|UFDC Home||myUFDC Home | Help|
This item has the following downloads:
1 PATTERNS OF CRANIOMETRIC VARIATION IN MODERN THAI POPULATIONS: APPLICATIONS IN FORENSIC ANTHROPOLOGY AND IMPLICATIONS FOR POPULATION HISTORY By LAUREL ELIZABETH FREAS 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 2011
2 2011 Laurel Elizabeth Freas
3 To my family, who have always believed and understood, even and perhaps, most especially when I have not
4 ACKNOWLEDGMENTS Though often held as the embodiment of solitary scholarly endeavor, anyone who has ever erous, yet largely silent, assistance on the parts of many, many people. For this dissertation, I extend my deepest thanks to the following: My deepest debt of gratitude is owed t o the members of my superlative advisory committee, Michael Warren (chair), David Daegling, John Krigbaum and Martha Burt, for providing a steadying hand, an endless supply of encouraging words, and the occasional motivating kick to the hindquarters, always right on time when I needed them most. If I am a better anthropologist to day, it is for having learned at their elbows and stood on their shoulders. I am additionally forever indebted to Mike for his outstanding mentorship, for being my biggest cheerleader and tireless advocate, for his generous support of my many, many trips to Thailand and, most of all, for his patient understanding of the peculiar workings of my brain, and for always knowing exactly what I needed to hear. It has been a very, very long road from first year graduate student to this completed dissertation, and he has never once allowed me to falter along the path. I am equally indebted t o the people of Thailand, for the extraordinary generosity, hospitality, and friendship they have continuously extended to this farang in their midst. Chief among these are Dr. Porntip Rojanasunan and her staff at the Central Institute of Forensic Sciences, for first suggesting this research project, and for opening the first doors; and Dr. Pasuk Mahakkanukrauh and her faculty in the De partment of Anatomy at Chiang Mai University, for for their constant support during my research trips. Additional special thanks are owed to Dr.
5 Pasuk for her enthusiasm for this project, for her indomitable spirit, for her kindness shown to a student far from home, and for her friendship. A huge goes also to her students, Aun, Khempith and Nim, for the countless dinners, bowling games, daytime sight seeing trips, and nighttime motorbike rides around Chiang Mai that they shared with me. Thanks are also owed to Dr. Sukkid Yasothornsrikul and Ajarn Chanasorn Poodendaen at Naresuan University, who, in lections also provided my first critical introductions into the Thai academic community. Dr. Panya Tuamsuk provided generous access to his collections at Khon Kaen University, allowing me to salvage the core question of this dissertation just when it seem ed all was lost. And, finally, I owe an immeasurable debt of gratitude to Noon for being my indefatigable tour guide, translator, cultural emissary, language coach, dinner company, research assistant and, above all else, my friend. Gratitude i n endless measure is owed to the gang from the Pound Lab Carlos, Nico, Katie, Traci and Ron for tailgatoring, MarioKart (Diddy Kong for the win!), Red vs. Blue, and jamming out on the lab floor; for the countless hours at the maceration hoods, the inside j okes when I was really, really cranky which was a lot ; and for always having my back. They have been my second family, always there to pick me up, dust me off and push me on down the road when the going got tough. Other members of my second family, to whom similar thanks are owed, include Carrie, Joe, Foxy, Derek, and Owen who know that when the going gets weird, the weird turn pro. If I have survived this with my sanity intact, it is through no strength of my own, but because of all of them. Many thanks go to Elizabeth Walters and Joe Hefner for their tireless help with the minutiae of proper usage ( both the grammatical and the statistical kind s),
6 for rea ding rough drafts and providing encouraging and thoughtful comments, and for at least havi ng pretended to enjoy the chore Gracious thanks are owed to the many parties who provided generous financial support for this project. This research has been supp orted in part by the Ellis R. Kerley Forensic Sciences Anthropology Additional travel support and access to the collections at Khon Kaen was provided by t he Joint Prisoner of War /M issing I n A ction (PO W/MIA) (JPAC CIL) 1 Last, but certainly not least, I own an immeasurable debt of gratitude to my family, for more things than I can possibly enumerate here. A partial list includes their unflagging support and unwavering confidence in me; their fearlessness in encouraging me to follow a path that has taken me far from home, and the sacrifices they have made that I might follow that path; and for their lessons that to do is to do whole heartedly. Most of all, I thank them for giving me roots and wings. 1 The opinions contained herein are solely those of the author and do not reflect the opinions or official position of the Joint POW/MIA Accounting Command or the Department of Defense.
7 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ ............... 4 LIST OF TABLES ................................ ................................ ................................ ........................... 9 LIST OF FIGURES ................................ ................................ ................................ ....................... 11 ABSTRACT ................................ ................................ ................................ ................................ ... 13 CHAPTER 1 INTRODUCTION AND STATEMENT OF QUESTION ................................ ..................... 15 2 PREHISTORY AND POPULATION HISTORY IN THAILAND, PART I: THE BASIC CULTURAL OUTLINE ................................ ................................ ............................ 24 The Pleistocene to the Paleolithic ................................ ................................ ........................... 27 Holocene Hunter Gatherers ................................ ................................ ................................ .... 29 The Arrival of Agriculture ................................ ................................ ................................ ...... 43 The Metal Age ................................ ................................ ................................ ........................ 48 The Bronze Age ................................ ................................ ................................ ............... 48 The Iron Age ................................ ................................ ................................ .................... 53 The Rise of the State ................................ ................................ ................................ ............... 56 3 PREHISTO RY AND POPULATION HISTORY IN THAILAND, PART II: THE VARIED LINES OF EVIDENCE ................................ ................................ .......................... 67 Archaeology ................................ ................................ ................................ ............................ 77 Linguistics ................................ ................................ ................................ ............................... 82 Population Genetics ................................ ................................ ................................ ................ 98 Physical Ant hropology ................................ ................................ ................................ ......... 112 4 MATERIALS AND METHODS ................................ ................................ ......................... 150 The Samples ................................ ................................ ................................ .......................... 151 Statistical Methods ................................ ................................ ................................ ................ 152 Basic Assumptions ................................ ................................ ................................ ........ 153 Data transformations and non parametric approaches ................................ ........... 156 Multivariate Statistical Analyses ................................ ................................ ................... 158 Analysis of variance ................................ ................................ ............................... 163 Discriminant function analysis ................................ ................................ ............... 164 Final Considerations ................................ ................................ ................................ ...... 171 5 RESULTS AND DISCUSSION ................................ ................................ ........................... 179
8 Discriminant Function Analysis By Sex ................................ ................................ ............ 179 Discriminant Function 1: All Variables ................................ ................................ ....... 179 Discriminant Function 2: 20 variables ................................ ................................ ......... 186 Discriminant Function 3: Forward Selected Stepwise Analysis ................................ .. 189 Discriminant Function Analysis By Regional Subpopulation ................................ ........... 192 6 SUMMARY AND CONCLUSIONS ................................ ................................ ................... 227 APPENDIX: DEFINITIONS OF CRANIOMETRIC LANDMARKS AND MEASURMENTS ................................ ................................ ................................ ................ 241 Definitions of Cranial Landmarks ................................ ................................ ........................ 241 Definitions of Cranial Measurements ................................ ................................ ................... 244 LIST OF REFERENCES ................................ ................................ ................................ ............. 246 BIOGRAPHICAL SKETCH ................................ ................................ ................................ ....... 265
9 LIST OF TABLES Table page 4 1 Standard cranial measurements used in this study. ................................ .......................... 172 4 2 Summary statistics of modern Thai population samples. ................................ ................ 176 4 3 Summary of Shapiro Wilk tests of normality for individual craniometric variables. ..... 178 4 4 ............ 178 5 1 Means, standard deviations, and number of missing cases, for 24 cranial measurements for the modern Thai sample, pooled by sex. ................................ ............ 208 5 2 Means and standard deviations for 24 cranial measurements for the modern Thai sample, by sex and population. ................................ ................................ ........................ 209 5 3 Independent samples t tests for equality of group means, by sex, for 24 craniometric variables. ................................ ................................ ................................ .......................... 210 5 3 Con tinued. ................................ ................................ ................................ ........................ 211 5 4 Rank transformed independent samples t tests for equality of group means, by sex, for 14 craniometric variables. ................................ ................................ .......................... 211 5 5 Discriminant Function Analysis (DFA) 1. ................................ ................................ ....... 212 5 6 Structure matrix for Discriminant Function 1. ................................ ................................ 212 5 7 Unstandardized canonical coefficients for Discriminant Function 1. .............................. 213 5 8 Cl assification results for Discriminant Function 1. ................................ ......................... 214 5 9 in Discriminant Function Analysis 2. ................................ ................................ .................. 214 5 10 Structure matrix for Discriminant Function 2. ................................ ................................ 215 5 11 Unstandardized canonical coefficients for Discriminant Function 2. .............................. 215 5 12 Classification results for Discriminant Function 2. ................................ ......................... 215 5 13 Stepwise selection matrix for variables included in Discriminant Function 3. ............... 216 5 14 Unstandardized canonical coefficients for Discriminant Function 3. .............................. 217 5 15 Classification results for Discriminant Function 3. ................................ ......................... 217
10 5 16 One way analysis of variance (ANOVA) results comparing from three regio nal Thai subpopulations, males only. ................................ ................................ ............................. 219 5 16 C ontinued. ................................ ................................ ................................ ........................ 220 5 17 Structure matrices for discriminant function analysis of regional subpopulations, by sex, based on 20 craniometric variables. ................................ ................................ ......... 221 5 18 Classification results for males, by regional subpopulation, from discriminant function analysis using 20 variables. ................................ ................................ ............... 223 5 19 Classification results for females, by regional subpopulation, from discriminant function analysis using 20 variables. ................................ ................................ ............... 224 5 20 Summary of stepwise discriminant function analyses of Culled Samples (CS) 1 through 4 (males only). ................................ ................................ ................................ .... 224 5 21 Unstandardized canonical coefficients for stepwise discriminant function analysis of Culled Sample 4. ................................ ................................ ................................ .............. 225 5 22 Cross validated classification results for stepwise discriminant function analyses of Culled Samples 1 through 4. ................................ ................................ ............................ 226
11 LIST OF FIGURES Figure page 1 1 Political map of East and Southeast Asia and Near Oceania ................................ ............ 23 2 1 Selected Paleolithic archaeological sites in Southeast Asia. ................................ ............. 63 2 2 Selected Neolithic archaeological sites in Southeast Asia. ................................ ................ 64 2 3 Selected Metal Age archaeological sites in Thailand. ................................ ....................... 65 2 4 Selected state level archaeological sites in Southeast Asia. ................................ .............. 66 3 1 Selected archaeological sites in China. ................................ ................................ ............ 148 3 2 Relationships among the various language families of Southeast Asia. .......................... 149 4 1 Anterior views of the skull, detailing cranial measurements. ................................ .......... 173 4 2 Left lateral view of the skull detailing cranial measurements. ................................ ....... 174 4 3 Inferior view of the skull, detailing cranial measurements. ................................ ............. 175 4 4 Vernacular sociopolitical regions of Thailand ................................ ................................ 177 5 1 Frequency distribution of male and female discriminant function scores for Discriminant Function 1. ................................ ................................ ................................ 213 5 2 Frequency distribution of male and female discriminant function scores for Discriminant Function 2. ................................ ................................ ................................ 216 5 3 Frequency distribution of male and female discriminant function scores for Discriminant Function 3. ................................ ................................ ................................ 2 18 5 4 Scatter plot of discriminant function scores for males, by regional subpopulation, for discriminant function analysis using 20 craniometric variables. ................................ ..... 222 5 5 Scatter plot of discriminant function scores for females, by regional subpopulation, for discriminant function analysis using 20 craniometric variables. ............................... 223 5 6 Scatter plot of discriminant function scores for males in Culled Sample 4, by regional subpopulation, for stepwise discriminant function analysis selecting 3 variables. ................................ ................................ ................................ .......................... 225 6 1 Sign in front of the International Center Hostel at Chiang Mai Univers ity, with text in the Thai (top), Lanna (middle), and Roman scripts ................................ .................... 239
12 6 2 Great Stupa of Wat Suandok, Chiang Mai; an example of Lanna style religious architecture ................................ ................................ ................................ ...................... 240
13 A bstract of D issertation P resented t o the Graduate School of the University o f Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy PATTERNS OF CRANIOMETRIC VARIATION IN MODERN THAI POPULATIONS: APPLICATIONS IN FORENSIC ANTHROPOLOGY AND IMPICATIONS FOR POPULATION HISTORY By Laurel Elizabeth Freas December 2011 Chair: Michael Warren Major: Anthropology Craniometric analyses are a long standing and well developed part of physical on patterns of intra and interpopulational human biological variation. Thus, craniometri c analyses find overlapping application to forensic anthropology identifications and the exploration of biological relationships among temporally and geographically distinct human populations. Yet the validity of such analyses depends upon having populati on specific reference datasets of sufficient robusticity to adequately characterize the true range of morphological variation within and among human populations. The 2004 Indian Ocean tsunami and other recent events have created the need for greater repre sentation of Southeast Asian populations in modern forensic craniometric databases. Additionally, it is argued that resolution of the central question in Southeast Asian prehistory that of the essential biological continuity or discontinuity between the p re and post Neolithic populations of the region will only be advanced through more focused sampling of the salient populations, to include consideration of patterns of regional and temporal variation within them.
14 To address these issues, 24 standard crani al measurements were collected on a large sample (n=429; 242 males/137 females) of modern Thai skeletons from three different regions within Thaila nd. These data were used to generate to generate linear discriminant functions for sex determination in unid entified remains from within the Thai population. These functions have cross validated accuracy rates of 77 83%, similar to the performance of other population specific sex determination standards, indicating their validity and broad utility to forensic anthropology. The Thai craniometric data were also used to explore patterns of regional within population skeletal variation in juxtaposition to observed patterns of phenotypic variation within the living Thai population. Comparison of the three regional subpopulations found no significant craniometric differ ences among them, despite their apparent ethnic, linguistic and physical diversity, indicating the relative craniometric homogeneity of the modern Thai population This finding is congruent with know n historical events that have clearly shaped the biological structure of this population but that are often overlooked by prehistorians. Given this greater understanding of craniometric variation within the modern Thai, it is hoped that commensurate gains in forensic anthropology applications and population history investigations will follow.
