that Bargalló and

Mosquera were correct in their claim that a combination of traits on a flake can indicate handedness far better than a single trait can. Still, the backwards regression process removed a majority of the dependent variable classes (eight of the original 11, see Table

87 1) and retained much of its predictive power. The three significant predictors, however, were quite surprising within the context of Bargalló and Mosquera’s original work.

The ridge location was originally deemed insignificant by Bargalló and Mosquera using a Mann-Whitney U test, and as discussed above, this characteristic had some methodological issues in my own replication that need to be addressed. Despite this,

Bargalló and Mosquera’s (2013) correspondence analysis and my own regression analysis both corroborate Rugg and Mullane’s (2001) theory on right- and left-skew predicting right- and left-handedness, respectively. Likewise, the platform inclination was a difficult characteristic to associate with handedness in my own assessment of this methodology, and it showed relatively poor agreement between observers. Bargalló and

Mosquera found that right- and left-platform slope associated with left- and right- handedness, respectively, which my regressions partially support. The left-sloped platform did have a relatively strong beta value indicating a right-handed knapper

(-1.9732, see Table 27), but all other platform inclinations also had negative beta values, and right-sloped platforms did not associate with left-handed knappers in my meta- assemblage. Regarding fracture location C, its significance is perhaps the most surprising result of regression analysis, as it is the only neutral fracture location of the seven listed by Bargalló and Mosquera, and none of the other six were found to indicate handedness.

Still, this flake characteristic showed relatively high agreement between observers, and it was also the only variable that showed clearer separation by handedness than by knapper.

The use of just these three significant independent-variable predictors only added a few cases to a final regression, and despite the efficacy of the final regression, I advocate the saturated model, much like Bargalló and Mosquera encourage a holistic approach.

88 Additional Considerations and Recommendations

Although flintknapping is a hobby for many, expert flintknappers interested in this type of research are rare, especially those who are left-handed (see Bargalló and

Mosquera 2013). Additionally, studies that infer handedness from material culture in particular have such low signal-to-noise ratios that they seem impractical to many, and the low number of assignable flakes produced from a single knapping event are discouraging (Patterson and Sollberger 1986; Pobiner 1999; Uomini 2001). Combining these inherent limitations with the complexity of lithic analysis and the breadth of

Paleolithic assemblages, it is no surprise that there have only been a handful of studies ever conducted. In my review of these studies, I have confirmed many of the previously suggested weaknesses, but I have also introduced several new issues.

First is that evidently, the theoretical skew of flake characteristics does not correlate very well with actual assemblages, and it is clear that personalized knapping styles lead to highly variable data. In terms of handedness, the influence of knapping style on technical characteristics in flakes was heavily discussed by Uomini (2006), but it is relatively unaddressed in the wider literature. Anecdotally, however, many of the knappers I interacted with as part of this thesis showed immense skepticism regarding the detection of handedness in debitage, and some claimed that they could make a left- skewed flake just as easily as a right-skewed one, depending on how they positioned their core. My finding that flake characteristics separate better by knapper than by handedness corroborates the fact that knapping, much like handwriting, is very specific to each individual, and future works on detecting individual knappers may be extremely fruitful.

89 Perhaps another reason for the variability in the data is a lack of experimental control. As mentioned earlier, some measures for experimental control were taken in the design of this experiment, but in reality, it was quite difficult to recruit knappers in the first place, and many decisions regarding data collection were made in response to unforeseen issues. In terms of material, I used nodules form a single source in order to ensure consistent material types, but even the variability within that source was surprising, with colors ranging from grey to brown to orange, and some nodules being extremely grainy, others being quite smooth, and one being completely unworkable (see

Figure 2). In terms of this thesis, it is unclear whether the frequencies of some features

(e.g. fractures, ripples, hackles) may be related to the material of the nodule vs. the knapper. Some recent studies have suggested using alternate materials in knapping experiments, which I believe could be extremely useful with regard to studying handedness in flake debitage.

As discussed earlier, silicate-based bricks have been used to compare novice and expert knapping styles, in order to accommodate novices’ inability to knap actual stone

(Geribàs et al. 2010), but more researchers are now advocating the use of alternative materials as assurance in experimental control. In a recent publication, N. Kreisheh, D.

Davies, and B. Bradley (2013) show that pre-molded porcelain blanks are softer and easier to knap and more manageable in terms of consistency, but they still preserve all aspects of worked stone (including traits of conchoidal fracture from manufacture, and even use-wear) (Kreisheh et al. 2013). Large chert nodules are relatively rare in the continental U.S., and thus they are quite hard to come by, but one could manufacture one’s own porcelain bricks with access to materials and a kiln. The use of perfectly

90 uniform porcelain blanks could eliminate, or at least reduce, the effect of material inconsistency on flake characteristics, which would be extremely helpful in studies on differential fracture mechanics of right- and left-handed knappers. Future study focused on the mechanics of flintknapping in general could benefit from controlled material sources.

Another source of variability in the meta-assemblage is the generally relaxed manufacture guidelines I gave to my knappers, which left styles largely in the control of the knappers themselves. I chose to allow my knappers freedom for several reasons. In many of the methodologies I reviewed, knappers were also given simple guidelines, such as to produce as many flakes as they could, or to only make tools that they were capable of. Additionally, a main motivation of my thesis was to produce results that could be applied to actual assemblages, so I did not want to impose strict guidelines that would result in a meta-assemblage unlike those we find at typical sties. The results of this relaxed experimental design, however, may have had larger consequences than I expected, especially after I had to solicit two additional left-handed knappers, which affected my plan to have each knapper make three handaxes. Overall, my intended 50-50 distribution was not achieved, but larger differences by knapper may have skewed my analysis as well. For example, the only left-handed knapper that made all three handaxes produced 249 of the flakes in my analysis (almost 40% of the meta-assemblage), whereas the only right-handed knapper that made all three handaxes only produced 91 flakes

(roughly 14%) above 2 x 2 cm (see Table 2). For future research, a method like Uomini’s dissertation work (2006), where she filmed knappers and analyzed their manufacture techniques in conjunction with the materials they produced, should be used whenever

91 possible. Knappers could be given either a set number of flakes to create and no goal of a finished product, or they could be given a target product of a uniform size, perhaps using uniform blanks made of porcelain or another silicate-based material.

Additional controls on the creation and collection of materials may help future researchers obtain a better data set for analysis, which may lead to better associations between flake characteristics and handedness through a reduction of variability.

However, one remaining issue still stands: the nominal nature of the classification system and inter-observer differences in its use. Rugg and Mullane (2001) attempted to address the issue of subjectivity by using the protractor method and assessing each flake twice, but they were unsuccessful in these endeavors (Rugg and Mullane 2001). As indicated by the Fleiss’ and weighted Cohen’s Kappa values for each characteristic (see Tables 35 and

36), multiple raters do not typically characterize a single flake as having the same characteristics. Considering the nature of the classification system, where all assessments are made by simple eye judgment and the options within each independent variable

(center, left, right, or none, in most cases) are hard to distinguish for many flakes, this result is not surprising. Future replications of Bargalló and Mosquera’s (2013) methodology should have three observers classify all flakes (instead of just a sample), and in the case of a one-way disagreement, the modal classification should be used for that flake. In a case where all three observers disagree on a characteristic, it should not be used in analysis, although this was relatively rare in our sample.

My other proposed solution to the issue of unreliability is much like Uomini’s final recommendation in her dissertation, although she was not proposing it within the same context. I recommend that future studies use a more rigorous morphotechnical

92 measurement system, such as 3D scanning, in lieu of simple observation for characteristics on flakes. In general, nominal data are hard to analyze, especially using statistics, and outside of analyzing frequencies, little has been done to assess the strength of the existing data collection schemes (consider that Bargalló and Mosquera used an inappropriate statistical measure as a major evaluation of their data). Using 3D scanning technology on flakes first eliminates the issue of using nominal data, as it generates spatial data instead. As mentioned earlier, binary logistic regression is certainly the most appropriate test for the current methodology, but my results indicate that much work still needs to be done. Spatial data are relatively complex, but they can be manipulated and statistically evaluated in a multitude of ways, such as geometric morphometrics, which may lead to better associations between traits and handedness, or even knapper. Second,

3D scanning would significantly reduce the effect of observer bias in data collection, although some subjectivity may still be inherent in rendering scans. Overall, an approach using 3D scanning would require a major modification of the existing classification schemes, if not an entirely new system, and in terms of some characteristics, 3D scanning may not prove to be a valid approach. Due to the involved nature of scanning, this approach should first be used on a small sample of flakes, perhaps with the simple goal of identifying how well Bargalló and Mosquera’s (2013) nominal classifications are exhibited in scans, and checking if these scans improve multi-rater agreement or regressions. If improvements are clear, a more involved scanning project could be viable.

