<%BANNER%>

The Statistical Determination of Ancestry Using Cranial Nonmetric Traits

Permanent Link: http://ufdc.ufl.edu/UFE0021200/00001

Material Information

Title: The Statistical Determination of Ancestry Using Cranial Nonmetric Traits
Physical Description: 1 online resource (126 p.)
Language: english
Creator: Hefner, Joseph Timothy
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2007

Subjects

Subjects / Keywords: ancestry, anthropology, forensic, nonmetric, quantitative
Anthropology -- Dissertations, Academic -- UF
Genre: Anthropology thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: Anthropologists have recognized the value of nonmetric cranial traits for the study of human variation for over three hundred years. A historical reliance on nonmetric cranial traits may explain why most practicing forensic anthropologists rely directly or indirectly on an anthroposcopic approach to skeletal analysis. Although the historical organization of humans into groups has been a central tenet in anthropological studies, the roots of the nonmetric approach are deeply imbedded in an early typological approach to race determination. Two anthropologists, Franz Boas and Earnest Hooton, changed the way anthropologists think about race in America, but in two very different ways. The Boasian approach did not support the prevailing paradigm and the belief in human races, but rather emphasized the study of human variation. The Hootonian approach, however, was rooted in the polygenist idea that races did indeed exist and that race could be defined by cranial measurements and nonmetric cranial traits. To that end, Hooton produced a list of nonmetric cranial traits he used to define each race. This list, now known as the Harvard Blanks, remains in use today in several manifestations, yet empirical support for the traits he selected has never been produced. Moreover, proper statistical methods for this type of data have never been explored. The nonmetric traits used to predict ancestry have historically relied on the experience of the observer and the typological trait lists from Hooton, which are both without empirical support. Assessing ancestry in this manner results in a priori conclusions drawn from some overall impression of the cranial Gestalt. This, the current method of nonmetric, macromorphoscopic trait analysis, is used almost exclusively today by forensic anthropologists, but the method has changed very little from Hooton?s method. This historical approach is ineffective and does not meet the guidelines established by the Daubert ruling. Previous research has demonstrated that the cranial nonmetric traits previously associated with ancestral groups are not found in frequencies high enough to permit visual observation alone. In other words, the extreme trait expressions for these nonmetric traits are not very reliable for estimating ancestry. The purpose of this research is to test the application of classification statistics to nonmetric traits for discriminating between populations frequently encountered during forensic investigations. Using a statistical framework, which provides the certainty, replicability, and reliability needed to meet the Daubert challenge, removes the subjectivity inherent in the human judgment process. In fact, all of the statistical methods used worked better than the traditional approach to ancestry determination?which is little more than visual estimation and guesswork. Each of the statistical methods correctly classified more than 83% of the sample, and several nearly 90%. The OSSA statistic and the CAP method show the greatest promise for forensic anthropological case work, correctly classified 83% and 86% of the entire sample correctly. The current approach to ancestry determination using nonmetric cranial traits is typological and relies too heavily on the human judgment process. This is an uncertain method, but this research will demonstrate that the uncertainty inherent in this method can be understood and explained using several statistical methods and by acknowledging the variation observed among populations. Recognizing the shortcomings of the current method and applying new approaches is both scientific and effective and should serve as a basis for the replacement of the traditional approach: classifications based on general impressions from the cranial Gestalt with little more than post hoc validation of the nonmetric traits.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Joseph Timothy Hefner.
Thesis: Thesis (Ph.D.)--University of Florida, 2007.
Local: Adviser: Warren, Michael W.
Electronic Access: RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2009-08-31

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2007
System ID: UFE0021200:00001

Permanent Link: http://ufdc.ufl.edu/UFE0021200/00001

Material Information

Title: The Statistical Determination of Ancestry Using Cranial Nonmetric Traits
Physical Description: 1 online resource (126 p.)
Language: english
Creator: Hefner, Joseph Timothy
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2007

Subjects

Subjects / Keywords: ancestry, anthropology, forensic, nonmetric, quantitative
Anthropology -- Dissertations, Academic -- UF
Genre: Anthropology thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: Anthropologists have recognized the value of nonmetric cranial traits for the study of human variation for over three hundred years. A historical reliance on nonmetric cranial traits may explain why most practicing forensic anthropologists rely directly or indirectly on an anthroposcopic approach to skeletal analysis. Although the historical organization of humans into groups has been a central tenet in anthropological studies, the roots of the nonmetric approach are deeply imbedded in an early typological approach to race determination. Two anthropologists, Franz Boas and Earnest Hooton, changed the way anthropologists think about race in America, but in two very different ways. The Boasian approach did not support the prevailing paradigm and the belief in human races, but rather emphasized the study of human variation. The Hootonian approach, however, was rooted in the polygenist idea that races did indeed exist and that race could be defined by cranial measurements and nonmetric cranial traits. To that end, Hooton produced a list of nonmetric cranial traits he used to define each race. This list, now known as the Harvard Blanks, remains in use today in several manifestations, yet empirical support for the traits he selected has never been produced. Moreover, proper statistical methods for this type of data have never been explored. The nonmetric traits used to predict ancestry have historically relied on the experience of the observer and the typological trait lists from Hooton, which are both without empirical support. Assessing ancestry in this manner results in a priori conclusions drawn from some overall impression of the cranial Gestalt. This, the current method of nonmetric, macromorphoscopic trait analysis, is used almost exclusively today by forensic anthropologists, but the method has changed very little from Hooton?s method. This historical approach is ineffective and does not meet the guidelines established by the Daubert ruling. Previous research has demonstrated that the cranial nonmetric traits previously associated with ancestral groups are not found in frequencies high enough to permit visual observation alone. In other words, the extreme trait expressions for these nonmetric traits are not very reliable for estimating ancestry. The purpose of this research is to test the application of classification statistics to nonmetric traits for discriminating between populations frequently encountered during forensic investigations. Using a statistical framework, which provides the certainty, replicability, and reliability needed to meet the Daubert challenge, removes the subjectivity inherent in the human judgment process. In fact, all of the statistical methods used worked better than the traditional approach to ancestry determination?which is little more than visual estimation and guesswork. Each of the statistical methods correctly classified more than 83% of the sample, and several nearly 90%. The OSSA statistic and the CAP method show the greatest promise for forensic anthropological case work, correctly classified 83% and 86% of the entire sample correctly. The current approach to ancestry determination using nonmetric cranial traits is typological and relies too heavily on the human judgment process. This is an uncertain method, but this research will demonstrate that the uncertainty inherent in this method can be understood and explained using several statistical methods and by acknowledging the variation observed among populations. Recognizing the shortcomings of the current method and applying new approaches is both scientific and effective and should serve as a basis for the replacement of the traditional approach: classifications based on general impressions from the cranial Gestalt with little more than post hoc validation of the nonmetric traits.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Joseph Timothy Hefner.
Thesis: Thesis (Ph.D.)--University of Florida, 2007.
Local: Adviser: Warren, Michael W.
Electronic Access: RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2009-08-31

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2007
System ID: UFE0021200:00001


This item has the following downloads:


Full Text

PAGE 1

THE STATISTICAL DETERMINATION OF ANCESTRY USING CRANIAL NONMETRIC TRAITS By JOSEPH T. HEFNER A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2007

PAGE 2

2007 Joseph T. Hefner 2

PAGE 3

To my inamorata, Natalie Marie Uhl, for her love, support, and encouragementI could not have done it without you 3

PAGE 4

ACKNOWLEDGMENTS I would like to acknowledge the support and guidance of my committee members. Michael Warren served as the chair for my Masters degree and my doctoral work. His relentless efforts to make me a better anthropologist will never be forgotten. Anthony Falsetti, David Daegling, John Krigbaum, and Thomas Hollinger continually gave me sound advice, constructive criticism, and hours of their own research time to assist me in mineI am forever grateful for their efforts. Finally, Stephen Ousley taught me outside of the classroom in so many ways that I am sure I will never be able to thank him enough. From one contrarian to the another, thank you Steve. This dissertation and my life would not be the same had it not been for our chance encounter at a baseball game. Dennis Dirkmaat of Mercyhurst College deserves special thanks for his tireless support and guidance. As a friend and mentor he has been invaluablewithout his creativity and insight, I would not be where I am today. I am also grateful for David Hunt, whose permission and assistance in obtaining skeletal material from the Smithsonian Institution has made this all possible. Lee Meadows Jantz and Richard Jantz also deserve special acknowledgment for their help with the William M. Bass Skeletal Collection and for answering my many questions, even when I was not always sure what I was asking. For their generous giving of time, ideas, and support, I would like to thank John Byrd, Paul Emanovsky, Laurel Freas, Nicholas Herrmann, Stephen Nawrocki, Nicholas Passalaqua, M. Kate Spradley, and Carlos Zambrano. For their participation in the inter-observer error study, I thank Thomas Gore, Anthony Koehl, and several anonymous participants. I would like to thank my family for instilling in me a love of learning, the need for creativity, and the belief in the power of words. My father suggested a stint in college nearly 12 years ago; I hope that I have made him proud with this accomplishment. My mother also 4

PAGE 5

deserves recognition. Like her, I saw a goal that would forever change my life and I reached out, took hold, and have never let go. My sister Nicole and my brother David taught me a lot when we were growing up, but more importantly, I learn from them even today. I want to thank you both for your love, your friendship, and your support. Finally, I would like to acknowledge and thank Natalie Uhl, for the hours she and I spent discussing our research, editing each others work, and just having fun together. Everyday Natalie reminds me of the things that are important in life. 5

PAGE 6

TABLE OF CONTENTS page ACKNOWLEDGMENTS ...............................................................................................................4 LIST OF TABLES ...........................................................................................................................8 LIST OF FIGURES .......................................................................................................................10 ABSTRACT ...................................................................................................................................12 CHAPTER 1 RACE AND PHYSICAL ANTHROPOLOGY.....................................................................15 Anatomy, Systematics, and the Search for Order...................................................................18 Typology, Polygenism, and the American School.................................................................23 2 THE FORENSIC APPROACH TO ANCESTRY DETERMINATION...............................34 3 NONMETRIC CRANIAL VARIATION...............................................................................44 Factors Affecting Trait Expression.........................................................................................45 Genes and the Environment.............................................................................................45 Sex, Age, and Trait Correlation.......................................................................................48 General Patterns of Variation.................................................................................................49 4 MATERIALS AND METHODS...........................................................................................52 Samples...................................................................................................................................52 African Sample................................................................................................................52 European Sample.............................................................................................................52 East Asian Sample...........................................................................................................52 Native American Sample.................................................................................................53 Anatomical Collection.....................................................................................................53 Bass Collection................................................................................................................53 Macromorphoscopic Traits.....................................................................................................54 Statistical Methods..................................................................................................................54 Correlations.....................................................................................................................55 Ordinal Regression..........................................................................................................55 Correspondence Analysis................................................................................................57 OSSA...............................................................................................................................58 Nave Bayesian................................................................................................................60 Linear Discriminant Function Analysis...........................................................................60 k-Nearest Neighbor.........................................................................................................62 Logistic Regression.........................................................................................................62 Canonical Analysis of Principal Coordinates..................................................................63 6

PAGE 7

Interand Intra-Observer Error.......................................................................................65 5 THE STATISTICAL DETERMINATION OF ANCESTRY................................................68 Statistical Approaches to Ancestry Determination.................................................................68 Ordinal Regression..........................................................................................................68 Correspondence Analysis................................................................................................69 Optimized Summed Score Attributes (OSSA)................................................................70 Nave Bayesian................................................................................................................71 Discriminant Function Analysis......................................................................................71 k-Nearest Neighbor..........................................................................................................73 Logistic Regression.........................................................................................................74 Canonical Analysis of the Principal Coordinates............................................................75 Interand Intra-Observer Error...............................................................................................76 6 CONCLUDING REMARKS..................................................................................................94 A Legacy and A Solution.......................................................................................................94 APPENDIX A MACROMORPHOSCOPIC TRAITS....................................................................................99 Anterior Nasal Spine (ANS)............................................................................................99 Inferior Nasal Morphology (INA)...................................................................................99 Interorbital Breadth (IOB).............................................................................................100 Malar Tubercle (MT).....................................................................................................100 Nasal Aperture Shape (NAS)........................................................................................101 Nasal Aperture Width (NAW).......................................................................................101 Nasal Bone Contour (NBC)...........................................................................................101 Nasal Bone Shape (NBS)..............................................................................................102 Nasal Overgrowth (NO)................................................................................................102 Post-bregmatic Depression (PBD)................................................................................103 Posterior Zygomatic Tubercle (ZT)...............................................................................103 Supranasal Suture (SPS)................................................................................................103 Transverse Palatine Suture Shape (TPS).......................................................................103 Zygomaticomaxillary Suture Shape (ZS)......................................................................104 B R CODE................................................................................................................................113 Correspondence Analysis..............................................................................................113 Nave Bayesian..............................................................................................................114 k-Nearest Neighbor.......................................................................................................114 LIST OF REFERENCES.............................................................................................................116 BIOGRAPHICAL SKETCH.......................................................................................................126 7

PAGE 8

LIST OF TABLES Table page 4-1 Frequency showing original scores and compressed OSAA scores.................................59 5-1 Parameter estimates and significance levels for Interorbital Breadth................................79 5-2 Classification matrix for the ORA two-group analysis......................................................79 5-3 Classification matrix for the ORA three-group analysis....................................................79 5-4 Frequency distribution of INA for American Blacks and Whites.....................................79 5-5 Values for Inferior Nasal Aperture derived from correspondence analysis......................79 5-6 Values for Anterior Nasal Spine derived from correspondence analysis..........................80 5-7 Values for Interorbital Breadth derived from correspondence analysis............................80 5-8 Values for Malar Tubercle derived from correspondence analysis...................................80 5-9 Values for Nasal Aperture Width derived from correspondence analysis.........................80 5-10 Values for Nasal Bone Structure derived from correspondence analysis..........................80 5-11 Values for Nasal Overgrowth derived from correspondence analysis...............................80 5-12 Values for Post-bregmatic Depression derived from correspondence analysis.................81 5-13 Values for Posterior Zygomatic Tubercle derived from correspondence analysis............81 5-14 Values for Supranasal Suture derived from correspondence analysis...............................81 5-15 Values for Zygomaticomaxillary Suture derived from correspondence analysis..............81 5-16 Frequency of summed scores for American Whites and Blacks.......................................81 5-17 Classification matrix for the OSSA method......................................................................82 5-18 Conditional probabilities for the Nave Bayesian two-group analysis..............................82 5-19 Clasification matrix from Nave Bayesian two-group analysis.........................................82 5-20 Conditional probabilities for the Nave Bayesian three-group analysis............................83 5-21 Classification matrix for three-group Nave Bayesian.......................................................83 5-22 Jackknifed classification matrix for discriminant function................................................83 8

PAGE 9

5-23 Classification matrix for two-group k-NN.........................................................................83 5-24 Classification matrix for multi-way k-NN.........................................................................84 5-25 Likelihood ratio tests for the two-Way logistic regression................................................84 5-26 Classification matrix for two-way logistic regression.......................................................84 5-27 Classification matrix for two-way logistic regression.......................................................84 5-28 Full model four-way logistic regression............................................................................84 5-29 Forward stepwise four-way logistic regression.................................................................85 5-30 Classification rates of the two-way CAP analysis.............................................................85 5-31 Classification rates ofr a three-way CAP analysis.............................................................85 5-32 Comparison of CAP analysis to LDF, k-NN, and LR........................................................85 5-33 Interand intra-observer error analysis using Fleiss and Cohens kappa.........................85 6-1 Classification rates for two-way analysis of each method.................................................98 9

PAGE 10

LIST OF FIGURES Figure page 3-1 Threshold values for a three character state trait in three environments...........................48 4-1 Screen-capture of the computer program Macromorphoscopics.......................................54 4-2 Summed scores derived from the OSSA method..............................................................59 5-1 A two-dimensional plot of the Inferior Nasal Aperture frequency table...........................86 5-2 A two-dimensional plot of the Anterior Nasal Spine frequency table...............................87 5-3 A two-dimensional plot of the Interorbital Breadth frequency table................................87 5-4 A two-dimensional plot of the Malar Tubercle frequency table........................................88 5-5 A two-dimensional plot of the Nasal Aperture Width frequency table.............................88 5-6 A two-dimensional plot of the Nasal Bone Structure frequency table..............................89 5-7 A two-dimensional plot of the Nasal Overgrowth frequency table...................................89 5-8 A two-dimensional plot of the Post-bregmatic Depression frequency table.....................90 5-9 A two-dimensional plot of the Posterior Zygomatic Tubercle frequency table.................90 5-10 A two-dimensional plot of the Supranasal Suture frequency table....................................91 5-11 A two-dimensional plot of the Zygomaticomaxillary Suture frequency table..................91 5-12 OSSA scores for all individuals.........................................................................................92 5-13 Plot of first two canonical scores for four-group and three-group.....................................92 5-14 Comparison of error rates k-Nearest Neighbor..................................................................93 5-15 Unconstrained PCOs and the constrained CAP analysis...................................................93 A-1 Character states for the Anterior Nasal Spine morphology.............................................106 A-2 Character states for the Inferior Nasal Aperture morphology.........................................106 A-3 Character states for the Interorbital Breadth....................................................................107 A-4 Character states for the Malar Tubercle morphology......................................................107 A-5 Character states for the Nasal Aperture shape.................................................................108 10

PAGE 11

A-6 Character states for the Nasal Aperture Width................................................................108 A-7 Character states for the Nasal Bone Contour...................................................................109 A-8 Character states for the Nasal Bone Shape......................................................................109 A-9 Character states for the Nasal Overgrowth......................................................................110 A-10 Character states for the Post-bregmatic Depression........................................................110 A-11 Character states for the Posterior Zygomatic tubercle.....................................................111 A-12 Character states for the Supranasal suture.......................................................................111 A-13 Character states for the Transverse Palatine suture.........................................................111 A-14 Character states for the Shape of the Zygomaticomaxillary suture.................................112 11

PAGE 12

Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy THE STATISTICAL DETERMINATION OF ANCESTRY USING CRANIAL NONMETRIC TRAITS By Joseph T. Hefner August 2007 Chair: Michael W. Warren Major: Anthropology Anthropologists have recognized the value of nonmetric cranial traits for the study of human variation for over three hundred years. A historical reliance on nonmetric cranial traits may explain why most practicing forensic anthropologists rely directly or indirectly on an anthroposcopic approach to skeletal analysis. Although the historical organization of humans into groups has been a central tenet in anthropological studies, the roots of the nonmetric approach are deeply imbedded in an early typological approach to race determination. Two anthropologists, Franz Boas and Earnest Hooton, changed the way anthropologists think about race in America, but in two very different ways. The Boasian approach did not support the prevailing paradigm and the belief in human races, but rather emphasized the study of human variation. The Hootonian approach, however, was rooted in the polygenist idea that races did indeed exist and that race could be defined by cranial measurements and nonmetric cranial traits. To that end, Hooton produced a list of nonmetric cranial traits he used to define each race. This list, now known as the Harvard Blanks, remains in use today in several manifestations, yet empirical support for the traits he selected has never been produced. Moreover, proper statistical methods for this type of data have never been explored. 12

PAGE 13

The nonmetric traits used to predict ancestry have historically relied on the experience of the observer and the typological trait lists from Hooton, which are both without empirical support. Assessing ancestry in this manner results in a priori conclusions drawn from some overall impression of the cranial Gestalt. This, the current method of nonmetric, macromorphoscopic trait analysis, is used almost exclusively today by forensic anthropologists, but the method has changed very little from Hootons method. This historical approach is ineffective and does not meet the guidelines established by the Daubert ruling. Previous research has demonstrated that the cranial nonmetric traits previously associated with ancestral groups are not found in frequencies high enough to permit visual observation alone. In other words, the extreme trait expressions for these nonmetric traits are not very reliable for estimating ancestry. The purpose of this research is to test the application of classification statistics to nonmetric traits for discriminating between populations frequently encountered during forensic investigations. Using a statistical framework, which provides the certainty, replicability, and reliability needed to meet the Daubert challenge, removes the subjectivity inherent in the human judgment process. In fact, all of the statistical methods used worked better than the traditional approach to ancestry determinationwhich is little more than visual estimation and guesswork. Each of the statistical methods correctly classified more than 83% of the sample, and several nearly 90%. The OSSA statistic and the CAP method show the greatest promise for forensic anthropological case work, correctly classified 83% and 86% of the entire sample correctly. T he current approach to ancestry determination using nonmetric cranial traits is typological and relies too heavily on the human judgment process. This is an uncertain method, but this research will demonstrate that the uncertainty inherent in this method can be understood and explained using several statistical methods and by acknowledging the variation observed among 13

PAGE 14

populations. Recognizing the shortcomings of the current method and applying new approaches is both scientific and effective and should serve as a basis for the replacement of the traditional approach: classifications based on general impressions from the cranial Gestalt with little more than post hoc validation of the nonmetric traits. 14