15 CHAPTER 1 INTRODUCTION AND STA TEMENT OF QUESTION A little neglect may breed mischief ... for want of a nail, the shoe was lost; for want of a shoe the horse was lost; and for want of a horse the rider was lost. Benjamin Franklin Poor Richard's Almanac (1758) Metric approaches to the analysis of human vari ation (i.e., osteometry and anthropometry) are among the oldest scientific traditions in the field of physical anthropology. Despite an ignoble intellectual adolescence, which saw metric analyses in general and craniometrics in specific applied tacitly an d overtly in support of racist sociopolitical and pseudoscientific agendas (Gould 1996), these methods have matured into a robust and respected part of modern Ellio t 1962, 1963; Howells 1973, 1989, 1995; Buikstra and Ubelaker 1994; Moore Jansen et al. 1994; Pietrusewsky 2000, 2006; Jantz and Ousley 2005; Buikstra and Komar 2008; SWGANTH 2011). Craniometrics now sees regular application to questions of human growth and development (Baughan and Demirjian 1978; Humphrey 1998), bioarchaeological analyses (Buikstra and Ubelaker 1994; Jantz and Owsley 2001; Pietrusewsky and Douglas 2002 Oxenham and Tayles 2006), human paleontology and fossil taxonomy (Kramer 1993; Matsumura and Pookajorn 2005), regional and global geographic and temporal patterns of human mo rphological variation and genetic diversity (Relethford 1994, 2002; Relethford and Harpending 1994; Jantz and Jantz 2000; Jantz and Owsley 2001; Roseman 2004; Roseman and Weaver 2004), population history and relationships among temporally and geographicall y distinct populations (Howells 1973, 1989; 1995; Pietrusewsky et al. 1992; Hanihara 1993; Nakbunlung 1994; Brace and Tracer 1995; Jantz 2004; Spradley 2006), and forensic identification (Moore Jansen et al. 1994; Jantz and Ousley 2005; Slice and Ross 2004 ). The dissertation research presented herein embraces the
16 last two of these uses of craniometric analyses, by considering the forensic applications and the population history implications of patterns of craniometric variation within the modern population of Thailand. Between these two uses of craniometric analyses, it is the matter of forensic analysis that has provided the immediate impetus for this research. Most of the modern analytical standards used by forensic anthropologists for determining biol ogical profiles (sex, ancestry, age at death and stature) were developed using data collected from populations originating wholly within the United States: primarily, the Terry and Hamann Todd skeletal collections, the remains of U.S. military personnel k (Buikstra and Komar 2008). However, as forensic anthropology has become increasingly active in the global arena identify the victims of natural disasters, wars and human rights violations and as our understanding of the true nature of human population variation has grown, so too has our realization that these American forensic anthropology standards are largely ina dequate for application to other populations around the world (Rankin and Moore 2004; Slice and Ross 2004; Ubelaker 2004; Schaefer and Black 2005; Steadman and Haglund 2005; Djuric et al. 2007; Buikstra and Komar 2008). As is true of all scientific reasoning, our ability to make predictions of sex, ancestry, age and stature using forensic anthropology methods is constrained by the representativeness of the skeletal samples which underpin these methods. Because the curren t forensic anthropology standards are based on a relatively small, restricted subsample of the total global human population, they often rapidly diminish in precision, accuracy and reliability the farther one moves from this population. This is especially true of discriminant function analyses of cranial
17 measurements (the preferred metric method for purposes of sex and ancestry determination), in which both the classification of an unknown individual and the statistical certainty of that classification (in terms of posterior and typicality probabilities) is driven entirely by patterns of variation within and among the reference populations from which the discriminant functions have been calculated (Howells 1973, 1989, 1995; Pietrusewsky 2000; Jantz and Owsl ey 2001; Buikstra and Komar 2008). Reliable sex determination standards that are appropriate for the population(s) under consideration are of the utmost importance to forensic anthropology, as the determination of all other parameters of the biological pr ofile is predicated on an accurate determination are critical to identification efforts, as ancestry or race is the second most dominant criterion, after biological sex, by which individuals are classified and identified. The 2004 Indian Ocean tsunami, which killed nearly 250,000 people in South and Southeast Asia, including approximately 8200 in Thailand, is but one of a series of natural disasters and human rights viola tions which have laid bare this need for regionally based, population specific forensic anthropology standards (Ubelaker 2004; Schaefer and Black 2005; Steadman and Haglund 2005; Djuric et al 2007; Buikstra and Komar 2008 ; Kenyhercz 2010 ). Although genet ic and dental comparisons are being applied with considerable success to the still ongoing identification efforts in Thailand, the lack of and/or loss of antemortem DNA (deoxyribonucleic acid) exemplars and dental records means that these methods are not a pplicable in all cases. This dilemma has led the Thai government to look to other areas of forensic science, including forensic anthropology, for alternative means of identifying the dead. To date, however, only three papers discussing the need for Thai specific forensic standards have et al. 1998; King et al. 1998; King
18 1999). Importantly, these studies clearly demonstrate that the available American standards are inaccurate when applied to the Thai population and in such applications will lead to systematic misclassifications of Thai males as females due to population differences in patterns of sexual dimorphism an d overall skeletal robusticity. Nevertheless, t hese studies are an insuffici ent remedy to lack of appropriate population specific standards for the Thai, as they only address sex determination from postcranial skeletal elements, which are generally regarded as less informative and reliable than the mor e robust craniometric analyse s (but see Spradley and Jantz 2011 for evidence to the contrary) The impact of the analytical constraints imposed by this lack of representative datasets is not limited to forensic anthropologists working in foreign contexts. These constraints also hav e significant influence on forensic anthropologists working to identify unknown individuals within the United States. This is particularly true if an unknown individual belongs to an ancestry group not represented in the FORDISC 3.0 reference database (Ja ntz and Ousley 2005). For example, the total Asian sample in the current FORDISC 3.0 reference database comprises only 100 Japanese males and females, 79 Chinese males and 51 Vietnamese males leaving a great many Asian populations completely undocumented, as well as grossly underrepresenting the total range of morphological variation within those populations that are documented. In such a diverse population as exists within the United States, with immigrant populations from nearly every nation and ethnic group on Earth, the potential for misidentifications simply for want of representative datasets looms large in much forensic anthropology case work. In light of these international and domestic considerations, there is a clear need for expanded representa tion of Asian populations in general, and the Thai population in specific, in the craniometric reference databases and analytical standards used by forensic anthropologists.
19 ute of Forensic Science, explicitly for the purposes of developing a Thai specific craniometric database for use both in Thailand and the United States, it also presents valuable opportunities for the investigation of broader physical anthropological quest ions pertaining to population history of the modern Thai people. There is little that is conclusively known about the origins of the modern Thai and other Southeast Asian populations; prevailing theories hold that these peoples either 1) evolved in situ from the indigenous prehistoric hunter gatherer populations; or 2) are descended from a wave of rice agriculturalists who expanded southward from the Yangzi Valley around 4500 years ago; and 3) that the Thai may be uniquely derived from peoples who fled in to Thailand from southern China in advance of the Mongol expa nsion in the thirteenth century AD, replacing or absorbing other existing populations (Briggs 1945; Rinehart 1981; Nakbunlung 1994; Higham and Thosarat 1998; Baker and Phongpaichit 2005; Lertrit et al. 2008). However, there is considerable debate and little agreement on this matter among archaeological, linguistic and historical lines of evidence. Anthropological and scientific common sense would suggest that this is reflective of a much more co mplex pattern of population origins and history. 1 (Figure 1 1), it has known cultural and linguistic, and therefore likely biological, ties with diverse peoples from throughout Asia. Among t hese are: the peoples of India, who, in addition to trade goods, brought with them the religious traditions of Hinduism and Buddhism and the linguistic traditions of Sanskrit and Pali, all of which are still evident in Thailand today; 1 Following most authorities working in this region, Southeast Asia is defined here as the area encompassing modern day Cambodia, Laos, Myanmar (Burma), Thailand, Vietnam (i.e., mainland southeast Asia), Malaysia, Indonesia, and the Philippines (i.e., island Southeast Asia) (Bellwood 1997; Higham 2002; Pietrusewsky 2006).
20 the peoples of China who exerted their influence through historically and archaeologically documented trade networks and population movements, and also though hypothesized linguistic influences. the Khmer civilization of the Mekong Valley itself heavily influenced by India w hich contributed documented and hypothesized political, architectural, religious and linguistic influences (Briggs 1945; Kaplan 1981; Rinehart 1981; Nakbunlung 1994; Bellwood 1997; Stark and Allen 1998; Higham and Thosarat 1998; Higham 2002; Baker and Phon gpaichit 2005). Today, the modern Thai population displays a surprising amount of phenotypic variation, much of which is said to follow broad regional patterns within the country. Thus, Thais often draw fairly reliable colloquial inferences about a pers within Thailand on the stated basis of their phenotypic appearance (skin tone, eye shape, facial features and build), in much the same way Americans use regional speech accents to ascertain where a person was raised. As noted above, craniometric analyses are often used to examine the biological relationships between geographically and temporally distinct populations via biological distance statistics. Such analyses draw broad methodological and theoretical sup port from research that has demonstrated the high precision and repeatability of cranial measurements; the moderate to strong heritability of craniometric dimensions; and the generally strong correlation between patterns of phenotypic and genetic variance when examined in adequately large samples (Howells 1973; Susanne 1977; Sjvold 1984; Cheverud 1988; Buikstra et al. 1990; Relethford and Harpending 1994; Konigsberg and Ousley 1995; Pietrusewsky 2000, 2006; Relethford 2002; Roseman 2004; Carson 2006). Thu s, biological distances based on craniometrics are frequently used and widely accepted as a proxy for genetic relationships among populations when dealing et al. 1990:1; Howel ls 1973; Pietrusewsky 2000, 2005; Relethford 2002; Roseman 2004). Previous studies have attempted to place archaeological and modern Thai skeletal series in relation to other Asian
21 populations using biological distance statistics, typically either Euclide 2 distances (Brace and Tracer 1992; Pietrusewsky et al. 1992; Hanihara 1992, 1993, 1996, 2006; Nakbunlung 1994; Pietrusewsky 2000, 2005, 2006; Matsumura and Hudson 2005; Matsumura 2006; among many others). The results of these studies, not surprisingly, generally group the Thai with other mainland Southeast Asian populations and place them close to East Asian and island Southeast Asian populations, yet most differ considerably in the fine details of the clustering patterns among the pop ulations. However, these studies frequently consider the archaeological or modern populations of Thailand as something of a monolith, using a single (often small) archaeological or modern sample from a single locale as a proxy for the entire population a nd paying no attention to the the reconstruction of population relationships. In only few studies are multiple contemporary populations from Thailand consi dered (Matsumura and Hudson 2005; Matsumura 2006; Pietrusewsky 2006), with multiple discriminant function and biological distance analyses reveal ing considerable differences among the three best represented archaeological skeletal series from Thailand (from the sites of Ban Chiang, Non Nok Tha and Khok Phanom Di). These results also suggest several different relationships between these individual arc haeological samples and geographically proximate non Thai populations, further suggesting that there may be significant intrapopulation diversity and regional geographic patterning within the modern Thai population. In light of the foregoing discussion, th is dissertation will seek to explore patterns of craniometric variation in the modern Thai population, including patterns of sexual dimorphism, as well as variation among regional subpopulations within the modern population of Thailand.
22 This exploration w ill address the proximate, practical need for Thai specific forensic anthropology standards and normative databases for sex and ancestry determination, which will find use within the forensic science community not only in Thailand but in the rest of Southe ast Asia and the United States, as well as hopefully provide some insight into physical anthropology questions of broader academic interest regarding the population history of the modern Thai.
23 Figure 1 1. Political map of East and Southeast Asia and Near Oceania; Thailand is highlighted in red ( Map sourc e: C entral Intelligence Agency (C IA ) World FactBook 2 010; image used with permission )
24 CHAPTER 2 PREHISTORY AND POPUL ATION HISTORY IN THA ILAND, PART I: THE BASIC CULTURAL OUTLINE As for wha t concerns the origine of the Siameses, it would be difficult to judge whether they are a single people, directly descended from the first men that inhabited the countrey of Siam, or whether in the process of time some other nation has not also settled the re, notwithstanding the first inhabitants. S. de la Loub re A New Historical Relation of the Kingdom of Siam (1693 :14 ) (Quoted in Higham 2002:18) Scholarly questions about the origins of the modern population of Thailand date back at least as far as the earliest European contact with the peoples of the Kingdom of Siam 1 in the sixteenth and early seventeenth centuries (Higham 1996, 2002; Baker and Phongpaichit 2005). As is broadly true of all questions of population origins, recurrent among inquiries int o the origins of the Thai is the consideration of whether the peoples of this country evolved in situ from long standing prehistoric populations, or whether they are the result, wholly or partly, of population migrations from other nearby regions (Nakbunlu ng 1994; Higham and Thosarat 1998; Higham 2002; Baker and Phongpaichit 2005; Matsumura and Hudson 2005; Pietrusewsky 2006). For example, F. Marcello de Ribadeneyra, a seventeenth century Portuguese missionary to the Siamese capital of Ayutthaya, remarked, came from the great city which is situated in the middle of a desert in the kingdom of i.e., Angkor (de Ribadeneyra 1601; quoted in Higham 2002:17). Indeed, the question of Thai population o rigins is firmly embedded within the larger discussion of similar indigenous versus migratory (or, endogenous versus exogenous) influences in the population history of Southeast Asia as a whole (Pietrusewsky 1981, 1990, 2006; Turner 1987, 1990; 1 T t this dissertation to refer to the land and its peoples, except when making is more appropriate and accurate, as it is here.
25 Bellwood 19 97; Matsumura and Hudson 2005; Hanihara 2006; Matsumura 2006). In turn, the examination of population history in Southeast Asia is particularly interesting because, in many ways, it is an intense distillation of the larger question that pervades the study of modern human origins in general: does the biological (i.e., fossils, modern skeletons, genetics) and artifactual continuity among populations through time debate between the Recent African Origins model and the Multiregional Evolution model of modern human origins structures much of the discourse surrounding the origins of the modern populations of Southeast Asia (Pietrusewsky 1990, 2006; Bellwood 1997; Howells 1997; Matsumura and Hudson 2005, Matsumura 2006). It would be folly, however, to assume that questions of the origins and population history of the modern Thai are solely the province of Western scholars, or are questions of purely academic interest. Because Thailand, alone among the countries of Southeast Asia, was never subjected to colonial rule by a European nation, Tha i scholars were able to take leading roles in throughout the nineteenth and early twentieth centuries. At a time when academic enterprise in neighboring countries wa s monopolized by the occupying European colonials and manipulated in the service of their imperial agendas, historic and archaeological research in Thailand was promulgated from within and most importantly, from above via royal patronage as a means of pres erving and celebrating the cultural heritage of the Siamese/Thai people in the face of increasing foreign pressures (Nakbunlung 1994; Higham and Thosarat 1998; Higham 2002; Baker and Phongpaichit 2005). At the same time, the output of this home grown scho larship
26 was nevertheless enlisted variously, first by the monarchy and later by subsequent military regimes and elected governments, in support of prevailing and/or desired notions of Siamese/Thai nationalism and ethnic identity, and in the furtherance of specific internal political agendas (Baker and Phongpaichit 2005). In this regard, the work of these Thai scholars is as colored by the zeitgeist albeit from their own unique reflexive and internal perspective as is that of their Western contemporaries. These potential fluctuating biases, driven by changing perceptions of the past and its relationship to the present, must be always borne in mind when considering the insights that both Western and Thai scholars contribute to the understanding of the origin s of the modern Thai (Wood 1989; Stark and Allen 1998; Baker and Phongpaichit 2005). In order to proceed with an examination of the origins and history of the modern Thai history and its early historic period, with some reference to contemporaneous developments in adjacent areas of Southeast Asia. This framework can then be juxtaposed against what is known and what is hypothesized of the broader pattern of population histo ry in Southeast Asia. While an exhaustive survey of the relevant literature on the prehistory of Southeast Asia is beyond the scope of this chapter, every attempt is made to highlight the salient discoveries throughout the region that inform our understan ding of the prehistory of Thailand itself. The authoritative summary of the Prehistoric Thailand: From Early Settlement to Sukhothai (1998); the following discussion draws heavily on info Early Cultures of Mainland Southeast Asia (2002).