Regardless of these issues, the ultimate result of this thesis is the addition of information to a field so new that even inconclusive data is received enthusiastically.

Despite the issues encountered, I believe that lithic analysis within the context of

93 handedness could provide exceptional insights into hominid evolution, if these methods continue to be improved upon. The inherent weaknesses of this thesis, as well as the methodologies herein reviewed, are secondary to the benefits that future studies may provide regarding hominid cognitive and behavioral evolution.

Implications of this Study: Applying Experimental Data to Fossil Assemblages

The use of experimental archaeology to determine handedness in fossil hominids is a relatively new approach in paleoanthropology, but its implications are vast. Out of the approaches that form the basis of this thesis, only three (Toth 1985; Uomini 2001,

2006) applied experimental data to fossil evidence, with extremely mixed results.

Gaining access to Paleolithic assemblages was outside the scope of this thesis, but the results of my study can be used on assemblages in the future. Since flakes are not unique to the manufacture of Acheulean tools, nor are they unique to modern-day knappers, the indicative technical features assessed in this thesis can also be assessed on varying

Paleolithic types, which could help track handedness estimates through time and space in the hominid fossil record. The best of my regressions only increased the overall predictive correctness of handedness classifications by 15.9%, which is not extremely promising. Still, the use of binary logistic regression on this data has resulted in equations that can be applied to singular flakes from any site in the world, which is unprecedented and important for several reasons.

As Bargalló and Mosquera noted (2013), many flakes do not display all 11 characteristics, but I have presented five subset regressions along with the saturated and unsaturated models, meaning that future researchers could pick the most appropriate

94 model for each flake they analyze, dependent upon which traits it showcases. Likewise, any combination of the 11 dependent variables can be used to run a new regression if needed. Bargalló and Mosquera, and many others, repeatedly claim that their methods work best on a cluster of associated flakes, especially refits, where a sequence of removals can be determined. Using the regression equations eliminates this problem, as each flake will get a value above or below 0.5, which will indicate its own handedness.

Considering this, however, it would still be useful to have associated flakes where multiple p-values could be assessed holistically to make an overall handedness assumption for the knapper. With predictive correctness of each model approaching- or over- 70% correct, we can predict knapper handedness of single flakes, clusters of flakes, or entire sites, with higher confidence than simple observation-based judgments.

In terms of data collection, it would be relatively easy to add some or all of the characteristic classifications to a standard lithic analysis, if the person was familiar with the literature. I have shown that unreliability is a large issue with many of the independent variable classes, so future works on paleo-archaeological assemblages would need to address this as well. First, I encourage a second evaluation of these regression formulas with other experimentally created flakes, for which handedness is known. If the p-values correlate relatively well with knapper handedness for other experimental assemblages, it would perhaps be wise to conduct the next few analyses of handedness using these formulas on Upper Paleolithic assemblages, where right-hand dominance has been repeatedly suggested, and thus p-values should be predominantly right (a majority of the flakes should have values below 0.5). If results of these analyses are congruent with right hand dominance, the method can be used on Middle- and eventually Lower-

95 Paleolithic assemblages, adding real data points to the evolutionary timeline of hominid handedness, before the time of definitive population-level right handedness in

Neanderthals and later Homo.

Cognitive Evolution and Language Acquisition

Centuries of research, triggered by advances in technology, has linked human handedness to the lateralization of the brain, and further associated fine motor movements involved in stone-tool production with neural areas responsible for language.

Still, more neuroscientific research on how handedness manifests in the brain, especially with regard to tool use and tool manufacture, could further support the use of handedness as a proxy for studying hominid cognitive evolution and language acquisition (Ruck

2014). The hominid fossil record is extremely limited, especially with respect to neurological and behavioral data, but fossil data still reflect the most direct source anthropologists have for understanding how human uniqueness evolved. The sample of natural brain endocasts is relatively small and direct evidence of language use is nonexistent before writing, so innovative research strategies, like the reviewed studies on flaked stone tools, are necessary. Paleoanthropologists interested in cognitive evolution need to direct their attention to the role of handedness in the fossil record, as it may be the best proxy for brain lateralization we have. Researchers interested in hominid handedness need to, in turn, focus more on lithic-based experimental approaches to inferring handedness, and tap in to the most naturally-preserved evidence our ancestors have left behind.

96 APPENDICES

97 Appendix A—Subset Handedness Inferences by Handaxe

Knapper N of N % N % CoP Handed- N IND % IND flakes Correct Correct Wrong Wrong ness A Right 62 2 3.23 54 87.10 6 9.68 B Right 16 1 6.25 12 75.00 3 18.75 C Right 89 3 3.37 83 93.26 3 3.37 D Left 100 7 7.00 84 84.00 9 9.00 E Right 64 11 17.19 42 65.63 11 17.19 F Right 11 3 27.27 5 45.45 3 27.27 G Left 62 9 14.52 48 77.42 5 8.06 H Left 57 7 12.28 44 77.19 6 10.53 I Left 92 17 18.48 54 58.70 21 22.83 J Left 78 12 15.38 47 60.26 19 24.36

Knapper N of N % N % ERA Handed- N IND % IND flakes Correct Correct Wrong Wrong ness A Right 62 5 8.06 53 85.48 4 6.45 B Right 16 5 31.25 11 68.75 0 0.00 C Right 89 6 6.74 80 89.89 3 3.37 D Left 100 13 13.00 87 87.00 0 0.00 E Right 64 4 6.25 53 82.81 7 10.94 F Right 11 2 18.18 4 36.36 5 45.45 G Left 62 5 8.06 53 85.48 4 6.45 H Left 57 2 3.51 51 89.47 4 7.02 I Left 92 8 8.70 74 80.43 10 10.87 J Left 78 12 15.38 61 78.21 5 6.41

Knapper N of N % N % Plat Handed- N IND % IND flakes Correct Correct Wrong Wrong ness A Right 62 4 6.45 50 80.65 8 12.90 B Right 16 2 12.5 14 87.50 0 0.00 C Right 89 7 7.86 71 79.78 11 12.36 D Left 100 14 14 77 77.00 9 9.00 E Right 64 15 23.43 39 60.94 10 15.63 F Right 11 4 36.36 7 63.64 0 0.00 G Left 62 10 16.12 46 74.19 6 9.68 H Left 57 7 12.28 50 87.72 0 0.00 I Left 92 11 11.95 68 73.91 13 14.13 J Left 78 17 21.79 56 71.79 5 6.41

98 Knapper N of N % N % CTX Handed- N IND % IND flakes Correct Correct Wrong Wrong ness A Right 62 0 0.00 57 91.94 5 8.06 B Right 16 2 12.50 13 81.25 1 6.25 C Right 89 7 7.87 79 88.76 3 3.37 D Left 100 4 4.00 84 84.00 12 12.00 E Right 64 9 14.06 51 79.69 4 6.25 F Right 11 1 9.09 6 54.55 4 36.36 G Left 62 8 12.90 49 79.03 5 8.06 H Left 57 6 10.53 47 82.46 4 7.02 I Left 92 10 10.87 72 78.26 10 10.87 J Left 78 9 11.54 58 74.36 11 14.10

Knapper N of N % N % Fract Handed- N IND % IND flakes Correct Correct Wrong Wrong ness A Right 62 10 16.13 44 70.97 8 12.90 B Right 16 8 50.00 8 50.00 0 0.00 C Right 89 18 20.22 61 68.54 10 11.24 D Left 100 15 15.00 74 74.00 11 11.00 E Right 64 10 15.63 48 75.00 6 9.38 F Right 11 1 9.09 9 81.82 1 9.09 G Left 62 13 20.97 46 74.19 3 4.84 H Left 57 1 1.75 50 87.72 6 10.53 I Left 92 9 9.78 77 83.70 6 6.52 J Left 78 11 14.10 61 78.21 6 7.69