PAGE 15

CHAPTER 1 RACE AND PHYSICAL ANTHROPOLOGY The value of nonmetric cranial traits for the study of human variability has been recognized for nearly three hundred years. The result is an exceptional number of academic articles devoted to these slight variations in cranial form for everything from population distance studies in a bioarchaeological context (Anderson 1969; Berry and Berry 1967; Brothwell 1981; Buikstra and Ubelaker 1995; Hanihara 1996; Ishida and Dodo 1997; Jantz 1970; Lahr 1996; Saunders 1989; Tyrell 2000; Tyrell and Chamberlain 1998) to the determination of sex and ancestry of unknown skeletal remains in a forensic setting (Angel and Kelly 1990; Baker et al. 1990; Brinton 1890; Brooks et al. 1990; Brues 1990; Burns 1999; Burris and Harris 1998; Duray et al. 1999; Finnegan and McGuire 1979; Gill and Rhine 1990; Hefner 2003; Hinkes 1990; Hooton 1931; Krogman 1939; Krogman and Iscan 1986; Rhine 1990; Stewart 1979). The historical reliance on nonmetric traits might explain why most practicing forensic anthropologists rely directly or indirectly on an anthroposcopic (from the Greek: 'a`nqrwpos [man] + skop(en) [to look]) approach to skeletal analysis. In fact, a metric analysis of skeletal material likely begins with a priori conclusions drawn from some overall impression of the cranial Gestalt (Ousley and Hefner 2005). This is not surprising because visual observation is a typical approach to organization, and organization of the physical worldfor objects as mundane as aluminum foil or as pertinent to the current research as groups of humansis a basic approach to human understanding. So, organization of objects using categorical terms is appealing because of the need for systematized structure and a simple way to describe that structure. Of course, the historical organization of humans into populations, ancestral groups, or races has been a central tenet in anthropological studies. Some researchers have argued that 15

PAGE 16

race studies are the original sin of anthropology (Levi-Strauss 1952), but we cannot deny the role of early race research in the establishment of the techniques used to analyze forensically significant human remains, particularly those methods used to determine ancestry from cranial material. Before delving into the historical development of the modern anthropological views of race, it is necessary to define several terms used throughout this work. The most socially provocative word used herein is race. Race has been defined by anthropologists, geneticists, historians, and philosophers and there are nearly as many definitions as there are those trying to define the word. The Oxford English Dictionary (OED) defines race (from the French word rasse) as a group of persons, animals, or plants, connected by common descent or origin. According to the OED, the term race may have first been used to describe human groups in 1570 by John Foxe in The Book of MartyrsThus was the outward race & stocke of Abraham after flesh refused. Today, the term race carries with it negative social connotations and is considered by many anthropologists to be a loaded expression (Brace 2005). In fact, most biological anthropologists consider race, or as it is often used in this context race, to be an arbitrary and racist line of research. Ousley and Jantz (2002) suggest the term bio-race for biologically-based taxonomic classifications of humans, like Mongoloid or Caucasoid. Social race on the other hand, is the official, or bureaucratic race, assigned to an individual by his or her peers. Social race may be self-prescribed or socially determined. Social race can be found on a drivers license, census form, or in an All Points Bulletin used by law enforcement officers to describe a suspect. However social race is used, it is important to remember that the social race of an individual is the determination made by the forensic anthropologist when attempting to establish ancestry 16

PAGE 17

from skeletal material (Ousley and Jantz 2002; Sauer 1992). Examples of social race include American White, Hispanic, and Native American. These terms encompass language groups (Hispanic) and breeding populations and they are a part of humans world view on human variation. Social race is also important because it structures mating choices and therefore preserves existing differences among groups. Ousley and Jantz (2002) suggest looking in the personal ads of any newspaper for evidence of this structures mating. It is clear that single Black professional females often seek single Black professional males and vice versa. Therefore, a correlation exists between biology (a priori group differences) and the social race classification used in the United States. Of course, as the mating structure of social race is removed and thus variation increases, the need for up-to-date data documenting the skeletal changes becomes essential. Another term often used, but rarely discussed in any detail is ancestry. Anthropologists speak of determining an individuals ancestry or predicting ancestry, but what does ancestry actually imply? In the United States, the tendency is to observe arbitrary, non-biological factors like ethnicity, tribe, language, or religion. But the Sioux are not a race, they are a tribe. The Slavic language does not equate a Slavic race. These terms refer to groups of people who historically shared a common geographic origin and thus some common genetic material. Thus, estimating ancestry from skeletal material is basically an attempt to assign that individual some ethnic group, from the analysis of a combination of metric and nonmetric variables. Ethnic group is correlated to social race in the United States due to genetic drift and assortative mating (Ousley and Jantz 2002). As a result, up-to-date data on the metric and nonmetric variation among these groups is necessary. While metric data for modern Americans is currently available from the Forensic Anthropology Databank, there is currently no databank 17

PAGE 18

for the nonmetric, macromorphoscopic data. In fact, the current standards for nonmetric trait analysis rely on small sample sizes which provide very little information on the actual observed variation among modern Americans. The historical development of ancestry prediction using nonmetric traits could explain some of the inadequacies of the nonmetric approach. The remainder of this chapter is dedicated to the development of ancestry prediction within biological anthropology. Although nonmetric traits, at least in their present form, only play a minor role in anthropology for the first two hundred years, these slight variations in cranial form were recognized by anatomists and physiologists when describing differences among races. Anatomy, Systematics, and the Search for Order The modern anthropological view on race was founded on the efforts of 18 th and 19 th century European scholars and naturalists attempting to classify everything in the natural world (Comas 1961; Montagu 1974; Shanklin 1994)from the inorganic to the organic, from the smallest of organisms to the largest. During the Age of Reason there was so much faith in the ability of the human mind to analyze and interpret the natural world, that any system of classification designed using the human mind must be right because those classifications are consistent. However, as emphasis was shifted to empirical evidence, the role of experience and experimental science in the formation of ideas became more important than reason and the human mind. Although experimental science was in its infancy, the emphasis on observation and, more importantly, the need for orderliness in science was at an unprecedented level. This new emphasis on orderliness acted as a catalyst for the rise of classification theory and, in turn, classification theory is inextricably linked to the age of European exploration. In the midst of exploration, Europeans encountered the other in a unique way. When travel on foot or on horseback was replaced with port-to-port travel, the clinally distributed differences among 18

PAGE 19

groups suddenly seemed much more dramatic and the differences among those groups seemed to suggest far greater differences in humans. In order to understand their place in the world, scientists speculated on the origins of the different people encountered by sailors and explorers. The physical, social, and cultural differences of these groups was also questioned. If the other looked and behaved differently, then they must be biologically different from Europeans. Classification of the other into discrete biological packages was first attempted by the father of modern taxonomy, Carolus Linnaeus (1707). Linnaeus includes his various subspecies of humans into the order Primates in the 10 th edition of Systema Naturae (1759), distinguishing four groups: Homo sapiens africanus, H. s. americanus, H. s. asiaticus, and H. s. europaeus. Within this order, Linnaeus also includes apes, monkeys and bats, based entirely on shared anatomical characteristics (Brace 2005). In doing so, Linnaeus effectively established the comparative anatomical approach to classification still pursued by todays scientists. Linnaeus did not rely solely on anatomical differences. He made his distinctions between subspecies of humans based on skin color and soft tissue morphology, as well as perceived differences in behavior and social interaction. For instance, he characterized H. s. asiaticus as a yellow melancholy people, ruled by belief (quoted in Shanklin 1994:26). The continentally-based terms Linnaeus used for classifying humans were borrowed from a 17 th -century French physician, Francois Bernier (1620). While Linnaeus was constructing his taxonomies, he borrowed heavily from the work of Bernier, who had suggested using the four quarters of the globe to attach a label to human differences (Brace 2005). While Berniers four quarters approach is a remnant of the belief in a flat world, Linnaeus recognized the utility of Berniers distinctions and applied them thusly. 19

PAGE 20

The Linnaean classification scheme arranged the world into hierarchies, and like most Enlightenment scholars, Linnaeus considered the hierarchy of those groups to be designed and intentional (i.e., The Great Chain of Being). In reality, Linnaeus was a classical creationist who saw a series of distinct, hierarchically arranged beings with God at the top and all other beings below, fixed in position and perfect in form (Brace 2005). Johann Blumenbach (1752), a German anatomist and naturalist, did not agree with Linnaeus. Blumenbach could not see how nature, given its unpredictable and erratic character, could be pigeon-holed using the arbitrary and artificial framework of medieval logic (Brace 2005: 45). This included humans, whom Blumenbach argued progressed from one variety to another. According to Blumenbach, the differences between the varieties of humans was the result of climate, nutrition, and mode of life, each having an effect on the nisus formativus, or vital force, of humans (Count 1950). Most naturalists during this time agreed that the environment could have lasting effects on the form of the human body. In turn, they believed the differences among human groups were linked to climatic adaptation. Noting the skin color of a group of Hottentots, Georges Louis Leclerc, Comte de Buffon (1707) argues that nothing can prove more clearly that climate is the principal cause of the varieties of mankind. In that same vein, Kant (1775) believed climatic adaptation most surely had an effect on skeletal structure. By means of natural disposition, parts of the facegradually become flattened by virtue of the foresight of Natureeven though they [the adaptations] may also be looked upon as natural effects of the climate (Kant 1775). This outlook typified the Monogenist assumption that all living beings are the lineal descendants of Adam and Eve, and that any difference between groups had to take place after the initial diaspora from Eden. 20

PAGE 21

The theoretical underpinnings for monogenism rest on Lamarcks theory of heredity. In Lamarckian terms, the "inheritance of acquired traits is best exemplified by giraffes, who stretched their necks to reach Acacias leaves high atop the trees, strengthening and gradually lengthening their necks. The offspring of these giraffes would subsequently be born with slightly longer necks. Like the neck of the giraffe, Blumenbach did not see the varieties of humans as a fixed part of creation, but rather, he saw the various forms of humans as degenerations (from Latin: degeneris [removed from ones origin]) from an original, perfect form to their current state. Thus the classifications of humans into five varieties by BlumenbachCaucasian, Mongolian, Ethiopian, American, and Malayan 1 did not denote order or hierarchy. According to Blumenbach, all humans had a single origin (i.e., the biblical Adam and Eve), but the differences between his observed varieties is attributable to population migrations and environmental shifts. These new environments caused soft tissue and skeletal changes, which, after a period of time, would become heritable (Brace 2005). Of course, this essentialist view of race led to Blumenbach to assume that a perfect form existed from which all other forms had degenerated. Blumenbach thought the ideal, perfect form was the: Caucasian variety. I have taken the name of this variety from Mount Caucasus, both because its neighbors, and especially its southern slope, produces the most beautiful race of men, I mean the Georgian; and because all physiological reasons converge to this, that in that region, if anywhere, it seems we ought with the greatest probability to place the autochthones of mankind (quoted in Count 1950: 36) 1 The classifications proposed by Blumenbach, particularly Caucasoid and Mongoloid remain in use in some circles today. A third term, Negroid, often erroneously attributed to Blumenbach, was first used to describe skeletal material by Samuel Morton in Crania Aegyptica (1844). 21

PAGE 22

Degeneration did not initially imply taxonomic rank or order, only that distances and circumstances had combined to change their appearance from what it had been in the beginning (Brace 2005: 46). Although Blumenbach had not affixed any significance to the original homeland for each race, later he, and others, did add a value to racial distinctions, establishing several criteria used as justification for modern racism and inequality. Slave owners justified slavery in part by arguing for the naturally inferior state of Africans (Popkin 1974: 143). The eugenics movement adopted the concept of degeneration and used it to justify the sterilization of supposedly weak and ailing individuals. The Nazis later adapted these eugenic ideas when attempting to exterminate the Jews, who, the Nazis feared, would corrupt future generations of Germans. Blumenbach not only used geographical origin and degeneration to make his distinctions of race; he also relied on absolute traits like skin color and hair form, as well as relative cranial dimensions and body measurements. Throughout his career, Blumenbach recorded a large number of measurements from living populations and skeletal collections. He was acutely aware of the skeletons utility for racial classification. Noting that when stripped of the soft parts and changeable parts, Blumenbach suggested crania exhibited the differentiating traits necessary for making correct assumptions of ancestry (quoted in Bendyshe 1969:234). A century later, forensic anthropologists depend on skeletal differences, both metric and nonmetric traits, to define and classify human groups and in some ways they still use Blumenbachs approach. As scientists and scholars became increasingly cynical of the biblical account of creation (Harris 1968), the belief in a single origin for all humans (i.e., monogenism) also faded into relative obscurity. The role of religious values to interpret natural phenomenon was losing ground to a new breed of scholars who looked to the scientific method for answers (Wolpoff and 22

PAGE 23

Caspari 1997). At the turn of the 19 th century, European scholars were gradually becoming more aware of the extent of human diversity and variation, but they continued to explain this diversity using a the typological approach of Linnaeus and Blumenbach. The typological approach focused on a small number of phenotypic traits, like hair form, skin color, etc. to assign an individual to a race, but typology certainly found more followers in the United States with its particular racial climate. Like the anatomists and natural historians of the 18 th century, scientists in 19 th century America emphasized the typological approach to, an approach that melded particularly well with the contemporary attitudes concerning race in the United States. Typology, Polygenism, and the American School The unique setting in the United States during the 19 th century should be apparent to the reader: indigenous Americans were displaced from their homelands, Europeans were expanding into and colonizing the United States, and the forced migration and/or enslavement of African slaves, Chinese laborers and Irish settlers was unprecedented. Taken as a whole, these diverse cultures were in close contact to one another for the first time in history. In this climate, Gould (1996) suggests that it is of little surprise that the polygenist movement was championed in the United States. Contrasting the single-origin, Genesis-based view of monogenism, polygenists held that human races were separate biological species (Gould 1996), each descended from its own Adam and Eve. For slave owners and proponents of the slave trade, the polygenist explanation of human variation was justification for inequality if American Indians, Africans, or the Irish are another race, then equal rights do not need to apply to them (Brace 2005, Gould 1996). The most prominent polygenist in the United States was Samuel G. Morton (1799), anatomist, physician, and craniologist. In all his endeavors, Morton made long-lasting contributions. As a physician, he worked on the diagnosis and treatment of tuberculosis (Morton 23

PAGE 24

1834a), and his research in paleontology (Morton 1834b) and anthropology (Morton 1839) essentially established each of these fields as a scientific discipline in the United States (Brace 2005; Harris 1968). However Mortons work in craniology is his most recognized contribution. Morton defended the polygenist explanation for the differences in human groups using thirteen cranial measurements collected from 250 individuals representing Caucasian, Malayan, American, Negroid, and Mongolian groups (Harris 1968: 90). These categories should look familiar; Morton borrowed his five groupings directly from Blumenbach, but unlike Blumenbach, Mortons races were unrelated to each other and had come about not through climactic adaptation, but through an act of divinityeach Race was adapted from the beginning to its peculiar local destination (Morton 1839, quoted in Brace 2005: 83). Any discussion of Morton and the American School of Anthropology must acknowledge Mortons cranial capacity studies. An in-depth discussion on this aspect of Mortons work can be found elsewhere (Gould 1996, Brace 2005) and need not be repeated here. Suffice it to say that Gould (1996) believed Morton had fudged and finagled his data to fit preconceived notions whereas Brace (2005), citing a little known Masters thesis by Michael (1988), supported the analytical abilities of Morton going so far as to praise him as an exemplary and fully in control researcher (Brace 2005: 85). Whichever the case may be, Mortons role in the development of anthropology should not be overshadowed by his cranial capacity research and support of the polygenist movement. While Mortons analytical abilities are debated, his conclusions concerning human intellect and cranial capacity have had a major impact on American anthropology and, unfortunately, racist research. Mortons results are still cited in the racist literature of Rushton (1999). 24

PAGE 25

Brace (2005) suggests that American anthropology after Morton, but prior to the modern era, is composed of researchers who were attempting to hold on to the last scraps of slavery in the United States. Not only did these anthropologists support slavery, but some, like Josiah Nott, actively fought to protect the southern bastion of inequality (Brace 2005). As the 19 th century was nearing its end, America was recovering from the Civil War and American anthropology was reeling from misconceptions about race (Gould 1996). However, two anthropologists were emerging who would have a major impact on the modern views of race in United States. The 20 th century witnessed the rise of two scholars with very different views on the scientific validity of race. Franz Boas at Columbia University and Earnest A. Hooton at Harvard University each trained a large number of students during their tenure (Spencer 1981), and because the relationship nurtured between student and mentor by both men was one of respect, admiration, and freedom in research interests (Spencer 1981, Washburn 1983, Wolpoff and Caspari 1997) the effect of their teachings can still be felt today. The career of each man has had a profound impact on the modern school of American anthropology and each deserves a closer look. Franz Boas (1858) was born in Minden, Germany in 1858. Like many scholars during this era, Boas was trained in several academic fields, a fact that would serve him well later in his career when he called on his training in the hard sciences and forever changed anthropology. At the University in Kiel, Boas was awarded a Doctorate in physics. He later conducted post-doctoral work on geography in the Baffin Islands, where he lived and worked closely with the native Inuit, and where he first began nurturing an interest in the way non-Western people lived. Due to the growing anti-Semitic climate in Germany, Boas left for America, where he was offered a position at Columbia University in 1896 following a curator 25

PAGE 26

position at the American Museum of Natural History. Once at Columbia, Boas consolidated the various professors of anthropology from other departments into one department, culminating in the establishment of the first Ph.D. program in anthropology in America. Given his background in physics and geography, it is little wonder he is renowned for applying the scientific method to the study of human culture and societies. One of Boas' greatest contributions to biological anthropology was the classic study, Changes in Bodily Form of Descendants of Immigrants (1910), in which Boas tested the influence of environment on body form (Relethford 2003). Prior to his study, Blumenbach, Morton, and other scientists measured crania to trace the relationships between races and to make inferences about the people. The scientists believed that the form of the skull was constant in each race and that measurements and calculated indices would allow them to draw conclusions about the race as a whole. Adamantly opposed to this concept of human variation, Boas, under a grant from the United States Immigration Commission, collected anthropometric, craniometric, and pedigree data from nearly 18,000 individuals from seven populations. From his analysis of this data, Boas concluded that the environment significantly influenced cranial morphology, even within a single generation (Boas 1910). The recent reanalysis of Boas original data by two research cohorts led to two very different interpretations of the same dataset (Sparks and Jantz 2002; Gravlee et al. 2003). While Gravlee and colleagues (2003) agree with Boas interpretation, Sparks and Jantz (2002) suggest cranial plasticity alone is not enough to explain modern human cranial variation. Relethford (2004) offered a very clear conclusion concerning the reanalysis of the Boas data. Although cranial plasticity is evident, the magnitude of the changes brought on by cranial plasticity are not enough to negate the underlying patterns of the population relationships. Relethford (2004: 381) 26

PAGE 27

explains the divergent conclusions reached by the two research cohorts as influences shaped to some extent by goals. Gravlee et al. (2003) demonstrated the existence of cranial plasticity with a statistical test, but Sparks and Jantz (2002) argued that the underlying pattern of the populations is not affected significantly by plasticity. In fact, a principal coordinate analysis (PCO) used by Sparks and Jantz (2002) separated Boas groups into East-West environmental grades, suggesting gene flow as a partial explanation for the patterns observed in the Boas dataset. Boas influence on American biological anthropology was matched only by E. A. Hooton (1887) at Harvard University. Hooton, however, provided an entirely different approach to ancestry and race. Hooton was born in the fall of 1887; the third child and only son of a Methodist minister (Spencer 1981). After earning a bachelors degree from Lawrence College, he went on to earn a Doctorate from the Classics department at the University of Wisconsin, presenting a dissertation on literary art from Pre-Hellenic Rome. His exceptional scholarship and academic record were rewarded with a Rhodes Scholarship in 1911, an undertaking which permitted him to work closely with Sir Arthur Keith, the preeminent Scottish anatomist and anthropologist at Oxford who was a leading figure in the study of the human fossil record (Spencer 1981). In 1930, Hooton accepted a position at Harvard University after his initial attempts to secure employment at the Smithsonian Institution under Ale Hrdlika failed to come to fruition (Spencer 1997). Hootons career at Harvard spanned four decades and comprised some of the most important contributions to current philosophies and methodologies for ancestry determination. In the same fashion Keith had so greatly influenced and shaped anthropology in the British Isles (Brace 2005), so too has Hooton shaped American anthropology. His most influential 27

PAGE 28

contribution to biological anthropology was the sheer number of doctoral students he trained. In fact, Hooton single-handedly trained more biological anthropologists than any other professor during this time, including Boas (Spencer 1981). While Hootons influence on forensic anthropology has been largely overshadowed by his eccentric work in criminal anthropology and the eugenics movement, the methods of analysis he initiated particularly his work with nonmetric traits, are still very much in use today (Brues 1990; Rhine 1990). Early in his career at Harvard, Hooton was looking for combinations of metric and nonmetric traits to define races. These traits Hooton was looking for were not to be affected by environmental factors, because he felt adaptive traits would not provide information on the common descent of individuals (Hooton 1926). He would later retract his insistenceon non-adaptive characters in human taxonomy [as] impractical and erroneous (Hooton 1946: 75), but he would always retain a polygenist outlook on human variation, with a particular interest in typological categories and biometric methods (Wolpoff and Caspari 1997). The typological approach to human variation was, of course, completely contrary to the emerging school of thought at Columbia University. At the end of the Holocaust attitudes about race and the study of race were changing, largely due to the efforts of Boas and his students (Wolpoff and Caspari 1997). In contrast, Hooton and his students continued their attempts to define and understand race. While Boas stressed human variation, Hooton and his students, particularly Carleton Coon, were trying to demonstrate the existence of distinct biological races (Comas 1961). Both insisted on collecting reams of data to justify their conclusions or theories. For example, Hooton collected anthropometric data on thousands of civilians and criminals for his Lombrosolike work in criminal anthropology, he collected craniometric and nonmetric data on all of the skeletal material from the Pecos Pueblo site, and he collected a large amount of data 28