27 The Pleistocene to the Paleolithic Despite the fact that expanding Homo erectus populations must have passed through Thailand in order to have reached Indone sia as early as 1.8 to 1.6 million years ago (Rinehart 1981; Swisher et al. 1994; Higham and Thosarat 1998), no fossils of this hominid species have been recovered from Thailand. Rather, the earliest traces of human occupation in Thailand are the putative pebble core chopper tools, ostensibly produced by H. erectus which were discovered by Pope and colleagues at the riverine sites of Ban Mae Tha and Ban Don Mun and the limestone karst rock shelter site of Khao Pah Nam, all in Lampang Province in northern Thailand (Figure 2 1) (Pope 1985; Pope et al. 1978, 1986; Higham and Thosarat 1998). These tools, which are dated to between 800,000 and 600,000 years ago (Pope et al. 1986; Nakbunlung 1994), considerably antedate the earliest known H. erectus fossils fro m mainland Southeast Asia, the fossilized teeth found in Tham Khuyen Cave in Lang Son Province, northern Vietnam, and dated to ca. 475,000 years ago (Ciochon et al. 1996). Even if one prefers to accept the more conservative age estimates of 1 million to 800,000 years ago for the earliest Indonesian H. erectus fossils (Bellwood 1997; Klein 1999), the scant evidence of contemporary H. erectus populations in Southeast Asia has long been puzzling. Surely the rock shelters and karstic caves that would shelter later groups of hunter gatherers archaeological record, would have appealed equally to the earlier populations of H. erectus inhabiting this region, and would have pr oven equally favorable for the preservation of their remains. As with all gaps in the fossil record, the possible causes of this lacuna are manifold and speculative. Perhaps the populations of H. erectus moved through the region too quickly to leave a su bstantial fossil record behind; or maybe the relevant fossils have been lost to the constant fluctuations of the Southeast Asian shoreline. Regardless of the circumstances, the lack
28 of H. erectus fossils in mainland Southeast Asia will likely continue to play a role in the debate over the true antiquity of the migrations of this most contentious of fossil hominids. Indeed, it is frequently used as a line of evidence in support of a middle Pleistocene date, rather than terminal Pliocene/early Pleistocene d ate for the Indonesian erectines (Klein 1999; Higham and Thosarat 1998). Following this indirect appearance of H. erectus in Thailand, the archaeological record is silent for nearly three quarters of a million years, until the arrival of anatomically moder n Homo sapiens populations in the late Pleistocene. The first evidence of these groups of hunter gatherers in Thailand comes from the Lang Rongrien rock shelter, in western peninsular Thailand. Here, charcoal hearths dated to ca. 38,000 to 27,000 years b efore present (BP) are found in three basal occupation layers, accompanied by faunal remains and a basic flake and pebble tool industry (scrapers, knives and some possible pebble core choppers) created from local stone (Anderson 1990; Bellwood 1997; Ha Van Tan 1997; Higham and Thosarat 1998; Higham 2002). These cultural deposits likely represent repetitive cycles of transitory occupation and abandonment of the rock shelter by small, mobile hunter gatherer groups over the course of many millennia, in tune w ith diachronic fluctuations in environmental conditions and resource availability (Higham and Thosarat 1998:24 25). A flake and pebble tool industry similar to that at Long Rongrien has been reported from the deepest cultural deposits at the nearby rock s helter site of Moh Khiew (Bellwood 1997:160; Pookajorn 1991, 1994; Nakbunlung 1994; Matsumura and Pookajorn 2005). A single extended burial was found at the interface between the two deepest cultural levels; accelerator mass spectrometry ( AMS ) dates of ca 26,000 BP for charcoal samples from the grave make this individual one of the earliest known modern humans from mainland Southeast Asia (Matsumura
29 and Pookajorn 2005). The biological affinities of this individual will be discussed in Chapter 3. Other i nland sites throughout Southeast Asia have also yielded analogous, roughly contemporary flake and pebble tool complexes that have been ascribed to various regional traditions. These include the Nguom and perhaps the early Sonviian industries of Vietnam, d ated to slightly before, and slightly after 23,000 BP, respectively (Ha Van Tan 1997; Nakbunlung 1994; Higham 2002:31); the lithic assemblages found in peninsular Malaysia at the sites of Bukit Bunuh, dated to ca. 40,000 BP (Saidin 2006); and possibly the assemblage from Kota Tampan, which may date volcanic ash from the eruption of Toba is accepted (Bellwood 1997; Saidin 2006). Holocene Hunter Gatherers During the ter minal Pleistocene, these regionally varying lithic traditions gave way to a Hoa Binh province in northern Vietnam where the type specimens were originally discover ed in the 1920s (Colani 1927; Barth 1952; Solheim 1972; Higham 2002). The Hoabinhian industry, which is broadly dated to between 18,000 and 3000 BP, is known from inland rock shelter sites spread from Burma to Vietnam and from southern China to peninsular Malaysia and northeastern Sumatra. Its occurrence throughout Southeast Asia is thought to represent the expansion of hunter gatherer groups into unoccupied or previously abandoned locations during the period of climatic amelioration following the Last Gl acial Maximum (ca. 21,000 to 18,000 years ago) (Bellwood 1997:158 161; Bellwood 1993; Ha Van Tan 1997; Higham and Thosarat 1998; Higham 2002). Hoabinhian lithic assemblages are characterized by a predominance of pebble core tools (rather than flake tools ), particularly the distinctive unifacially worked river cobbles and so
30 sumatrali ths (Bellwood 1997:158; Ha Van Tan 1997; Higham 2002:33; Higham and Thosarat 1998). Late Hoabinhian assemblages typically include an increasing proportion of edge ground stone tools a late ) and the possible presence of early pottery forms, although the latter is somewhat debated (Bellwood 1997; Higham 2002). The relationship between the Hoabinhian and preceding lithic traditions is complex, and appears to vary regionally: in Vietnam, the Hoabinhian is thought to have emerged directly from the preceding Sonviian tradition ca. 18,000 BP (Ha Van Tan 1997:37); similar local continuities between the Hoabinhian and earlier lithic assemblages are also suggested for southern China and northern an d central Thailand (Bellwood 1997; Higham 1996; Ha Van Tan 1997). In contrast, in peninsular Malaysia and southern/peninsular Thailand, Bellwood (1997:160) describes a Tampan, Lang Rongrien and Moh Khiew, and the Hoabinhian complex, which does not appear at these sites until at least 13,000 BP. Nevertheless, Higham and Thosarat (1998:63 65) see the Hoabinhian in Thailand as a more or less direct continuation of the earl ier hunter gatherer traditions such as those known from Long Rongrien. In Thailand, the principal Hoabinhian sites are clustered throughout the Chao Phraya River catchment in the northern and central part of the country. These include the cave sites of Sa i Yok, Tham Ongbah, Ment Cave, Khao Talu and Heap Cave in the southern reaches of the Chao Phraya and the Kwae Noi River valleys (see Figure 2 1) (van Heekeren and Knuth 1967; Srensen 1979; Higham and Thosarat 1998; Higham 2002). To the north, in Mae Hongson province, the rock shelter sites of Spirit Cave (dated to ca. 9000 to 5500 BC), Banyan Valley Cave (3500 to 2000 BC) and Steep Cliff Cave (5 500 to 3500 BC) have all yielded extensive
31 Hoabinhian deposits from their lowest occupational layers (Gorman 1969, 1971, 1972; Reynolds 1992; Higham and Thosarat 1998; Higham 2002). Spirit Cave is a particularly important site, not only because of its tim e depth, but also because a number of cord marked and incised pottery sherds, polished quadrangular stone adzes and ground slate knives were recovered from the interface between the two uppermost Hoabinhian layers here. At the time of their discovery, the se were believed to be the earliest such tools from mainland Southeast Asia, and the earliest known pottery in the world (Solheim 1972:37). Gorman (1969:671, 673; 1971) initially ween the local Hoabinhian hunter gatherers and an incipient agricultural group around the end of the seventh millennium BC. innovative decisions to use flotation analysis and to screen the excavated deposits through a fine wire mesh, resulted in the recovery of an unprecedented quantity and diversity of faunal and (Go rman 1969; Solheim 1972; Higham and Thosarat 1998; Higham 2002). Among the recovered plant and animal species from Spirit Cave are almonds, betel nuts, butternut, candle nut, beans, peas, bottle gourds, water chestnuts, pepper and cucumbers, as well as th e remains of fish, crustaceans, mollusks, deer, pigs, monkeys and an array of small to medium sized arboreal and terrestrial mammals (Gorman 1969, 1971; Higham 2002). Gorman (1969:672) saw the contents of this assemblage particularly the beans, peas, cuc umber, gourd and chestnut despite his original characterization of this evidence of agriculture, especially the pottery and ground stone tools, as intrusive elements in the Hoabinhian deposits, Gorman (1969:673) was
32 convinced that the nascent agricultural tradition was of Southeast Asian origin. Therefore, the de the passive recipient of advancing cul ture via diffusion from elsewhere (i.e., China) Solheim (1972) took this interpretation a step beyond and proclaimed the artifacts and biological remains from Spirit Cave to be proxy evidence that Southeast Asia was the location of agriculture in the Levant by several millennia (Higham and Thosarat 1998). This transition, Solheim argued (1 agricultural complex of domesticated cultivars and livestock plus pottery during the time of the occupation of Spirit Cave. Interestingly, though, Solheim (1972:39) acknowledges that the earliest inhabitants of Spirit Cave possessed only portions of this agricultural suite, and would s arrived at Spirit Cave from can clearly d etect a facet of the continuity ve rsus discontinuity debate in Southeast Asia. Despite the obvious and (by their own pens) admittedly intrusive nature of the pottery and polished stone tools in the uppermost cultural layers at Spirit Cave, Solheim and Gorman circumlocute their way into pr offering this material as evidence for the local development and advancement of already been mentioned above. Solheim (1972:39 40), mere paragraphs after stat ing that the pottery and ground stone tools at Spirit Cave were introduced from elsewhere in Southeast Asia,
33 cannot separate its elements from their late Hoabinh to draw parallels between the Spirit Cave material and similar artifacts from Taiwan Ta 1997) and a similarity that Gorman (1969) observed as well and notes that both the material from Spirit Cave and the Ta AD 0. As is evident from the name, the defining chara cteristic of the Extensionistic period is the movement and expansion of populations; a perhaps overly broad age range notwithstanding, this name is actually a fairly apt acknowledgement of a dominant agent in middle Holocene Southeast Asia. Unfortunately, Solheim limited his posited population movements to those in which people moved out of the mountains (where the majority of prehistoric hunter gatherer sites in Southeast Asia are found) and into the plains and coastal regions where agriculture would be f avored and where later Neolithic, Bronze and Iron age sites would be discovered. He did not allow for the possibility of a real biological and cultural connection between the populations of Taiwan and mainland Southeast Asia, noting both the large geograp hic distance separating the two, and the contrast between Hoabinhian rock shelters and the open air Ta (Solheim 1972). My intention, however, is not to make straw men out of Gorman and Solheim, to whom we owe much of our understanding of S outheast Asian prehistory, and who were writing without the benefit of the illumination provided by nearly 40 years of further research and field work that have occurred since these articles were published. Rather, my point here is to demonstrate how the un dercurrent of endogenous versus exogenous origins and influences pervades scientific
34 inquiry into population history in Southeast Asia, even when not dealing directly with the human populations themselves. First impressions are telling, and Gorman and S incongruent presentations and initial interpretations of the evidence from Spirit Cave the recognition of clearly intrusive, likely exogenous elements in the archaeological record, and the seemingly irresistible inclination to shoe horn these elements into a locally continuous, endogenous evolutionary sequence that establishes Southeast Asia as a region of special cultural and historical significance demonstrates exactly how powerful this undercurrent is. As it stands now, the pott ery and ground stone tools from the upper layers at Spirit Cave have been determined to be of a much younger age than their archaeological context would suggest. Higham (2002) reports that these artifacts are now dated to ca. 2000 to 1000 BC well after th e Hoabinhian period on the basis of AMS dating of organic residues on one of the sherds. Discarded too is the notion that the Hoabinhians of Spirit Cave were practicing any form of well developed agriculture: the densely forested and steep terrain and th in acidic soils in the area are poorly suited to plant cultivation. Additionally, there are no traces of rice, taro or dogs all hallmarks of the Southeast Asian agricultural complex in the botanical and faunal remains recovered from Spirit Cave. Rather, the suite of plant and animal remains recovered from the cave are indicative of a sophisticated broad spectrum foraging subsistence strategy operating in a resource rich subtropical montaine/riverine environment (Gorman 1971:315; Hutterer 1983, 1998:67; Gl over and Higham 1996; Higham and Thosarat 1998; Higham 2002). On the other hand, the possibility that the Hoabinhians were selectively fostering the growth of some types of plants, such as beans, within the native environment cannot yet be definitively ex cluded (Bellwood 1997; Hutterer 1998; Higham 2002). In any case, sites such as Heap Cave and Khao Talu in the Kwae Noi River valley speak to the success of this hunting and gathering
35 subsistence strategy, as the Hoabinhian tradition persisted there until as late as 2500 BC practically to the eve of the advent of agriculture in Thailand (Rinehart 1981; Pookajorn 1990; Bellwood 1997). Other Hoabinhian sites in Thailand have yielded similar evidence for small, mobile groups of broad spectrum hunter gatherer p opulations. These include the previously mentioned Steep Cliff Cave and Banyan Valley Cave, both of which were excavated by Gorman to the same exacting standards as Spirit Cave. Banyan Valley Cave generated significant interest when it was reported that carbonized rice husks were recovered from the uppermost cultural levels and in association with edge ground adzes, slate knives and cord marked pottery similar to that from the upper layer at Spirit Cave (Reynolds 1992; Higham and Thosarat 1998; Hutterer 1998; Higham 2002). It was subsequently established by Yen (1977) that this rice is of a wild type and probably dates to 900 BC to AD 1000, as does the uppermost cultural layer in general. These dates are only slightly younger than the AMS dates of the p ottery from Spirit Cave and lend weight to the interpretation that the pottery and polished stone tools from Spirit Cave were introduced into the Hoabinhian deposits from a much later occupation (Reynolds 1992). In the far south of Thailand, Hoabinhian ar tifacts are also found in the upper cultural layers at Lang Rongrien, dated to ca. 8000 to 4000 BP, and at Moh Khiew (Anderson 1990; Pookajorn 1991, 1994; Bellwood 1997; Higham and Thosarat 1998; Hutterer 1998; Higham 2002; Matsumura and Pookajorn 2005). A single, crouched burial excavated at Moh Khiew, found in association with Hoabinhian tools and dated to between 26,000 and 6000 BP, is similar to burials from Hoabinhian contexts at the sites of Hang Dang and Moc Long in Vietnam (Pookajorn 1992; Higham 2 002; Matsumura and Pookajorn 2005). Interestingly, this burial posture differs from the Hoabinhian period flexed burials from the much closer peninsular Malaysian sites of Gua
36 Cha (10,000 to 3000 BP) and Gua Gunung Runtuh (11,000 to 7500 BP) (Bellwood 199 3, 1997; Zuraina 1994; Matsumura and Zuraina 1999). Clearly, despite the long standing overall technological uniformity of the Hoabinhian tool industry throughout mainland Southeast Asia, such regional variations in other aspects of the Hoabinhian culture indicate that these hunter gatherer populations were neither static nor monolithic. Changing environmental conditions towards the end of the Hoabinhian period have permitted us further insights into patterns of cultural variation among hunter gatherer gro ups in Thailand. During the Last Glacial Maximum, ca. 21,000 to 18,000 BP, mean sea levels were as much as 100 to 150 meters lower than their current level, exposing the shallow continental shelf around much of mainland and island Southeast Asia (Chappell and Thom 1977; Chappell 1994; Bellwood 1997). During this period, the mainland of Southeast Asia was joined with the islands over the next 12,000 years, the sea levels rose, progressively inundating the lowest lying areas and obscuring any traces of early human habitation along the ever changing coastline. Around 6000 BP, the rising sea reached its maximum and then retreated several meters to the present day sho relines of Southeast Asia (Higham and Thosarat 1998:38; Bellwood 1997; Higham 2002). It is only from this point on that evidence of hunter gatherer populations living along the coast is accessible to archaeologists, though as Higham and Thosarat (1998:64) Excavations of coastal sites from this period reveal important contrasts with the Hoabinhian hunter gatherer groups from the inland uplands of Thailand. H ere, the biotic richness of the tidal swamps, mangroves and estuaries supported larger, predominantly sedentary groups of fisher hunter gatherers, who left behind a more expansive suite of cultural remains
37 than their more migratory late Hoabinhian contempo raries. At the important sites of Nong Nor (ca. 2500 BC) and Khok Phanom Di (2000 to 1500 BC), edge ground stone tools, worked bone and cord marked, burnished and incised pottery are common, and were likely produced locally, as suggested by intrasite patt erning in the distribution of these artifacts (Higham and Thosarat traded with the inland hunter gatherer groups, as suggested by the presence of marine faunal remains at the inland sites and of stones obtained from inland sources at the coastal sites. A single, crouched burial from Nong Nor echoes those found at Moh Khiew and at Hoabinhian sites in Vietnam, but with the added element of accompanying pottery vessels (High am and Thosarat 1998:43; Higham 2002). The cultural richness of these costal populations is exemplified by the many burials recovered from the mound site of Khok Phanom Di, near the head of the Gulf of Thailand (Higham and Bannanurag 1990, 1991; Higham a nd Thosarat 1994, 1998; Nakbunlung 1994; Bellwood 1993, 1997; Higham 2002). Within the three mortuary phases (MP) attributed to the early hunter gatherer occupation of the site (out of seven total mortuary phases), people were interred in a supine, extend ed position, often in clusters believed to represent family units. A wealth of grave goods heretofore unknown for the prehistoric inhabitants of Thailand included pottery vessels and pottery making tools, edge ground stone tools, bone fishhooks and harpoo ns, shell knives, bark cloth and asbestos fabric, red ocher, shell beads and other bone and shell ornaments, and selected faunal remains. The asymmetrical distribution of some of these items, particularly within the third mortuary phase, hints at the nasc ent expression of individual rank and social stratification, as well as the emergence of sex related differences in mortuary practices