99 Appendix B—Overall Frequency Data for Technical Features

Cone of Percussion (CoP) Frequency Percent Valid Center 311 49.3 Right 35 5.5 Left 40 6.3 None 65 10.3 Total 451 71.5 Missing 999 180 28.5 Total 631 100.0

Ridge Frequency Percent Valid Center 218 34.5 Positive (Right) 29 4.6 Negative (Left) 20 3.2 None 364 57.7 Total 631 100.0

Ripples Frequency Percent Valid Center 143 22.7 Right 66 10.5 Left 65 10.3 None 166 26.3 Total 440 69.7 Missing 999 191 30.3 Total 631 100.0

Extraction Axis Frequency Percent Valid Center 297 47.1 Right 70 11.1 Left 65 10.3 None 4 .8 Total 437 69.3 Missing 999 194 30.7 Total 631 100.0

100 Hackles Frequency Percent Valid Center 94 14.9 Right 70 11.1 Left 72 11.4 None 195 30.9 Total 431 68.3 Missing 999 200 31.7 Total 631 100.0

Eraillure Scar Location Present Absent % Present Center 61 570 9.7 Right 70 561 11.1 Left 46 585 7.3

Platform Type Frequency Percent Valid Linear 302 47.9 Punctiform 150 23.8 Total 452 71.6 Missing 999 179 28.4 Total 631 100.0

Platform Inclination Frequency Percent Valid Center 267 42.3 Right 93 14.7 Left 61 9.7 Sinuous 32 5.1 Total 453 71.8 Missing 999 178 28.2 Total 631 100.0

Impact Point Frequency Percent Valid Center 313 49.6 Right 79 12.5 Left 59 9.4 Total 451 71.5 Missing 999 180 28.5 Total 631 100.0

101 Cortex Frequency Percent Valid A 12 1.9 AB 16 2.5 AC 28 4.4 AD 2 .3 ABC 2 .3 ABD 3 .5 ACD 7 1.1 B 8 1.3 BC 3 .5 BD 30 4.8 BCD 10 1.6 C 41 6.5 CD 41 6.5 D 5 .8 ALL 71 11.3 NONE 247 39.1 Total 526 83.4 Missing 999 105 16.6 Total 631 100.0

Fracture Location Present Absent % Present A 10 621 1.6 B 27 604 4.3 C 24 607 3.8 D 15 616 2.4 E 11 620 1.7 F 25 606 4 G 27 604 4.3

102 Appendix C—Binary Logistic Regression Case Processing Summaries

Cone of Percussion Selected Cases N Percent Included in Analysis 411 65.1 Missing Cases 220 34.9 Total 631 100.0

Eraillure Scar Selected Cases N Percent Included in Analysis 631 100.0 Missing Cases 0 0.0 Total 631 100.0

Platform Selected Cases N Percent Included in Analysis 450 71.3 Missing Cases 181 28.7 Total 631 100.0

Cortex Selected Cases N Percent Included in Analysis 526 83.9 Missing Cases 105 16.6 Total 631 100.0

Fractures Selected Cases N Percent Included in Analysis 631 100.0 Missing Cases 0 0.0 Total 631 100.0

Saturated Model Selected Cases N Percent Included in Analysis 403 63.9 Missing Cases 228 36.1 Total 631 100.0 Backwards Model Selected Cases N Percent Included in Analysis 403 63.9 Missing Cases 228 36.1 Total 631 100.0

Unsaturated Model Selected Cases N Percent Included in Analysis 453 71.8 Missing Cases 178 28.2 Total 631 100.0

103 Appendix D—Technical Features Frequency Data by Handedness

Cone of Percussion Handedness Center Right Left None Total Right 120 11 17 36 184 65.22% 5.98% 9.24% 19.57% 100.00% Left 191 24 23 29 267 71.54% 8.99% 8.61% 10.86% 100.00% Total 311 35 40 65 451 68.96% 7.76% 8.87% 14.41% 100.00%

Hackles Handedness Center Right Left None Total Right 30 33 27 75 165 18.18% 20.00% 16.36% 45.45% 100.00% Left 64 37 45 120 266 24.06% 13.91% 16.92% 45.11% 100.00% Total 94 70 72 195 431 21.81% 16.24% 16.71% 45.24% 100.00%

Ripples Handedness Center Right Left None Total Right 52 29 30 61 172 30.23% 16.86% 17.44% 35.47% 100.00% Left 91 37 35 105 268 33.96% 13.81% 13.06% 39.18% 100.00% Total 143 66 65 166 440 32.50% 15.00% 14.77% 37.73% 100.00%

Extraction Axis Handedness Center Right Left None Total Right 108 28 31 2 169 63.91% 16.57% 18.34% 1.18% 100.00% Left 189 42 34 3 268 70.52% 15.67% 12.69% 1.12% 100.00% Total 297 70 65 5 437 67.96% 16.02% 14.87% 1.14% 100.00%

Ridge Handedness None Center Right Left Total Right 176 52 9 5 242 72.73% 21.49% 3.72% 2.07% 100.00% Left 188 166 20 15 389 48.33% 42.67% 5.14% 3.86% 100.00% Total 364 218 29 20 631 57.69% 34.55% 4.60% 3.17% 100.00%

104

Platform Type Handedness Linear Punctiform Total Right 117 68 185 63.24% 36.76% 100.00% Left 185 82 267 69.29% 30.71% 100.00% Total 302 150 452 66.81% 33.19% 100.00%

Platform Inclination Handedness Flat Right Left Sinuous Total Right 119 31 32 4 186 63.98% 16.67% 17.20% 2.15% 100.00% Left 148 62 29 28 267 55.43% 23.22% 10.86% 10.49% 100.00% Total 267 93 61 32 453 58.94% 20.53% 13.47% 7.06% 100.00%

Impact Point Handedness Center Right Left Total Right 130 31 24 185 70.27% 16.76% 12.97% 100.00% Left 183 48 35 266 68.80% 18.05% 13.16% 100.00% Total 313 79 59 451 69.40% 17.52% 13.08% 100.00%

Center Eraillure Scar Handedness Absent Present Total Right 215 27 242 88.84% 11.16% 100.00% Left 349 40 389 89.72% 10.28% 100.00% Total 564 67 631 89.38% 10.62% 100.00%

Right Eraillure Scar Handedness Absent Present Total Right 219 23 242 90.50% 9.50% 100.00% Left 347 42 389 89.20% 10.80% 100.00% Total 566 65 631 89.70% 10.30% 100.00%

105

Left Eraillure Scar Handedness Absent Present Total Right 230 12 242 95.04% 4.96% 100.00% Left 362 27 389 93.06% 6.94% 100.00% Total 592 39 631 93.82% 6.18% 100.00%

Cortex Handedness A AB AC AD ABC Right 2 5 12 1 1 0.93% 2.33% 5.58% 0.47% 0.47% Left 10 11 16 1 1 3.22% 3.54% 5.14% 0.32% 0.32% Total 12 16 28 2 2 2.28% 3.04% 5.32% 0.38% 0.38%

Cortex (Continued) Handedness ABD ACD B BC Right 1 2 3 1 0.47% 0.93% 1.40% 0.47% Left 2 5 5 2 0.64% 1.61% 1.61% 0.64% Total 3 7 8 3 0.57% 1.33% 1.52% 0.57%

Cortex (Continued) Handedness BD BCD C CD Right 13 5 15 15 6.05% 2.33% 6.98% 6.98% Left 17 5 26 26 5.47% 1.61% 8.36% 8.36% Total 30 10 41 41 5.70% 1.90% 7.79% 7.79%

Cortex (Continued) Handedness D ALL NONE Total Right 4 33 102 215 1.86% 15.35% 47.44% 100.00% Left 1 38 145 311 0.32% 12.22% 46.62% 100.00% Total 5 71 247 526 0.95% 13.50% 46.96% 100.00%

106

A Fracture B Fracture Handedness Absent Present Total Absent Present Total Right 237 5 242 232 10 242 97.93% 2.07% 100.00% 95.87% 4.13% 100.00% Left 384 5 389 372 17 389 98.71% 1.29% 100.00% 95.63% 4.37% 100.00% Total 621 10 631 604 27 631 98.42% 1.58% 100.00% 95.72% 4.28% 100.00%