PAGE 29

for the U.S. Air Force and the Quartermaster Corp relating to the design of their equipment, including helmets, seat belts and safety harnesses. Hooton used this data to formulate his larger-than-life authoritarian pronouncements, to validate his numerous theories and as justification for these theoretical leaps of faith. Unlike Hooton, Boas generally avoided making broad declarations from the large datasets he analyzed empirically (Ousley and Jantz 2002). Hooton could also be considered a biological determinist, as he included various aspects of behavior when deciding on the biological traits he used to define each race. However, Hooton (1946) took biological determinism a step further when he, like Alphonse Lombroso, looked for an association between various criminal acts and race. Lombroso believed criminals presented with inferior physiological differences and could therefore be detected, but Hooton took that a step further and supposed certain races may be more likely to commit certain criminal acts. Hooton published most of findings on criminals in The American Criminal: An Anthropological Study (1939). In that volume, Hooton analyzed the data from nearly 14,000 criminals and 3,000 civilians. He found that in 19 of his 33 measurements, criminals were significantly different from civilians. Like Lombroso, Hooton offered morphological criteria for distinguishing criminals (e.g., low foreheads, high pinched nasal roots, excess of nasal deflections), but unlike Lombroso, Hooton offered frequency data on race and criminal acts. In his book Crime and Man, Hooton explored the physiognomy of the Old American criminal, the New American criminal, and the Negro and Negroid criminal. Hooton differentiated the races in order to provide a more broad picture of deviant behavior as he saw it. His definition of the Old American criminal is a native born White of native parentage-from two or three generations of American nativity (Hooton 1939: 34). New Americans are those individuals who had recently immigrated to the 29

PAGE 30

United States (e.g., Italians, Jews, Chinese, Danes, Russians). Negro criminals were defined as those of African-American descent, while Negroids were simply members of the Negro race into which the incessant seepage of White and Indian blood...have brought about a multiplication of physical types (Hooton 1939: 291). It is obvious that Hooton (1926) believed in biologically discreet human races. In fact, he divided humans into three primary races (White, Negroid, and Mongoloid). Each of these primary races was further subdivided into secondary races. For example, Hooton divided the Negroid race into African Negro, Niolitic Negro, and Negrito. In The Indians of Pecos Pueblo (1930), Hooton introduces a third racial categorypseudo-types. Using the skeletal material recovered from the Pecos site, Hooton isolated eight morphological types, first by morphology and later using metric analysis (Woodbury 1932). As an example, individuals with morphological features (i.e., nonmetric traits) who resembled Africans were called Pseudo-Negroids. Hooton (1930) did not see a genetic relationship of these Native American individuals to Africans, but he did see the discordant nature of these morphologies to be evidence of heterogeneity for the peoples of Pecos Pueblo. Hootons interpretation of these discordant morphologies is further evidence of his typological approach to race. Hootons theoretical approach to race was adopted by many of his students (Brace 2005). For instance, Carleton Coons (1904) dissertation explored the adaptive significance of racial features and typical racial forms (Wolpoff and Caspari 1997: 155) and Harry Shapiros (1902) research centered on hybrid vigor in a racially mixed population on the Pitcairn Islands in the Southern Pacific Ocean (Spencer 1981). Carlton Coons view on race deserves closer inspection. Coons view was founded on the polygenist assumption that there are a fixed number of races corresponding to divisions that 30

PAGE 31

occurred prior to the evolution of Homo sapiens (Coon 1962). In Coons opinion, the fossil record provided direct evidence of the five categories of humans: Australoids, Mongoloids, Caucasoids, Congoids, and Capoids. The fossil evidence for each of these distinct races also suggested cultural evolution to Coon. He believed the five races had each evolved at different rates to the present Homo sapiens-form and hence differences existed on the level of their cultural development. Of course, this view of the evolution of humans and human variation is inconsistent with both the fossil record (Stringer and McKie 1996) and modern genetic evidence (Sykes 2002). Another Hooton student who studied the classification of human variation in a completely different approach to Coon, but very similar to Hooton was Stanley Garn (1922). Garn noted that groups of people in the same geographical area resemble each other more closely than they do to people in other geographical areas. Based on this observation, Garn divided humans into nine all encompassing geographical races: Amerindian, Asiatic, Australian, Melanesian, Micronesian, Polynesian, Indian, African, and European. The criterion for group membership was geography alone, not skin color, or head form, or nasal breadth. As Garn (1961) saw it, gene flow takes place more often within a geographical area, rather than between geographical areas. However, Garn (1961) also made two distinctions within the geographical races, which he called local races. According to Garn, local races are either 1) distinct, isolated groups (remnants of larger units i.e., the Ainu of Japan) or 2) large local races with greater levels of gene flow (i.e., Eastern Europe). Garn apparently recognized the large amount of variation in these local races, because he divided the large local races into arbitrary smaller units which he called microraces (Garn 1961). The actual definition of a microrace is unclear. Garn (1961) reports that precise boundaries can not be drawn and in fact members of one microrace may have phenotypes more 31

PAGE 32

similar to another microrace, but Garn saw marriage [as a] mathematical function of distance (Garn 1961: 24). This line of thinking is not unlike the pseudo-races used by Hooton at Pecos Pueblo, but Garn (1961) recognized the role geography plays on gene flow and the dynamic nature of human variation, at least more so than Hooton. The impact of Hootons students, particularly Coon and Garn, on the modern views of race in the anthropological community can not be understated. But it was condemnation and criticism to Coons approach that most affected anthropological studies (Montague 1963). In fact, Marks (1995) suggests that the general reaction to Coons work in the anthropological community precipitated the change from studies of human races to the study of human variation. The switch to the study of human variation rather than race did not completely happen within forensic anthropology. In fact, it was another of Hootons students who had more of an impact on forensic anthropology and the methods used to determine ancestryAlice M. Brues (1913). Brues was born in 1913 in Boston, Massachusetts, the daughter of Charles T. Brues, a professor of entomology at Harvard University. Brues completed graduate work in anthropology at Radcliffe College, where she was originally interested in studying comparative religions. However, she was so greatly influenced by Hooton that she later changed her line of study to physical anthropology. As a woman, Brues was not allowed to sit in the classroom with the other graduate students and actually Brues had to ask her father to speak with Hooton in order to gain permission to take his class (pers. comm. Giles 2007). Nevertheless, in 1940 Brues became the second woman in the United States to earn a Ph.D. in anthropology with a specialization in physical anthropology. Brues is best known for her work in human variation and genetics, especially two papers on genetic and blood group analysis (Brues 1954, 1959). Her book, People and Races was 32

PAGE 33

published in 1977, and exemplifies the Hootonian approach to race with broad assertions concerning human variability within races, but very few empirically supported models. Her interest in aspects of morphology and evolution, also included the application of physical anthropological techniques to criminal investigation (Gill and Rhine 1990), an interest she passed on to many of her students. One of those students was Stanley Rhine. Rhine received his PhD from the University of Colorado in 1969 under the tutelage of Brues. Rhine inherited much of his thinking on race from Brues. The approach Rhine inherited and subsequently passed on to a new generation of forensic anthropologists, however, is typological, and difficult to work with considering the modern views on human variation. Rhines approach to race is exemplified in The Skeletal Attribution of Race (1990), a modern collection of edited articles on the determination of race in a forensic setting. Rhine does not describe human variation per se, rather he presents a collection of morphological traits (i.e., nonmetric cranial traits) useful for identifying the race of a skull, a typological approach very much in line with the Hootonian thinking. The typological approach to the forensic determination of ancestry is common. So common in fact that the majority of active forensic anthropologists work within some typological framework (Gill and Rhine 1990). When identifying unknown skeletal remains, forensic anthropologists historically have not consider the range of variation present within populations, particularly when they use simplified grades of a feature supposedly explicitly linked to a specific race. Instead they merely provide answers to forensic questions, not insights into human variation. As we shall see, forensic anthropologists have difficulty determining ancestry using nonmetric traits, in part because they have ignored the true nature of human variability and relied solely on an inherited, typological approach. 33

PAGE 34

CHAPTER 2 THE FORENSIC APPROACH TO ANCESTRY DETERMINATION In a forensic setting, the goal of ancestry determination is seemingly simple: provide a prediction of the peer-perceived ancestry of an individual from metric and nonmetric variables of the cranial and postcranial skeleton. In practice, however, the determination of ancestry is much more difficult, particularly when that determination is made using nonmetric cranial traits. There are two widely accepted methods of ancestry determination in forensic anthropology each contrasting the approaches of Boas and Hooton. As an empiricist, Boas preferred metric data, which more accurately reflects the continuous distribution of human variation. Hooton, on the other hand, preferred nonmetric, categorical traits. As a keen observer of human variation, Hooton preferred nonmetric traits to metric because they [nonmetric traits] are capable of classification according to presence or absence, [and] grade of development and form, if the observer is experienced and is able to maintain a consistent standard for his morphological appraisals (Hooton 1926emphasis added). The standardization of observations, particularly of the morphological variants like nonmetric traits, is difficult, even for the expert. Hooton noted the same: Even veteran anthropologists have difficulty in maintaining consistency in these subjective ratings and still greater difficulty in equating their standards with those of equally experienced observers (Hooton 1946: 715). Metric analysis is historically characterized by standard definitions of measurements collected using instruments and analyzed statistically. Methods used to analyze this data vary from indices and ratios to univariate and multivariate analysis, but the cranial measurements collected at present essentially conform to the standards established at the Frankfurter Verstndigung (Frankfort Convention) of 1882 (Hursh 1976). The Frankfurter Verstndigung met to standardize cranial landmarks and data collection methods of anthropometric data. There 34

PAGE 35

have been subsequent refinements and alterations of the definitions established at this convention (Pearson 1925; Howells 1975) and because of conventions like the Frankfurter Verstndigung, metric methods remain fairly standardized. Likewise, nonmetric traits have been collected from skeletal material since the time of Blumenbach (1776), but unlike metric analysis, nonmetric traits have never been subjected to the same level of scientific inquiry as their metric counterparts. These slight variations in cranial form are difficult to measure on an interval scale and have generally been arranged into one of four discrete classes: extra-sutural bone, bony bridging, ossification failure, and foramina variation (Buikstra and Ubelaker 1995). Although nonmetric methods have never been empirically supported, they have been preferred over metric methods for centuries. In physical anthropological studies, nonmetric traits are used because: 1) Nonmetric cranial traits have a long standing tradition in anthropological studies; 2) Even when remains are fragmented nonmetric cranial traits may still be observed; 3) Nonmetric cranial traits can be scored and collected with minimal difficulty; and, 4) best of all: THEY WORK (Buikstra 1974: 1her emphasis). All of these are correct. For instance, like craniometric data, nonmetric traits show familial inheritance (Cheverud 1981; Laughlin and Jrgensen 1956) and have been applied to questions of biological distance (Laughlin and Jrgensen 1956). The thorniest of issues, however, is the application of nonmetric traits in a forensic setting. The nonmetric traits used to answer forensic questions of identity differ from the nonmetric traits traditionally used by physical anthropologists, an artifact of the Harvard school of anthropology and Hooton. The typological approach Hooton used to identify racial characteristics was transformed by his students into the modern techniques used to determine ancestry. 35

PAGE 36

Hooton recognized the subjectivity of nonmetric traits and the need for standardization. In fact, he attempted to standardize observations with a list of traits that would become known as the Harvard Blanks (Brues 1990). Although it is unclear how many institutions use similar versions of this list. Brues (1990) suggests that the majority of Hooton's students used or borrowed from it when establishing their own laboratories across the country. In A Handbook on Anthropometry (1960), Ashley Montague, a Boas student from Columbia University, includes one version of the Harvard blanks which he attributes to J. Lawrence Angel, a Hooton student and later a curator in the Physical Anthropology department at the Smithsonian Institution. The link between Hooton and the Harvard blanks is an important one, because it demonstrates the origins of the use of a very distinct scoring system for nonmetric cranial traits used to predict ancestry. The traits used in forensic anthropological research have an underlying quasi-continuous structure, but they are measured on a categorical scale. These traits include assessing bone shape (e.g., the nasal bone structure), bony feature morphology (e.g., the inferior nasal aperture morphology), suture shape (e.g., the zygomaticomaxillary suture), presence/absence data (e.g., the presence or absence of a post-bregmatic depression), and feature prominence/protrusion (e.g., the anterior nasal spine). Ousley and Hefner (2005) noted this discrepancy in trait selection between physical and forensic anthropologists and suggest the term macromorphoscopic traits to describe those nonmetric traits used in a forensic context. Unfortunately, the subjectivity and lack of standardization noted by Hooton has never been addressed. An element of subjectivity may always be present in the nonmetric method, but minimizing that subjectivity is one of the goals of science. The current approach to ancestry determination from nonmetric traits is not scientific. Rhine (1990) described the nonmetric method of ancestry determination as as much art as sciencecontribut[ing] to the mistrust of 36

PAGE 37

some who feel that to quantify is to do science, while to evaluate is to produce a system wholly experiential, and thus unverifiable by retest (Rhine 1990: 18). The method used currently is an art. Moreover, it is an art that is unverifiable and therefore unscientific. For example, Rhine (1990) examined 45 nonmetric (macromorphoscopic) cranial traits in four groups (Whites [n = 53], Blacks [n = 7], Hispanics [n = 15], and Amerindians [n = 12]). Based on his analysis of the frequency distribution of this dataset, he concluded that nonmetric traits are useful for predicting ancestry. Yet, he came to this conclusion despite less than ideal sample sizes, in the midst of what are indeed paltry results, and with little empirical support for his conclusions. Although Rhine (1990) clearly intimates that his sample, particularly American Blacks (n = 7), is very small; his list of expected trait values can be found in most forensic anthropology textbooks (Byers 2002). In other words, even with small sample sizes and inadequate results, the Rhine (1990) study continues to be used as an almost exclusive reference for nonmetric, macromorphoscopic trait analysis. Perhaps a closer look at the some of the frequency distributions for expected trait values (derived from Rhines lists) is noteworthy. For instance, in most textbooks a post-bregmatic depression is reported for American Blacks. However, only 33% of the Rhine sample, or one individual (American Black, n = 3, for this trait), actually showed a post-bregmatic depression. In fact, of the 45 traits collected by Rhine, nineteen had at least one unexpected observed value (Rhine 1990:16). In other words, over 40% of the expected values in the Rhine trait lists were not the observed values for his sample. This approach to nonmetric trait analysis relies on the observers experience and an expert-level surety of human variation (Gill 1998; Rhine 1990), rather than on the interaction of traits and their association with ancestral groups. Recent research on the judgment process has 37

PAGE 38

concluded that expert judgments may not be as accurate as the expert would like to believe and judgment theory increasingly indicates that the experts evaluation of his or her own performance is often at odds with reality. The honest expert must therefore ask, am I doing any better than flipping pennies (Meehl 1986). The ability to accurately predict unknown events from visual information is subject to certain systematic flaws; perhaps the most prominentis simple overconfidence (Meehl 1986). Researchers from fields like psychometrics and medicine have demonstrated several problems inherent in the human judgment process, which, if correct, greatly diminish the value of the current analytical approaches to nonmetric trait analysis. Psychological experiments have shown that experts tend to rely on very little information to make a prediction of an unknown event, in part because feedback is often not available until long after a judgment has been made (Hastie and Dawes 2001). The same is true in forensic anthropological analyses; only after a determination of ancestry is made and a positive identification is established does the forensic anthropologist learn the accuracy of the judgment. Information obtained after the fact leads to adjustments of the relative importance of each trait used to determine ancestry. This leads to post hoc trait selection for ancestry dependent on the cranial Gestalt, which Hefner and Ousley (2006) have suggested has no empirical basis. In fact, this Hootonian approach to ancestry determination presents a paradox: although traits have been historically associated with certain groups, the actual trait frequencies in modern populations may not be as high as previously suggested. These differences in trait frequencies are likely a reflection of the lack of standard observations, poor critical evaluation, and the inadequate sample sizes of earlier studies (Hefner 2003b). 38

PAGE 39

Like experts in other fields, forensic anthropologists also seem to lack insight into their judgment systemthey do not know, nor can they explain, how they weight different traits when determining ancestry because it is a mental process. According to Hastie and Dawes (2001) this is particularly true when the judge is considered an expert and thought to be highly experienced. Studies in psychometrics and medicine have shown that experts can and generally do select the variables that are most important in making correct predictions, however, the expert has a difficult time describing the decision making process (i.e., variable weighting) and can only support their decisions with verbal claims and post hoc data analysis. Gill (1998) touches on the issue of variable weighting when he suggests the forensic anthropologists should not give equal weight [to traits] in assessing ancestry. Unfortunately, but for one example (Cranial Form), he does not suggest what the differential weights for these traits should be or how these weights should be calculated. Several researchers, including Gill (1998) and Rhine (1990) have stated that experience is the essential criteria for the successful interpretation of macromorphoscopic traits. Meehl (1986) demonstrated that experts do only about as well as novices when attempting to clinically weight and compute some function as complex as macromorphoscopic traits. Meehl (1986: 372) illustrates the navet of relying on experience and the human mind for computing even a simple function: Surely we all know that the human brain is poor at weighting and computing. When you check out at the supermarket, you dont eyeball the heap of the purchase and say to the clerk, Well, it looks to me as if its about $17.00 worth; what do you think? The clerk just adds it up. Thiele (1993) suggests low interjudge agreement as another weakness of the judgment process. Despite the fact that standardization would remedy this problem, trait selection criterion do not exist for nonmetric traits, but rather trait selection is adjusted on a case by case basis. One explanation may be that forensic anthropologists believe a common filter exists among them 39

PAGE 40

which can be used to discov[er] morphological charactersin an objective and repeatable process even without listed criteria (Thiele 1993:42), but Hefner and colleagues (2007) confirmed that this was not the case. Instead, they found a high degree of interjudge disagreement in both nonmetric trait selection and specimen evaluation (Hefner et al. 2007). In that study, participants were asked to predict sex and ancestry of several crania, listing the traits used to make the prediction. The results of that study suggest low agreement in trait selection and even lower agreement on the selection of the character states for each nonmetric trait. This suggests a lack of standardization and definitions, as previously suggested by Hooton (Hooton n.d.). Finally, a study of fingerprint experts by Dror et al. (2005) suggests that extraneous information influenced the decision of fingerprint experts concerning a fingerprint match. When presented with information of no analytical value, the fingerprint experts actually reversed previous decisions they had made and came to a conclusion directly opposite their original findings. Other researchers studying how experts make decisions confirm that judges, when presented with irrelevant information, become more confident in the accuracy of their judgments, although true accuracy does not increase (Hastie and Dawes 2001). The inclusion of ad hoc and quasi-Bayesian analysis to back up any judgment leads to overconfidence and unrealistic classification rates, particularly when the judgment is based on what may be irrelevant information. Judgment theory researchers have demonstrated the difficulty of prediction in areas like medicine, psychology, and engineering, and they are realizing that conclusions should be based only on the relationship between the data and the results. The same is true for forensic anthropology. 40

PAGE 41

If the judgment process is flawed, and if nonmetric trait analysis, in its current state, relies solely on expert judgments, then it is safe to assume that nonmetric trait analysis in its current state is also fundamentally flawed. Nonmetric trait analysis is an uncertain process, at best. However, by understanding the amount of uncertainty (error rate) inherent in the method, the nonmetric approach can be applied in a scientific manner rather than as it has been used in the pastclassifications based on general impressions from the cranial Gestalt with post hoc validation of the method. Nonmetric cranial traits are useful for predicting ancestry, but not in the manner they were used in the past and not as they are applied today. Although these traits are thought to be as accurate as metric methods when the analyst is experienced and using a large number of traits (Rhine 1990:18emphasis added), the actual error rates associated with predicting ancestry from nonmetric traits have never been established. This subjective method of nonmetric trait analysis does not lend itself to objective or defensible explanations for ancestry determination on the witness stand (Hefner et al 2004). In 1975 a uniform set of rules known as the Federal Rules of Evidence (FRE), were enacted to govern evidentiary standards in both civil and criminal courts. One of these rulesFederal Evidence Rule 702directs the qualifications of an expert based on his or her knowledge, skill, experience, training, or education (Federal Rules of Evidence 1975). Amendments to the FRE, including Rule 702, state that testimony of any expert 1) must be based on facts, 2) it must be based on dependable methods, and 3) those methods had to have been applied in a reliable manner (Federal Rules of Evidence 2000). These amendments were prompted by the ruling submitted in Daubert v. MerrellDow Pharmaceuticals, Inc., (1993). The Daubert decision was based in part on Frye v. United States, 54 App.D.C. 46, 47, 293F. 1013, 41