38 (Higham and Thosarat 1998:55 56; Higham and Thosarat 1994; Nakbunlung 1994; Bellwood 1997; Higham 2002). The apparent s edentism and cultural richness (especially in terms of the volume and variety of pottery forms) of the population of Khok Phanom Di has led some to suggest that these people were among the earliest people to practice agriculture in Thailand (Higham and Tho sarat 1998; Higham 2002). This suggestion has been lent additional weight by the recovery of domesticated rice chaff from coprolites and the abdominal regions of burials belonging to the third mortuary phase (Higham and Thosarat 1998:56; Higham 2002:63; T hompson 1992; Glover and Higham 1996). The presence of dog bones in the second and third mortuary phases also hints at agriculture, since the dog is not native to Southeast Asia and is almost universally found accompanying agricultural communities elsewhe re in the world (Higham and Thosarat 1998:54; Higham 2002:59, 63; Higham 1996). Sauer (1952) suggested that resource rich marine coastal environments such as that at Khok Phanom Di could have fostered experimentation with plant cultivation and horticultur e by providing both 1) an abundance of useful plant species with which to experiment; and 2) an abundance of resources in general, providing the inhabitants of these areas with sufficient free time away from their hunting and gathering activities to permit such experimentation (Bellwood 1997). As Bellwood (1997:203 hypothesis has been discounted on several grounds. Not least among these is the unlikelihood that hunter gatherer populations living in resource rich environme nts, with seemingly endlessly renewable food sources, would ever feel the need to experiment with plant cultivation in order increase the yields of already abundant plants. Rather, Bellwood (1997:204) and others (McCorriston and Hole 1991; Higham and Thos arat 1998) favor a model of agricultural development wherein increasing plant cultivation must have been induced indeed, was
39 probably ultimately necessitated by environmental or demographic stresses on existing food sources. Such stresses do not appear to have affected the early inhabitants of Khok Phanom Di. Although their skeletal remains indicate that they suffered the short life expectancy, high infant mortality, and parasitic and infectious disease burdens common to all hunter gatherer populations, the people appear to have been well nourished and in overall good health (Bellwood 1997; Higham and Thosarat 1998; Tayles 1999; Higham 2002). This further supports the hypothesis that the rich marine coastal environment provided adequate nutritional reso urces for this population, without the need for active cultivation of additional food sources. Nor do the climate or environment appear to have been contributing factors to any nascent agricultural development here. Rather, as Higham and others (Higham a nd Thosarat 1998:53; Thompson 1996; Higham 2002) note, the soil and groundwater at Khok Phanom Di during the period of the first three mortuary phases were likely too saline to support rice cultivation, or even the growth of wild rice. It is more likely, they surmise, that the domesticated rice consumed at Khok Phanom Di during this period was obtained from elsewhere in Thailand or Southeast Asia via trade networks (Higham 2002:77; Hutterer 1983; Higham and Thosarat 1994, 1998; Glover and Higham 1996; High am 1996). This contact with other groups may be evinced by the appearance of new styles of pottery, including new types of clay and temper and new patterns of decoration and shape, which occurred during the second mortuary phase and again during the third (Higham 1996; Higham 2002; Vincent 2003, cited in Higham 2002). Whatever the nature of this early agricultural influence, both it and the cultural richness of Khok Phanom Di appear to have reached an apogee during the time when burials of the fourth and f ifth mortuary phases (often referred to as the middle cultural layers of the site; Bellwood
40 1993, 1997) were laid down. During this period, it is believed that a climatic or other environmental shift resulted in the area around Khok Phanom Di becoming mor e of a freshwater riverine or estuarine environment. This shift is evidenced by changes in the composition of the faunal assemblages from these phases, to include more freshwater fish and shellfish species (Mason 1991, 1996; Higham and Thosarat 1994, 199 8; Bellwood 1997; Higham 2002). The first appearance of the remains of water buffalo, an increasing prevalence of granite hoes, and a dramatic increase in shell knives (believed to have been used for harvesting rice) during this period have led many to co nclude that the people of Khok Phanom Di had by this time begun to actively cultivate domesticated rice in the more agriculturally favorable freshwater swamps and floodplains which surrounded the site (Higham and Thosarat 1998:62 63; Higham and Thosarat 19 94; Glover and Higham 1996; Higham 1996; Bellwood 1997; Higham 2002). This period witnessed changes in mortuary practices as well: some of these, such as the differentiation between the graves of males and females by the inclusion of turtle carapaces an d clay anvils, respectively, were intensifications of practices established in earlier mortuary phases. Others, including a relaxation of the tight clustering of graves by family group and an increase in individual graves, to the point where clustering ce ased altogether during the fifth mortuary phase, mark a departure from previous mortuary practices (Higham and Thosarat 1998; Higham 2002). These shifts are indicative of a dynamic and malleable sociocultural environment at Khok Phanom Di, one that was re sponding to fluctuating climatic, geophysical, social and economic conditions, and they may be further evidence of contact with other cultural groups. A distinct hiatus in interments between the third and fourth mortuary phases, during which over a third of a meter of non mortuary cultural deposits accumulated, gives some sense of the upheaval
41 experienced by the inhabitants of the site during this time period (Higham and Thosarat 1998; Higham 2002). The greatest changes in mortuary practice, suggesting t he great est changes in the lives of the people, are observed in the fifth mortuary phase, which also contains perhaps the most famous prehistoric burial in all of Thailand: the so middle aged female believed to be a n accomplished potter based on the amount of pottery, raw she was interred with an unprecedented wealth of shell bangles, ear ornaments, discs and other jewelry of novel design, including a necklace c ontaining over 120,000 shell beads. All of these were made from the shells of marine mollusks that by this time were exotic to the freshwater environs of Khok Phanom Di, further ity. Adjacent to her, two infants thought to be her daughters were interred in similarly rich graves, one of which sized ealth of their graves, the location of their burials in an area that had never before been used for interments, and that was set apart from the primary mortuary zone, provides additional emphasis of the high status accorded these individuals (Higham and Th osarat 1998:61 63; Bellwood 1997:256; Higham 2002:68 73). Although we do revered artisan; matriarch of a prominent family; tribal leader; head of a prosperous and influential imm igrant agricultural community recently arrived at Khok Phanom Di; or something else altogether her exaltation in death foreshadows the emergence of increasing social stratification during the Neolithic and Metal Ages.
42 The provision of special burial rite s for selected individuals continued and was elaborated in the sixth mortuary phase, during which a few rich burials were enclosed within a unique clay structure adjacent to a wooden structure containing more pedestrian burials. But such practices were ab andoned during the seventh and final mortuary phase of Khok Phanom Di, the overall depauperate nature of which indicates a site in decline. Paleoenvironmental reconstructions indicate that beginning late in the fifth mortuary phase, conditions at Khok Pha nom Di had returned to a more marine, coastal environment, the salinity of which forced the inhabitants to abandon rice cultivation and return to procuring this staple via trade contacts (Higham and Thosarat 1998:61 63; Higham 2002:74 75). The people may have struggled to return to their pre agricultural lifeways of fishing and gathering coastal food resources, and when the environment shifted again, this time moving the seacoast far from the site, Khok Phanom Di was abandoned (Bellwood 1997:258; Higham an d Thosarat 1998; Higham 2002). Although the remains of domesticated rice are present throughout nearly the entire occupation of Khok Phanom Di, it appears that it was grown locally only during a short interval al groundwater and soil conditions became less saline and thus more suitable for rice cultivation. As was noted above, it is most likely that at all other times, the inhabitants of Khok Phanom Di obtained their rice via trade with other groups, who must h ave been practicing agriculture elsewhere in Thailand and the surrounding areas of Southeast Asia. The identity and origins of these agricultural groups, however, as well as the means by which they came to possess their knowledge and techniques of rice cu ltivation, periods in Thai prehistory, and our understanding of the N eolithic period in Thailand suffers at
43 the hands of a host of archaeological misfortunes, including the destruction (by both natural and artificial processes) of critical sites, inaccurate or otherwise unreliable dating estimates, and, remarkably, the abse nce of direct evidence for rice agriculture from key sites that otherwise undoubtedly belong to the Neolithic. Small wonder then, that this most important of questions has dominated the modern era of archaeological research in Thailand. Excavations of Ne olithic sites elsewhere in Southeast Asia have helped clarify the picture in Thailand, but the debate remains active today and it is central to the understanding of the origins of the modern Thai population. The broader archaeological, linguistic, genetic and skeletal data that inform and drive this debate will be considered in the following chapter; for now, let us continue our focus on what we have learned from the salient archaeological sites in Thailand. The Arrival of Agriculture The secure dating of Khok Phanom Di and several other sites allows us to determine that rice agriculturalists were present in Thailand by at least 2300 BC, and most authorities concur that these groups probably inhabited sites situated on inland river flood plains favorable fo r rice agriculture (Glover and Higham 1996). Among the most important of these inland sites is the Neolithic cemetery of Ban Kao, in the Kwae Noi River valley northwest of Bangkok, which was occupied between 2300 and 1500 BC (Figure 2 2) (Srensen and Hat ting 1967; Bellwood 1997; Higham and Thosarat 1998; Higham 2002). Here, Srensen and colleagues found 44 single and paired extended burials interred with quantities of pottery; polished adzes and other stone tool forms; bone tools and projectile points; s tone and shell sickles and knives used for harvesting rice; and stone, animal bone, antler and shell worked into bangles, ornaments and beads (Srensen and Hatting 1967; Srensen 1972; Bellwood 1997; Higham and Thosarat 1998; Higham 2002).
44 1997:258). The presence of rice agriculture at Ban Kao has also been inferred from the quantity of stone and shell sickles and reaping knives recovered from the site. But with regards to the questions of the origins of agriculture in Thailand and in Southeast Asia, the most important materials recovered from Ban Kao are the ceramics. Srensen (1972 ) described twelve distinct pottery forms from the graves there, divided into early and late phases and including carinated jars with tripod bases, goblets, ring pedestalled jars, and flat bottomed jars and bowls (Bellwood 1997; Higham and Thosarat 1998; H igham 2002). Most of the pottery forms are red or black wares that are burnished or cord marked on their external surfaces, but some bear a distinctive motif of sharply delineated sinuous bands of incised or impressed dentate surface textures (Srensen 19 72; Bellwood 1993, 1997; Higham 1996, 2002; Higham and Thosarat 1998). This impressed dentate motif and the carinated tripod vessels the defining features of have become something of a hallmark of expanding agricultural popula tions in Southeast Asia (Srensen 1972; Glover and Higham 1996; Higham 1996, 2002; Bellwood 1993, 1997; Rispoli 1997; Higham and Thosarat 1998). For example, such ceramic wares appear as novel elements in the grave good assemblages from the second and thi rd mortuary phases at Khok Phanom Di (Higham 1996, 2002; Bellwood 1997; Higham and Thosarat 1998). As noted above, these finds prompted Vincent (2003, cited in Higham 2002) to suggest that the new pottery forms were indicative, at a minimum, of contact an d trade with outside groups i.e., the agriculturalists from whom the hunter gatherers of Khok Phanom Di are believed to have obtained their domesticated rice but they may also indicate an influx of peoples to Khok Phanom Di from other areas (Higham 2002:58 65).
45 Elsewhere in the Chao Phraya basin, similar impressed dentate motif pottery, along with a broadly similar material culture in general, has been found at the Neolithic sites of Ban Tha Kae (poorly dated but probably ca. 2500 to 1500 BC), Non Pa Wai (2300 to 1500 BC) and Khok Charoen (1400 to 800 BC, and thus a late Neolithic/early Metal Age site), as well as in the upper layers overlying the Hoabinhian deposits at Heap Cave and Khao Talu (Higham 1996, 2002; Loofs Wissowa 1997; Pigott et al. 1997; Be llwood 1997; Higham and Thosarat 1998). The presence at these inland sites of beads and other ornaments made from marine shells lends further weight to the interpretation that these agriculturalists were engaged in exchange relationships with the coastal hunter gatherers from sites such as Nong Nor and Khok Phanom Di (Higham and Thosarat 1998:81 82). On the Khorat Plateau, this same pottery motif has been found in conjunction with definitive evidence of rice cultivation in the Neolithic levels at the key sites of Non Nok Tha (ca. 2000 to 1500 BC) and Ban Chiang (2300 to 1500 BC), (Higham and Thosarat 1998:83; Higham 2002:91 92; Bayard 1972; Gorman and Charoenwongsa 1976; Hutterer 1988, 1993; Higham 1996). As at the Neolithic sites in the Chao Phraya basin the deceased at Non Nok Tha and Ban Chiang were interred in a supine, extended position, along with grave good assemblages composed of pottery vessels, polished stone tools, shell beads and faunal remains. The bones of domesticated pigs, dogs and cattle recovered during excavations help round out the picture of these sites as full fledged agricultural settlements. This uniformity of practice in pottery styles and mortuary ritual throughout Neolithic Thailand suggests a sole source, whether exogenous or endogenous, for the advent of agriculture in the area. Additionally, these common pottery forms and mortuary practices are not unique to agricultural settlements in Thailand. Rather, they have a broad distribution throughout Southeast Asia and speak to the spread of agriculture throughout the region (Leong 1991; Glover and
46 Higham 1996; Higham 1996, 2002; Bellwood 1997; Higham and Thosarat 1998). In Vietnam, the earliest agricultural settlements are found on flood plains of the lower reaches of the Red R iver valley, and date, as in Thailand, to the late third to mid second millennium BC (Higham 2002:86; Hutterer 1983; Higham 1996; Higham and Thosarat 1998). At the sites of Phung Nguyen, Co Loa, Trang Kenh and Lung Hoa, the suite of dentate stamped and i ncised pottery forms known locally as the Phung Nguyen culture is found in conjunction with mortuary material culture assemblages similar to those from Neolithic Thailand. Significantly, the Phung Nguyen culture also features the added, distinctive elemen ts of bangles, beads and other ornaments carved of nephrite and jadeite and reminiscent of similar artifacts from Neolithic Chinese sites (Ha Van Tan 1985; Higham 1996, 2002; Higham and Thosarat 1998; Hoang Xuan Chinh and Nguyen Ngoc Bich 1978, Ha Van Tan 1980, Lai Van Toi 1999, cited in Higham 2002). Similar incised and dentate stamped ceramic wares and/or ring pedestalled and tripod vessels have also been found in the Mekong Valley at Samrong Sen, in Cambodia (Higham 1996, 2002), and in peninsular Thail and and Malaysia at Lang Rongrien, Gua Cha (in association with extended burials at both sites), Gua Kecil and Jenderam Hilir, among others (Sieveking 1954, 1987; Anderson 1990; Leong 1991; Bellwood 1997; Higham and Thosarat 1998). It is worth noting that a millennium later in the Malay peninsula than it does in Thailand, Vietnam and Cambodia, suggesting both a direction and a time frame for the spread of agriculture throughout Southeast Asian mainland (Bellwood 1993, 1997). Bellwood (1997:265) also notes that there is typically a sharp discontinuity between the earlier Hoabinhian and later Neolithic deposits at the peninsular sites, suggesting, as at Khok Phanom Di, that agricu lture was carried to these sites by an influx
47 of people from elsewhere. Both of these points, and their impact on our understanding of population history in Southeast Asia, will be considered further in Chapter 3. With the establishment of agriculture in the resource rich Southeast Asian mainland, the eventual emergence of metallurgy there would seem to be a foregone conclusion. Indeed, in a number of ways, the Bronze Age in Thailand appears to be an intensification and elaboration of many cultural trends begun during the Neolithic. Rice agriculture expanded and was made more efficient by the use of metal tools to clear the land, till the soil and reap the harvest. Growing populations fed by the increased yields continued to settle on low mounds along fe rtile river flood plains and tributaries. Many of the key Neolithic sites from the Chao Phraya basin and Khorat Plateau were also thriving Bronze Age settlements, indicating successful adaptation to technological and social innovations by longstanding com munities. Material culture expanded to include new forms and styles of both utilitarian and ornamental items. Exchange relationships among groups, for items both required and desired, became more complex and expanded in economic scale and geographic scop e to include a broader range of ordinary and exotic goods and materials. In some places, the differential distribution of these items in grave good assemblages indicates that the nascent social stratification observed in late hunter gather and early agric ultural sites began to crystallize during the Bronze Age, as individuals and lineages began to distinguish themselves, perhaps through auspicious descent, leadership ability, finesse as craftsmen, or prowess as traders within an increasingly complex societ y. Everywhere, the pace of social and cultural change quickened (Bellwood 1997; Higham and Thosarat 1998; Higham 2002; Baker and Phongpaichit 2005).