C Fracture D Fracture Handedness Absent Present Total Absent Present Total Right 238 4 242 238 4 242 98.35% 1.65% 100.00% 98.35% 1.65% 100.00% Left 369 20 389 378 11 389 94.86% 5.14% 100.00% 97.17% 2.83% 100.00% Total 607 24 631 616 15 631 96.20% 3.80% 100.00% 97.62% 2.38% 100.00%

E Fracture F Fracture Handedness Absent Present Total Absent Present Total Right 237 5 242 235 7 242 97.93% 2.07% 100.00% 97.11% 2.89% 100.00% Left 383 6 389 371 18 389 98.46% 1.54% 100.00% 95.37% 4.63% 100.00% Total 620 11 631 606 25 631 98.26% 1.74% 100.00% 96.04% 3.96% 100.00%

G Fracture Handedness Absent Present Total Right 235 7 242 97.11% 2.89% 100.00% Left 369 20 389 94.86% 5.14% 100.00% Total 604 27 631 95.72% 4.28% 100.00%

107

Appendix E—Technical Features Frequency Data by Knapper

Cone of Percussion Knapper (Handedness) Center Right Left None Total EM (L) 122 18 13 20 173 70.5% 10.4% 7.5% 11.6% 100.0% MC (R) 55 5 9 8 77 71.4% 6.5% 11.7% 10.4% 100.0% OS (L) 22 3 6 4 35 62.9% 8.6% 17.1% 11.4% 100.0% RC (L) 47 3 4 5 59 79.7% 5.1% 6.8% 8.5% 100.0% SH (R) 65 6 8 28 107 60.7% 5.6% 7.5% 26.2% 100.0% Total 311 35 40 65 451 69.0% 7.8% 8.9% 14.4% 100.0%

Hackles Knapper (Handedness) Center Right Left None Total EM (L) 35 18 26 86 165 21.2% 10.9% 15.8% 52.1% 100.0% MC (R) 19 13 13 34 79 24.1% 16.5% 16.5% 43.0% 100.0% OS (L) 13 4 9 11 37 35.1% 10.8% 24.3% 29.7% 100.0% RC (L) 16 15 10 23 64 25.0% 23.4% 15.6% 35.9% 100.0% SH (R) 11 20 14 41 86 12.8% 23.3% 16.3% 47.7% 100.0% Total 94 70 72 195 431 21.8% 16.2% 16.7% 45.2% 100.0% Ripples Knapper (Handedness) Center Right Left None Total EM (L) 60 27 25 56 168 35.7% 16.1% 14.9% 33.3% 100.0% MC (R) 21 8 12 39 80 26.3% 10.0% 15.0% 48.8% 100.0% OS (L) 11 3 4 19 37 29.7% 8.1% 10.8% 51.4% 100.0% RC (L) 20 7 6 30 63 31.7% 11.1% 9.5% 47.6% 100.0% SH (R) 31 21 18 22 92 33.7% 22.8% 19.6% 23.9% 100.0% Total 143 66 65 166 440 32.5% 15.0% 14.8% 37.7% 100.0%

108

Extraction Axis Knapper (Handedness) Center Right Left None Total EM (L) 113 30 23 2 168 67.3% 17.9% 13.7% 1.2% 100.0% MC (R) 59 8 12 1 80 73.8% 10.0% 15.0% 1.3% 100.0% OS (L) 30 3 3 1 37 81.1% 8.1% 8.1% 2.7% 100.0% RC (L) 46 9 8 0 63 73.0% 14.3% 12.7% 0.0% 100.0% SH (R) 49 20 19 1 89 55.1% 22.5% 21.3% 1.1% 100.0% Total 297 70 65 5 437 68.0% 16.0% 14.9% 1.1% 100.0%

Ridge Knapper (Handedness) None Center Right Left Total EM (L) 131 101 11 6 249 52.6% 40.6% 4.4% 2.4% 100.0% MC (R) 39 42 7 3 91 42.9% 46.2% 7.7% 3.3% 100.0% OS (L) 29 23 5 5 62 46.8% 37.1% 8.1% 8.1% 100.0% RC (L) 28 42 4 4 78 35.9% 53.8% 5.1% 5.1% 100.0% SH (R) 137 10 2 2 151 90.7% 6.6% 1.3% 1.3% 100.0% Total 364 218 29 20 631 57.7% 34.5% 4.6% 3.2% 100.0%

Platform Type Knapper (Handedness) Linear Punctiform Total EM (L) 110 61 171 64.3% 35.7% 100.0% MC (R) 52 26 78 66.7% 33.3% 100.0% OS (L) 28 9 37 75.7% 24.3% 100.0% RC (L) 47 12 59 79.7% 20.3% 100.0% SH (R) 65 42 107 60.7% 39.3% 100.0% Total 302 150 452 66.8% 33.2% 100.0%

109

Platform Inclination Knapper (Handedness) Center Right Left Sinuous Total EM (L) 101 33 20 17 171 59.1% 19.3% 11.7% 9.9% 100.0% MC (R) 45 13 19 1 78 57.7% 16.7% 24.4% 1.3% 100.0% OS (L) 19 12 4 2 37 51.4% 32.4% 10.8% 5.4% 100.0% RC (L) 28 17 5 9 59 47.5% 28.8% 8.5% 15.3% 100.0% SH (R) 74 18 13 3 108 68.5% 16.7% 12.0% 2.8% 100.0% Total 267 93 61 32 453 58.9% 20.5% 13.5% 7.1% 100.0%

Impact Point Knapper (Handedness) Center Right Left Total EM (L) 121 31 19 171 70.8% 18.1% 11.1% 100.0% MC (R) 50 12 16 78 64.1% 15.4% 20.5% 100.0% OS (L) 27 2 8 37 73.0% 5.4% 21.6% 100.0% RC (L) 35 15 8 58 60.3% 25.9% 13.8% 100.0% SH (R) 80 19 8 107 74.8% 17.8% 7.5% 100.0% Total 313 79 59 451 69.4% 17.5% 13.1% 100.0%

Center Eraillure Scar Knapper Absent Present Total (Handedness) EM (L) 228 21 249 91.60% 8.40% 100.00% MC (R) 80 11 91 87.90% 12.10% 100.00% OS (L) 56 6 62 90.30% 9.70% 100.00% RC (L) 65 13 78 83.30% 16.70% 100.00% SH (R) 135 16 151 89.40% 10.60% 100.00% Total 564 67 631 89.40% 10.60% 100.00%

110

Right Eraillure Scar Knapper Absent Present Total (Handedness) EM (L) 228 21 249 91.60% 8.40% 100.00% MC (R) 77 14 91 84.60% 15.40% 100.00% OS (L) 56 6 62 90.30% 9.70% 100.00% RC (L) 63 15 78 80.80% 19.20% 100.00% SH (R) 142 9 151 94.00% 6.00% 100.00% Total 566 65 631 89.70% 10.30% 100.00%

Left Eraillure Scar Knapper Absent Present Total (Handedness) EM (L) 234 15 249 94.00% 6.00% 100.00% MC (R) 83 8 91 91.20% 8.80% 100.00% OS (L) 61 1 62 98.40% 1.60% 100.00% RC (L) 67 11 78 85.90% 14.10% 100.00% SH (R) 147 4 151 97.40% 2.60% 100.00% Total 592 39 631 93.80% 6.20% 100.00%

Cortex Knapper A AB AC AD ABC (Handedness) EM (L) 7 9 11 0 0 3.70% 4.70% 5.80% 0.00% 0.00% MC (R) 1 2 6 0 1 1.10% 2.30% 6.90% 0.00% 1.10% OS (L) 2 2 1 0 0 4.00% 4.00% 2.00% 0.00% 0.00% RC (L) 1 0 4 1 1 1.40% 0.00% 5.70% 1.40% 1.40% SH (R) 1 3 6 1 0 0.80% 2.30% 4.70% 0.80% 0.00% Total 12 16 28 2 2 2.30% 3.00% 5.30% 0.40% 0.40% 111

Cortex (Continued) Knapper BC ABD ACD B (Handedness) EM (L) 1 1 3 2 0.50% 0.50% 1.60% 1.00% MC (R) 0 0 0 2 0.00% 0.00% 0.00% 2.30% OS (L) 0 0 1 3 0.00% 0.00% 2.00% 6.00% RC (L) 1 1 1 0 1.40% 1.40% 1.40% 0.00% SH (R) 1 1 2 1 0.80% 0.80% 1.60% 0.80% Total 3 3 7 8 0.60% 0.60% 1.30% 1.50%