PAGE 42

1014, which says admissible expert testimony requires techniques generally accepted by the scientific community in which the methods were developed (Christensen 2004). The court further interpreted and defined the terminology in Rule 702. The reliability of a method, they interpreted, most be grounded in scientific knowledge, and not certainty based on subjective conjecture. From the rulings in the Daubert case, four guidelines were established for determining whether evidence (including expert testimony) is admissible in court: 1) Has the method been tested using the scientific method; 2) Has the method been subjected to peer review, preferably in peer reviewed literature; 3) Are there standards for the application of the method, with known error rates for that method; 4) Is the method generally accepted in the scientific community? The techniques used to establish a biological profile of human skeletal remains (which includes ancestry, but also age, sex and stature) must meet the four guidelines established by the Daubert Ruling (Christensen 2004). These guidelines embody superior scientific method, yet predicting ancestry using nonmetric cranial traits does not currently meet these guidelines. First, nonmetric traits have not been tested scientifically. In place of science, forensic anthropologists have relied on the experience of the observer and on an historical method. Second, nonmetric traits, particularly the macromorphoscopic traits used to predict ancestry, have only minimally been subjected to peer review. Predicting ancestry using nonmetric traits has, to the authors knowledge, never been tested. Third, there are currently no standards for the application of this method, let alone standard approaches with known error rates. Rhines (1990) study is frequently cited, but the methods introduced in that text do not meet the Daubert guidelines. Finally, the acceptance of nonmetric traits for predicting ancestry has historically been very strong. However, as forensic anthropology matures, the techniques established during the development of the field 42

PAGE 43

are being tested and it is being discovered that these methods do not work as well as was originally thought. The Daubert ruling does not mean forensic anthropologists should discontinue the use of nonmetric traits when determining ancestry. In fact, nonmetric cranial traits can be used to predict ancestry using methods that meet the Daubert guidelines. The remainder of this dissertation is devoted to exploring the underlying biological basis of the cranial nonmetric traits used ancestry prediction and exploring several analytical methods of analysis that meet the Daubert guidelines. 43

PAGE 44

CHAPTER 3 NONMETRIC CRANIAL VARIATION The usefulness of nonmetric cranial traits for determining ancestry is based on the assumption that these traits have an underlying genetic basis. While we know that multiple genes play a role in the formation and development of nonmetric variables it is difficult to say with certainty which genes and which environmental factors are responsible for a given nonmetric trait. Nevertheless, there is evidence of a genetic, an epigenetic, and an environmental component to nonmetric trait development (Hauser and De Stefano 1986; Moss 1997d). Before exploring these components of nonmetric traits some generalizations concerning the underlying biology of the cranium are warranted. The skull comprises a complex set of 24 bones which functions as the skeletal framework for the head. The skull serves as the foundation for most of the senses and serves as the housing and protective case for the brain, all the associated sense organs, and it is the location of the primary structures associated with mastication and breathing. The skull is actually the combination of the cranium and mandible. Although variation in aspects of the cranium is important for determining ancestry, it is important to remember that the skull must also be fairly conservative. For instance, the distance between the eyes must fall within a certain range to maintain the stereoscopic vision of humans and the position of the external ear is maintained to ensure that the brain can reliably and accurately process outside auditory cues. The development of the skull is a fairly complex process. There are two ways the bones of the skull can ossify: directly in mesenchymal tissue or indirectly within cartilage derived from mesenchymal tissue (Enlow 1975). The result of both processes is bone with the same material properties and structure. In humans, the primary bones of the vault develop intramembranously (i.e., within mesenchymal tissue) and the major bones of the face and basicranium develop 44

PAGE 45

endochondrally (i.e., within cartilage). During endochondral bone development, a cartilaginous model forms in the location where the bone will eventually develop. The cartilaginous model gradually becomes bone through osteoblast (bone forming cells) and osteoclast (cells responsible for breaking down previously formed bone) activity in the area (England 1996). Intramembranous ossification of bone is the direct process of bone development from mesenchymal tissue and does not require a cartilaginous model. The cells in intramembranous ossification derive from specialized mesoderm cells which form sheets of bone with focused osteoblastic and osteoclastic regions (Moss 1997c). Factors Affecting Trait Expression Genes and the Environment Whatever their mode of development, the bones of the skull change dramatically during the first few years of an individuals life. In fact the skull continues to grow and develop through adulthood and any variation during the normal development process can have a dramatic impact. For instance, craniofacial anomalies like Binders Syndrome (Mulhern 2001) or cleft palate disorders (Mooney and Siegel 2002) are primarily the effect of genetic factors during the developmental process. Environmental and epigenetic factors likewise affect normal skull development. Moss (1997abcd) recommends a dynamic approach to cranial analysis that emphasizes functional cranial features rather than a sole reliance on genetic or epigenetic explanations for a feature. According to Moss (1997a) the functional cranial features are independent to one another in size, shape, and position. Therefore, each of the functional cranial matrices should not be treated as osteological units (e.g., mandible, maxilla), but rather each should be treated independently. As Enlow (1975) explains the FCM model, the developing embryo and the 45

PAGE 46

growing child alike have varying levels of sensitivity to environmental influences based on genetic factors, which ultimately affect the phenotype (Enlow 1975). A timetable for the growth and development of the skull and, as a result, all of the associated nonmetric cranial traits, follows a theoretical model of growth where each of the skeletal components of the skull have growth rates integrating both an epigenetic controlling factor (e.g., growth is slow, rapid, or sustained depending on the condition) and a specific, genetically determined growth rate for each structure comprising that functional region of the skull. In other words, Moss (1997a) proposes that the genetic component of a tissue produces a given phenotype, but the vital cells are irritable or perturbed by and respond to alterations in their external environment. The skeletal tissues of the skull associated with nonmetric cranial traits develop within this matrix and are subject to these guiding factors. For example, the formation and position of the malar tubercle on the maxilla may have a very specific genetic origin, but the extent to which the tubercle actually develops is likely an interaction of bony development, functional morphology, and adaptive processes. Establishing the genetic basis for an individual nonmetric traits along with any environmental effects on that trait is difficult, at best. There is currently very little information on the environmental factors affecting nonmetric trait expression. Ossenberg (1970) examined the influence of cranial deformation on the expression of extra-sutural bone in a collection of Hopewell Indian crania. She found that localized stress (i.e., artificial deformation) may produce an increase in the number of sutural bones and ossicles; unfortunately she could not rule out other environmental factors not explained by her model. Since the development of most nonmetric traits is assumed to be an interaction between genetic and epigenetic factors (Berry and Berry 1967; Grneberg 1963), Moss (1997a), and by 46

PAGE 47

extension Falconer (1965), suggest a threshold model to explain the expression of some nonmetric trait. Moss (1997b: 339) functional matrix hypothesis suggests more involvement of epigenetic factors for the development of shape variations, contrary to a strictly genetic basis which posits that all (phenotype) features are ultimately determined by the DNA sequence of the genome. According to Moss (1997c: 411), assuming a strictly genetic origin for all anatomical structures is reductionist and molecular and is akin to suggesting morphogenesis is as simple as passing directly from DNA molecules to adult gross morphology. This approach ignores the role of epigenetic processes and, as an extension, intrinsic human variation. Moss is not implying that the genome does not play a role in the development of form. Rather, he synthesizes the two paradigms and argues that morphogenesis is regulated (controlled, caused) by the activity of both genomic and epigenetic processes and mechanismsand only their integrated activities provides the necessary and sufficient causes of growth and development (Moss 1997d: 413). The threshold model followed in this dissertation accounts for both the genomic and the epigenetic mechanisms proposed by Moss. To clarify, for any feature, there are threshold values which correspond to the expression of a given character state along a normally distributed curve. When a stimulus parameter (Moss processes) exceeds a threshold value, the tissue has an osteoblastic response (i.e., deposition, resorption, maintenance). The probability of an individual expressing some trait or character state is an interaction between the individuals genes and a suite of epigenetic factors influencing the likelihood of the expression of that trait. The individuals position relative to each threshold value dictates the expression of that trait. Of course, position on the curve is based on the loadings associated with whichever genes are influencing a traits expression as well as the environment in which that genetic component is being expressed. For example, in Figure 3-1 47

PAGE 48

three environments are represented for some nonmetric trait. The threshold values do not shift in a new environment, so the frequency of expression for the various character states are different for each environment (e.g., different loadings). The genetic factors (gene flow) determine the individuals position relative to the threshold values, but the environment establishes loadings which cause an osteoblastic response (i.e., deposition, resorption, or maintenance). Sex, Age, and Trait Correlation Differences in the expression of traits which are not explained by genetic or environmental factors include sex, age, and trait collinearity. Research on the effects of each of these sources of variation is extensive in the physical anthropological literature. Unfortunately, research on this variation is virtually nonexistent regarding the traits used in forensic anthropology. Even within the physical anthropological literature, sex differences in trait expression are ambiguous and show very little consistency among researchers (Hauser and De Stefano 1986). Figure 3-1. Threshold values for a three character state trait in three environments. 48

PAGE 49

Berry (1975) believes the lack of consistency of sex dimorphism observed for many of the traits confirms the idea that they are the outward manifestation of the activity of genetic, epigenetic, and even overtly environmental forces. Hefner (2003b) found similar results for sex differences in trait expression using macromorphoscopic traits. In that study, there were no significant differences in trait expression for males and females, with the possible exception of post-bregmatic depression which was moderately, but significantly, expressed more frequently in black females. The age regressive nature of nonmetric traits is a concern, but like sexual dimorphism, the literature is inconsistent on the actual affect of age on trait incidence. Some traits will obviously be affected by age. For instance, the supranasal suture is a secondary center of ossification and would not be expected in juvenile remains (Hefner 2003). Buikstra (1972) documented several traits affected by age, but like many others (Perizonius 1979a), she concluded that age is only a factor in younger individuals. So, if only adult crania are used for analysis, age regression may be ignored (Hauser and De Stefano 1986). Hauser and De Stefano (1986) surveyed a large portion of the literature on nonmetric traits and found uncertainty and inconsistency in the literature on trait correlations. They concluded that trait associations are likely because of the common processes affecting cranial growth/development and nonmetric trait manifestation. Hefner (2003a) recently explored trait independence for macromorphoscopic traits and found a moderate and significant correlation of traits in sample of a modern American Blacks and Whites. General Patterns of Variation Modern humans are distinct from other species in their ability to expand to and occupy virtually every environment on earth. Yet, the global pattern of human variation today is the result of the expansion of modern humans out of Africa some two hundred thousand years ago 49

PAGE 50

(Stringer and McKie 1996) Although the evolution of modern humans is beyond the scope of this dissertation, it is important to understand modern human variation in this context. The current variation we see when walking down the street resulted from very specific evolutionary mechanisms acting on populations. Human biologists recognize the mechanisms acting to create this variation includes the genetic effects of selective pressures, the effects of gene flow between groups, and the random effects of drift and founding (Lahr 1996; Mielke et al. 2006). Each of these effects played a role in the development of modern human diversity, but the influence of natural selection alone does not explain the entire range of human variability. So what explains the levels of variation observed today? Evidence in the fossil record suggests slight regional variation, but the major theme in hominid evolution has been a trend to gracilization in cranial dimensions from early moderns to more recent humans (Howells 1973; Lahr 1996). Today, size-related variation is negligible and likely stems from a combination of genetic homogeneity, or common ancestry, environmental selective pressures, and/or developmental restrictions on the size of the cranium (Howells 1989; Moss1997a). This implies that different evolutionary mechanisms were acting on populations asynchronously, an assumption supported by the fossil record. In the Upper Pleistocene, various levels of specialization appears at different times and at variable rates (Lahr 1996). Therefore, the different levels of gene flow, selection, and drift acting on the various populations of humans is closely linked to the different the unique geographical circumstances of that group (Lahr 1996). These early geographic divisions account for the high degree of variation observed among modern humans today, particularly in the United States where these groups are juxtaposed in close proximity. 50

PAGE 51

In terms of how that modern variation is distributed, Howells (1989) demonstrates that the majority of shape differences between ancestries is located in the midface, including: variation in the upper face, variation in the nasal region, and variation in the orbital borders. These are the same regions of the craniofacial skeleton relied heavily upon by forensic anthropologists today. For example, Brues (1990) came to the same conclusions as Howells, but her results were based on her experience in the osteology laboratory and not on empirically supported evidence. Nonmetric cranial traits can be used to make a prediction of ancestry because the underlying genetic basis of these traits and the epigenetic loadings acting to create these features are the result of the unique environment where they initially developed. The resulting nonmetric traits are the effects of selective pressures from particular environments, gene flow between groups, and the random effects of drift and founding during the evolution of modern humans. Today, assortative mating and social race maintain these a priori group differences and the underlying patterns of the population relationships. 51

PAGE 52

CHAPTER 4 MATERIALS AND METHODS To explore underlying patterns of population variation in the expression of 16 nonmetric, macromorphoscopic traits, data were collected from the skeletal groups described below (Total N = 747). Hefner (2003a) previously found no significant sex differences in macromorphoscopic traits, with the exception of post-bregmatic depression, so males and females are pooled within groups for analyses. Except where noted, samples were combined using either the traditional three-ancestry model of forensic anthropology (African, European, Asian) or the four-ancestry model (Black, White, Asian, Native American). Samples African Sample The African sample consists of both native Africans and American Blacks. The native African sample comprises individuals from East and West Africa (n = 15 and n = 17, respectively) retrieved during the Smith African Expedition and purchased by the Smithsonian through Frederick Muller & Co., Amsterdam, Holland. They are currently housed at the National Museum of Natural History (NMNH), Smithsonian Institution in Washington, DC. European Sample The European sample consists of both native Europeans and American Whites. The sample is composed of individuals from The Netherlands and Germany (n = 7 and n = 8, respectively) purchased by the NMNH from anatomical houses or other museum trades. The American whites from the Terry Collection are described in detail below. East Asian Sample The East Asian sample consists of individuals from Japan and China (n = 15 and n = 59, respectively) housed at the NMNH. The Chinese specimens were collected by Hrdlicka from a 52

PAGE 53

cannery in Karluk, Alaska and they represent individuals who worked in the salmon canneries in the late 1800's. The Japanese specimens are from the Tokyo prefecture and were all donated by Tokyo Imperial University. Native American Sample The Native American sample consists of the following groups (all housed at the NMNH): Arikara, SD (n = 42), Hawikuh, NM (n = 40), Doyon Eskimo (n = 39), Pastolik Eskimo (n = 12), Pueblo Bonito, NM (n = 7), Santa Barbara Island, CA (n = 57), Almeda, CA (n = 26), Perico Island, FL (n = 17), Cape Canaveral, FL (n = 19), and St. Lawrence Eskimo (n = 9). All of these groups are preand protohistoric, ranging in chronological age from 1000 years before present to the early 19 th century. Anatomical Collection The Robert J. Terry Collection is also housed at the NMNH and consists of approximately 1730 individuals of known age, sex, ancestry, and stature. This collection represents a cross-section of 19 th century American Whites and American Blacks. A sample of American Blacks (n = 150) and American Whites (n = 170) was collected from the Terry Collection. Bass Collection The William M. Bass Donated Skeletal Collection is curated by the Forensic Anthropology Center in the Department of Anthropology at The University of Tennessee, Knoxville, Tennessee. The Bass Collection consists of approximately 500 skeletons from throughout the United States of known age, sex, ancestry, and stature. The Bass Collection is predominantly modern American Whites and Blacks. A sample of 38 American Blacks was collected from the Bass Collection. 53

PAGE 54

Macromorphoscopic Traits Sixteen nonmetric traits were collected using Macromorphoscopics, a program designed by Hefner and Ousley (2005) for data imputation ( Figure 4-1 ). The definitions are modified from Hefner (2003a). Each trait definition is presented along with individual character states and black and white line drawings in Appendix A Figure 4-1. Screen-capture of the computer program Macromorphoscopics. Statistical Methods The greatest advantage of a statistical analysis is that the inductive reasoning inherent in the human thought process can be made precise by specifying the amount of uncertainty involved in the conclusions drawn (Rao 1989: ix). But what does that really mean to the forensic anthropologist working in the laboratory, trying to identify a set of skeletal remains? In short, everything. The methodological approaches used to make a determination of sex, age, ancestry, or stature should rely on one or more statistical methods. A restatement on the utility of macromorphoscopic traits and the development and application of statistical methods appropriate 54

PAGE 55

for this type of data is needed. In particular, forensic anthropologists need statistical methods for classifying an unknown cranium based on the presence of a suite of macromorphoscopic traits. To address this issue, the following statistical approaches were utilized. Correlations Hefner (2003a) used the polychoric correlation coefficient to measure the association of nonmetric traits. For categorical, ordinal variables (0,1,2,3,4), the polychoric correlation coefficient is appropriate because variable correlations are not affected when the latent continuous variable underlying the trait is compressed into a categorical response (Coenders and Saris 1995). Tetrachoric correlation coefficients are the recommended measures of association for binary (0,1) variables. LISREL 8.51 was used to estimate polychoric and tetrachoric correlation coefficients. LISREL uses Limited Information Maximum Likelihood analysis following a two-step procedure (Joreskorg and Sorbom 2001). First, thresholds are estimated from the raw frequency distributions for y i and y j the two variables of interest. Correlations are then estimated using a Restricted Maximum Likelihood method conditional on each threshold value calculated during the first step of analysis (Coenders and Saris 1995: 132). These methods are appropriate for use when the underlying character states that form the fundamental scoring methodology can be viewed as continuous, a condition met with the large sample size and the quasi-continuous nature of macromorphoscopic traits. Ordinal Regression Ordinal regression analysis (ORA) measures the association of an ordinal response from a set of predictor variables. In traditional linear regression, the sum-of-squared differences between a continuous dependent variable and the weighted combination of the independent variables is minimized prior to calculating regression coefficients. This is not the case when the dependent variable is ordinal. Ordinal regression calculates coefficients based on the assumption 55

PAGE 56

56 [...]iipij that the response variable is a categorical variable with some underlying continuous distribution. In most cases, there is a valid theoretical basis for assuming such an underlying distribution. However, even when this is not case, the model can still theoretically produce valid results. Rather than predicting the actual cumulative probabilities, the ORA predicts a function of those values using a process known as a link function. After the analysis, the predicted probability of each response category can be used to assign an unknown individual to a group. An ORA can be expressed as: 1122 ()ijjlink where link( ) is the link function for the current analysis, ij is the cumulative probability of the j th category for the i th case, j is the threshold for the j th category, p is the number of regression coefficients, i1 ip are the values of the predictors for the i th case, and 1 p are regression coefficients. An added benefit of an ORA is that not only is significance testing of the individual variables possible, but it is also possible to test for significant interaction among all response variables. For example, the ORA allows one to determine if sex, ancestry, or the interaction of sex and ancestry, significantly affects inferior nasal aperture morphology. All ordinal regression analyses were carried out using the PLUM function in SPSS 14.0 for Windows. The purpose of the ORA was twofold. First, the ORA was used to determine the significance of sex and ancestry, and the interaction of the two, on the expression of each nonmetric, macromorphoscopic trait. In other words, the ORA was used to see how sex and ancestry actually affects trait expressions. If ancestry is not significantly influenced by a trait, then that trait was dropped from the subsequent ORA analyses. Significance was assessed at the

PAGE 57

= 0.05 level, using the Wald statistic, a measure similar to the F-value in a traditional analysis of variance test. Parameter estimates for each trait were assessed for significance. Correspondence Analysis Correspondence analysis is an ordination procedure that transforms contingency table information into a two-dimensional graphical presentation used to visualize patterns, relative dispersions, and total variance. To do this, correspondence analysis relies on three interrelated concepts of contingency tables: profiles, masses, and chi-square distances. The profile of a correspondence analysis is the computed percentages for the response variables and each category, so each row profile is then a unique point in multidimensional space. Reducing this dimensionality facilitates visualization of these relationships into a format we can understandtwo or three dimensional graphs, or correspondence maps. Each cell frequency divided by the grand total is the mass, and is used to assign a weight to each cell proportional to the number of respondents for that category. Finally, the chi-square distance is used to measure the distance between profile points in multivariate space and to facilitate the visual display of each profile as a correspondence map. Correspondence analysis cannot be used to test hypotheses, but it can be used to map out the relationships of variables to one another in multivariate space and, perhaps more practically, the method can be used to make complex frequency tables more clear. For instance, a frequency table may show the distribution of a given traits character states for three groups, but the correspondence map demonstrates the relationship of each character state to each group. All correspondence analyses were conducted using R 2.4.0. For each trait, cross-tabulation tables of trait frequencies within populations were calculated and standardized to create a table of relative frequencies representing the distance between individual rows and columns. Row and column mass values were calculated (R-scores and C-scores) for the relative frequencies. Next, 57

PAGE 58

inertia values were calculated as the total Pearson X 2 for the two-way table divided by the total sum of the table. Plots of the column points and the row points were generated to summarize the information contained in each two-way table. These new dimensions were calculated to maximize the distances between traits, so succeeding dimensions will explain less and less of the overall X 2 value. In this way, the extraction of the dimensions is similar to the extraction of principal components in a Principal Components Analysis. The code for the correspondence analysis can be found in Appendix B along with an example dataset. OSSA The Optimized Summed Scoring Attributes method (OSSA) was developed by Hefner et al. (2004). OSSA is a nonparametric method which compresses morphological variation into two classes. The method is appropriate for the majority of American forensic cases, which are generally American Whites or American Blacks. OSSA is a simple device to maximize the between group differences of two groups by compressing the original ordinal values of a trait (0,1,2,3,4) to a new binary score (0,1). Optimization of the compressed trait score is achieved by ordering the original trait values to maximize the differences among groups. For example, the INA ( Table 4-1 ) was originally scored 0 to 4; Hefner et al. (2004) determined heuristically that the best cutoff point for the new binary scores is the conversion of all ordinal scores of 2 or less to 0 and all scores of 3 or more into ones. Using inferior nasal aperture alone, ancestry could be correctly determined 78% of the time. For each of the five remaining traits, ordinal trait scores more common in American Blacks were optimized as a score of 1 and those more common in American Whites were optimized as a score of 0. Once all traits have been transformed to these new binary variables, the sum total of all the traits is calculated. This summed score provides each individual with a unique score which ranges from 0 to 6. A plot of the individual summed scores is presented in Figure 4-2 58