48 The Metal Age The Bronze Age Many of the same research questions that pertain to earlier prehistoric era s in Thailand have also been applied to the Bronze Age: did the practice of metallurgy have an independent, in situ origin in mainland Southeast Asia (Nakbunlung 1994), or was it imported from elsewhere in Asia (i.e., from China), and if so, via what mean s the diffusion of ideas, the trade of artifacts, or the movement of peoples (Bellwood 1997; Higham and Thosarat 1998; Higham 2002; Baker and Phongpaichit 2005)? As with the origins of agriculture in Southeast Asia, early researchers (Solheim 1968; Gorman and Charoenwongsa 1976; inter alia ) were quick to suggest that the advent of metallurgy in Southeast Asia occurred not only independent of, but also prior to, its emergence in China and the Near East (Higham and Thosarat 1998:92 93; Nakbunlung 1994). As was the case with the debate over the origins of agriculture, the excavation of additional sites and the acquisition of better radiocarbon dates have resolved some, but not all, of the questions surrounding the emergence of metallurgy in Southeast Asia. Although initially thought to date as far back as the fourth millennium BC (Solheim 1968; Gorman and Charoenwongsa 1976; Rinehart 1981; White 1982), the Bronze Age in Thailand and most of mainland Southeast Asia is now considered to have begun no earlier t han 1500 BC (Higham and Thosarat 1998; Higham 2002). This makes Southeast Asian metallurgy much younger than the Bronze Age in China, which likely dates back to the early third millennium BC (Renfrew and Bahn 1996; Higham and Thosarat 1998; Diamond 1999; Higham 2002). Additionally, similarities in smelting and casting technologies and artifact forms between sites in southern China and Southeast Asia suggest that the idea, if not the actual physical practice, of bronze metallurgy spread throughout the regi on from an ultimate source in China, via increasingly expansive and important trade networks. Further evidence for such trade networks
49 can be found in the Chinese style jades that have been recovered from late Neolithic and Bronze Age contexts in Thailand (Bellwood 1997; Higham and Thosarat 1998; Higham 2002). In general, two types of Bronze Age sites are encountered in Thailand: those where copper ores were mined, smelted and then cast into ingots for trade; and those where the ingots were alloyed with tin or lead to create bronze that was then worked into tools and artifacts. The majority of bronze artifacts are known from mortuary contexts at both types of sites. Important mining and smelting sites have been found near copper ore sources along the M ekong River in northeastern Thailand, and in the Chao Phraya basin in central Thailand (Figure 2 3) In the latter area, the smelting slag, clay smelting furnaces, casting crucibles and ingot molds found at Non Pa Wai and dated to 1500 to 1000 BC provide the earliest evidence for bronze metallurgy in Thailand (Natapintu 1991; Higham and Thosarat 1998). The nature of the cultural deposits suggests that while ore processing sites such as Non Pa Wai, Nil Kham Haeng and Phu Lon were in use for many centuries, they were only occupied seasonally, as smelting fires would have been easier to maintain during the annual dry season hiatus in monsoon rains. The raw copper and bronze ingots produced at these sites would then have been traded with other settlements clu commerce and exchange. And it is these sites, where bronze was cast into tools and ornaments for work and for wear, and included in grave good assemblages, th at provide the greatest insight At Ban Na Di (1400 to 400 BC), on the Khorat Plateau, excavation in two separate areas of the site (designated Areas A and B) revealed not only extensive evidence for bronze casting including an alloying furnace, crucibles and fragments of bivalve casting molds but also a rich and diverse material culture. Interestingly, there appears to have been a significant disparity in
50 the distribution of this material wealth among the population of Ban Na Di, as evidenced by a stark contrast in the provisioning of grave goods between the groups of burials in the two excavation areas. Despite a shared basic mortuary assemblage of pottery vessels, shell beads and faunal remains, throughout three Bronze Age m ortuary phases, the graves in Area A are distinctly more modest than those at Area B, containing hundreds, rather than thousands, of shell beads, fewer ritual offerings of cattle and pig forelimbs, and almost no bronze artifacts. By contrast, in Area B, i ndividuals were buried with miniature clay figurines of cattle, deer, elephants and people; bangles cast in bronze and carved from shell, marble and other exotic stones; and the first iron artifacts known from the site. A few individuals, including a chil d, were buried with ornaments created from the skin and bone of crocodiles. Higham and colleagues (Higham and Kijngam 1984; Higham and Thosarat 1998; Higham 2002), however, have argued that while this striking difference in the allocation of grave goods c ertainly indicates a higher social status for the people buried in Area B, the asymmetry is of a magnitude insufficient to suggest that these people belonged to a distinct ruling class. Rather, they suggest that the two groups were members of different, a nd differently favored, lineal groups within the same general social class. Elsewhere in the Khorat Plateau, similar evidence of bronze casting including furnaces, crucibles and bivalve molds has been found at the sites of Ban Chiang and Non Nok Tha (both 1500 to 1000 BC or later). Likewise, the burials found at these sites share many of the same mortuary traditions with Ban Na Di, such as the inclusion of pottery, shell beads and bangles, axes, projectile points and bangles cast in bronze, miniature cattl e figurines made of clay, and ritual offerings and ornaments derived from both domesticated and wild animals, as well as the alignment of graves in clusters or rows (Bayard 1972; Gorman and Charoenwongsa 1976;
51 Higham and Thosarat 1998; Higham 2002; Pietrus ewsky and Douglas 2002). Somewhat surprisingly, though, neither of these sites appear to display the type of patterned asymmetry in the distribution of grave goods within the population as was observed at Ban Na Di (Higham and Thosarat 1998; Higham 2002; Pietrusewsky and Douglas 2002). There are several possible explanations for this apparent difference in social and mortuary practice among these contemporary and geographically proximate settlements. The simplest of these is that the perceived patternin g in grave good distribution at Ban Na Di is due to sampling error. Had more of the burials from a greater area of the site been excavated, the sharp distinctions in mortuary assemblages between Area A and Area B might been subsumed within a broader conti nuum of relative grave wealth, and thus not seem so sharp after all. This explanation is largely unsatisfying, though, not least because the burials at Ban Chiang were also excavated from two widely separated areas within the site, and therefore might rea sonably be expected to display a similar dichotomy in the provisioning of grave goods. A much more likely and appealing hypothesis is that these intersite contrasts in mortuary practice reflect intersite differences in social structure and/or the means of reification of social structure through cultural practice. White (1995:105) and Pietrusewsky and Douglas (2002:188), for example, note that it may be more accurate to regard the burgeoning social complexity of Bronze Age societies in Southeast Asia as he terarchical, rather than hierarchical, in nature. In a heterarchical society, community may be defined along fluid lines that cross cut the society from a variet y of social and cultural angles (Pietrusewsky and Douglas 2002:188; Crumley 1995). Thus, the status relationships signified by the asymmetrical distribution of grave goods at Ban Na Di may not
52 have existed in the same form within the communities at Non No k Tha or Ban Chiang, or may have been expressed through some other facet of their cultural traditions. While the Bronze Age saw the continued growth of successful Neolithic communities in the Chao Phraya basin and the northern Khorat Plateau, it also wit nessed the first large scale settlement of the Mun River valley, in the southern half of the plateau. These new settlements, such as Ban Lum Khao (1400 to 500 BC), followed very much the same cultural pattern as their long established counterparts in the northern region of the plateau: settlements located on low rises near river floodplains made favorable for rice agriculture by annual flooding; burials laid out in rows and including grave good assemblages composed of pottery, faunal remains, a few bronze tools and jewelry items, and beads, bangles and other ornaments made of exotic shell and stone; and the presence of furnaces, crucibles and molds for alloying and casting bronze. As in the north, there is also some local differentiation in cultural tradi tions in the southern Khorat, as evidenced by the presence at these sites of a distinctive red slipped, trumpet rimmed pottery tradition. While a few individuals at Ban Lum Khao were buried with particularly rich grave good assemblages, as at Ban Chang an suggestion of the type of broad intrasite dichotomy in the allotment of grave goods as observed at Ban Na Di (Higham One other locale that does appear to show significant intrasite and intrapopulation pattering in the distribution of grave goods is the Bronze Age cemetery at the important pre Neolithic coastal site of Nong Nor (1100 to 600 BC). Here, small groups of individuals with distinctly richer grave good assemblages have been found in both the earlier, western sector and the later, eastern sector of the cemetery. The eastern sector of the cemetery is of partic ular interest, for
53 grave good assemblages from this area also include ornaments carved from jade, serpentine and talc, and the earliest known carnelian artifacts from Thailand. As noted above, the jades were most likely obtained via trade with Bronze Age peoples in southern China. The carnelian, on the other hand, is tantalizing evidence of early, albeit likely remote and indirect, trade contacts with India (Bellwood 1997; Higham and Thosarat 1998; Higham 2002). During the Iron Age and beyond, economic and social contact with emergent Indian civilizations would only intensify, and would exert far reaching and long lasting influences on the people and cultures of Southeast Asia in general, and of Thailand in specific (Higham and Thosarat 1998; Stark and A llen 1998). Although little is known about the precise nature and magnitude of the earliest contact with Indian civilizations, it is no surprise that the first evidence for this contact comes from a coastal site such as Nong Nor, which is separated from t he Bay of Bengal only by the narrow Kra Isthmus of the Malay Peninsula. Carnelian artifacts are also known from terminal Bronze Age contexts at Ob Luang and Nil Kham Haeng, located in the northern and central reaches of the Chao Phraya basin, respectively and directly upstream from Nong Nor. Just as the Chao Phraya and Mekong rivers and their tributaries had served as critical internal avenues of trade within Southeast Asia during the Neolithic and Bronze Ages, so would they also prove to be important th oroughfares for the spread of truly external and exotic ideas and goods during both the Iron Age and the period of the rise and rule of early state level civilizations in Southeast Asia (Bellwood 1997; Higham and Thosarat 1998; Higham 2002; Baker and Phong paichit 2005). The Iron Age Most authorities agree that the Iron Age (500 BC to ca. AD 500) in Thailand and the rest of Southeast Asia was a period of great social and cultural upheaval and change, due in no small part to burgeoning contacts with the truly foreign cultures of India and the Han and Qin dynasties
54 of China (Stark and Allen 1998 ). While, as noted above, it is likely that the first agriculture and, later, the first bronze metallurgy were introduced from southern China, it can be (and has been) argued that, in terms of climate, physical geography and prehistoric culture, southern China is more properly considered a part of greater Southeast Asia, particularly during the Neolithic and early Metal Age (Barth 1952; Bellwood 1993, 1997; Matsumura and Hudson 2005; Pietrusewsky 2006). The same cannot be said for the cultures of the Han and Qin dynasties, who expanded from their heartland in central China to colonize most of southern China and even the northern parts of Vietnam in the last centuries BC (Bellwood 1993, 1997; Higham and Thosarat 1998; Higham 2002; Baker and Phongpaichit 200 5). Whether iron forging was introduced to Thailand via these expanding cultural and economic ties is, again, both unknown and the subject of much debate. What is certain is that along with new types of material goods, Southeast Asian populations would a lso have been exposed to new languages, belief systems and religious practices (specifically those of Buddhism and Hinduism), sociocultural norms and political structures. Other social changes were surely wrought by expanding population sizes fueled by ev er more efficient and elaborate agricultural practices that taxed supplies of natural resources and strained cultural mechanisms for resolving conflict and mitigating distress within society. New economic, spiritual and political apparatuses would have pr ovided additional, more robust means for exceptional individuals and lineages to reinforce the incipient societal segmentation begun in the Bronze Age, and to begin to convert heterarchy into hierarchy (Nakbunlung 1994; Bellwood 1997; Higham and Thosarat 1 998; Stark and Allen 1998; Higham 2002; Baker and Phongpaichit 2005). comes from mortuary contexts, which have yielded the bulk of the evidence for foreign trade
55 during this period, and which, as a whole, are richer than their Bronze Age precursors. At Ban Don Ta Phet, in the southern Chao Phraya basin (see Figure 2 3) mortuary contexts have yielded cotton textiles, etched jades and a carnelian lion figurine, all of und oubted Indian origin or influence. This last item is of particular significance, as the lion was an early avatar of the Buddha, and thus it represents a concrete manifestation of the exchange of not just material items, but also of ideas between the peopl es of Thailand and India (Glover 1996; Bellwood 1997; Higham and Thosarat 1998; Higham 2002). With the advent of forged iron tools, bronze became predominantly a luxury material and was cast into a variety of new ornamental forms, including bangles, earri ngs, torcs, beads, bells, rings, figurines and decorated bowls, all of which were included in the grave assemblages. Along with these bronzes, grave good assemblages at Ban Don Ta Phet and other Iron Age sites include novel glass, jade and agate beads and pendants, iron tools and weapons, and the ever present marine shell jewelry, faunal offerings and ornaments, and locally distinctive ceramics (Higham and Thosarat 1998; Higham 2002). For some individuals, there was also an elaboration of the mortuary rit ual itself during the Iron Age, as Srensen (1973, 1988; Higham and Thosarat 1998; Higham 2002) uncovered at Tham Ongbah a group of burials contained within boat shaped coffins featuring carved bird heads at the prow and stern. Several of these graves als o included rare, elaborate bronze drums from the Iron Age Dong Son culture of northern Vietnam. At other sites, the dead were placed in plaster or clay lined, rice filled grave chambers stocked with grave good assemblages of extraordinary richness. At e very Iron Age site, archaeologists have found some small group of individuals singled out by some unique facet of the mortuary practice, and have concluded that these individuals must be the emergent elites (Higham and Thosarat 1998; Higham 2002).