Cortex (Continued) Knapper BD BCD C CD (Handedness) EM (L) 11 3 14 17 5.80% 1.60% 7.30% 8.90% MC (R) 6 3 6 6 6.90% 3.40% 6.90% 6.90% OS (L) 5 0 2 2 10.00% 0.00% 4.00% 4.00% RC (L) 1 2 10 7 1.40% 2.90% 14.30% 10.00% SH (R) 7 2 9 9 5.50% 1.60% 7.00% 7.00% Total 30 10 41 41 5.70% 1.90% 7.80% 7.80%

Cortex (Continued) Knapper D ALL NONE Total (Handedness) EM (L) 0 26 86 191 0.00% 13.60% 45.00% 100.00% MC (R) 0 14 40 87 0.00% 16.10% 46.00% 100.00% OS (L) 1 2 29 50 2.00% 4.00% 58.00% 100.00% RC (L) 0 10 30 70 0.00% 14.30% 42.90% 100.00% SH (R) 4 19 62 128 3.10% 14.80% 48.40% 100.00% Total 5 71 247 526 1.00% 13.50% 47.00% 100.00%

112

A Fracture B Fracture Knapper Absent Present Total Absent Present Total (Handedness) EM (L) 248 1 249 238 11 249 99.60% 0.40% 100.00% 95.60% 4.40% 100.00% MC (R) 90 1 91 88 3 91 98.90% 1.10% 100.00% 96.70% 3.30% 100.00% OS (L) 62 0 62 61 1 62 100.00% 0.00% 100.00% 98.40% 1.60% 100.00% RC (L) 74 4 78 73 5 78 94.90% 5.10% 100.00% 93.60% 6.40% 100.00% SH (R) 147 4 151 144 7 151 97.40% 2.60% 100.00% 95.40% 4.60% 100.00% Total 621 10 631 604 27 631 98.40% 1.60% 100.00% 95.70% 4.30% 100.00%

C Fracture D Fracture Knapper Absent Present Total Absent Present Total (Handedness) EM (L) 238 11 249 244 5 249 95.60% 4.40% 100.00% 98.00% 2.00% 100.00% MC (R) 89 2 91 88 3 91 97.80% 2.20% 100.00% 96.70% 3.30% 100.00% OS (L) 56 6 62 60 2 62 90.30% 9.70% 100.00% 96.80% 3.20% 100.00% RC (L) 75 3 78 74 4 78 96.20% 3.80% 100.00% 94.90% 5.10% 100.00% SH (R) 149 2 151 150 1 151 98.70% 1.30% 100.00% 99.30% 0.70% 100.00% Total 607 24 631 616 15 631 96.20% 3.80% 100.00% 97.60% 2.40% 100.00%

113

E Fracture F Fracture Knapper Absent Present Total Absent Present Total (Handedness) EM (L) 239 10 249 244 5 249 96.00% 4.00% 100.00% 98.00% 2.00% 100.00% MC (R) 88 3 91 88 3 91 96.70% 3.30% 100.00% 96.70% 3.30% 100.00% OS (L) 56 6 62 62 0 62 90.30% 9.70% 100.00% 100.00% 0.00% 100.00% RC (L) 74 4 78 77 1 78 94.90% 5.10% 100.00% 98.70% 1.30% 100.00% SH (R) 147 4 151 149 2 151 97.40% 2.60% 100.00% 98.70% 1.30% 100.00% Total 604 27 631 620 11 631 95.70% 4.30% 100.00% 98.30% 1.70% 100.00%

G Fracture Knapper Absent Present Total (Handedness) EM (L) 239 10 249 96.00% 4.00% 100.00% MC (R) 84 7 91 92.30% 7.70% 100.00% OS (L) 58 4 62 93.50% 6.50% 100.00% RC (L) 74 4 78 94.90% 5.10% 100.00% SH (R) 151 0 151 100.00% 0.00% 100.00% Total 606 25 631 96.00% 4.00% 100.00%

114

Appendix F—Overall Frequency Data for Technical Features: Observer B

Cone of Percussion Frequency Percent Valid Center 8 16.0 Right 11 22.0 Left 21 42.0 None 9 18.0 Missing 999 1 2.0 Total 50 100.0

Hackles Frequency Percent Center 12 24.0 Right 19 38.0 Left 7 14.0 None 12 24.0 Total 50 100.0

Ripples Frequency Percent Center 7 14.0 Right 11 22.0 Left 24 48.0 None 8 16.0 Total 50 100.0

Extraction Axis Frequency Percent Center 9 18.0 Right 14 28.0 Left 26 52.0 None 1 2.0 Total 50 100.0

Ridge Frequency Percent Left 9 18.0 Right 41 82.0 Total 50 100.0

Eraillure Scar Location Present Absent % Present Center 3 47 6.0 Right 24 26 48.0 Left 5 45 10.0

115

Platform Type Frequency Percent Valid Linear 41 82.0 Punctiform 2 4.0 Missing 999 7 14.0 Total 50 100.0

Platform Inclination Frequency Percent Valid Center 17 34.0 Right 13 26.0 Left 13 26.0 Missing 999 7 14.0 Total 50 100.0

Impact Point Frequency Percent Valid Center 24 48.0 Right 11 22.0 Left 8 16.0 Missing 999 7 14.0 Total 50 100.0

Cortex Frequency Percent A 3 6.0 AC 1 2.0 ACD 3 6.0 B 2 4.0 BC 2 4.0 BD 2 4.0 BCD 1 2.0 C 4 8.0 CD 3 6.0 D 2 4.0 ALL 2 4.0 NONE 25 50.0 Total 50 100.0

116

Fracture Location Present Absent % Present A 1 49 2.0 B 2 48 4.0 C 7 43 14.0 D 10 40 20.0 E 8 42 16.0 F 10 40 20.0 G 10 40 20.0

117

Appendix G—Overall Frequency Data for Technical Features: Observer C

Cone of Percussion Frequency Percent Center 19 38.0 Right 13 26.0 Left 15 30.0 None 3 6.0 Total 50 100.0

Hackles Frequency Percent Center 7 14.0 Right 19 38.0 Left 19 38.0 None 5 10.0 Total 50 100.0

Ripples Frequency Percent Center 12 24.0 Right 18 36.0 Left 17 34.0 None 3 6.0 Total 50 100.0

Extraction Axis Frequency Percent Center 13 26.0 Right 15 30.0 Left 21 42.0 None 1 2.0 Total 50 100.0

Ridge Frequency Percent None 26 52.0 Left 9 18.0 Right 15 30.0 Total 50 100.0

118

Eraillure Scar Location Present Absent % Present Center 12 38 24.0 Right 7 43 14.0 Left 13 37 26.0

Platform Type Frequency Percent Valid Linear 42 84.0 Punctiform 6 12.0 Missing 999 2 4.0 Total 50 100.0

Platform Inclination Frequency Percent Valid Center 36 72.0 Right 4 8.0 Left 4 8.0 Sinuous 4 8.0 Missing 999 2 4.0 Total 50 100.0

Impact Point Frequency Percent Valid Center 25 50.0 Right 16 32.0 Left 7 14.0 Missing 999 2 4.0 Total 50 100.0

119

Cortex Frequency Percent A 1 2.0 AB 1 2.0 AC 1 2.0 AD 1 2.0 ABC 2 4.0 ACD 6 12.0 B 4 8.0 BD 2 4.0 BCD 3 6.0 C 2 4.0 CD 1 2.0 D 2 4.0 ALL 6 12.0 NONE 18 36.0 Total 50 100.0

Fracture Location Present Absent % Present A 1 49 2.0 B 3 47 6.0 C 8 42 16.0 D 6 44 12.0 E 2 48 4.0 F 6 44 12.0 G 5 45 10.0

120

Appendix H—IRB Approval

121

REFERENCES

Ambrose, S. 2001 Paleolithic Technology and Human Evolution. Science 291:1748–1753.

Aiello, L. C., and Dunbar, R. I. M. 1993 Neocortex Size, Group Size, and the Evolution of Language. Current Anthropology 34(2):184–193.

Amunts, K., Schlaug, G., Schleicher, A., Steinmets, H., Darbinghaus, A., Roland, P. E., and Zilles, K. 1996 Asymmetry in the Human Motor Cortex and Handedness. NeuroImage 4:216–222.