PAGE 59

Table 4-1. Frequency table showing original scores and compressed OSAA scores. New White (N = 179) Original Black (N = 211) New Score Cumul. % % Score % Cumul. % Score 0 38 0 6.6 0 0 77.1 39.1 1 13.7 0 1 20.1 2 21.8 1 1 2.2 3 28 79.7 1 1 0.6 4 29.9 1 To use the OSSA method, macromorphoscopic traits scores are collected for an unknown cranium using the original character states (original ordinal categories). Once these values are compressed to their respective OSSA state and summed, they can be compared to the cutoff point of 3 and above for Africans and 2 and below for Europeans. This cutoff point (red dashed line in Figure 4-2) is important because it indicates that even with this binary scoring system, only half of the traits expected for Blacks were actually present. Figure 4-2. Summed scores derived from the OSSA method. 59

PAGE 60

Nave Bayesian Nave Bayesian analysis is considered nave because all variables are equally weighted in the model. In other words, the strength, or conversely the weakness, of a variable is not considered when making a classification. To run a Nave Bayesian analysis, a training sample of the dataset is removed to calculate the prior and the conditional probabilities. Using these parameters, the method then estimates posterior probabilities for the remainder of the original dataset using a probabilistic classifier based on Bayes theorem. The Nave Bayesian method can be expressed in the form of the following conditional model: 21(,,...)n p CFFF where the probability, p, of a given class variable, C, is conditional on several features, F. Using the Bayesian model, we can express the model as: 122112()(,,...)(,,...)(,,...)nnn p CpFFFCpCFFFpFFF Stated plainly, the Nave Bayesian model above can be expressed as the probability a cranium is of some ancestry given the values expressed for a set of nonmetric traits. Nave Bayesian assumes that the features of interest are independent of one another. Yet, because the model is probabilistic, Nave Bayesian, like discriminant function analysis, correctly allocates crania as long as the correct population is more probable than any other. Put another way, if the overall classification rate is high enough, serious violations in underlying model assumptions can be ignored with little consequence to the overall model. Linear Discriminant Function Analysis Giles and Elliot (1962) first used a linear discriminant function analysis (DFA) to determine race from the cranial measurements of a sample of American White, American Black, 60

PAGE 61

and Native Americans. In the 60s and 70s DFA had to be done by hand, including the calculation of the variance/covariance matrix. Today, however, DFA is made much more simple because of desktop computers and statistical software packages. This had led to an increase in the use of DFA to answer anthropological questions. Linear discriminant function analysis was initially developed to classify a target individual (e.g., unknown crania) into one of several reference groups. Discriminant function analysis incorporates an approach similar to regression analysis (Krzanowski 2002). However, in regression analysis, a weighted combination of predictor variables (X 1 X 2 X p ) is used to calculate an objects value (e.g., stature from measurements of the postcranial skeleton). Likewise, DFA uses a weighted combination of predictor variables to classify an unknown object into a reference group, but in discriminant function analysis, the predictor variable is a derived discriminant function score (Krzanowski 2002). This DF score is equal to the weighted sum of the values for each variable. The assumptions associated with DFA include: multivariate normality and equality of covariance matrices. As Finnegan and McGuire (1979) point out, these assumptions are not met when using categorical variables. However, Krzanowski (2002: 312) suggests that any categorical variable can be re-expressed as a set of binary variables and that these derived binary variables can be used in a discriminant analysis. For example, nasal aperture width (NAW) is scored as 1 = narrow, 2 = intermediate, and 3 = wide. These values may be re-expressed as two binary variables x 1 and x 2 as follows: x 1 = 0 x 2 = 0 if NAW is narrow; x 1 = 1 x 2 = 0 if NAW is intermediate, and x 1 = 0 x 2 = 1 if NAW is wide. 61

PAGE 62

All discriminant function analyses were carried out using SYSTAT 9.0 and the DISCRIM function. Except where noted all analyses were cross-validated using a jackknife procedure. k-Nearest Neighbor Nearest Neighbor classification methods are some of the easiest clasification statistics to use and understand, in part because nearest neighbor works similarly to the way people think. In short, objects that are close to each other are similar to each other. k-Nearest Neighbor (k-NN) methods classify objects on the basis of a training set of individuals of known class plotted as vectors in multivariate feature space. An unknown individual placed in this space is assigned to the most frequent class among the k nearest individuals in the training sample. During a supervised learning phase, the training sets vectors and class labels are stored. For a classification, the unknown individual is also plotted as a vector and the distance to each of the stored vectors from the training set are computed using some distance measure (e.g., city-block, Euclidean). The k closest individuals in the training set are selected and the new unknown individual is classified with the most frequent group within the training set. Choosing the optimal number of neighbors for a classification depends principally on the dataset; generally, larger values of k blur boundaries between the groups, but also reduces any noise inherent in a dataset. The k-Nearest Neighbor analyses was carried out in R 2.4.0. Code for the k-Nearest Neighbor statistic can be found in Appendix B Logistic Regression Logistic regression (LR) is a semi-parametric method similar to linear regression, but LR is useful in the current study because it enables the researcher to overcome many of the assumptions of linear regression. Like linear regression, logistic regression predicts an outcome for a binary variable Y from one or more X variables, however, these X variables may be 62

PAGE 63

categorical or continuous, because the model does not strictly require continuous data. To predict group membership, LR uses the log odds-ratio and an iterative maximum likelihood method, unlike traditional linear regression which uses the least squares to fit the model. This means the investigator has slightly more freedom when using LR and the method is more appropriate when analyzing data with a non-normal distribution or unequal covariance matrices, because LR does not assume homoscedasticty or normality. Logistic regression assumes that each variable contributes independently to the odds-ratio, however, Hefner (2003b) demonstrated this assumption is likely violated when using nonmetric data because many of the macromorphoscopic variables are strongly correlated. Unfortunately, logistic regression does not produce typicality probabilities, which are useful for biological anthropological studies. However, Ousley and Hefner (2005) suggest the typicality probabilities could be substituted with nonparametric estimates such as ranked probabilities or ranked interindividual similarity measures. Also, cross-validation methods can be used to test the performance of a logistic regression and to obtain a sense of the reliability of estimate for the model. Logistic regression analysis was carried out in SPSS 14.0 using the NOMREG procedure. Except where noted, all LR analyses used a forward stepwise selection for variable inclusion. Canonical Analysis of Principal Coordinates Legendre and Legendre (1998) first proposed that a canonical discriminant analysis could be preformed on the transformed values of the principal coordinates. Anderson and Willis (2003) implemented the method for numerical ecological studies. The CAP method offers several promising qualities to biological anthropologists and forensic anthropologists alike. First, the method is highly flexible. Several distance and similarity/dissimilarity measures can be selected 63

PAGE 64

for an analysis. These include, but are not limited to: Euclidean distance, Chi-square distance, Bray-Curtis dissimilarity, Manhattan Block distance, and the Canberra distance. The second advantage of CAP is that the method accounts for correlations of the data, unlike Bayesian statistics which assumes that all response variables are independent. By accounting for the correlative structure in the response matrix, significant patterns in the data set can be highlighted, particularly in reference to a priori hypotheses, like group membership. The complex relationship of cranial traits, environment, genetics, and head form should and can be explored more thoroughly if the inter-relationships of the data are properly understood. The final advantage of the CAP method is that unknown individuals can be assigned to a group through a generalized discriminant analysis based on distances (Anderson and Robinson 2003). This is particularly useful to forensic anthropologists who seek to identify the ancestry of an unknown skeleton using categorical variables from the cranium. Although traditional discriminant function analysis produces excellent results using categorical variables (Ousley and Hefner 2005), the underlying assumptions of multivariate normality and equal variance-covariance are not met. Krzanowski (2001) reviews the use of categorical data in multivariate discriminant analysis and acknowledges that underlying assumptions will be violated, but he also points out that the method may be appropriate for descriptive purposes and as an exploratory method. However, the CAP method circumvents this issue by transforming data values from categorical variables into principal coordinate scores; in a sense, the data are changed from categorical responses to continuous variables (Anderson and Willis 2003). The CAP method applies a principal coordinate analysis (Anderson and Willis 2003) using any one of several distance measures. The principal coordinate analysis (PCO) is used to explain a multi-dimensional dataset through fewer dimensions although the resulting dimensions are still 64

PAGE 65

linear combinations of the original variables. Once the PCOs are obtained, an appropriate number of axes is selected to maximize the classification rate. The final step in the CAP method is a canonical discriminant analysis on the first m axes of the PCO, followed by a leave-one-out cross-validation procedure. Assumptions of CAP deal strictly with the selection of a distance/dissimilarity measure. The measure giving the highest classification rate is assumed to be the best fitting model, regardless of any underlying assumptions that may be violated. The CAP was carried out using CAP, a DOS-based computer program by Anderson (2004) For the full CAP analysis, the sample was reduced to two and three groups (American Blacks, American Whites, and a pooled Native American/Asian sample). Interand Intra-Observer Error In order to assess both interand intra-observer variation (i.e., observer error) 5 observers were asked to score 7 crania using Macromorphoscopics (Hefner and Ousley 2005). One of the observers (JTH) scored the crania on two separate occasions to assess intra-observer error. Two separate but related statistics were used to measure observer agreement. The statistical tests generally applied to continuous, quantitative data usually comprise a standard ANOVA mixed model to document differences in the mean response among observers. However, such approaches can not be used appropriately when the dataset in question is composed of categorical, qualitative traits. Therefore, intra-observer variability was assessed using Cohens kappa statistic (Cohen 1960). Cohens kappa measures the agreement between two observations, while taking into account any agreement that would occur by chance. If the observations are in perfect agreement, a kappa value of 1 would be expected. Cohens kappa is calculated as: 1eePPP 65

PAGE 66

where P is the relative observed agreement among raters, and eP is the probability that the observation was agreed by chance. Inter-observer variability (e.g., multiple observers) can not be assessed using Cohens kappa, as that measure is appropriately applied to two observers only. Fortunately, a similar (and related) measure exists for multiple observers. Fleiss' kappa (Fleiss 1971) works for any number of observers and is meant for use with categorical, qualitative traits. Like Cohens kappa, Fleiss kappa is a measure of observer agreement expressed as a value between 0 and 1. Fleiss kappa can be expressed as: 1eePPP where 1eP is the agreement likely above chance and ePP is the observed agreement. Thus, Fleiss kappa is roughly equivalent to the ratio of observed versus expected values. The interpretation of both Cohens and Fleiss kappa has been debated. For the current study, the table of Landis and Kochs (1977) significance values for the individual kappa score was used. Landis and Koch (1977) point out that the kappa statistic has several advantages over other measures of observer agreement. First, kappa is easily calculated, whether in a spreadsheet or a commercially available statistical software program. In the current study, both types of kappa statistics were calculated using a spreadsheet. A second advantage of the kappa statistic is the appropriateness of the method for testing whether observer agreement exceeds chance for binary and nominal categorical classes. In this way, the kappa statistic tests both observers independent of chance agreement and it quantifies the level of agreement among observers. Of course there are also disadvantages to the kappa statistic. First, a kappa test does not make distinctions about the various types and sources of disagreement. As stated, the kappa merely 66

PAGE 67

quantifies the level of agreement without denoting why there is disagreement. Second, when working with ordered, categorical variables, the researcher has to determine appropriate weights to calculate a weighted kappa (Maclure & Willet, 1987). In the current study, all weights were set to equal 1, with the understanding that these arbitrary weights may not actually represent the appropriate weight for each response. Finally, because trait distributions are used to calculate the scores, kappa studies are seldom comparable and can rarely be used in cross-study comparisons (Feinstein & Cicchetti, 1990). Regardless of these complications, using each of these kappa statistics is necessary and should shed light on the range of variability in observations. 67

PAGE 68

CHAPTER 5 THE STATISTICAL DETERMINATION OF ANCESTRY The utility of nonmetric, macromorphoscopic cranial traits depends on the application of statistical methods that can correctly classify a cranium and ideally provide the probability of a correct classification. The following methods are not inclusive of the classificatory statistics available, but they represent several novel and innovative approaches to determine ancestry. Statistical Approaches to Ancestry Determination Ordinal Regression A complete examination of the ORA analysis is not necessary here, but as an example, the ORA parameter estimates for interorbital breadth are presented in Table 5-1 Eleven traits were assessed for significance. Four were not significantly influenced by ancestry (MT, NO, PZT, and ZS) and were removed from the ORA. Quite surprisingly, three traits were significantly influenced by sex (ANS, NBS, and SPS). The sexual dimorphism noted in these three traits needs to be addressed in a future project, but will not be assessed here. Only one trait, post-bregmatic depression, was significantly influenced by the interaction of ancestry and sex, and only for one groupBlack femalesconfirming Hefners (2003b) earlier estimate on the occurrence of post-bregmatic depression. Once those traits significantly influencing ancestry were determined, the next step was an ORA using those variables as the independent (predictor) variables for ancestry. The ordinal regression analysis worked well separating American Blacks and Whites in a two-way analysis, correctly classifying nearly 90% overall (cross-validated). Table 5-2 presents the clasification matrix for the two group analysis. A Goodness-of-fit test for the two group analysis is significant (X 2 = 190.709; df = 3; p < .0001) using four variables (i.e., INA, IOB, NAW, and PBD). 68

PAGE 69

The multiple group ORAs were not as successful. Despite less than ideal results, the ORA still correctly classified 70% overall using the sample of American Whites, American Blacks, and Native Americans. Table 5-3 presents the results of that analysis. As more groups are added the classification rate is drastically reduced, although Native American sample does consistently well. This is likely a result of the sensitivity of this method to sample size, so the results of multi-way classification seem spurious. The American Black sample was misclassified most often; nearly 40% were misclassified in the three group analysis. Most of the misclassifications were classified as Native American further demonstrating the sensitivity to sample size of this method. Correspondence Analysis A complete examination of the correspondence analysis (CA) is not necessary here, but as an example, the Inferior Nasal Aperture (INA) will be used to show the value of this method. The presence or absence of a nasal sill has been used for ancestry determination. Data on the inferior nasal aperture morphology is scored on an ordinal scale as: sill, partial sill, straight, incipient guttered, and guttered. Table 5-4 presents the cross-tabulation table for the inferior nasal aperture morphology. Frequencies do not actually fall within the range of expected counts (e.g., American Blacks = [4] GutteredFreq ~ 30%). Table 5-5 presents the extracted values of INA calculated during the CA. The correspondence map highlights the relationship of INA morphologies to ancestry. The first two dimensions, or axes, explain over 98 percent of the variation observed for the inferior nasal aperture morphology. The CA map ( Figure 5-1 ) demonstrates that the first axis separates inferior nasal morphologies with any vertical projection of bone (i.e., a nasal sill) from those with out a projection. American Whites correspond to a sill, supporting previous literature, however, their relative position to the cloud of points representing sill and partial sill suggests slightly more variation than was previously suggested. The same is true for the sample of American Blacks, Native Americans, and Asians. 69

PAGE 70

The obvious advantage of a correspondence analysis is that theoretical interpretations of the extracted dimensions can be achieved by plotting the column and row values and observing the correspondence of one to the other. Two-dimensional plots and the associated CA values for the remaining traits are presented below (Tables 5-6 to 5-15 and Figures 5-2 to 5-11). Although the other traits will not be discussed here, a correspondence map of the frequency table data is provided to highlight some aspects of the datasets. The reader is encouraged to peruse these maps and observe the variation observed (particularly the variation observed between the Asian and Native American samples). Optimized Summed Score Attributes (OSSA) Only two groups were used for the OSSA method, American Blacks and American Whites. In order to test the OSSA method, six nonmetric traits were used: Anterior Nasal Spine, Inferior Nasal Aperture, Interorbital Breadth, Nasal Aperture Width, Nasal Bone Structure, and Post-bregmatic Depression. Table 5-16 presents a summary of the summed score values for the 390 individuals used for the OSSA analysis. The bimodal distribution is obvious in the table, but for clarity the same data is presented in Figure 5-12 The OSSA method was nearly 84% accurate using these 6 traits ( Table 5-17 ). The group classification rate for the OSSA method is 78% for American Blacks and nearly 90% for American Whites. Despite the absence of "typical," or expected, values the OSSA method facilitates the determination of ancestry, in part because the method diminishes interand intraobserver errors by compressing trait scores and still accounting for a large amount of the total variability. The OSSA method compresses morphological variation into two classes for a two-way classification, but multiple group classifications are more challenging. However, the method looks quite promising for the bulk of forensic cases. 70

PAGE 71

Nave Bayesian For the two-group Nave Bayesian analysis a random 20% subset of the entire dataset was removed to calculate the models parameters. The prior probabilities were roughly equal for the two groups and were calculated as 0.501 for American Blacks and 0.498 for American Whites. Conditional probabilities for the two-group analysis are presented in Table 5-18 The classification matrix calculated from the Nave Bayesian model is presented in Table 5-19 American Black and White crania were correctly classified nearly 87% of the time. This method performs slightly better than the OSSA method, as both groups have nearly equal error rates, whereas in the OSSA method, error rates for American Blacks are slightly larger than American Whites. Unlike the OSSA method, the Nave Bayesian method is as stable for multi-group classifications as for a two-way classification. A three-group analysis using six traits is presented below. Prior probabilities were calculated as: Native American = 0.41; American Black = 0.36; American White = 0.23. The slightly higher probability for Native American reflects the somewhat larger sample size. Overall, the multi-group analysis was nearly 80% correct. The function performed slightly better for American Blacks (84%) than for either American Whites (77%) or the Native American sample (77%), but the method does show promise for multiple classifications. As we shall see in the section on discriminant function analysis, when multiple groups are used in an analysis, one group does not perform as well as the others. This phenomenon has not been explained, but it is likely the product of using ordinal variables with only as many character states as there are classes. Discriminant Function Analysis Several methods were used to test discriminant function analysis. A complete discriminant function analysis using all eleven variables was used to test DFA on a sample of American 71

PAGE 72

Whites and American Blacks. One benefit statistical methods like DFA is that classification rates may be cross-validated. Cross-validation methods partition a sample of the dataset and initially perform the analysis on a single subset, retaining the others for subsequent validation of the initial analysis. SYSTAT uses a leave-one-out jackknife procedure, which removes a single observation from the original dataset as the validation data, processes the remaining individuals in the DFA, and then repeats the process over again using another individual. This is repeated until each observation in the original dataset is used and a classification matrix has been constructed. By using this approach, the discriminant function is calculated without using the object of interest in the original function. All classification rates presented herein are the cross-validated results. The complete DFA for two-groups performed exceptionally well, classifying 88% correctly. Next, a forward step-wise DFA (p = .15 enter/remove) using only five variables (Inferior Nasal Aperture, Interorbital Breadth, Nasal Aperture Width, Nasal Bone Structure, and Post-bregmatic Depression) was used. This function performed slightly better (89% correct) than the full model. Table 5-22 presents the results for both the complete and the step-wise analysis. A four-way DFA using all 11 variables correctly classified 67% of the sample. A plot of the first two canonical scores ( Figure 5-13 ) shows the close correspondence of the Native American and Asian groups in the first plot. In the second plot, the two groups are pooled. The three-way DFA using a pooled Asian/Native American sample classified 79% correctly using nine variables (INA, ANS, IOB, MT, NAW, NBS, NO, PBD, and SPS). Several of the variables selected in this analysis, but not the two-way analysis, are important nonmetric traits for discriminating American Whites and Blacks from Asian and Native Americans. These include: the malar tubercle, the nasal overgrowth, and the supranasal suture. The results of this DFA 72

PAGE 73

correspond with the conclusions drawn by Finnegan and McGuire (1979) who used a traditional linear discriminant function analysis to classify crania from nonmetric traits and reported a mean classification rate of nearly 78%. Overall, the DFA worked well for classifying twoand three-groups, particularly when the Asian and Native American samples were pooled. These results do not coincide with the results of the correspondence analysis, which suggest much more variation between the Asian sample and the Native American sample. More research is needed to explain this discrepancy. k-Nearest Neighbor A two-group k-NN analysis of American Blacks and Whites was first performed. A random twenty percent of the dataset (American Black = 43; American White = 33) was drawn to serve as an initial training set Next, thirteen analyses were performed using an increasing number of nearest neighbors, including: 1; 10, 13, 15, 19, and 21 neighbors. Figure 5-14 compares the error rates (misclassification percentage) for each number of k nearest neighbors. Based on this plot, k was set at 4 neighbors for the remaining analyses. As Figure 5-14 illustrates, the number of k nearest neighbors used affects the overall classification rate, but perhaps more importantly, the plot also confirms that setting k = 4, error over the entire model is reduced for all groups to equal values. In a two group analysis and 4 neighbors, k-NN correctly classified approximately 88% of the individuals ( Table 5-23 ). The three-way k-NN did not perform as well. Table 5-24 presents the classification matrix for Native Americans, American Blacks, and American Whites. In the multi-way analysis, the same phenomenon noted in the Nave Bayesian method emerges for k-NN. Two of the samples (Native American and American White) perform quite well (~72% correctly classified); the third group (American Blacks), however, performed miserablyonly 46% of the sample was correctly classified. Overall, k-NN classified 64% of the individuals, approximately 30% higher 73