56 The se ttlements themselves also underwent changes and elaboration in size, configuration and location: mounded sites near waterways were still preferred, but during the Iron Age they came to be ringed with canals or moats, perhaps for defense or to ensure a sup ply of water to the growing populations. Many of the main sites are believed to have had populations of many hundreds or even a few thousand people far larger than could have been supported by the agricultural yields of the surrounding lands. Thus, the I ron Age saw the first establishment of settlements situated at higher elevations in the terraces above the floodplains, at sites that were smaller and less favorable for rice agriculture. These sites were likely subordinate to the main settlements around which they clustered, and if so, represent the first emergence of settlement hierarchies in Thailand. The appearance of numerous local pottery traditions such as Phimai Black, Ban Chiang Red on Buff and Roi Et cord marked red banded wares, further reinfor ces this notion of peripheral settlements organized around long standing local centers. This wealth of localized mortuary and material traditions contributes to the picture of Iron Age Thailand as a period of flourishing cultural diversity and social comp lexity, building inexorably to the appearance of the first state level civilizations (Nakbunlung 1994; Higham and Thosarat 1998; Higham 2002; Baker and Phongpaichit 2005). The Rise of the State The first state level civilization to emerge in Thailand prope r was the kingdom of Dvaravati, comprising numerous settlements along the Chao Phraya basin (Figure 2 4) and probably initially established during the late prehistoric period, around the sixth century AD (Briggs 1945; Rinehart 1981; Nakbunlung 1994; Higha m and Thosarat 1998; Higham 2002; Baker and Phongpaichit 2005). A less well known sister state or perhaps, rather, a colony of Dvaravati, with which the latter appears to have had favorable economic, political and cultural relations, developed at Hariphu nchai in the northern reaches of the Chao Phraya basin beginning
57 in the late seventh century AD (Briggs 1945; Rinehart 1981; Higham and Thosarat 1998). The rise of Dvaravati and Hariphunchai in Thailand followed the waning of the civilization of Funan ( AD 100 to 550 ), whose influence extended from its center on the Mekong Delta in southern Vietnam, across Cambodia and into the area encompassed by modern day Thailand. The first definitive state level society in Southeast Asia, Funan laid the foundations of extensive maritime and overland trade with China and India and across Southeast Asia; the enthronement of hereditary rulers; the further development of populous urban centers supported by an agricultural hinterland; the adoption of Indic religious practic es and the construction of monumental public architecture tied closely to the practice of Buddhism and Hinduism that would come to characterize the varied civilizations of Thailand right up through the modern era (Rinehart 1981; Bellwood 1997; Higham and T hosarat 1998; Stark and Allen 1998; Higham 2002; Baker and Phongpaichit 2005). The society of Dvaravati was likewise heavily influenced by Indian culture. By as early as the seventh century AD, its rulers are known from inscriptions the earliest in Thai land, written in Sanskrit, Pali and Mon, and using scripts borrowed from southern India to have adopted Sanskrit names, the veneration of the Buddha and of the Hindu pantheon, the conventions of Buddhist and Hindu religious and monumental architecture, and the material and ritual trappings and symbology of the Indic elite. Theravada Buddhism, the predominant form of Buddhism in modern day Thailand, was possibly introduced from Sri Lanka during this period. Also present at Dvaravati period sites are a numb er of artifacts, including items of Greco Roman, Byzantine, reaching trade networks in the Old World. Dvaravati period sites also frequently continued the Iron Age pattern o f constructing defensive moats around the settlements, and artwork from this period
58 suggests increasing levels of conflict among, and possibly within, the emerging states of Southeast Asia (Briggs 1945; Rinehart 1981; Bechert 1984; Bunnag 1984; Higham and Thosarat 1998; Stark and Allen 1998; Higham 2002; Baker and Phongpaichit 2005). One of the states with w hich Dvaravati may have clashed 2 was the early Khmer civilization of Chenla (or Zhenla), which emerged in the seventh century AD in the middle Mekong Valley, also following the collapse of Funan. Initially centered in northeastern Cambodia and central Vietnam, the Chenla civilization was at first a collective of semi autonomous local polities, each organized around a central settlement that was surroun ded and supported by smaller agricultural communities, much like Dvaravati. However, by the early ninth century, these individual polities were increasingly consolidated into a true empire under the rule of the dynamic and aggressive kings of the Jayavarm an dynasty, who in the early tenth century established a new capital at Angkor, in northwestern Cambodia. The growing early political and economic influence of Chenla also extended northward, into the Mun River valley of northeastern Thailand, for at seve ral sites here, Khmer cultural deposits are found overlying layers containing Dvaravati style artifacts. This influence intensified during the Angkor Period (tenth to fifteenth centuries AD), when the ruling dynasty was supplanted by Jayavarman VI, who, d espite his promulgation of the dynastic name, had ancestral ties not to Chenla, but to the Iron Age settlement of Phimai, in the Mun Valley of the Khorat Plateau He and his successors would develop northeastern Thailand into an important and prosperous p rovince of the Angkor state, many of the most important Angkor period temples and archaeological sites are found not in 2 Given the pres ence of Khmer influenced artifacts at Dvaravati period sites, the two civilizations probably enjoyed favorable political and economic relations at times, but they appear to have come increasingly into conflict with one another as each state grew in power a nd size.
59 Cambodia, but in northeastern Thailand. Angkor further extended its reach into Thailand during the tenth and eleventh centuries, when it came to completely dominate the Dvaravati heartland in the Chao Phraya basin (Briggs 1945; Rinehart 1981; Higham and Thosarat 1998; Higham 2002; Baker and Pho ngpaichit 2005). The majority of Thailand would thus remain under the control of the Angkor until the fall of that city in AD 1431. Remaining beyond the sway of Angkor, however, were northern Thailand and the adjacent Laos uplands north of the Mekong Ri ver. In this area, just as in the rest of Thailand during the muang clustered along the waterways. As these polities grew in size and influence over the course of th e first millennium AD, so their leaders grew in stature and ambition. By the second half of the thirteenth century, two independent but allied kingdoms, each composed of numerous muang had emerged in the region and begun to stand in resistance to the exp anding Khmer empire. The Mai, just north of the old Dvaravati period capital of Hariphunchai. The second was Sukhothai, ruled by the legendary King Ramakhamh aeng and situated in the northern central part of the Chao Phraya basin near the modern city of Phitsanulok. It was Ramakhamhaeng who reaffirmed Theravada Buddhism as the predominant religion of the state, and who is also believed to be the author of the oldest known inscription (dated to AD 1292) written in the Thai language and script, the latter of which wa s derived from the Khmer script 3 The art and architecture at Sukhothai incorporated Indian, Chinese, Khmer and Dvaravati influences, and in so doin g would presage the cultural diversity of modern Thailand (Briggs 1945; Rinehart 1981; Bunnag 1984; Higham and Thosarat 1998; Baker and Phongpaichit 2005). Indeed, today the modern Thai 3 The Khmer script was itself adapted from scripts borrowed from the languages of southern India (Higham 2002; Baker and Phongpaichit 2005)
60 people look to the founding of Sukhothai in 1238 as the birth of the ur Thailand: the first nation that was truly Thai, and the first from which the modern Thai nation can claim direct descent (Rinehart 1981; Higham and Thosarat 1998; Baker and Phongpaichit 2005). As the kingdoms of Lanna and Sukhothai expanded during th e thirteenth and fourteenth the far flung rural fringes of the Khmer empire had suffered under the heavy taxation required to less growth, but they had received few of the benefits of citizenship in the flourishing and rapidly advancing civilization. The emergent prosperity of Sukhothai and Lanna must have been a tempting alternative for communities that were of perhaps less tha n wholehearted and steadfast allegiance to distant Angkor (Rinehart 1981; Higham 2002; Baker and Phongpaichit 2005). By the middle of the fourteenth century, Sukhothai and Lanna had come to control most of the northern and central Chao Phraya catchment, w hile the new kingdom of Ayutthaya had risen in the Chao Phraya delta and southern portion of the basin, leaving only northeastern Thailand under Khmer hegemony. Ayutthaya, founded in AD 1350, soon outstripped its northern neighbors in economic and demogr aphic strength, owing largely to a strategic location in the Chao Phraya delta that allowed this kingdom to control the bulk of maritime trade with China and India (and later, Europe) (Rinehart 1981; Stark and Allen 1998; Baker and Phongpaichit 2005). In AD 1378, Ayutthaya conquered Sukhothai, and then turned its attentions towards Angkor and Lanna. Over the next century, many wars and rebellions would rage back and forth between these three kingdoms, with Ayutthaya, bolstered by its near monopoly on acce ss to foreign trade, mercenaries and superior foreign weaponry, ultimately emerging victorious. By the sixteenth century, the
61 kingdom of Ayutthaya (known to t 4 European contact in 1511) controlled a terr itory encompassing the area of modern Thailand as 1981). Like Angkor bef ore it, the society and culture of the kingdom of Ayutthaya flourished on within its borders and of the vast commercial opportunities to be found in foreign trade with empires beyond them. And like Sukhothai before it, Ayutthaya would draw its cultural richness from both without and within. Its kings looked to India, the powerful gods of the Hindu pantheon, the righteous benevolence of Buddhist philosophy and prac tice, and the ritual and spiritual grandeur of the Brahmans as sources for their power, authority, legitimacy and apotheosis. Much of architecture in the city, particularly the temples and royal buildings, invoked the Khmer styles of Angkor Wat, and the K hmer empire as a whole was reimagined as a gilded and exalted ancestor of Ayutthayan civilization. Trade with China and Europe brought new styles of politics, art, literature, philosophy, music and dress to the kingdom, and the foreign traders and merchan ts who settled in Ayutthaya enriched the life of the city (Rinehart 1981; Stark and Allen 1998; Baker and Phongpaichit 2005). Ayutthaya blossomed into a mature, sophisticated and cosmopolitan society which attracted both admiration and envy from all corne rs of the world. 1767, the city was sacked by the Burmese, and its wealth, cultural treasures, and even its people 4 tra nsliteration into Portuguese (Baker and Phongpaichit 2005).
62 were hauled away en masse by the conquering army. After the first attack, the city was left to stand as a vassal to the Burmese empire at Pegu until the reign of Naresuan the Great (1590 siege left Ayutthaya in r uins and was the prelude to nearly a half century of relentless aggression by the Burmese, intended to utterly destroy the kingdom of Ayutthaya and its people. During this time, nearly every major settlement between Chiang Mai and the Isthmus of Kra was d estroyed and depopulated; many cities and towns would not recover to their prewar condition until the mid to late nineteenth century. With characteristic Thai resilience, however, the capital and along with it, the first shreds of renewed Siamese indepen dence would be restored much more quickly, and in 1782, both the modern capital of Bangkok and the ruling Chakri dynasty were founded (Rinehart 1981; Baker and Phongpaichit 2005). The Kingdom of Siam thus established would continue to grow and evolve into the modern Kingdom of Thailand, a land and a people now, as seemingly at all times throughout their history and prehistory, very much at the center of events in Southeast Asia, and deeply interconnected to the wider networks of commerce and culture in the East and beyond
63 Figure 2 1. Selected P aleolithic archaeological sites in Southeast Asia. ( Redrawn from original map data obtained from the CIA World FactBook ( 2010 ) and used with permiss ion. Site locations following Bellwood 1997; Higham and Thosarat 1998; Higham 2002 ). A. Ban Mae Tha, Ban Don Mun, Khao Pah Nam B. Tham Khuyen C. Lang Rongrien, Moh Khiew D. Hoabinhian type sites E. Sai Yok, Tham Ongbah F. Heap Cave, Khao Talu, Ment Cave G. Spirit Cave, Steep Cliff Cave Banyan Valley Cave H. Gua Cha I. Gua Gunung Runtuh J. Nong Nor K. Khok Phanom Di
64 Figure 2 2. Selected Neolithic archaeological sites in Southeast Asia. ( Redrawn from original map data obtained from the C IA World FactBook (2010) and used with permission. Site locations following Bellwood 1997; Higham and Thosarat 1998; Higham 2002 ). A. Ban Chiang B. Ban Kao C. Ban Tha Kae, Non Pa Wai, Khok Charoen D. Gua Cha E. Heap Cave, Khao Talu F. Khok Phanom Di G. Lang Rongrien H. Non Nok Tha I. Phung Nguyen sites
65 Fi gure 2 3 Selected Metal Age archaeological sites in Thailand ( Redrawn from original map data obtained from the CIA World FactBook (2010) and used with permission. Site locations following Higham and Thosarat 1998; Higham 2002 ). A. Ban Chiang B. Ban Don Ta Phet C. Ban Lum Khao D. Ban Na Di E. Non Nok Tha F. Non Pa Wai, Nil Kham Haeng G. Nong Nor H. Ob Luang I. Tham Ongbah
66 Figure 2 4. Selected state level archaeological sites in Southeast Asia. ( Redrawn from original map data obtai ned from the CIA World FactBook (2010) and used with permission. Site locations following Higham and Thosarat 1998; Higham 2002 ). A. Angkor B. Ayutthaya C. Core Dvaravati s ites D. Funan s ites E. Haripunchai F. Isanapura (Chenla) G. Phimai H. Sukhothai
67 CHAPTER 3 PREHISTORY AND POPUL ATION HISTORY IN THA ILAND, PART II: THE VARIED LINES OF EVIDENCE In truth we know nothing, for truth lies in the depth. Democritus The discussion of the foregoing chapter has, hopefully, provided the reader with a useful prcis of the history of human occupation and endeavor in Thailand. But while an appreciation of when and how the p rehistoric communities of Thailand attained the critical milestones of cultural evolution provides us with important landmarks for understanding the origins of the modern Thai nation and its people, it tells us little about who these people are who their ancestors were and whence these forebears came. Archaeological investigations can, of course, provide some insight into these questions, as the spread of artifacts of a given type can also be used to chart the spread of peoples. But these are only indir ect insights: artifacts can be traded widely among groups with little in common culturally, historically or genetically. More direct evidence comes from the spread of languages, which requires closer and more extended contact, and from the spread of gene s, which requires closer contact still. Such data are not easily obtained, though, particularly for prehistoric populations. Researchers may find themselves hard pressed to reconstruct the basic features of an ancient language, or to convincingly connec t that language to a given ancient population absent written texts (which appear comparatively late in Southeast Asia). Genetic evidence from ancient populations is similarly difficult to come by, as curators and site directors are understandably reluctan t to allow destructive sampling of scarce and precious ancient skeletal remains. Even if such permission is granted, the financial expense and technological difficulties of extracting and sequencing ancient DNA samples may prove equally prohibitive.
68 On the other hand, however, physical anthropology analyses of craniometric and dental morphology have long been recognized as valuable proxy measures of biological relationships among geographically and temporally disparate populations, in circumstances where appropriate genetic data are unavailable (Howells 1973, 1989; 1995; Spielman and Smouse 1976; Van Vark and Howells 1984; Turner 1987, 1990; Cheverud 1988; Buikstra et al. 1990; Li et al. 1991; Pietrusewsky et al. 1992; Hanihara 1993; Nakbunlung 1994; Brac e and Tracer 1995; Pietrusewsky 2000, 2005, 2006; Relethford 2002; Jantz 2004; Roeseman 2004; Matsumura and Hudson 2005; Matsumura 2006; Spradley 2006). As was noted in the introduction, these physical anthropology analyses have the additional multiple ad vantages of being 1) non destructive; 2) inexpensive and technologically easy to execute; and 3) founded upon traits known to be of high heritability and therefore reasonable surrogates for genetic markers of population affinities (Howells 1973; Spielman and Smouse 1976; Susanne 1977; Sjvold 1984; Turner 1987, 1990; Cheverud 1988; Buikstra et al. 1990; Li et al. 1991; Relethford and Harpending 1994; Konigsberg and Ousley 1995; Bellwood 1997; Pietrusewsky 1994, 2000, 2005, 2006; Relethford 2002; Roseman 20 04; Carson 2006). Although obviously not as direct a measure of population relationships as genetic analyses are, reconstructions of population history based on studies of craniometric and dental morphology are nevertheless considered superior to those ba sed on linguistic and archaeological data (Hudson 1999; Matsumura and Hudson 2005). They are not without their limitations, however, as the multivariate statistical methods underpinning these analyses derive their power from the consideration of moderatel y large numbers of variables from moderately large sample sizes, both of which can be difficult to come by in small, typically fragmentary and incomplete archaeological skeletal populations. Reconstructions of population relationships may be further compl icated by the frequent lack of
69 secure dates for archaeological skeletal populations included in the analyses (Bellwood 1997; Pietrusewsky 2000; Matsumura 2006). There is thus no single analytical panacea for resolving questions of population history and th e relationships among ancient and modern peoples. Indeed, Cavalli Sforza and colleagues (1988:6005; 1991), as well as other researchers (Bellwood 1993, 1997; Diamond and Bellwood is to consider archaeological, linguistic, and genetic (and by extension, biological ) lines of evidence in conjunction, and that the most powerful and reliable conclusions will be drawn when these varied lines of evidence are all in accord Although the available studies of archaeology, linguistics, genetics and craniometrics in the myriad ancient and modern populations of Southeast Asia do not provide full coverage of the relevant geographic regions, time periods, ethnic groups or skeleta l collections, there are substantial areas of overlap. These are sufficient to provide meaningful insights into the questions of population history within the region and to inform the debate over population continuity versus discontinuity and endogenous v ersus exogenous origins. Given the scant evidence for Homo erectus the Recent African Origins model of modern human origins, it is held here that H. erectus in the origins of the modern populations of Southeast Asia in general is negligible at best. We will thus forgo any further consideration of this fossil species, and the voluminous debate over the nature of its relationship to H. sapiens and instead turn our attention to the first anatomically mod ern humans in Thailand. As noted in Chapter 2, these populations are primarily known from sites in peninsular Thailand, such as Moh Khiew and Lang Rongrien, and are dated to ca. 40,000 to 25,000 years BP (before present) (Anderson 1990; Pookajorn 1991, 19 94; Nakbunlung 1994;
70 Bellwood 1997; Ha Van Tan 1997; Higham and Thosarat 1998; Higham 2002; Matsumura and Pookajorn 2005). Skeletal evidence for these early inhabitants is scarce in the utmost, and genetic and linguistic data are, of course, all but nonex istent. The ages for the few early H. sapiens that have been found in mainland Southeast Asia make them roughly contemporary with or perhaps a little younger than the majority of the earliest known H. sapiens fossils from island Southeast Asia and Austral ia (Kline 1999; Bellwood 1997; Howells 1997; Matsumura and Pookajorn 2005). Since, as nearly all authorities agree, the expanding populations of anatomically modern H. sapiens would have had to pass through Thailand to reach island Southeas t Asia and ult imately Australia 1 and Melanesia, it has been hypothesized that early H. sapiens from each of these regions should display mutual biological affinities. Bellwood (1997), Howells (1997), Matsumura (2006) and others (Brace and Hunt 1990; Matsumura and Hudson 2005; Matsumura and Pookajorn 2005) have argued that such affinities are demon aboriginal populations of Australia and Melanesia. Skeletally, the australoid phe limbed body build. In living populat ions, the skeletal phenotype is typically accompanied by dark skin and straight to wavy hair (Howells 1997:171; Matsumura and Hudson 2005:182 183; Matsumura 2006:38; Comas 1960). Outside of the earliest fossil material from Australia, such as the Lake Mun go, Willandra 1 Most likely colonized ca. 40,000 to 30,000 years ago, but possibly as early as 60,000 years ago (Bellwood 1997; Howells 1997; Higham and Thosarat 1998; Kline 1999; Thorne et al. 1999; Bowler and Magee 2000; Brown 2000; Gillespie and Roberts 2000).