Andrefsky, W. 1998 Lithics: Macroscopic Approaches to Analysis. New York, NY: Cambridge University Press.

Annett, M. 1985 Left, Right, Hand and Brain: The Right Shift Theory. London, UK: Lawrence Erlbaum Associates Ltd.

1998 Stories about hands, brains, and minds. Brain and Language 65:356-358.

Balzeau, A., Gilissen, E., and Grimaud-Hervé, D. 2011 Shared pattern of endocranial shape asymmetries among great apes, anatomically modern humans, and fossil hominins. PloS one 7:e29581.

Bax, J. S., and Ungar, P. S. 1999 Incisor Labial Surface Wear Striations in Modern Humans and their Implications for Handedness in Middle and Late Pleistocene Hominids. International Journal of Osteoarchaeology 198:189–198.

Bargalló, A., and Mosquera, M. 2013 Can hand laterality be identified through lithic technology? Laterality 18:1- 27.

Ben-David, A. 2008 Comparison of classification accuracy using Cohen’s Weighted Kappa. Expert Systems with Applications 34(2): 825-832.

122

Bermúdez de Castro, J., Bromage, T. G., and Jalvo, Y. F. 1988 Buccal striations on fossil human anterior teeth: evidence of handedness in the middle and early Upper Pleistocene. Journal of Human Evolution 17:403–412.

Bishop, D. V. M., Watt, H., and Papadatou-Pastou, M. 2009 An efficient and reliable method for measuring cerebral lateralization during speech with functional transcranial Doppler ultrasound. Neuropsychologia, 47(2):587–90.

Bordes, F. 1968 The Old Stone Age. (trans J. E. Anderson). New York, NY: World University Library.

Bromage, T. G., and Boyde, A. 1984 Microscopic criteria for the determination of directionality of cutmarks on bone. American Journal of Physical Anthropology 65:359–66.

Bryden, M. P., Roy, E. A., McManus, I. C., and Bulman-Fleming, M. B. 1997 On the Genetics and Measurement of Human Handedness. In Hemispheric Specialisation in Animals and Humans: A Special Issue of Laterality (eds J. Fagot, L. Rogers, J. Ward, B. Bulman-Fleming, and W. D. Hopkins). pp. 247-266. East Sussex, UK: Psychology Press, Ltd.

Byrne, R. W. 2005 The Maker Not the Tool: the Cognitive Significance of Great Ape Manual Skills. In Stone Knapping: The Necessary Conditions for a Uniquely Hominin Behaviour (eds V. Roux and B. Bril). pp. 159-170. Cambridge, UK: McDonald Institute of Archeological Research.

Cantalupo, C., and Hopkins, W. D. 2008 Theoretical Speculations on the Evolutionary Origins of Hemispheric Specialization. Current Directions in Psychological Science 17: 233-237.

Cantalupo, C., Freeman, H., Rodes, W., and Hopkins, W. D. 2008 Handedness for tool use correlates with cerebellar asymmetries in chimpanzees (Pan troglodytes). Behavioral Neuroscience 122:191–198.

Cantalupo, C., Oliver, J., Smith, J., Nir, T., Taglialatela, J. P., and Hopkins, W. D. 2009 The chimpanzee brain shows human-like perisylvian asymmetries in white matter. The European Journal of Neuroscience 30:431–8.

Cashmore, L., Uomini, N., and Chapelain, A. 2008 The evolution of handedness in humans and great apes: a review and current issues. Journal of Anthropological Sciences 86:7–35.

123

Cashmore, L. 2009 Can hominin “handedness” be accurately assessed? Annals of Human Biology 36:624–41.

Çokluk, Ö. 2010 Logistic regression: Concept and application. Kuram Ve Uygulamada Egitim Bilimleri, 10(3): 1397-1407.

Coolidge, F. L., and Wynn, T. 2009 The Rise of Homo sapiens: The Evolution of Modern Thinking. West Sussex, UK: Wiley-Blackwell.

Corballis, M. C. 1983 Human Laterality. New York, NY: Academic Press.

1991 The Lopsided Ape: Evolution of the Generative Mind. Oxford, UK: Oxford University Press.

2002 Laterality and Human Speciation. In The Speciation of Modern Homo sapiens: Proceedings of the British Academy 106 (ed T. J. Crow). pp. 137-152. Oxford, UK: Oxford University Press.

2003 From mouth to hand: gesture, speech, and the evolution of right-handedness. Behavioral and Brain Sciences 26:199–260.

2007 Cerebral Asymmetry and Human Uniqueness. In The Evolution of Hemispheric Specialization in Primates (ed W. D. Hopkins). pp. 1-21. Norman, OK: Elsevier.

2009 The evolution and genetics of cerebral asymmetry. Philosophical transactions of the Royal Society of London. Series B, Biological sciences 364:867–79.

Cornford, J. M. 1986 Specialised resharpening techniques and evidence of handedness. In La Cotte de St. Brelade: Excavations by C.B.M. McBurney (eds P. Callow and J. M. Cornford). pp. 337-35. Cambridge, UK: Geo Books.

Cotterell, B., and Kamminga, J. 1979 The Mechanics of Flaking. In Lithic Use-wear Analysis (ed B. Hayden). pp. 97-112. New York, NY. Academic Press, Inc.

1990 Stone tools. In Mechanics of Preindustrial Technology (B. Cotterell and J. Kamminga). pp. 125-155. New York, NY. Cambridge University Press.

124

Crow, T. J. 2009 A theory of the origin of cerebral asymmetry: Epigenetic variation superimposed on a fixed right-shift. Laterality 15:289–303.

Crow, T. J., Close, J. P., Dagnall, A. M., and Priddle, T. H. 2009 Where and What is the right-shift Factor or Cerebral Dominance Gene? A critique of Francks et al. (2007). Laterality 14:3-10.

Davidson, I., and McGrew, W. C. 2005 Stone Tools and the Uniqueness of Human Culture. Journal of the Royal Anthropological Institute 11:793–817. de Beaune, S. A. 2004 The Invention of Technology: Prehistory and Cognition. Current Anthropology 45:139–162.

Fagot, J., and Vauclair, J. 1991 Manual laterality in nonhuman primates: a distinction between handedness and manual specialization. Psychological bulletin 109:76–89.

Falk, D. 1980 Hominid brain evolution: The approach from paleoneurology. American Journal of Physical Anthropology 23:93–107.

1987 Brain Lateralization in Primates and Its Evolution in Hominids. Yearbook of Physical Anthropology 30:107–125.

Faurie, C., and Raymond, M. 2004 Handedness frequency over more than ten thousand years. Proceedings of the Royal Society Series B: Biological Sciences 271:S43–S45.

Field, A. 2005 Discovering statistics using SPSS. London, UK: Sage Publications.

Fleiss, J. L., Cohen, J., and Everitt, B. S. 1969 Large sample standard errors of kappa and weighted kappa. Psychological Bulletin 72(5):323-327.

Fox, C. L., and Frayer, D. W. 1997 Non-dietary Marks in the Anterior Dentition of the Krapina Neanderthals. International Journal of Osteoarchaeology 7:133–149.

Gannon, P. J., Holloway, R. L., Broadfield, D. C., and Braun, A .R. 1998 Asymmetry of Chimpanzee Planum Temporale: Humanlike Pattern of Wernicke’s Brain Language Area Homolog. Science 279:220–222.

125

Geribàs, N., Mosquera, M., and Vergès, J. M. 2010 The gesture substratum of stone tool making: an experimental approach. Annali dell’Universita di Ferrara Museologia Scientifica e Naturalistica 6:155– 162.

Geschwind, N., and Galaburda, A. M. 1987 Cerebral Lateralization: Biological Mechanisms, Associations, and Pathologies. Cambridge, MA: MIT Press.

Guiard, Y. 1987 Asymmetric Division of Labor in Human Skilled Bimanual Action: The Kinematic Chain as a Model. Journal of Motor Behavior 19:1–23.

Harrell, F. E. 2001 Regression Modelling Strategies: With Applications to Linear Models, Logistic Regression, and Survival Analysis. New York, NY: Springer.