PAGE 74

than random chance, but the disparity in correct classification for the American Blacks suggests a weakness in the k-NN model that needs to be furthered explored. Logistic Regression All logistic regression analyses were carried out using a stepwise variable selection. An initial two-group LR using five variables (Inferior Nasal Aperture, Interorbital Breadth, Nasal Aperture Width, Nasal Bone Structure, and Post-bregmatic Depression) is first presented. The likelihood ratio test ( Table 5-25 ) is significant and demonstrates that the reduced model is equivalent to the final LR model (i.e., no loss of power with the reduction of variables). The Cox and Snell Pseudo R-Squared statistic equals 0.553 and suggests that approximately 56% of the variation in the macromorphoscopic traits can be explained by ancestry. The LR accurately classified 90% of the sample ( Table 5-26 ). This suggests that for two-group analysis, the LR works exceptionally well. The multi-way LR also worked well, but this may have resulted from LRs sensitivity to sample size. Overall, the three-way LR correctly classified 81% of the samples. Table 5-27 presents the classification matrix. Unlike the Nave Bayesian and the k-NN method, the LR appears to inflate one group (Native Americans) with a higher accuracy rate. This is likely attributable to the slightly larger sample size of the Native American group, rather than a more homogenous population. The next step in the LR analysis was a fourand five-way analysis, first using the full model (i.e., all variables) and then with a forward, stepwise variable selection. Using 11 variables, the four-way LR correctly classified 75% of the sample. However, only 40% of the Asian sample was correctly classified. Forty-three percent of the Asian sample (31/72) was misclassified as Native American. This suggests a level of similarity in macromorphoscopic trait expression between Native Americans and Asians, an idea previously discussed in the literature, 74

PAGE 75

but never empirically demonstrated. However, the correspondence analysis (see above) suggests this is not the case for all of the macromorphoscopic traits. Again, this is an area worth further research. A forward stepwise LR analysis selected six variables: Inferior Nasal Aperture, Interorbital Breadth, Nasal Aperture Width, Nasal Bone Structure, Nasal Overgrowth and Post-bregmatic Depression. The inclusion of Nasal Overgrowth is likely an artifact of the relative frequency of this trait in Asian and Native American populations. Unfortunately, the LR classification rate for the Asian sample dropped even lower with the stepwise analysis ( Table 5-29 ); 61% were misclassified as Native American. While sample size does influence the LR analysis, a two-way LR using just the Native American and Asian samples classified 85% of the entire sample (265/313) as Native American. The Native American sample had a classification rate of 94%, but the Asian sample was only correctly classified in 48% of the cases. In other words, the function did classify well, but only because the sample size for Native Americans was so much larger. Overall, it appears that the LR approach works well when the groups are heterogeneous. Multi-way LR works well when the Asian and Native American groups are pooled, following a three-ancestry model favored by most practicing forensic anthropologists, but this approach may result in a loss of fine-tuning. A LR using equal samples and with more homogeneity among the samples may resolve some of these issues. Canonical Analysis of the Principal Coordinates For the two-group CAP analysis, classification rates are similar to those obtained using logistic regression (89%) and linear discriminant function analysis (85%) ( Table 5-30 ). Using 11 variables, the method performed well classifying approximately 84% of the entire sample correctly. Using 5 variables and 3 PCOswhich explained over half of the total observed variationthe CAP method correctly classified 87% of the sample. These results were obtained 75

PAGE 76

using the Chi-Square Distance measure, which produced the lowest misclassification error for all of the distance measures used. A three-group classification is also quite promising using the CAP method ( Table 5-31 ). The Chi-square distance measure (with m = 12) worked extremely well, correctly classifying 75% of the sample using a leave-one-out cross-validation method. This method appears to perform far better at three-way discrimination than either logistic regression or k-Nearest Neighbor. Note that group separation in the unconstrained analysis is not as well defined as in the CAP, which clearly demonstrates group separation ( Figure 5-15 ). The different criteria used to obtain the unconstrained plot (based on a DFA using the PCOs) versus the CAP plot demonstrated how the ordination procedure can radically affect multivariate patterns. Remember: the cloud of points in both plots are derived from the same dataset, but the view of has been changed by the ordination procedure chosen. This further demonstrates the usefulness of the CAP method. Several of the inter-individual distance measures produced acceptable results, but the chi-square distance measure performed best overall. Chi-square distance measures have been used extensively in anthropology and as mentioned above, they are the basis of correspondence analysis. In this instance, Chi-Square distance likely works better than other measures because this measure emphasizes compositional differences rather than accumulation differences, as in the Bray-Curtis measure, which is commonly used for genetic data, the CY distance measure (basically a log transformed Bray-Curtis), and the Gower and Canberra measures, which standardize each variable by the range. Interand Intra-Observer Error Table 5-33 presents the results of the intraand inter-observer error analysis. Each of the Cohens kappa values were moderately significant or higher for the intra-observer test. This 76

PAGE 77

suggest very little intra-observer error. Only two traits (Anterior Nasal Spine and Supranasal Suture) was significantly lower than the others (ANS k = 0.422; SPNS k = 0.468). Assessing anterior nasal spine morphology relies on a measure of the length of the anterior nasal spine relative to the facial skeleton. Why this trait was only moderately significant is unclear. Several explanations are possible, including error in recording the score, difficulty assessing the length of a small structure relative to a larger structure, or inadequacies in the morphological variants included in the Macromorphoscopic program. The Supranasal Suture may be difficult to assess with in true reliability, in part because the observer has to determine whether the suture is obliterated, closed, but visible, etc. The difficulty inherent in this trait may suggest dropping it from further analyses, at least until it has been more systematically defined and illustrated. The results of the analysis of inter-observer variation and error suggest slightly more disagreement among observers. Two of the sixteen traits (Posterior Zygomatic Tubercle and Nasal Aperture Shape) have extremely low k values and suggest one of two scenarios. First, all but one of the observers (JTH) had never used the Macromorphoscopic program and thus were unfamiliar with the application. Second, PZT and NAS are not often used in a standard forensic analysis of human skeletal material and therefore are not readily known. However, given the standard definitions and illustrations provided in the Macromorphoscopic program, the latter should not be an issue. The first cause can not be ruled out as a possible source of error, and obviously the second suggests these two traits may need to be defined more carefully. A third source of error may be that these traits can not be observed with any repeatability, however, the results of the intra-observer error do not support this hypothesis. The suggestion henceforth is the need for published standards for scoring these traits. 77

PAGE 78

Five of the remaining fourteen traits had only a fair level of agreement among observers, including Post-bregmatic Depression (k = 0.232), Nasal Bone Contour (k = 0.231), Inferior Nasal Aperture (k = 0.376), Interorbital Breadth (k = 0.325), Naso-Frontal Suture (k = 0.210). The discordance among observers for two of the five traits is surprising. Post-bregmatic depression (PBD) is scored on a binary scale (presence/absence) in the Macromorphoscopics program, a feature generally thought to reduce the level of observer error. However, PBD, like most nonmetric traits, is distributed along a continuum, so slight depressions may be missed by the novice and expert alike. The low agreement level among the observers for Nasal Bone Contour (NBC) is also surprising. The Macromorphoscopics program guides the observer on the use of a contour gage (Hefner 2003b) to assess NBC, unlike the traditional approach, which is not clearly defined and has no standardized assessment method. The use of a contour gage to assess NBC is a relatively recent phenomenon (Hefner 2003b) and so there may be a learning curve associated with the use of this tool. However, the advantage of using a contour gage has always been the reduction of observer error. The author demonstrated a high level of agreement (k = 0.810) for NBC, so perhaps as the use of a contour gage becomes more accepted the level of observer agreement will also increase. 78

PAGE 79

Table 5-1. Parameter estimates and significance levels for Interorbital Breadth. Ind. Variable Estimate Std. Error Wald df Sig. Ancestry 2.492 0.340 53.723 1 0.000 Sex -1.113 1.250 0.792 1 0.373 Ancestry*Sex 0.929 1.299 0.512 1 0.420 Table 5-2. Classification matrix for the ORA two-group analysis. Black White Total % correct Black 203 15 218 93.12 White 22 124 146 84.93 Total 225 139 364 89.03 2 =190.709; p < .000 Table 5-3. Classification matrix for the ORA three-group analysis. N. Amer Black White Total % correct N. American 206 46 10 262 78.63 Black 59 130 29 218 59.63 White 10 33 103 146 70.55 Total 275 209 142 626 69.60 2 =287.765; p < .000 Table 5-4. Frequency distribution of INA for American Blacks and Whites. INA Score Black White Total 0 14 68 82 1 29 70 99 2 46 36 82 3 59 4 63 4 63 1 64 total 211 179 390 Table 5-5. Values for Inferior Nasal Aperture derived from the correspondence analysis. Dim Singular Values Eigenvalue % of Inertia Cum. % Chi-Square 1 6.2540120 0.3911 65.4561 65.4561 282.3742 2 0.4410263 0.1945 32.5523 98.0084 140.4290 3 1.0890590 0.0119 1.9916 100.0000 8.5918 4 2.5700160 0.0000 0.0000 100.0000 0.0000 79

PAGE 80

Table 5-6. Values for Anterior Nasal Spine derived from the correspondence analysis. Dim Singular Values Eigenvalue % of Inertia Cum. % Chi-Square 1 0.3678779 0.1353 92.2920 92.2920 97.6866 2 0.0843752 0.0071 4.8431 97.1351 5.1262 3 0.0650769 0.0042 2.8649 100.0000 3.0324 4 0.0000000 0.0000 0.0000 100.0000 0.0000 Table 5-7. Values for Interorbital Breadth derived from the correspondence analysis. Dim Singular Values Eigenvalue % of Inertia Cum. % Chi-Square 1 0.6176239 0.3815 84.1606 84.1606 275.4430 2 0.2680077 0.0718 15.8394 100.0000 51.8396 3 0.0000000 0.0000 0.0000 100.0000 0.0000 Table 5-8. Values for Malar Tubercle derived from the correspondence analysis. Dim Singular Values Eigenvalue % of Inertia Cum. % Chi-Square 1 0.0898999 0.0081 62.7907 62.7907 5.8482 2 0.0689100 0.0047 36.4341 99.2248 3.3934 3 0.0104118 0.0001 0.7752 100.0000 0.0722 4 0.0000000 0.0000 0.0000 100.0000 0.0000 Table 5-9. Values for Nasal Aperture Width derived from the correspondence analysis. Dim Singular Values Eigenvalue % of Inertia Cum. % Chi-Square 1 0.5874727 0.3451 60.9394 60.9394 249.1622 2 0.4703165 0.2212 39.0606 100.0000 159.7064 3 0.0000000 0.0000 0.00000 100.0000 0.0000 Table 5-10. Values for Nasal Bone Structure derived from the correspondence analysis. Dim Singular Values Eigenvalue % of Inertia Cum. % Chi-Square 1 0.5659286 0.3203 66.0821 66.0821 231.2566 2 0.3706733 0.1374 28.3474 94.4295 99.2028 3 0.1643220 0.0270 5.5705 100.0000 19.4940 4 0.0000000 0.0000 0.0000 100.0000 0.0000 Table 5-11. Values for Nasal Overgrowth derived from the correspondence analysis. Dim Singular Values Eigenvalue % of Inertia Cum. % Chi-Square 1 0.2746571 0.0754 98.4334 98.4334 54.4388 2 0.0350569 0.0012 1.5666 100.0000 0.8664 3 0.0000000 0.0000 0.0000 100.0000 0.0000 80

PAGE 81

Table 5-12. Values for Post-bregmatic Depression derived from the correspondence analysis. Dim Singular Values Eigenvalue % of Inertia Cum. % Chi-Square 1 0.3974335 0.158 94.215862 94.21586 114.076 2 0.0983287 0.0097 5.784138 100.0000 7.0034 3 0.0000000 0.0000 0.0000 100.0000 0.0000 Table 5-13. Values for Posterior Zygomatic Tubercle derived from the correspondence analysis. Dim Singular Values Eigenvalue % of Inertia Cum. % Chi-Square 1 0.2821514 0.0796 89.03803 89.03803 57.4712 2 0.0989274 0.0098 10.96197 100.0000 7.0756 3 0.0036079 0.0000 0.0000 100.0000 0.0000 4 0.0000000 0.0000 0.0000 100.0000 0.0000 Table 5-14. Values for Supranasal Suture derived from the correspondence analysis. Dim Singular Values Eigenvalue % of Inertia Cum. % Chi-Square 1 0.2595223 0.0674 97.5398 97.5398 48.6628 2 0.0326237 0.0011 1.5919 99.1317 0.7942 3 0.0247800 0.0006 0.8683 100.0000 0.4332 4 0.0000000 0.0000 0.0000 100.0000 0.0000 Table 5-15. Values for Zygomaticomaxillary Suture derived from the correspondence analysis. Dim Singular Values Eigenvalue % of Inertia Cum. % Chi-Square 1 0.1596046 0.0255 67.8191 67.8192 18.4110 2 0.1095887 0.0120 31.9149 99.7340 8.6640 3 0.0115912 0.0001 0.2660 100.0000 0.0722 4 0.0000000 0.0000 0.0000 100.0000 0.0000 Table 5-16. Frequency of summed scores for American Whites and Blacks Group Summed Score White Black 0 64 4 1 61 13 2 36 30 3 9 39 4 7 41 5 2 66 6 0 18 Total 179 211 81

PAGE 82

Table 5-17. Classification matrix for the OSSA method. White Black % correct White 161 18 89.94 Black 47 164 77.73 Total 208 182 83.83 Table 5-18. Conditional probabilities for the Nave Bayesian two-group analysis. INA 0 1 2 3 4 Black 0.071066 0.147208 0.208122 0.269036 0.304569 White 0.400000 0.374194 0.200000 0.019355 0.006452 NBS 0 1 2 3 4 Black 0.040609 0.101523 0.101523 0.248731 0.507614 White 0.329032 0.283871 0.161290 0.180645 0.045161 ANS 1 2 3 Black 0.091371 0.223350 0.685279 White 0.354839 0.277419 0.367742 NAW 1 2 3 Black 0.040609 0.431472 0.527919 White 0.574194 0.335484 0.090323 IOB 1 2 3 Black 0.096447 0.360406 0.543147 White 0.316129 0.625806 0.058065 PBD 0 1 Black 0.548223 0.451777 White 0.812903 0.187097 Table 5-19. Clasification matrix from Nave Bayesian two-group analysis. Black White % correct Black 172 19 87.31 White 25 136 87.74 Total 197 155 87.50 82

PAGE 83

Table 5-20. Conditional probabilities for the Nave Bayesian three-group analysis. ANS 0 1 2 3 4 Native American 0.213115 0.491803 0.196721 0.065574 0.032787 Black 0.198157 0.502304 0.202765 0.078341 0.018433 White 0.058394 0.313869 0.262774 0.284672 0.080292 INA 0 1 2 3 4 Native American 0.036885 0.22541 0.577869 0.155738 0.004098 Black 0.294931 0.290323 0.21659 0.133641 0.064516 White 0.007299 0.029197 0.211679 0.423358 0.328467 IOB 1 2 3 Native American 0.610656 0.356557 0.032787 Black 0.096774 0.345622 0.557604 White 0.306569 0.649635 0.043796 NAW 1 2 3 Native American 0.081967 0.811475 0.106557 Black 0.036866 0.410138 0.552995 White 0.562044 0.343066 0.094891 NBS 0 1 2 3 4 Native American 0.086066 0.266393 0.237705 0.368852 0.040984 Black 0.525346 0.230415 0.096774 0.105991 0.041475 White 0.043796 0.160584 0.175182 0.270073 0.350365 PBD 0 1 Native American 0.893443 0.036885 Black 0.516129 0.414747 White 0.708029 0.145985 Table 5-21. Classification matrix for the three-group Nave Bayesian analysis. Native American Black White % Correct Native American 205 32 30 76.78 Black 25 170 8 83.74 White 14 15 99 77.34 Total 244 217 137 79.30 Table 5-22. Jackknifed Classification Matrix for the discriminant function. Black White % correct Black 153/156 24/21 86/88 White 15/14 120/121 89/90 Total 168/170 144/142 88/89 Table 5-23. Classification matrix for two-group k-NN. Black White % correct Black 121 17 87.68 White 14 100 87.72 Total 135 117 87.70 Note: k = 4 83

PAGE 84

Table 5-24. Classification matrix for multi-way k-NN. Asian Black White % correct Asian 165 14 48 72.69 Black 57 70 24 46.36 White 27 1 72 72.00 Total 63.68 Note: k = 15 Table 5-25. Likelihood ratio tests for the two-way logistic regression. Model Fitting Criteria Likelihood Ratio Tests Effect -2 Log Likelihood of Reduced Model Chi-Square df Sig. Intercept 183.2665 0 0 INA 238.2002 54.9337 5 0.00000 IOB 207.4619 24.1953 2 0.00001 NAW 193.9302 10.6637 2 0.00484 NBS 199.0345 15.7680 4 0.00335 PBD 191.4299 8.1634 2 0.01688 Table 5-26. Classification matrix for the two-way logistic regression. Black White % correct Black 200 17 92.17 White 19 117 86.03 Total 89.80 Table 5-27. Classification matrix for the three-way logistic regression. Native American Black White % Correct Native American 275 30 11 87.03 Black 38 166 13 76.50 White 29 9 98 72.06 Total 80.57 Table 5-28. Classification matrix for the four-way logistic regression. Native American Asian Black White % Correct Native American 200 11 20 11 82.64 Asian 31 29 9 3 40.28 Black 24 5 138 10 77.97 White 22 1 8 104 77.04 Total 75.24 84

PAGE 85

Table 5-29. Classification matrix for the stepwise logistic regression. Native American Asian Black White % Correct Native American 202 8 20 12 83.47 Asian 44 14 11 3 19.44 Black 28 3 135 11 76.27 White 26 5 7 97 71.85 Total 71.57 Table 5-30. Classification rates of the two-way CAP analysis for eight distance measures. Dist. Measure # of Vars m Black White Overall % correct Chi-Square 5 3 81 93 87 Chi-Square (metric) 13 10 88 80 84 Euclidean 5 3 83 84 84 Bray-Curtis 5 3 83 84 84 Manhattan 5 3 82 84 83 Orloci's 5 3 80 84 82 Canberra 5 3 73 88 81 Jaccard 5 3 63 85 74 Table 5-31. Classification rates of the three-way CAP analysis. Native Americans American Black American White % correct Native Americans 297 43 57 74.81 American Black 24 146 14 79.35 American White 130 24 26 72.22 Total 75.46 m = 12 Table 5-32. Comparison of CAP analysis to LDF, k-NN, and LR. Method # of vars Native American Black White ORA 7 79 60 71 Nave 6 77 84 77 DFA 10 77 73 80 k-NN 11 73 46 72 LR (Ordinal) 6 89 7 64 CAP 13 80 75 72 85

PAGE 86

Table 5-33. Interand intra-observer error analysis using Fleiss and Cohens kappa. Inter-Observer Error Intra-Observer Error Trait Fleiss' k Cohen's k Anterior Nasal Spine 0.506 0.422 Inferior Nasal Aperture 0.376 0.964 Interorbital Breadth 0.325 0.857 Malar Tubercle 0.470 0.929 Nasal Aperture Shape 0.056 0.767 Nasal Aperture Width 0.732 0.929 Nasal Bone Contour 0.231 0.810 Nasal Bone Shape 0.504 0.813 Nasal Overgrowth 1.000 1.000 Naso-Frontal Suture 0.210 0.905 Orbit Shape 0.442 0.857 Post-Bregmatic Depression 0.232 0.820 Posterior Zygomatic Tubercle 0.059 1.000 Supranasal Suture 0.650 0.468 Transverse Palatine Suture 0.700 1.000 Zygomaticomaxillary Suture 0.541 0.857 mean 0.440 0.837 sd 0.256 0.170 Moderately significant or higher following Landis and Koch (1977) -1.0-0.50.00.51.01.52.0 -1.0-0.50.00.51.01.52.0 PCA1PCA2 GutInc GutPa SlSill StrWhiteAsianAfricanAmerInd Figure 5-1. A two-dimensional plot of the Inferior Nasal Aperture frequency table. See Appendix A for an explanation of the individual character states. 86

PAGE 87

-1.0-0.50.00.51.01.52.0 -1.0-0.50.00.51.01.52.02.5 PCA1PCA2 b sentsmallmediummedium larg e White A sianAfricanAmerInd Figure 5-2. A two-dimensional plot of the Anterior Nasal Spine Frequency table. See Appendix A for an explanation of the individual character states. -2.0-1.5-1.0-0.50.00.5 -1.5-1.0-0.50.00.51.0 PCA1PCA2narro w intermediate w ide WhiteAsianAfricanAmerI n Figure 5-3. A two-dimensional plot of the Interorbital Breadth Frequency table. See Appendix A for an explanation of the individual character states. 87

PAGE 88

-10123 -1.0-0.50.00.51.0 PCA1PCA2 absentweakmediumstron W hiteAsianAfricanAmerInd Figure 5-4. A two-dimensional plot of the Malar Tubercle Frequency table. See Appendix A for an explanation of the individual character states. -1.0-0.50.00.51.01.52.0 -1.0-0.50.00.51.01.5 PCA1PCA2 narro w inte termedia w ide WhiteAsian f ricanAmerInd Figure 5-5. A two-dimensional plot of the Nasal Aperture Width Frequency table. See Appendix A for an explanation of the individual character states. 88