71 Lakes, Keilor and Coobool Creek specimens, the australoid phenotype has also been attributed to the early fossil H. sapiens the oldest H. sapiens in Southeast Asia) and from Tabon Cave on the Philippine island of Palawan (ca. 23,000 to 16,500 years BP) (Brothwell 1960; Macintosh 1978; Bellwood 1997; Howells 1997; Krigbaum and Datan 1999; Dizon et al 2002; Matsumura and Hudson 2005; Matsumura and Pookajorn 2005; Matsumura 2006 ). An extensive analysis of the Moh Khiew skeleton from peninsular Thailand establishes the concurrent presence of the australoid phenotype in mainland Southeast Asia populations during the late Pleistocene. This conclusion is based on findings of signif icant cranio and odontometric, as well as general morphological, similarities between the Moh Khiew skeleton and both the Coobool Creek fossils and skeletal samples from modern Australian Aborigines (Matsumura and Pookajorn 2005; Matsumura 2006). Althoug h Lang Rongrien has yielded no skeletal remains for analysis, it is broadly assumed that the stone tool assemblage there, similar as it is to that from Moh Khiew, also Melanesians of the late Plei stocene (Higham and Thosarat 1998:24). Other studies have established the australoid affinities of a number of Hoabinhian (i.e., terminal Pleistocene to early to mid Holocene) skeletal samples as well, including those from Gua Cha, Gua Gunung Runtuh and ot her sites in peninsular Malaysia, Ma Da Dieu and Mai Da Nuoc in Vietnam, and the early Holocene sites of Tam Hang and Tam Pong in Laos (Barth 1952; Trevor and Brothwell 1962; Cuong 1986; Bellwood 1993, 1997; Matsumura and Zuraina 1995, 1999; Matsumura and Hudson 2005; Matsumura and Pookajorn 2005; Matsumura 2006). As part of their broader analyses of the Moh Khiew skeleton, Matsumura and Pookajorn (2005) demonstrated that several skeletons from early to middle Holocene contexts in Laos and Vietnam
72 also cl uster closely with both Moh Khiew and Coobool Creek. Even the contentious Wadjak fossils from Java (dated to ca. 6500 years ago) have been argued to be of generalized australoid affinities (Dubois 1922; Weidenreich 1945; Bellwood 1997; Howells 1997; Matsu mura and Pookajorn 2005; Matsumura 2006). These morphological conclusions are also somewhat reinforced by the work of Oota et al. (2001), if the results of their analyses of ancient DNA (aDNA) from the Moh Khiew skeleton can be accepted as valid (always a dicey consideration where aDNA sequence data from fossilized remains is concerned). Their analyses demonstrated that the mitochondrial DNA (mtDNA) sequence of the late Pleistocene Moh Khiew skeleton and that of another specimen from later Hoabinhian cont exts within the cave (dated to ca. 12,000 years ago; early Holocene) both cluster predom inantly with modern Australian Papua New Guinean and Semang Negrito samples all of which are considered populations of australoid affinities. The persistence of the australoid phenotype in mainland Southeast Asia until at least the early Holocene suggests that these skeletons are representative of a well established and long surviving population within the region, one that must be accounted for in any consideration o f the development of the modern populations of Southeast Asia (Bellwood 1997; Howells 1997; Matsumura 2006). Most authorities consider the Holocene australoid populations in Southeast esia, joined by common descent from a shared late Pleistocene ancestor (Bellwood 1997:91; Howells 1997; Matsumura and Hudson 2005; Matsumura 2006). Further, it is generally believed that the varied modern hunter gatherer populations of Southeast Asia, inc luding the Semang and other orang asli or aboriginal peoples, of peninsular Malaysia, the Andamanese, and similar populations in the Philippines actually
73 constitute relict populations o f Australo Melanesians in Southeast Asia descended, like the modern Australians and Melanesians themselves, from the late Pleistocene parent population (Barth 1952; Brace et al 1991; Bellwood 1993; 1997; Howells 1997; Matsumura and Hudson 2005; Matsumura 2006; but see Hanihara (1992,1993) for a widely divergent interpretation of the phenotypic affinities and population history of these groups). Juxtaposed against the ancient and modern Australo Melanesians of Southeast Asia are the current predominant in habitants of the region, who possess a phenotype freque ntly termed 2 (Barth 1952; Comas 1960; Bellwood 1993, 1997; Howells 1997; Matsumura and Hudson 2005; Matsumura 2006). This phenotype, distinctly different from that of the australo ids, is characterized skeletally by a short, light to medium build and gracile, brachicephalic crania, and is accompanied in living individuals by a medium complexion, straight dark hair, and low to moderate frequencies of the epicanthic fold (Comas 1960; Coon 1966; Bellwood 1997; Howells 1997). In a narrow sense, the southern mongoloid phenotype can typically have lower frequencies and less pronounced expressions of the m orphological features 2 ailing from the Far East (Brace 1995:171; Li et al. 1991; Brace and Tracer 1992; Sauer 1992; Kennedy 1995). However, in the main, and ( and particularly cranial) morphology of the populations of Southeast Asia. This understandably objectionable terminology is necessitated both by the fact that the southern mongoloid phenotype is a distinct sub type within the to the peoples of the Indian subcontinent. To use any of these to refer to the physical characteristics of the modern and archaeological populations of Southeast Asia and neighboring areas would therefore be both misleading and populations characterized by this phenotype, since the distribution of these popul ations both in archaeological and modern contexts does not neatly correspond to a specific geographic region of origin that could provide a convenient descriptive term. For example, many populations long resident in southern China possess the southern mon goloid phenotype, but southern China is not generally considered a part of Southeast Asia (but more on that n to refer to the phenotypes characteristic of certain populations modern populations, where permitted by concerns of clarity a nd accuracy, will generally be referred to by the more
74 considered stereotypical of the Asian phenotype: these include the prominent zygomatic bones shaped and winged incisors, which are often col mongoloid phenotypes that repeated craniometric analyses have arrived at the conclusion that the Pleistocene inhabitan (Jantz and Owsley 2001:153; Kamminga and Wright 1988; Van Vark and Dijkema 1988; Bellwood 1997; Sarich 1997; Matsumura and Zuraina 1999). Thus, the physical anthropological distinction between the australoids and southern mongoloids is widely recognized as an ancient and fundamental division of humanity, one that avalli Sforza et al. 1988; Brace and Hunt 1990; Li et al. 1991; Pietrusewsky et al. 1992; Howells 1989, 1997; Pietrusewsky 1990, 2000, 2005, 2006; Jantz and Owsley 2001; Matsumura and Hudson 2005; Matsumura 2006; Turner 1990, 2006). However, at this point it must be said, of course, that the overall pattern of human variation over the vast expanse encompassing mainland and island Southeast Asia, Australia, New Guinea and Melanesia must be clinal in nature just as it is across broad swaths of geography the world over (Bellwood 1997; Howells 1997). Indeed, Bellwood (1997:75) singles out the modern population of Indonesia as singularly demonstrative in the west and nor south and east. It is only when the populations at either end of this continuum are brought into apposition that the contrast between the australoid and southern mongoloid phenoty pes is thrown into sharp relief as it is in modern Southeast Asia. This contrast leads inescapably to the
75 question of how it is that the entirety of Southeast Asia, once home to a long resident population of clear australoid affinities, came to be dominat ed by the markedly different southern mongoloids. When and where and by what means did this population transition take place? What is the relationship between the various ancient populations of Southeast Asia and the modern inhabitants of the region? It is here that the great debate of Southeast Asian population history the debate of exogenous versus endogenous origins, of essential discontinuity versus essential continuity is centered. The debate is traditionally framed in terms of two competing hypothe 34; Jacob 19967; Nakbunlung 1994; Bellwood 1997; Pietrusewsky 2006). Both models acknowledge the presence of an indigenous late Paleolithic (i.e., terminal Pleistocene and early to middle Holocene) australoid population, resident in mainland Southeast Asia from at least 26,000 years ago. Both models also agree that the Neolithic revolution and the advent of wet rice agriculture in Southeast Asia is the nexus around which th e entire debate revolves (Bellwood 1993, 1997; Matsumura and Hudson 2005; Pietrusewsky 2005, 2006). But beyond this small common ground, the two models diverge widely. In brief, the former holds that this original population of australoids was subsequent ly displaced in some fashion whether by direct replacement or indirectly via admixture and gene flow by expanding populations of southern mongoloid Neolithic rice agriculturalists from southern China, following the development of rice agriculture in the ce ntral and lower Yangzi River valley approximately 8000 years ago. These Neolithic populations are believed to have spread from southern China south through mainland and island Southeast Asia approximately 5000 years ago, and eventually on to Micronesia, P olynesia and Remote Oceania during the following millennia (Brace et al. 1991; Brace and Tracer 1992;
76 Bellwood 1993, 1997; Glover and Higham 1996; Higham 1996; Howells 1997; Higham and Thosarat 1998; Matsumura and Hudson 2005; Matsumura 2006). Thus, the c haracter of the southern mongoloid agricultural populations on the existing australoid hunter gatherer population (Barth 1952; Coon 1966; Bellwood 1997; Matsumur a and Hudson 2005; Matsumura 2006). Thus, under the Two Layer model, Southeast Asian populations are locally continuous only back to the time of the Neolithic expansion. In contrast, the Regional Continuity model holds that there was no such Neolithic e xpansion of southern mongoloid populations into Southeast Asia. Rather, the modern populations of the region evolved in situ in conjunction with the localized, endemic development of rice agriculture, independent of the advent of rice agriculture in China The modern populations of Southeast Asia are therefore directly regionally continuous not only with the their immediate Neolithic forebears, but also with all the populations who have inhabited Southeast Asia since the terminal Pleistocene (Turner 1987, 1990; Nakbunlung 1994; Hanihara 1992, 1993, 2006; Pietrusewsky 2006). The Two Layer model is the elder of the two hypotheses, having been the sole model of population origins in Southeast Asia for much of the twentieth century (Matsumura and Hudson 2005 :182). As such, it is underpinned by a considerably larger body of research, stemming from a greater diversity of scientific fields, than is the Regional Continuity model (Bellwood 1997). Lines of evidence in support of the Two Layer model come from arch aeology (Bellwood 1993, 1997; Glover and Higham 1996; Higham 1996, 2002; Higham and Thosarat 1998), linguistics (Benedict 1942; Higham 1996,2002), physical anthropology (Mijsberg 1940; Barth 1952; von Koenigswald 1952; Coon 1966; Brace et al. 1991; Brace a nd Tracer 1992; Matsumura and
77 Hudson 2005; Matsumura 2006), and genetics/biochemistry (Ballinger et al. 1992; Nei and Roychoudhury 1993; Melton et al. 1995; Omoto and Saitou 1997; Oota et al. 2001 ; Lertrit et al. 2008). While a comprehensive summary of al l the relevant literature would require several volumes, the following discussion will attempt to summarize the most salient research from each of these fields. Archaeology While it may have been interactions with the living peoples of Southeast Asia dur ing the Age of Exploration and the Colonial Era that prompted the first questions among Western observers about their origins, it was the excavations of the sites inhabited by their ancestors that has propelled these questions to the forefront of modern sc ientific inquiry in the region (Higham and Thosarat 1998; Baker and Phongpaichit 2005). The archaeological excavations have also provided the raw materials including the human remains, the artifacts and the earliest inscriptions through which those questi ons are addressed. Although these are often presented separately, and merit consideration as independent lines of evidence (Bellwood 1993), it must be remembered that these fields nevertheless represent different facets of the same broad phenomenon of hum an agency, and therefore are necessarily overlapping and mutually reticulate in nature (Bellwood 1996:882; Cavalli Sforza et al. 1988; Higham 1996; Diamond and Bellwood 2003). As was noted previously, it is the search for the origins of rice agriculture in Southeast Asia that has galvanized much of the debate surrounding the origins of the populations of Southeast Asia. It is an understatement in the utmost to say that the inception of agriculture is a revolutionary event regardless of when and how it oc curs within a given region and one that always engenders much excitement and debate among researchers. It has long been
78 ies have led many to wonder whether agriculture did not also have an independent origin here (Bellwood 1997; Higham and Thosarat 1998). On the other hand, if rice agriculture came to Southeast Asia from elsewhere, then perhaps its people did, too. Many o f the earliest serious archaeological excavations in Thailand, such as those at Spirit Cave and Banyan Valley Cave, led to exuberant proclamations of an early and independent origin of rice agriculture in Thailand, and much of the effort in these excavatio ns was focused on recovering any trace of domesticated rice (Gorman 1969; Solheim 1972; Higham and Thosarat 1998; Higham 2002). But at each turn, the rice recovered from these early to mid Holocene sites has been shown to be either a wild type or an intru sive deposit from later Neolithic occupations (Higham and Thosarat 1998; Higham 2002). It is not until the mid third millennium BC that we find unequivocal evidence of rice agriculture in Southeast Asia. Today, as Higham and Thosarat (1998:76) emphaticall y state, No excavated site in [Thailand] has any evidence for agricultural settlement before about 2300 BC, nor have concentrations of pottery which might be earlier than this date ever been found during site surveys. Nor do we know of an archaeological s equence suggesting a local transition from hunting and gathering to the Neolithic in Thailand. The pottery concentrations to which they are referring, of course, are the large, distinctive ceramics assemblages from the contemporary sites of Khok Phanom Di, Ban Tha Kae, Non Pa Wai, Non Nok Tha, Ban Chiang, and, most importantly, Ban Kao. It nearly goes without saying that pottery is an important hallmark of a burgeoning agricultural tradition, in which people have a newfound need to store surpluses of food and stockpiles of seed for the future. Styles of pottery manufacture and decoration frequently differ from site to site or region to region, and are the moveme nts of people. The Ban Kao pottery tradition identified by Srensen (1963; Srensen and Hatting 1967) is particularly provocative because it appears across nearly the entire extent of
79 Southeast Asia at precisely the same time and in the same locations as much of the earliest undisputed evidence for domesticated rice and rice agriculture. Upon his analysis of the Ban Kao pottery tradition, with its signature motif of curvilinear bands infilled with impressed or stamped dentate textures and distinctive cari nated tripod vessels, Srensen (1963; Srensen and Although this vi ew was unpopular at the time, as were most theories of cultural change via style pottery vessels have been found in conjunction with the remains of domesticated rice at sites thro ughout Thailand and the rest of Southeast Asia (Higham and Thosarat 1998:70; Sieveking 1954, 1987; Bayard 1972; Gorman and Charoenwongsa 1976; Ha Van Tan 1985; Hutterer 1988, 1993; Anderson 1990; Leong 1991; Bellwood 1993, 1997; Glover and Higham 1996; Hig ham 1996, 2002; Loofs Wissowa 1997; Pigott et al. 1997; Rispoli 1997; Vincent 2003). At the majority of these sites, the appearance of Ban Kao style ceramics typically marks a sharp discontinuity with the preceding cultural deposits, whether they be the aceramic assemblages of inland Hoabinhian sites or the locally produced pottery forms found at thriving late hunter gatherer estuarine and coastal settlements such as Khok Phanom Di and Nong Nor (Bellwood 1997; Higham and Thosarat 1998; Vincent 2003). At other sites, such as Non Nok Tha and Ban Chiang, B an Kao style pottery accompanied the first settlers of these newly established communities. Yet nowhere in Southeast Asia is there a ceramic culture that can be considered a direct and immediate precursor to the Ban Kao style, as would be expected in a lo cally evolving pottery tradition. Additionally, it was realized that there is a significant correlation between the temporal and geographic distributions of the Ban Kao pottery forms within Southeast Asia: they arrive first in
80 the northern part of the re gion, appearing as the Phung Nguyen culture of the Red River valley in northern Vietnam, around 2500 to 2000 BC. A few centuries later, the pottery appears farther to the south at Ban Kao itself, as well at Ban Chiang (both dated to 2300 to 1500 BC) and o ther sites throughout the Khorat Plateau and Chao Phraya River valley (Bellwood 1993, 1997; Higham 1996, 2002; Higham and Thosarat 1998). Somewhat later and farther south still, as Khok Phanom Di and Ban Kao begin to appear in sites all down the Malay peninsula, southward for 1500 km, into Collectively, these observations strongly suggest that the Ban Kao pottery forms represent an invasive element thro ughout the region, one with distinctly different origins than the foregoing home grown material culture of the Paleolithic and Mesolithic hunter gatherer populations of Southeast Asia. One of the most striking features of the Ban Kao culture, which greatl y reinforces the interpretation of a single, external origin and subsequent widespread dispersal throughout the region, is the overall homogeneity of the culture (Bellwood 1993, 1997). archaeological similarities right down the peninsula, that agriculture and the distinctive associated artifact forms were simply adopted wholesale by preexisting Hoabinhian foragers, without considerable Where, then, is th e ultimate source of the ur Ban Kao pottery tradition, and along with it, rice agriculture? Phung Nguyen and other contemporary sites along the Red River valley in Vietnam have provided tantalizing clues to this question, in the form of nephrite and jadei te bangles, beads and other ornaments that have clear stylistic and technological links with the material culture of the Chinese Neolithic (Higham 2002). The presence of Chinese influenced
81 artifacts at sites along the Red River valley is significant, beca use the importance of rivers as avenues of commerce and population movement is a recurrent thread in Southeast Asian prehistory. Many researchers have noted that the major rivers of South and Southeast Asia, including the Red, Mekong, Chao Phraya, Salween Irrawaddy and Brahmaputra Rivers, are southern China, along the upper reaches of the Yangzi River (Figure 3 1) (Higham 1996:115; Chang 1976; Watanabe 1984; Blust 19 93a, cited in Higham 1996; Blust 1996; Higham and Thosarat 1994, 1998; Glover and Higham 1996; Higham 2002). Following these spokes back to the hub leads to a number of sites in Yunnan, such as Baiyankun, that are also dated to the mid third millennium BC and that have yielded both domesticated rice and Ban Kao style dentate stamped pottery (Higham 1996, 2002). From these sites in the upper Yangzi, one needs only to travel to the middle and lower Yangzi Valley to reach the sites of Pengtoushan and Fensh anbao in Hunan province and Hemudu in Zhejiang province, where domesticated rice was unquestionably cultivated as early as 8000 BC (Glover and Higham 1996; Bellwood 1997; Higham 2002; Matsumura and Hudson 2005) Working forward, then, it is easy to see how rice agriculture could have spread from its place of origin in the lower Yangzi, first upstream along the great river to Yunnan province and beyond, and then from this hub downstream along the major rivers of Southeast Asia to the rich, fertile lowlands of the Red, Mekong and Chao Phraya basins and deltas, the Khorat Plateau, and the great flood plains of central Cambodia (Chang1976; Watanabe 1984; Spriggs 1989; Blust 1993a, cited in Higham 1996; Glover and Higham 1996; Higham 1996; 2002; Matsumura and Huds on 2005).