Hecht, E. E., Gutman, D. A., Khreisheh, N. N., Taylor, S. V., Kilner, J., Faisal, A. A., Bradley, B.A., Chaminade, T., Stout, D. 2014 Acquisition of Palaeolithic tool making abilities involves structural remodelling to inferior frontoparietal regions. Brain Structure and Function. doi: 10.1007/s00429- 014-0789-6

Hewes, G. H. 1993 A History of Speculation on the Relation between Tools and Language. In Tools, Language and Cognition in Human Evolution (eds K. R. Gibson and T. Ingold). pp. 20-32. New York, NY: Cambridge University Press.

Hilbe, J. M. 2009 Logistic regression models. Boca Raton, FL: CRC Press.

Holloway, R. L. 2008 The Human Brain Evolving: a Personal Retrospective. In The Human Brain Evolving: Paleoneurological Studies in Honor of Ralph L. Holloway (eds D. Broadfield, M. Yuan, K. D. Schick, and N. Toth). pp. 1-14. Gosport, IN: Stone Age Institute Press.

Holloway, R. L., Broadfield, D. C., and Yuan, M. S. 2003 Morphology and histology of chimpanzee primary visual striate cortex indicate that brain reorganization predated brain expansion in early hominid evolution. The Anatomical Record. Part A, Discoveries in molecular, cellular, and evolutionary biology 273:594–602.

126

Hopkins, W. D., Phillips, K. A., Bania, A., Calcutt, S. E., Russell, J., Schaeffer, J., Lonsdorf, E. V., Ross, S. R., and Schapiro, S. J. 2011 Hand Preferences for Coordinated Bimanual Actions in 777 Great Apes: Implications for the Evolution of Handedness in Hominins. Journal of Human Evolution 60:605–611.

Hopkins, W. D., and Rilling, J. K. 2000 A comparative MRI study of the relationship between neuroanatomical asymmetry and interhemispheric connectivity in primates: Implication for the evolution of functional asymmetries. Behavioral Neuroscience 114:739–748.

Hopkins, W. D., Russell, J. L., and Cantalupo, C. 2007 Neuroanatomical Correlates of Handedness for Tool Use in Chimpanzees (Pan troglodytes). Psychological Science 18:971–977.

Hopkins, W. D., Russell, J. L., Lambeth, S., and Schapiro, S. J. 2007 Handedness and Neuroanatomical Asymmetries in Captive Chimpanzees: A Summary of 15 Years of Research. In The Evolution of Hemispheric Specialization in Primates (ed W. D. Hopkins). pp. 147-181. Norman, OK: Elsevier.

Hopkinson, T., and White, M. J. 2005 The Acheulean and the Handaxe: Structure and Agency in the Paleolithic. In The Hominid Individual in Context: Archaeological Investigations of Lower and Middle Paleolithic Landscapes, Locales, and Artefacts (eds C. Gamble and M. Porr). pp. 13-28. New York, NY: Routledge.

Hosmer, D. W., Lemeshow, S, and Sturdivant, R. X. 2013 Applied Logistic Regression, 3rd Edtion. New York: Wiley.

Keeley, L. H. 1980 Experimental Determination of Stone Tool Uses: A Microwear Analysis. Chicago, IL: The University of Chicago Press.

Kochetkova, V. 1978 Paleoneurology. (trans. H. J. Jerison). Washington, DC: V. H. Winston and Sons.

Khreisheh, N., Davies, D. and Bradley, B. 2013 Extending Experimental Control: The Use of Porcelain in Flaked Stone Experimentation. Advances in Archaeological Practice: A Journal of the Society for American Archaeology 1(1):37–46.

Lazenby, R. A. 2002 Skeletal Biology, Functional Asymmetry and the Origins of “Handedness.” Journal of Theoretical Biology 218:129–138.

127

Liguria, G. 1997 A Case of Marked Bilateral Asymmetry in the Upper Limbs of an Upper Paleolithic Male from Barma. International Journal of Osteoarchaeology 7:18–38.

Lozano, M., Mosquera, M., Bermúdez de Castro, J. M., Arsuaga, J. L., and Carbonell, E. 2009 Right handedness of Homo heidelbergensis from Sima de los Huesos (Atapuerca, Spain) 500,000 years ago. Evolution and Human Behavior 30:369– 376.

MacNeilage, P. F. 2005 Evolution of Whole-Body Asymmetry Related to Handedness. Cortex 42:94-95.

Marshack, A. 1976 Some Implications of the Paleolithic Symbolic Evidence for the Origin of Language. Current Anthropology 17:274–282.

Matsuzawa, T. 2001 Primate Foundations of Human Intelligence: A View of Tool Use in Nonhuman Primates and Fossil Hominids. In Primate Origins of Human Cognition and Behavior (ed T. Matsuzawa). pp. 3-25. Hong Kong, Japan: Springer-Verlag.

McGrew, W. C., and Marchant, L. F. 1996 On which side of the Apes? Ethological Study of Laterality of Hand Use. In Great Ape Societies (eds W. C. McGrew, L. F. Marchant, and T. Nishida). pp. 255-272. New York, NY: Cambridge University Press.

1997 On the Other Hand: Current Issues in and Meta-Analysis of the Behavioral Laterality of Hand Function in Nonhuman Primates. Yearbook of Physical Anthropology 40:201–232.

McManus, I. C. 1985 Right- and left-hand skill: Failure of the right shift model. British Journal of Psychology 76:1–16.

McPherron, S. P. 2000 Handaxes as a Measure of the Mental Capabilities of Early Hominids. Journal of Archaeological Science 27(8): 655–663.

Menard, S. 2002 Applied logistic regression analysis. Thousand Oaks, CA: Sage Publications.

128

Moore, M. W. 2011 The design space of stone flaking: implications for cognitive evolution. World Archaeology 43:702–715.

Neter, J., Wasserman, W., Nachtsheim, C. J., and Kutner, M. H. 1996 Applied Linear Regression Models, 3rd Edition. Chicago: Irwin.

Nichols, T. R., Wisner, P. M., Cripe, G., and Gulabchand, L. 2011 Putting the Kappa Statistic to Use. The Quality Assurance Journal 13(3):57- 61.

Noble, W., and Davidson, I. 1996 Human Evolution, Language and Mind: A Psychological and Archaeological Inquiry. New York, NY: Cambridge University Press.

Noll, M. P., and Petraglia, M. D. 2003 Acheulean Bifaces and Early Human Behavioral Patterns in East Africa and South India. In Multiple Approaches to the Study of Bifacial Technologies (eds M. Soressi and H. L. Dibble). pp. 31-54. Philadelphia, PA: The University of Pennsylvania.

Nowell, A., and White, M. 2010 Growing Up in the Middle Pleistocene: Life History Strategies and Their Relationship to Acheulean Industries. In Stone Tools and the Evolution of Human Cognition (eds A. Nowell and I. Davidson), pp. 67-81. Boulder, CO: University Press of Colorado.

Odell, G. H. 2001 Stone Tool Research at the End of the Millennium: Classification, Function, and Behavior. Journal of Archaeological Research 9:45–100.

Patterson, K., and Sollberger, J. P. 1986 Comments on Toth’s right-handedness study. Lithic Technology 15:109- 111.

Pelegrin, J. 2009 Cognition and the Emergence of Language: A Contribution from Lithic Technology. In Cognitive Archaeology and Human Evolution (eds S. A. de Beaune, F. L. Coolidge, and T. Wynn). pp. 95-108. New York, NY: Cambridge University Press.

Phillips, K. A., and Hopkins, W. D. 2007 Exploring the relationship between cerebellar asymmetry and handedness in chimpanzees (Pan troglodytes) and capuchins (Cebus apella). Neuropsychologia 45:2333–9.

129

Phillips, K. A., and Thompson, C. R. 2012 Hand Preference for Tool-Use in Capuchin Monkeys (Cebus apella) is Associated with Asymmetry of the Primary Motor Cortex. American Journal of Primatology 440:435–440.

Pickering, T. R., and Hensley-Marschand, B. 2008 Cutmarks and hominid handedness. Journal of Archaeological Science 35:310–315.

Pobiner, B. 1999 The Use of Stone Tools to Determine Handedness in Hominids. Current Anthropology 40:90–92.

Provins, K. A. 1997 Handedness and speech: a critical reappraisal of the role of genetic and environmental factors in the cerebral lateralization of function. Psychological Review 104:554–71.