PAGE 89

-1.5-1.0-0.50.00.51.0 -1.0-0.50.00.51.01.52.02.5 PCA1PCA2 R ouOvalVaultedSemi_triTriang nd u WhiteAsianAfricanAmerInd Figure 5-6. A two-dimensional plot of the Nasal Bone Structure Frequency table. See Appendix A for an explanation of the individual character states. -1.0-0.50.00.51.01.5 -2-101 PCA1PCA2absen tpresentuno b White A sianAfricanAmerInd Figure 5-7. A two-dimensional plot of the Nasal Overgrowth Frequency table. See Appendix A or an explanation of the individual character states. 89

PAGE 90

-2.0-1.5-1.0-0.50.00.51.0 -10123 PCA1PCA2 ab p resunob WhiteAsianAfricanAmerI n Figure 5-8. A two-dimensional plot of the Post-bregmatic Depression Frequency table. See Appendix A for an explanation of the individual character states. -1.0-0.50.00.51.01.52.0 -1.0-0.50.00.51.01.52.0 PCA1PCA2 absentweakmstron mediu W hiteAsianAfricanAmerInd Figure 5-9. A two-dimensional plot of the Posterior Zygomatic Tubercle Frequency table. See Appendix A for an explanation of the individual character states. 90

PAGE 91

-1012 -1.5-1.0-0.50.00.51.0 PCA1PCA2 open c l_vcl_ooblit W hiteAsia n AfricanAmerInd Figure 5-10. A two-dimensional plot of the Supranasal Suture Frequency table. See Appendix A for an explanation of the individual character states. -1012 -101234 PCA1PCA2 Obliteratedangledsmooths-sha p WhiteAsianAfrican m erInd Figure 5-11. A two-dimensional plot of the Zygomaticomaxillary Suture Frequency table. See Appendix A for an explanation of the individual character states. 91

PAGE 92

Figure 5-12. OSSA scores for all individuals. Note: Cutoff value at 2 (Whites) and 3 (Blacks). Figure 5-13. Plot of first two canonical scores for four-group and three-group DFA. Kernel density histogram is superimposed on axes. 92

PAGE 93

211915131087654321 k-Nearest Neigbors 0.240.210.180.150.120.090.06 Clasification Error (%) TotalWhiteBlack Figure 5-14. Comparison of Error Rates of k nearest neighbors. Figure 5-15. Unconstrained PCOs and the constrained CAP analysis. 93

PAGE 94

CHAPTER 6 CONCLUDING REMARKS A Legacy and A Solution The legacy of nonmetric cranial traits in the determination of ancestry historically relied on the experience of the observer and a set of typological trait lists never empirically supported. Approaching the analysis of nonmetric traits in this way results in a priori conclusions drawn from some overall impression of the cranial Gestalt. The current method of nonmetric, macromorphoscopic trait analysis has changed very little from the efforts of Hooton and his students. This historical approach is ineffective and does not meet the Daubert guidelines, especially as the method is currently employed. Part of this may be that the traits used by forensic anthropologists simply can not be described by extreme trait values generally associated with a particular race. Hefner (2003b) demonstrated that when using only five traits (INA, IOB, NAW, NBS, and PBD) individuals expressing all of the trait values associated with their ancestral group ranged from 17% for American Blacks to 51% for American Whites. This suggests that either these traits do not work or the true extent of variation in these traits has never been properly explored. The suggestion by some researchers (Rhine 1990) that idiosyncratic variation and racial admixture is a possible cause for some of the low observed frequencies of these traits is unfounded and represents a typological approach. Couple to this the inclusion of expected trait values contradicting the actual observed frequencies, features like the post-bregmatic depression remain in the suite of traits used to determine ancestry, but without validation. This research has established empirically that the typical nonmetric traits of African-, Asian-, and European-derived groups are not found at the frequencies suggested by forensic anthropologists such as Bass, Byers, Gill, and Rhine. While these traits were valued by Hooton 94

PAGE 95

for their use in ancestry prediction, neither he nor any of his students published trait frequencies for any of the groups, let alone were there attempts to describe human variation. The extreme trait expressions for these nonmetric traits are not very reliable for estimating ancestry using the traditional visual approach. Using classification statistics and nonmetric traits shows great promise, particularly for discriminating between American Blacks and Whites. In the last chapter, the nonmetric approach to ancestry determination using traits explicated by Hooton and Rhine was tested within a statistical frameworka framework which provides the certainty, replicability, and reliability needed to meet the Daubert challenge. More importantly, using these statistical approaches to determine ancestry removes some of the subjectivity inherent in the human judgment process, a process that is fundamentally flawed when practiced without some objective way to assess the value of individual traits or character states. All of the statistical methods presented herein work better than the traditional approach of visual estimation and guesswork. Each of these methods highlights some important aspect of the multivariate data while enabling the practicing forensic anthropologist to predict the ancestry of a cranium from nonmetric traits. The samples selected for analysis reflect a conscious effort to document variation. The inclusion of modern American Blacks and Whites was a conscious one, while still including historic samples like the Terry Collection in order to document as much variation in nonmetric traits as possible for a large and diverse a sample. It is not surprising that in many of these statistical approaches, the Native American and East Asian samples were often classified into each other when using a four-way analysis. Many of their trait frequencies are similar, but the correspondence analysis can be used to highlight the traits where discrimination between the two groups would be more possible. These results also reflect the poor resolution of 95

PAGE 96

nonmetric traits compared to metrics analysis which can separate East Asian and Native American groups exceptionally well. More research is needed in this area of nonmetric trait analysis. Although the effects of polygenic traits on the expression of each character state are still not fully understood, the significance of nonmetric traits in the determination of ancestry is clearthey work well within a statistical framework relying neither on the experience of the observer nor on a typological, paradigmatic form of an ancestry. In a statistical approach, the variation and frequency of each trait is considered to be a part of the larger statistical model. A large number of these traits are significantly correlated (Hefner 2003), and so the models which assume independence among the variables (e.g., Nave Bayesian, OSSA) may be weakened. In that same vein, the linear discriminant function analysis assumes multivariate normality and equality of variances. While the sample in the current study is large enough to assume normality, the DFA may produce unrealistic posterior and typicality probabilities because non-normal distribution does not guarantee a sectioning point positioned to optimize group differences. However, if the function completely separates groups, it may be used no matter what the distribution is, with the caveat that the posterior and typicality probabilities may be ambiguous, at best. However, the cross-validated classification rates support the use of each of these models, since quite simply, they classify unknown individuals well ( Table 6-1 ). Each of these statistical approaches has a classification rate above 83%, and several achieve nearly 90%. Although the CAP analysis classifies 97% of the American Blacks, the method misclassifies 19% of the American Whites, and thus lowers the overall percent correct. This pattern emerged for several of the analyses. The Nave Bayesian and k-NN approaches were the exception, classifying 87% and 88% of both samples correctly. However, the Bayesian 96

PAGE 97

approach assumes independence and weighs each trait equally. In so doing, the Bayesian approach fails to account for traits like the Post-bregmatic Depression, which is present in 44% of the American Blacks and 17% of the American Whites. More robust methods, like DFA and LR, consider the contribution of each trait and adjust the function coefficients accordingly. The OSSA statistic correctly classified 83% of a sample of American Blacks and American Whites, by weighing each variable equally and simply summing the binary trait scores which were heuristically derived from the original ordinal trait scores. This method shows the greatest promise here because of the simplicity of the approach and because ordinal variables are compressed to a binary scale, interand intra-observer error is minimized interobserver error. As it stands now, the nonmetric approach to ancestry determination is a typological approach that relies on human judgment. While this is an uncertain process, this research has shown that the uncertainty inherent in this method can be understood and explained using several statistical methods and by acknowledging the variation observed among populations. Recognizing these shortcomings of the current method and applying new approaches is both scientific and effective and should serve as a basis for the replacement of the traditional approach: classifications based on general impressions from the cranial Gestalt with little more than post hoc validation of the nonmetric traits. 97

PAGE 98

Table 6-1. Classification rates for two-way analysis of each method. Percent Correct Method Black White Overall ORA 93 85 89 OSSA 78 90 83 Nave 87 87 87 DFA (sw) 88 90 89 k-NN 88 88 88 LR (ord) 92 86 89 CAP 97 81 86 sw = Step-wise variable selection ord = Ordinal model 98

PAGE 99

APPENDIX A MACROMORPHOSCOPIC TRAITS Anterior Nasal Spine (ANS) One of the problems confronting the biological anthropologist when assessing an anterior nasal spine (ANS) is its extreme fragility. This area is often damaged either perior postmortem. Those crania exhibiting trauma, pathology (including alveolar resorption), or postmortem damage to the overall inferior nasal margin will be excluded from the analysis. The anterior nasal spine ( Figure A-1 ) is scored progressively as slight, intermediate, and marked. 1) Slight: minimal-to-no projection of the anterior nasal spine beyond the inferior nasal aperture. 2) Intermediate: a moderate projection of the anterior nasal spine beyond the inferior nasal aperture. 3) Marked: a pronounced projection of the anterior nasal spine beyond the inferior nasal aperture. Inferior Nasal Morphology (INA) Inferior nasal morphology is defined as the most inferior portion of the nasal aperture, which, when combined with the lateral alae, constitutes the transition from nasal floor to the vertical portion of the maxillae, superior to the anterior dentition. Inferior nasal aperture (INA) is an assessment of the shape of the inferior border of the nasal aperture just lateral to the anterior nasal spine, which defines the transition from nasal floor to the vertical portion of the maxillae. Bilateral asymmetry may occur. If so, the left side is used. The morphology of INA ranges from an inferior slope with no delineation of the inferior border (1) to a sharp, vertical ridge of bone, or nasal sill (5). INA ( Figure A-2 ) is scored as follows: 1) An inferior sloping of the nasal floor which begins within the nasal cavity and terminates on the vertical surface of the maxilla, producing a smooth transition. The morphology is distinct from INA 2 regarding the more posterior origin and the greater slope of INA 1; 2) Sloping of the nasal aperture beginning more 99

PAGE 100

anteriorly than in INA 1, and with more angulation at the exit of the nasal opening; 3) The transition from nasal floor to the vertical maxilla is not sloping, nor is there an intervening projection, or sill. Generally, this morphology is a right angle, although a more blunted form may be observed; 4) Any superior incline of the anterior nasal floor, creating a weak (but present) vertical ridge of bone that traverses the inferior nasal border (partial nasal sill); 5) A pronounced ridge (nasal sill) obstructing the nasal floor-to-maxilla transition. Interorbital Breadth (IOB) Interorbital breadth is a macromorphoscopic trait that could be measured with calipers using the defined measurement dacryon to dacryon (Howells 1973, 1989), rather than scored nonmetrically. Interorbital breadth ( Figure A-3 ) is assessed in the following nonmetric states: 1) narrow, 2) intermediate, and 3) broad. This assessment is made relative to the facial skeleton. Malar Tubercle (MT) The malar tubercle (MT) ( Figure A-4 ) is a caudally protruding tubercle located on the inferior margin of the maxilla and zygomatic bone in the region of the zygomaticomaxillary suture. The presence of malar tubercle is scored by placing a transparent ruler at the intersection of the zygomaticomaxillary suture and the inferior margin of the malar to the deepest point on the curvature of the maxilla. An assessment is then made on the extent of protrusion beyond the ruler's edge. In instances where the suture is directly on the tubercle, the ruler is placed from the deepest curvature of the maxilla to the deepest anterior curvature on the zygomatic. It should be noted that a malar tubercle may be present on the maxilla, the zygomatic, or along the zygomaticomaxillary suture. Observers should not consider the tubercles on the lateral zygomatic arch. A completely absent MT is rare. MT is scored as follows: 0) No projection of bone; 1) A trace tubercle below the ruler's edge (roughly 2 mm or less); 2) A medium protrusion 100

PAGE 101

below the ruler's edge (roughly 2 to 4 mm); 3) A pronounced tubercle below the ruler's edge (roughly 4mm or more). Nasal Aperture Shape (NAS) The shape of the nasal aperture (NAS) is assessed by observing both the lateral contours of the nasal aperture and the position of greatest lateral projection of the margin [see red arrows in illustrations]. NAS ( Figure A-5 ) is scored as follows: 1) Teardrop lateral projection intermediate to 2 and 3 (see illustration). 2) Bell shape greatest lateral projection at the inferior margin. 3) Bowed greatest lateral projection at midline. Nasal Aperture Width (NAW) The width of the nasal aperture ( Figure A-6 ) is assessed relative to the facial skeleton. It is scored as 1) narrow, 2) medium, or 3) broad. Nasal Bone Contour (NBC) Nasal Bone Contour (NBC) ( Figure A-7 )is defined as the contour of the midfacial region (particularly the contour of the nasal bones and the frontal process of the maxilla) approximately 1 cm below nasion. Visual interpretation of nasal contour is not the most effective manner of analysis due to high interand intra-observer error. The use of a contour gauge permits a more reliable and consistent assessment of nasal contour. To assess NBC, the cranium is placed in a position that allows the observer to gently, but with consistent and balanced pressure, place the contour gauge directly on the nasal bones approximately 1 cm inferior to nasion, while maintaining the gauge roughly perpendicular to the palate and parallel to the orbits. NBC is scored as follows: 0) Low and rounded nasal bone contour. 1) An oval contour, with elongated, high, and rounded lateral walls; NBC 1 presents a circular shape and lacks steep walls. Brues (1990) suggests the term Quonset-hut to describe this shape, although the term is somewhat dated; 2) Steep lateral walls and a broad (roughly 7 mm or more), flat superior surface "plateau," 101

PAGE 102

noted on the contour gage as a flat cluster of needles in the midline; 3) Steep-sided lateral walls and a narrow superior surface "plateau"; 4) Triangular cross section, lacking a superior surface "plateau". Nasal Bone Shape (NBS) Nasal bone shape (NBS) ( Figure A-8 ) is assessed from the anterior view with the cranium positioned in approximate anatomical position. A determination is made regarding 1) the position of a nasal pinch, if any, and 2) the amount of lateral bulging. While making the assessment, the observer should not consider the frontonasal suture, the nasal suture, or the symmetry of the nasal bones. Rather, an assessment is made of the lateral contours of the nasal bones. Nasal bone shape is assessed as follows: 1) None: Nasal bones with no nasal pinch. The nasal bones may be wide or narrow. 2) Minimal Pinch: Nasal bones with a superior pinch and minimal lateral bulging. 3) Pronounced: Nasal bones with a superior pinch and pronounced lateral bulging of the inferior region. 4) Triangular: Triangular-shaped nasal bones. Nasal Overgrowth (NO) Nasal overgrowth (NO) ( Figure A-9 )is defined as an inferior projection of the lateral border of the nasal bones beyond the maxillae at nasale inferious. Assessment of nasal overgrowth does not include anterior bulging of the nasal bones. Observations should be made of the left side if it is undamaged. If the left side is damaged, the right side may be substituted. If both nasal bones are missing or fractured (ante-, peri-, or postmortem), the trait is not scored. It is often useful to run a finger along the borders of the maxilla and nasal bones near nasale inferious to determine whether a projection is present. Nasal overgrowth is scored dichotomously as 0) absent or 1) present. 102

PAGE 103

Post-bregmatic Depression (PBD) Post-bregmatic depression ( Figure A-10 )is a slight to broad depression along the sagittal suture, posterior to bregma, which is not the result of pathology. Observed in lateral profile, the trait is scored as either 0) absent (no depression) or 1) present. Posterior Zygomatic Tubercle (ZT) The posterior zygomatic tubercle (ZT), or marginal process, ( Figure A-11 )is a posterior projection of the zygomatic bone at approximately midorbit as viewed in lateral plane. To observe the various degrees of expression, a small, transparent ruler is placed on the frontal process of the zygomatic from the landmarks frontomalare posteriale to jugale. The extent of bony protrusion beyond the ruler's edge is then assessed. ZT is scored as follows: 0) No projection of bone; 1) A weak projection of bone (less than 4 mm); 2) A moderate projection of bone (approximately 4 to 6 mm); 3) A marked projection of bone (generally > 6 mm). Supranasal Suture (SPS) In adult crania, a secondary complex suture may persist, which is generally referred to as the supranasal suture, or sutura supranasalis ( Figure A-12 ). This suture does not represent the nasal portion of a persistent metopic suture, which is generally a single, non-oscillating line, but rather is the fusion of the nasal portion of a frontal suture. SPS appears as a complex of interlocking bony spicules at glabella. SPS is scored as follows: 0) Completely obliterated; 1) Open (unfused); 2) Closed, but visible. Transverse Palatine Suture Shape (TPS) The course of the transverse palatine suture (TPS) ( Figure A-13 ) is highly variable, although certain themes persist. TPS is not scored unilaterally, although asymmetrical sutures are not uncommon. The entire suture is observed, but the medial one-half in the region of the palatine suture is most closely scrutinized. When an asymmetrical suture is present (the two 103

PAGE 104

branches of the suture do not come into contact at midline) the general theme is recorded (e.g., straight, jagged, etc.). Slight undulations of the suture should not be considered when making a determination. If the suture is obliterated, it is not scored. TPS is scored as follows: 1) The suture crosses the palate perpendicular to the median palatine suture, with no significant anterior or posterior deviations. If the right and left halves of the suture do not contact each other at midline, but the suture is otherwise straight, score the suture as a 1; 2) The suture crosses the palate perpendicular to the median palatine suture, but near this juncture a significant anterior deviation, or bulging, is present. If the right and left halves of the suture do not contact each other, but the suture is otherwise bulging anteriorly, a score of 2 is used; 3) The suture crosses the palate, but deviates anteriorly and posteriorly (e.g., M-shaped) in the region of the median palatine suture. If the right and left halves of the suture do not contact each other, but the suture is otherwise jagged, a score of 3 is used; 4) The suture crosses the palate perpendicular to the median palatine suture, but near this juncture a posterior deviation, or bulging, is present. Zygomaticomaxillary Suture Shape (ZS) The zygomaticomaxillary suture (ZS) ( Figure A-14 ) is the suture between the maxilla and the zygomatic. The course of the suture is best observed in the anterior view. In instances of asymmetrical manifestations, the left side is preferred. The infraorbital suture should be ignored when making a determination. Assessment of ZS is based primarily on the approximate location of greatest lateral projection of the suture, and also on the number of major angles present. ZS is scored as follows: 0) A suture with no angles and greatest lateral projection at the inferior margin of the malar. Sutures having greatest lateral projection at the inferior margin, but a slight angle near the midpoint of the suture should be scored as 0; 1) A suture with one angle and greatest 104

PAGE 105

lateral projection near the midline; 2) A suture with two or more angles (presenting a jagged and/or S-shaped appearance) with variable greatest lateral projection. The figure shows both S-shaped and jagged courses of the suture. 105

PAGE 106

Figure A-1. Character states for the Anterior Nasal Spine morphology. Figure A-2. Character states for the Inferior Nasal Aperture morphology. 106

PAGE 107

Figure A-3. Character states for the Interorbital Breadth. Figure A-4. Character states for the Malar Tubercle morphology. 107

PAGE 108

Figure A-5. Character states for the Nasal Aperture shape. Figure A-6. Character states for the Nasal Aperture Width. 108

PAGE 109

Figure A-7. Character states for the Nasal Bone Contour. Figure A-8. Character states for the Nasal Bone Shape. 109

PAGE 110

Figure A-9. Character states for the Nasal Overgrowth. Figure A-10. Character states for the Post-bregmatic Depression. 110

PAGE 111

Figure A-11. Character states for the Posterior Zygomatic Tubercle. Figure A-12. Character states for the Supranasal Suture. Figure A-13. Character states for the Transverse Palatine suture. 111

PAGE 112

Figure A-14. Character states for the shape of the Zygomaticomaxillary Suture. 112

PAGE 113

APPENDIX B R CODE Correspondence Analysis #Correspondence Analysis of nonmetric data #By: Joseph T. Hefner #GetCA Correspondence Analysis Is modified from Kenneth Portier's GetCa #Feel free to use it in any capacity library(MASS) GetCa
PAGE 114

Nave Bayesian ###Naive Bayesian ###Script adapted from R code on CRAN website bayes
PAGE 115

res
PAGE 116

LIST OF REFERENCES Agur, AMR (1991) Grant's Atlas of Anatomy. 9th ed. Cambridge: William & Wilkins Press. Anderson, JE (1969) Skeletal "anomalies" as genetic indicators. In, Skeletal Biology Symposia for the Study of Human Biology. Edited by D. Brothwell. Elmsford, New Jersey: Pergamon Press. Anderson, MJ (2004) CAP: A computer program. Auckland, New Zealand. Anderson, MJ and TJ Willis (2003) Canonical analysis of principal coordinates: a useful method of constrained ordination for ecology. Ecology 84: 511. Angel, JL and JO Kelly (1990) Inversion of the posterior edge of the jaw ramus: New race trait. In Skeletal Attribution of Race: Methods for Forensic Anthropology, edited by Gill and Rhine, pp. 33. Albuquerque, New Mexico: Maxwell Museum of Anthropology. Armelagos, G (1992) The Concept of Race, Racism, and Anthropology. Paper presented at the WennerGren Foundation for anthropological research symposium. Baja California, Mexico. Baker, SJ, GW Gill and DA Keiffer (1990) Race and sex determination from the intercondylar notch of the distal femur. In Skeletal Attribution of Race: Methods for Forensic Anthropology, edited by Gill and Rhine, pp. 91. Albuquerque, New Mexico: Maxwell Museum of Anthropology. Bass, WM (1987) Human Osteology: A Laboratory and Field Manual. 3 rd Edition. Columbia: Missouri Archaeological Society. Bendyshe, T (1969) The Anthropological treatises of Johann Blumenbach. Publications for the Anthropological Society. London: Longman Green Press. Berry AC (1975) Factors affecting the incidence of nonmetrical skeletal variants. J Anat 120: 519. Berry, RJ and AC Berry (1967) Epigenetic variation in the human cranium. J Anat 101:361. Blumenbach, JF (1775) De Generis Humani Varietate Nativa Liber. Gottingae, Germany: Bergman Publishers. Blumenbach, JF (1781) De Generis Humani Varietate Nativa Liber. 2 nd Edition. Gottingae, Germany: Bergman Publishers. Blumenbach, JF (1795) De Generis Humani Varietate Nativa Liber. 3 rd Edition. Gottingae, Germany: Bergman Publishers. Boas, F (1910) Changes in bodily form of descendants of immigrants, United States Immigration Commission (p. 113). Washington, D.C.: Government Printing Office. 116