Reynolds, P. C. 1993 The Complementation Theory of Language and Tool Use. In Tools, Language and Cognition in Human Evolution (eds K. R. Gibson and T. Ingold). pp. 407-428. New York, NY: Cambridge University Press.

Rilling, J. K. 2006 Human and Non-Human Primate Brains: Are They Allometrically Scaled Versions of the Same Design? Evolutionary Anthropology 77:65–77.

2008 Neuroscientific approaches and applications within anthropology. American Journal of Physical Anthropology 47:2–32.

Ruck, L. 2014 Manual Praxis and Stone Tool Manufacture: Implications for Language Evolution. Brain and Language 139C: 68-83. doi: 10.1016/j.bandl.2014.10.003.

Rugg, G., and Mullane, M. 2001 Inferring handedness from lithic evidence. Laterality 6:247–259.

Sarkar, S. K., and Midi, H. 2010 Importance of assessing the model adequacy of binary logistic regression. Journal of Applied Sciences 10(6): 479-486.

Semenov, S. A. 1964 Prehistoric Technology: An Experimental Study of the Oldest Tools and Artefacts from Traces of Manufacture and Wear. (trans M. W. Thompson). London, UK: Cory, Adams, and Mackay Ltd.

130

Schick, K. D., and Clark, J. D. 2003 Biface Technological Development and Variability in the Acheulean Industrial Complex in the Middle Awash Region of the Afar Rift, Ethiopia. In Multiple Approaches to the Study of Bifacial Technologies (eds M. Soressi and H. L. Dibble). pp. 1-30. Philadelphia, PA: The University of Pennsylvania.

Schick, K. D., and Toth, N. 2006 Making Silent Stones Speak: Human Evolution and the Dawn of Technology. New York, NY: Simon and Schuster.

Shaw, C. N. 2011 Is “hand preference” coded in the hominin skeleton? An in-vivo study of bilateral morphological variation. Journal of Human Evolution 61:480–7.

Sherwood, C. C., Broadfield, D. C., Holloway, R. L., Gannon, P. J., Hof, P. R. 2003 Variability of Broca’s area homologue in African great apes: implications for language evolution. The Anatomical Record. Part A, Discoveries in molecular, cellular, and evolutionary biology 271:276–85.

Springer, S. P., and Deutsch, G. 1981 Left Brain, Right Brain. New York, NY: W. H. Freeman and Company.

Steele, J. 2000 Handedness in past human populations: Skeletal markers. Laterality 5:193– 220.

Steele, J., and Uomini, N. T. 2005 Humans, Tools, and Handedness. In Stone Knapping: The Necessary Conditions for a Uniquely Hominin Behaviour (eds V. Roux and B. Bril). pp. 217-255. Cambridge, UK: McDonald Institute of Archeological Research.

2009 Can the Archaeology of Manual Specialization Tell Us Anything About Language Evolution? A Survey of the State of Play. Cambridge Archaeological Journal 19:97.

Stout, D. 2005 Neural Foundations of Perception and Action in Stone Knapping. In Stone Knapping: The Necessary Conditions for a Uniquely Hominin Behaviour (eds V. Roux and B. Bril). pp. 273-286. Cambridge, UK: McDonald Institute of Archeological Research.

Stout, D., and Chaminade, T. 2012 Stone tools, language and the brain in human evolution. Philosophical Transactions of the Royal Society of London—Series B: Biological sciences 367:75–87.

131

Stout, D., and Semaw, S. 2006 Knapping Skill of the Earliest Stone Toolmakers: Insights from the Study of Modern Human Novices. In The Oldowan: Case Studies into the Earliest Stone Age (eds N. Toth and K. D. Schick). pp. 267-306. Gosport, IN: Stone Age Institute Press.

Stout, D., Toth, N., and Schick, K. D. 2006 Comparing the Neural Foundations of Oldowan and Acheulean Toolmaking: A Pilot Study Using Positron Emission Tomography (PET). In The Oldowan: Case Studies into the Earliest Stone Age (eds N. Toth and K. D. Schick). pp. 267-306. Gosport, IN: Stone Age Institute Press.

2009 Understanding Oldowan Knapping Skill: An Experimental Study of Skill Acquisition in Modern Humans. In The Cutting Edge: New Approaches in the Archaeology of Human Evolution (eds N. Toth and K. D. Schick). pp. 247-266. Gosport, IN: Stone Age Institute Press.

Stout, D., Toth, N., Schick, K. D., and Chaminade, T. 2008 Neural correlates of Early Stone Age toolmaking: technology, language and cognition in human evolution. Philosophical Transactions of the Royal Society of London. Series B, Biological sciences 363:1939–49.

Stout, D., Toth, N., Schick, K. D., Stout, J., and Hutchins, G. 2000 Stone Tool-Making and Brain Activation: Position Emission Tomography (PET) Studies. Journal of Archaeological Science 27:1215–1223.

Tjur, T. 2012 Coefficients of Determination in Logistic Regression Models—A New Proposal: The Coefficient of Discrimination. The American Statistician: 63(4):366-372.

Toth, N. 1985 Archaeological Evidence for Preferential Right-handedness in the Lower and Middle Pleistocene, and Its Possible Implications. Journal of Human Evolution 14:607–614.

Toth, N., and Schick, K. D. 1986 The First Million Years : The Archaeology of Protohuman Culture. Advances in Archaeological Method and Theory 9:1–96.

1993 Early Stone Industries and Inferences Regarding Language and Cognition. In Tools, Language and Cognition in Human Evolution (eds K. R. Gibson and T. Ingold). pp. 346-362. New York, NY: Cambridge University Press.

132

Trinkaus, E., Churchill, S. E., and Ruff, C. B. 1994 Postcranial robusticity in Homo. II: Humeral bilateral asymmetry and bone plasticity. American Journal of Physical Anthropology 93(1):1-34.

Ubelaker, D. H., and Zarenko, L. M. 2012 Can handedness be determined from skeletal remains? A chronological review of the literature. Journal of Forensic Sciences 57:1421–6.

Uomini, N. T. 2001 Lithic indicators of handedness: assessment of methodologies and the evolution of laterality in hominids. Unpublished M.Sc. dissertation, University of Durham, UK.

2006 In the knapper's hands: testing markers of laterality in hominin lithic production, with reference to the common substrate of language and handedness. Unpublished Ph.D. thesis, University of Southampton.

2009 Prehistoric Handedness and Prehistoric Language. In Cognitive Archaeology and Human Evolution (eds S. A. de Beaune, F. L. Coolidge, and T. Wynn). pp. 37-56. New York, NY: Cambridge University Press.

2011 Handedness in Neanderthals. In: Neanderthal Lifeways, Subsistence and Technology (eds N. J. Conard and J. Richter). pp. 139-158. Heidelberg, Germany: Springer.

Uomini, N. T. and Meyer, G. F. 2013 Shared Brain Lateralization Patterns in Language and Acheulean Stone Tool Production: A Functional Transcranial Doppler Ultrasound Study. PloS one 8(8): e72693.

Vallortigara, G., and Rogers, L. J. 2005 Survival with an asymmetrical brain: advantages and disadvantages of cerebral lateralization. Behavioral and Brain Sciences 28:575–633. van der Heijden, H. 2012 Decision support for selecting optimal logistic regression models. Expert Systems with Applications 39(10): 8573-8583.

Vandermeersch, B., and Trinkaus, E. 1995 The Postcranial Remains of the Regourdou 1 Neandertal: The Shoulder and Arm Remains. Journal of Human Evolution 28: 439–476.

133

Volpato, V., Macchiarelli, R., Guatelli-Steinberg, D., Fiore, I., Bondoli, L., and Frayer, D. W. 2012 Hand to mouth in a Neandertal: right-handedness in Regourdou 1. PloS one 7:e43949.

Warrens, M. J. 2011 Cohen’s kappa is a weighted average. Statistical Methodology 8(6):473-484.

Westergaard, G. C., and Suomi, S. J. 1996 Hand preference for stone artefact production and tool-use by monkeys: possible implications for the evolution of right-handedness in hominids. Journal of Human Evolution 30:291–298.

Whittaker, J. C. 2004 American Flintknappers: Stone Age Art in the Age of Computers. Austin, TX: The University of Texas Press.

Wynn, T. 2002 Archaeology and cognitive evolution. Behavioral and Brain Sciences 25:389–438.

134