PAGE 117

Brace, CL (1992a) A four letter word called "race." In, Race and Other Miscalculations and Mismeasures: Papers in Honor of Ashley Montague, edited by LJ Reynolds and L Lieberman. New York: General Hall Publishers. Brace, CL (1992b) Modern human origins and the dynamics of regional continuity. In, Prehistoric Mongoloid Dispersals, Symposium 1992, edited by T. Akazawa and E. Szathmary. New York: Oxford University Press. Brace, CL (2005) race is a FourLetter Word. London: Oxford Press. Brinton, DG (1890) Races and People. New York: NDC Hodges Press. Brooks, S, RH Brooks, and D France (1990) Alveolar prognathism contour, an aspect of racial identification. In Skeletal Attribution of Race: Methods for Forensic Anthropology, edited by G. C. Gill and S. Rhine, pp. 41. Albuquerque, New Mexico: Maxwell Museum of Anthropology. Brothwell, DR (1981) Digging Up Bones. 3 rd Edition. Ithaca, New York: Cornell University Press. Brues, AM (1990) The once and future diagnosis of race. In Skeletal Attribution of Race: Methods for Forensic Anthropology, edited by Gill and Rhine, pp. 1. Albuquerque, New Mexico: Maxwell Museum of Anthropology. Buikstra, J (1972) Techniques for coping with the ageregressive nature of nonmetric traits. Paper presented before the American Association of Physical Anthropologists. Buikstra, J (1974) NonMetric Traits: The Control of Environmental Noise. Paper presented before the American Association of Physical Anthropologists, Amherst, Massachusetts. Buikstra, J and D Ubelaker (1995) Standards for Data Collection from Human Skeletal Remains. Proceedings of a Seminar at the Field Museum of Natural History. Arkansas Archeological Survey Research Series #44, Arkansas. Burns, K (1999) Forensic Anthropology Training Manual. New Jersey: PrenticeHall. Burris, BG and EF Harris (1998). Identification of race and sex from palate dimensions. J Forensic Sci 43:959. Byers, SN, SE Churchill, and B Curran (1997) Identification of EuroAmericans, AfroAmericans, and Amerindians from palatal dimensions. J Forensic Sci 42:3. Cheverud, J (1981) Phenotypic, Genetic, and Environmental Morphological Integration in the Cranium. Evolution 36:499. Christensen, A (2004) The Impact of Daubert : Implications for Testimony and Research in Forensic Anthropology (and the Use of Frontal Sinuses in Personal Identification). J Forensic Sci 49:1. 117

PAGE 118

Clemente, CD (1975) Anatomy: A Regional Atlas of the Human Body. Lea & Feibeger: Philadelphia. Coenders, G and WE Saris (1995) Categorization and measurement quality. The choice between Pearson and Polychoric correlations. In, The MultitraitMultimethod Approach to Evaluate Measurement Instruments, edited by WE Saris & Mnnich, pp. 125. Etvs University Press: Budapest. Cohen, J (1960) A coefficient of agreement for nominal scales, Educational and Psychological Measurement 20: 37. Comas, R (1961) Scientific Racism Again? Current Anthropology 2(4): 310. Coon, CS (1962) The Origins of Races. New York: Knopf. Count, EW (1950) This is Race. New York: Henry Schuman. Craig, EA (1995) Intercondylar shelf angle: a new method to determine race from the distal femur. J Forensic Sci 40:777. Darwin, C (1859) On the Origin of Species. John Murray Press: London. Darwin, C (1871) The Descent of Man. London. Daubert v. Merrell Dow Pharmaceuticals, Inc., 509 U.S. 579 (1993). DiBennardo, R and JV Taylor (1983). Multiple discriminant function analysis of sex and race in the postcranial skeleton. Am J Phys Anthropol 61: 305. Dror, IE, D Charlton, and AE Peron (2005) Contextual information renders experts vulnerable to making erroneous identifications. For Sci Int 156: 74. Duray, SM, HB Morter and FJ Smith (1999) Morphological variation in cervical spinous processes: potential applications in the forensic identification of race from the skeleton. J Forensic Sci 44: 937. Enlow, DH (1975) Handbook of Facial Growth. Philadelphia: Saunders. England, M (1996) Facial growth and the ontogeny of morphological variation. Journal of Zoology, London 264:37. Falconer, DS (1965) The inheritance of liability to certain diseases, estimated from distances among relatives. Ann Hum Genet 29: 51. Federal Rules of Evidence (1975); 2000 Finnegan, M and SA McGuire (1979) Classification systems for discrete variables used in forensic anthropology. Am J Phys Anthropol 51: 547. 118

PAGE 119

Fleiss JL. (1971) Measuring nominal scale agreement among many raters. Psychological Bulletin. 76:378-81. Gill, GW (1990) Introduction. In, Skeletal Attribution of Race: Methods for Forensic Anthropology, edited by G. W. Gill and S. Rhine, pp. viixii. Albuquerque, New Mexico: Maxwell Museum of Anthropology. Gill, GW (1995). Challenge on the frontier: discerning American Indians from whites osteologically. J Forensic Sci 40: 783. Gill, GW (1998) Craniofacial Criteria in the Skeletal Attribution of Race. In, Forensic Osteology, 2 nd edition, ed. Kathleen Reichs. Springfield, Illinois: Charles C. Thomas. Gill, GW and R Gilbert (1990) Race identification from the midfacial skeleton: American blacks and whites. In, Skeletal Attribution of Race: Methods for Forensic Anthropology, edited by G. W. Gill and S. Rhine, pp. 47. Albuquerque, New Mexico: Maxwell Museum of Anthropology. Gill, GW and S Rhine (1990) Skeletal Attribution of Race: Methods for Forensic Anthropology. Albuquerque, New Mexico: Maxwell Museum of Anthropology. Goss, CM (1973) Gray's Anatomy. 29th ed. Philadelphia: Lea & Febiger Press. Gravlee CC, Bernard HR, and Leonard WR (2003) Heredity, environment, and cranial form: a reanalysis of Boas's immigrant data. American Anthropologist 105:125. Grneberg, H (1963) The Pathology of Development: A Study of Inherited Skeletal Disorders in Animals. Blackwell, Oxford Press. Hanihara, T (1996) Comparison of craniofacial features of major human groups. Am J Phys Anthropol 99: 389. Hanihara, T (2000) Frontal and facial flatness of major human populations. Am J Phys Anthropol 111:105. Harris, M (1968) Rise of Anthropological Theory. New York: Thomas Y. Crowell Company. Hastie R and RM Dawes (2001) Rational Choice in an Uncertain World: The Psychology of Judgment and Decision Making. 2 nd Edition. California: Sage Publications. Hauser, G and GF De Stefano (1989) Epigenetic Variants of the Human Skull. Stuttgart, Germany: E. Schweizerbart'sche Verlagsbuchhandlung. Hefner, JT (2003a) Assessing Nonmetric Cranial Traits Currently used in the Forensic Determination of Ancestry. Masters Thesis, The University of Florida, Gainesville, FL. 119

PAGE 120

Hefner, JT (2003b) The utility of nonmetric cranial traits in ancestry determination Part II. Proceedings of the 55th Annual meeting of the American Academy of Forensic Sciences, Chicago, IL Hefner, JT, PD Emanovsky, J Byrd, SD Ousley (2007) The Value of Experience, Education, and Methods in Ancestry Prediction. Proceedings of the 59 th Annual meeting of the American Academy of Forensic Sciences, San Antonio, TX. Hefner, JT and SD Ousley (2006) Morphoscopic Traits and the Statistical Determination of Ancestry II. Proceedings of the 58 th Annual meeting of the American Academy of Forensic Science, Seattle, WA. Hinkes, MJ (1990) Shovelshaped incisors in human identification. In, Skeletal Attribution of Race: Methods for Forensic Anthropology, edited by Gill and Rhine, pp. 21. Albuquerque, New Mexico: Maxwell Museum of Anthropology. Holliday, TW and AB Falsetti (1999). A new method for discriminating AfricanAmerican from EuropeanAmerican skeletons using postcranial osteometrics reflective of body shape. J Forensic Sci 44: 926. Hooton, EA (no date) The Test. Unpublished manuscript by EA Hooton. Peabody Museum archives. Cambridge, MA. Hooton, EA (1926) Lecture notes of EA Hooton. Peabody Museum archives. Cambridge, MA. Hooton, EA (1930) The Indians of Pecos Pueblo. Yale University Press, New Haven. Hooton, EA (1931) Lecture notes of EA Hooton. Unpublished manuscripts of the Peabody Museum archives of E. A. Hooton. Cambridge, MA. Hooton, EA (1946) Lecture notes of EA Hooton. Unpublished manuscripts of the Peabody Museum archives of E. A. Hooton. Cambridge, MA. Howells, WW (1970) Multivariate analysis for the identification of race from the crania. In Personal Identification in Mass Disasters, edited by T. D. Stewart, pp. 111. Washington, D. C: Smithsonian Institution Press. Howells, WW (1973) Cranial Variation in Man: A Study by Multivariate Analysis of Patterns of Difference Among Recent Human Populations. Papers of the Peabody Museum of Archaeology and Ethnology, Harvard University, vol. 67. The Museum: Cambridge, MA. Howells, WW (1989) Skull Shapes and the Map: Craniometric Analyses in the Dispersion of Modern Homo. Papers of the Peabody Museum of Archaeology and Ethnology, Harvard University, vol. 79. The Museum: Cambridge, MA. Howells, WW (1995)Who's Who in Skulls: Ethnic Identification of Crania from Measurements. Papers of the Peabody Museum of Archaeology and Ethnology, Harvard University, vol. 82. The Museum: Cambridge, MA. 120

PAGE 121

Iscan, MY (1983). Assessment of race from the pelvis. Am J Phys Anthropol 62: 205. Isan, MY (1990) A comparison of techniques on the determination of race, sex, and stature from the Terry and HamannTodd Collections. In, Skeletal Attribution of Race: Methods for Forensic Anthropology, edited by Gill and Rhine, pp. 73. Albuquerque, New Mexico: Maxwell Museum of Anthropology. Ishida, H and Y Dodo (1997) Cranial variation in prehistoric human skeletal remains from the Marianas. Am J Phys Anthropol 104: 399. Jantz, RL (1970) Change and variation in skeletal populations of Arikara Indians. Unpublished Doctoral Dissertation, Department of Anthropology, University of Kansas, Lawrence. Jantz, RL (2001) Cranial change in Americans:1850. J Forensic Sci 46: 784. Kant, E (1775) On the Different Races of Man. In, This is Race, ed. Count, EW. New York: Henry Schuman. Konigsberg, LW (1990) Analysis of prehistoric biological variation under a model of isolation by geographic and temporal distance. Hum Bio 62: 49. Krogman, WM (1939) A guide to the identification of human skeletal remains. FBI Law Enforcement Bulletin 3: 3. Krogman, WM and MY Isan (1986) The Human Skeleton in Forensic Anthropology. 2 nd Edition. Springfield, Ill: Charles C. Thomas Press. Krzanowski, WJ (2002) Principles of Multivariate Analysis: A Users Perspective. Oxford University Press: London. Lahr, MM (1996) The Evolution of Modern Human Diversity: A Study of Cranial Variation. Cambridge: Cambridge University Press. Landis, JR and Koch, GG (1977) The measurement of observer agreement for categorical data. Biometrics. 33:159. Laughlin, WS and JB Jorgenson (1956) Isolate Variation in Greenlandic Eskimo Crania. Acta Genet Stat Med 6: 3. Lewis, CS (1952) The Chronicles of Narnia, The Voyage of the Dawn Treader. Loeb Classical Library. LviStrauss, C (1952) Race et histoire. Paris, UNESCO. Paris: Gonthier. Linnaeus, CV (1759) Systema Naturae. Lugduni Batavorum: Theodorum Haak. MacBride, R. (1967). The Automated State: Computer Systems as a New Force in Society. Philadelphia: Chilton Book Company. 121

PAGE 122

Marks, J (1995) Human Biodiversity: Genes, Race, and History. New York: Aldine de Gruyter. Meadows, L and RL Jantz (1995). Allometric secular change in the long bones from the 1800s to the present. J Forensic Sci 40: 762. Meehl, P (1986) Clinical Versus Statistical Prediction: A Theoretical Analysis and Review of the Literature. Minneapolis: University of Minnesota Press. Michael, JS (1988) A New Look at Mortons Craniological Research. Current Anthro 29(2): 349. Montagu, A (1963) The Concept of Race. London: CollierMacmillan Limited. Montagu, A (1966) A Handbook of Anthropometry. Springfield: Charles C. Thomas. Montagu, A (1974) Man's Most Dangerous Myth: The Fallacy of Race. 5 th Edition. London: Oxford University Press. Mooney, M and M Siegel (2002) Understanding Craniofacial Anomalies: The Etiopathogenesis of Craniosynostoses and Facial Clefting. Canada, Wiley and Sons. Moore, KL (1992) Clinically Oriented Anatomy. 3 rd Edition. Baltimore: Williams & Wilkins. Morton, S.G. 1834 Illustrations of pulmonary tuberculosis. Philadelphia, Key and Biddle. Moss, M (1956) The functional matrix concept and its relationship to temporomandibular joint dysfunction and treatment. Dent Clin North Am 27(3):445. Mulhern, D (2001) Probable Case of Binder Syndrome in a Skeleton From Quarai, New Mexico. Am J Phys Anthropol 118: 371. Ossenberg, NS (1970) The Influence of Artificial Cranial Deformation on Discontinuous Morphological Traits. Am J Phys Anthropol 33: 357. Osterburg, JW and JF Coates (1967) Crime, science, and technology. Science 158: 1079. Ousley, SD and JT Hefner (2005) The Statistical Determination of Ancestry. Proceedings of the 58th Annual meeting of the American Academy of Forensic Sciences, Seattle, WA. Ousley, SD and RL Jantz (1996) FORDISC 2.0: Personal Computer Forensic Discriminant Functions. Knoxville: The University of Tennessee. Ousley, SD and RL Jantz (1998) The forensic data bank: Documenting skeletal trends in the United States. In, Forensic osteology: Advances in the Identification of Human Remains, edited by KJ Reichs, pp. 441. Springfield, IL: Charles C. Thomas. Ousley, SD and RL Jantz (2002) Social races and human populations: Why forensic anthropologists are good at identifying races. Paper presented at the 71st annual meeting of the American Association of Physical Anthropologists, Buffalo, NY. April 2002. 122

PAGE 123

Pearson, K (1925) The Definition of the Alveolar Point. Biometrika 17: 53. Perizonius, WRK (1979a) Nonmetric cranial traits: Sex difference and age dependence. J Hum Evol 8: 679. Popkin, RH (1974) The philosophical basis of modern racism. In, Philosophy and the Civilizing Arts, eds. C Walton and JP Anton, pp126 R 2.4.0 (2007) The R Foundation for Statistical Computing. CRAN. http://cran.rproject.org/ Rao, CR (1989) Statistics and Truth: Putting Chance to Work. 2 nd Edition. London, World Scientific. Relethford JH (2003) Boas and beyond: migration and craniometric variation. Am J Hum Biol. 16: 379. Rhine, S (1990) Nonmetric Skull Racing. In Skeletal Attribution of Race: Methods for Forensic Anthropology, edited by Gill and Rhine, pp. 7. Albuquerque, New Mexico: Maxwell Museum of Anthropology. Rhine, JS and HR Campbell (1980). Thickness of facial tissues in American blacks. J Forensic Sci 25: 847. Richtsmeier, JT, VB Deleon, SR Lele (2002) The promise of geometric morphometrics. Yearbook of Phys Anthropol 45:63. Rohlf, F and L Marcus (1993) A revolution in morphometrics. TREE 8:129 Ross, AH, RL Jantz and WF McCormick (1998). Cranial thickness in American females and males. J Forensic Sci 43: 267. Ross, AH, H McKeown and LW Konigsberg (1999). Allocation of crania to groups via the "new morphometry." J Forensic Sci 44: 584. Ross, AH, DH Ubelaker, and AB Falsetti (2002) Craniometric variation in the Americas. Human Biology. In press. Rushton, JP (1999) Race, evolution, and behavior: A lifehistory perspective (2nd special abridged edition). Port Huron, MI: Charles Darwin Research Institute. Sampson, HW, JL Montgomery, and GL Henryson (1991) Atlas of the Human Skull. College Station, TX: Texas A&M University Press. Sauer, NJ (1992) Forensic anthropology and the concept of race: If races don't exist why are forensic anthropologists so good at identifying them? Social Science and Medicine 34(2):107. Saunders, S (1989) Nonmetric skeletal variation. In, Reconstruction of Life from the Skeleton, edited by M.Y. Isan and K. A. R. Kennedy, pp. 95. New York: Alan R. Liss. 123

PAGE 124

Sciulli, PW and RM Gramly (1989). Analysis of the Ft. Laurens, Ohio, skeletal sample. Am J Phys Anthropol 80: 11. Shanklin, E (1994) Anthropology & Race. Belmont, CA: Wadsworth Publishing Company Smedley, A (1993) Race in North America: Origin and Evolution of a Worldview. Boulder, CO: Westview Press. Snow, CC, S Hartman, E Giles, and FA Young (1979) Sex and race determination of crania by calipers and computer: A test of the Giles and Elliot discriminant functions in 52 forensic science cases. J Forensic Sci 24: 448. Sparks CS, and Jantz RL (2003) A Reassessment of Human Cranial Plasticity: Boas Revisited. Proceedings of the National Academy of Sciences 99:14636. Spencer, F (1981) Charter members of the American Association of Physical Anthropologists. Am J Phys Anthropol 56(4): 531. Spencer, F (1997) Anthropometry. In, History of Physical Anthropology: An Encyclopedia, edited by F. Spencer, pp. 80. New York: Garland Publishing, Inc. SPSS 14.0 (2005) SPSS, Inc. Student Version. Steele, DG and CA Bramblett (1988) The Anatomy and Biology of the Human Skeleton. College Station, TX: Texas A&M University. Stewart, TD (1979) Essentials of Forensic Anthropology. Springfield, IL: Charles C. Thomas. SYSTAT 9.0 (1998) SPSS, Inc. Standard Version. Szathmary, EJ and NS Ossenberg (1978) Are the biological differences between North American Indians and Eskimos truly profound? Current Anthrop 19: 673. Thiele, K (1993) The holy grail of the perfect character: The cladistic treatment of morphometric data. Cladistics 9: 275. Thompson, DD (1982) Forensic anthropology. In, A History of American Physical Anthropology, edited by F. Spencer. New York: Academic Press. Tyrell, AJ (2000) Skeletal nonmetric traits and the assessment of inter and intra population diversity: Past problems and future potential. In, Human Osteology in Archaeology and Forensic Science, edited by M Cox and S Mays. London: Greenwich Medical Media. Tyrell, AJ and AT Chamberlain (1998) Nonmetric trait evidence for modern human affinities and the distinctiveness of Neanderthals. Journal Human Evol 34: 549. Ubersax, JS (2000) Statistical modeling of expert ratings on medical treatment appropriateness. J Am Stat Assoc 88: 421. 124

PAGE 125

Washburn, S (1983) Evolution of Teacher. Annual Review Anthrop 12:1. Walker, GF (1976). The computer and the law: Coordinate analysis of skull shape and possible methods of postmortem identification. J Forensic Sci 21: 357. Whittaker, DK and MM Bakri (1996). Racial variations in the extent of tooth root translucency in aging individuals. Arch Oral Biol 41: 15. Wolpoff, MH and R Caspari (1997) Race and Human Evolution: A Fatal Attraction. New York: Simon and Schuster. Wright, S (1930) Review of Fisher's "The Genetical Theory of Natural Selection." J Heredity 21: 349. 125

PAGE 126

BIOGRAPHICAL SKETCH Joseph Timothy Hefner was born in Hickory, NC, on the 15 th day of September in 1973. Eighteen years later, he graduated from Hunter Huss High School. In 1997, Joseph received a B.S. in anthropology from Western Carolina University in Cullowhee, NC. Following three years of work at Mercyhurst College under Dr. Dennis C. Dirkmaat, Joseph matriculated to the University of Florida in Gainesville to study human variation and forensic anthropology with Dr. Michael W. Warren. While at Florida, Joseph served as a research assistant for the C. A. Pound Human Identification Laboratory and received both his M.A. and Ph.D. in anthropology. As of this writing, Joseph is looking for a job. 126