82 Linguistics It is one thing to use an archaeologically durable pottery tradition as an index for tracking the spread of the more archaeologically ephemeral phenomenon of rice agriculture. It is something quite different to infer the movement o f populations solely from the spread of a ceramic typology and/or subsistence strategy. The Ban Kao pottery tradition and wet rice agriculture are memes (Dawkins 1976) that require no large scale, permanent movement of peoples to account for their spread across Southeast Asia. Their distribution within the region can be and has been explained simply, if somewhat unsatisfactorily, by mechanisms of passive cultural exchange and diffusion, whereby the patterns and processes spread but the people stay put. I nstead, if we are making the argument of the Two Layer Model, that agriculture was brought to Southeast Asia by southern mongoloid populations who displaced or replaced the indigenous australoid populations, then we are in need of lines of evidence involvi ng sensu Dawkins 1976) that are more implicitly reliant on population migrations to carry them into new areas. One such system of replicators is language. Cavalli Sforza and colleagues (Cavalli Sforza 1991; Cavalli Sforza et al. 1988) have established that cluster analyses of linguistic and population genetic data tend to produce strikingly similar phylogenetic trees, indicating a strong correlation between the flow of genes and the flow of languages among populations (Cavalli Sforza 1991; Cavalli Sforza et al. 1988). This correlation is attributed to the fact that languages behave in much the same way as gene pools do (Cavalli Sforza 1991). When a single parent population is divided into two separate daughter populations via allopatric me chanisms such as migration and geographic isolation, their gene pools become increasingly divergent due to the actions of genetic drift, mutation and natural selection of existing variation. Their languages will diverge in similar fashion, due to similar mechanisms: a sort of linguistic drift, characterized by
83 distinct initial dialects in each daughter population (including differences in accent, lexicon and idiom); the generation of novel variation via neologisms; and ongoing evolution of the dialect/pro tolanguage in response to changing social and environmental contexts (Bryson 1990). As Cavalli linguistic and genetic patterns that bear the marks of shared branching points. He nce, some measured and used to reconstruct a timeline for these branching events a linguistic clock akin to the molecular clock (Cavalli Sforza et al. 1988; Cavalli Sforza 1991; Bellwood 1997). Of course, languages can spread to new areas through diffusion without accompanying large scale population movements. However, this phenomenon is most common in the areas of interface between different linguistic popu lations, where the speakers of different languages are already in frequent contact with one another, and therefore have an economic or social need to bridge their mutual communication gap. This mode of language transfer seldom results in the wholesale rep lacement of one language with another; rather, it more commonly leads to the creation of new languages, beginning as pidgins and creoles, which draw on elements of both languages for their structure, syntax and vocabulary (Bryson 1990). Correspondingly, g ene flow the respective gene pools, but will not amount to a total replacement of one gene pool with another. True in situ language replacement generally occurs o nly in conjunction with significant population migrations, via two mechanisms: elite dominance where the language of an invasive ruling class is imposed upon the subjugated populace; and numerical superiority where some technological or economic advanta ge allows the immigrant population and their language to outstrip and eventually absorb or otherwise overwhelm the indigenous population and their
84 language (Renfrew and Bahn 1996:447; Renfrew 1987; Cavalli Sforza et al. 1988; Cavalli Sforza 1991). Both me chanisms are likely at play in the linguistic history of Thailand, but for now, we are primarily concerned with the latter. Today, the primary languages of Southeast Asia fall into four families: Austroasiatic: includes Vietnamese; the Khmer language o f Cambodia; the Munda language of northeastern India; and Mon, which is currently spoken only by small populations on the Andamanese coast and in the hills between Burma and Thailand, but which once had a much wider distribution across much of Thailand. Au stronesian: includes Formosan, Malay, Filipino, and the languages of island Southeast Asia, coastal Papua New Guinea, Polynesia and Micronesia; also includes Cham, which is currently spoken only along the central Vietnamese coast and in a small area of th e adjacent uplands, and in isolated pockets in southeastern Cambodia. Like Mon, Cham once had a much wider distribution throughout Vietnam. This family also includes Malagasy, the language of Madagascar. Daic (or Tai Kadai): includes Thai, Laotian, Kada i (small linguistic isolates spoken primarily in Hainan and Southern China) and the languages of some of the Hill Tribes of northern Thailand and Laos (Hmong Mien) (Benedict 1942; Briggs 1945; Diffloth 1991; Bellwood 1997; Howells 1997; Higham 1996, 2002; Baker and Phongpaichit 2005). Sino Tibetan: includes Chinese as well Burmese, Karen and the languages of other Hill Tribes living in the mountainous region between Burma and Thailand), and, with the exception of Chinese, is perhaps the least well understo od of the four groups. It will be of little further concern here. Not surprisingly, the interrelationships of these languages are rather complex, and are neither fully understood nor agreed upon. We can gain some sense of their time depth within Southeas t Asia from the various inscriptions that have been recovered from archaeological sites in the region. We know that in the Chao Phraya basin, the inhabitants of Dvaravati and Hariphunchai, the earliest state level societies in Thailand (sixth to twelfth c entury AD), spoke Mon (Briggs 1945; Rinehart 1981; Bechert 1984; Bunnag 1984; Higham 1996, 2002; Higham and Thosarat 1998). To the east, in the lower Mekong Valley, the people of Angkor (ninth to fifteenth century AD) and its predecessor, Chenla (seventh to ninth century AD), spoke Khmer. Central and southern coastal Vietnam were dominated by the Cham speakers of the kingdom of
85 Champa (fifth to fifteenth century), while the Red River valley to the north forms the core of the area inhabited by Vietnamese s peakers (Higham 1996; 2002). Over the intervening centuries, as these various states waxed and waned, the speakers of these languages ebbed and flowed across Southeast Asia before settling into their modern geographic distributions. The Thai language its elf is a relative latecomer to the area, known certainly only from the late thirteenth century AD onwards, and superimposed upon the linguistic landscape of mainland Southeast Asia by the rapid ascendancy and expansion of kingdoms of Sukhothai, Lanna and A yutthaya (Barth 1952; Bayard 1979; Briggs 1945; Rinehart 1981; Higham and Thosarat 1998; Higham 2002; Baker and Phongpaichit 2005). But what of the prehistoric populations of Southeast Asia what languages did they speak? Since we know that the scripts in which Mon, Cham, Khmer and Thai were written were derived from Sanskrit (itself most likely introduced along with Buddhism during the late Iron Age), we can reasonably assume that these languages had been resident in Southeast Asia for some time. Less wel l established languages would likely have been easily and thoroughly overwhelmed by a linguistic tradition as robust and sophisticated as Sanskrit (Bellwood 1997). Instead, as far as we can tell, the populations of Southeast Asia co opted the script and n umerous loan words, but retained the central character of their respective indigenous languages. This resilience is a common occurrence in the evolution of languages: a given language will rarely absorb all of the influence of an unrelated language, rega rdless of how intensive and wide ranging that influence may be. Similarly, divergent languages descended from a common predecessor will, despite profound changes in structure and vocabulary over the course of millennia, nevertheless retain seemingly immut able traces of their common ancestry. This being the case, we can look to the lexical content of individual languages and broader phylogenetic
86 relationships between these languages and language families to understand when and where they are likely to have emerged and how they have spread (Cavalli Sforza 1991; Higham 1996; Bellwood 1997). To begin with, it has been suggested that the Daic and Austronesian families can be united in a superfamily known as Austro Tai (Figure 3 2) (Benedict 1942, 1975; Higham 1996; Bellwood 1997; Howells. 1997). This grouping is based on the existence of similarities in grammar and syntax between Thai and the Austronesian languages of Indonesia, as well as extensive cognates for a number of fundamental objects and concepts, in cluding the sun, moon and stars; water, fire, ice and other environmental elements; parts of the body and bodily functions; the words for eat die this I and grandfather ; and, tellingly, the word for rice fields (Benedict 1942:591 592). Via additional r esearch, Benedict (1975) has argued that proto Austro Tai, the common linguistic ancestor of the Daic and Austronesian languages, contained numerous terms related to the wet wet field, le, water buffalo to sow, [to] winnow the proto Austro Tai vocabulary leads, almost ineluctably, to the conclusion that this lan guage has close ties to the emergent southern Chinese Neolithic contexts of the middle and lower Yangzi Valley, and that the descendent languages must have thus spread to their current distributions from this ancient homeland (Benedict 1942, 1975; Higham 1 996; Bellwood 1997). Additionally, it is estimated that the Austronesian and Austro Tai language families may have begun to diverge as much as 7000 years ago, which corresponds well with the best dates for rice agriculture in China (Glover and Higham 1996 ; Higham 1996).
87 As the Daic language family is somewhat less than well studied, owing to the minor status of its member languages (except for Thai), the much more completely reckoned Austronesian language family provides critical insights into the possible developmental a nd migratory paths of agriculture and language in prehistoric Southeast Asia (Higham 1996:112; Bellwood 1997; Matsumura and Hudson 2005). By comparing the points of divergence among the various languages that constitute the Austronesian family, linguists can reconstruct a language the Oceanic language group (the languages of Polynesia and Micronesia) and the languages of eastern Indonesia (Bellwood 1997:103). Be cause the coastal and archipelagic geography of the western Pacific acts to channel and constrain human migration routes (even during periods of lowered sea levels), it is much easier here to trace the spread of languages and speakers back to their place o f origin (Higham 1996; Diamond and Bellwood 2003). Thus, the aboriginal home of the Austronesian language family has been traced back to the island of Taiwan, from whence it spread and diversified: first southward to the Philippines, then arcing simultan eously westward, through Borneo, Java and Sumatra to the Malay peninsula and coastal Vietnam (the Cham), and eastward, to the Lesser Sunda islands of eastern Indonesia, the coastal regions of Papua New Guinea and the islands of western Melanesia, and then onward across the whole of Oceania (Bellwood 1997:105; Benedict 1942; Blust 1984 1985; Higham 1996; Howells 1997). Based on these linguistic reconstructions, Bellwood (1997:117) places the initial Austronesian colonization of Taiwan around 4000 to 3000 BC If we are to entertain the argument that the spread of agriculture is commensurate with the expansion of the Austronesians, then the archaeological record of Taiwan should provide some corroborating evidence. The early Neolithic culture of Taiwan is k nown as the Ta
88 (or TPK) culture, and is dated to 4500 to 2500 BC (Chang 1969, 1970; Spriggs 1989; Bellwood 1997). However, no rice remains have been recovered from TPK contexts; it does not conclusively appear in Taiwan until the following Suo kang period. Instead, the presence of agriculture in general during the TPK period has been suggested both by environmental data (i.e., palynology cores) and by the nature of the material culture itself. TPK assemblages feature incised and cordmarked pot tery, quadrangular stepped stone adzes, bark cloth beaters and ground stone projectiles. These artifacts contrast sharply with the Hoabinhian like flaked pebble tool ] In addition to some broad similarities in artifact type and form between TPK and the Chinese Neolithic site of Hemudu, which unquestionably had rice agriculture as ea rly as 5200 BC, it has been observed that some elements of the TPK pottery assemblage bear a striking resemblance to pottery from other Neolithic sites in coastal southern China. These sites are dated to ca. 5200 to 4200 BC and feature pottery assemblages characterized by decorative bands of dentate stamping (Bellwood 1997:208 213; Chang 1977). By now, this motif should be familiar to the reader as the hallmark of the Ban Kao pottery culture and its posited correlation with the spread of rice agriculture from southern China into mainland Southeast Asia (Higham 1996). These coastal sites are tantalizing candidates for the jumping off point of the initial Austronesian colonization of Taiwan, based both on their dating, which places them as immediate anteced ents of the TPK sites, and on their locations in Fujian and Guangdong provinces, south of the lower Yangzi Valley and directly opposite Taiwan across the Taiwan Strait (see Figure 3 1)
89 It will be recalled from the previous chapter that Solheim (1972) also saw parallels among the TPK culture, pottery he had previously observed in the Philippines, and the pottery he recovered from Spirit Cave. Solheim saw this connection as supporting his hypothesis of an on migrations and expansions, tied to a growing reliance on agriculture, between 8000 BC and AD off by some millennia the Spirit Cave pottery is now dated to 2000 to 1000 BC and is clearly intrusive to the Hoabinhian conte xts from which it was recovered (Higham 2002) there is a great unrecognized kernel of truth in his assessment of the relationship between the movements of agricultural populations and the distribution of this pottery style throughout Southeast Asia. Unfor tunately, Solheim misinterpreted this pottery distribution as evidence of a once widespread population of agriculturalists fragmented into separate island and mainland populations by rising sea levels, rather than as evidence of precisely the type of radia ting population expansion implicit in his framing of the Extensionistic Period and currently hypothesized for the Austronesians (Solheim 1972; Higham 1996). At this point, the hypothesized link between the expansion of rice agriculture and the expansion of the Austronesians may seem perilously close to a circular argument archaeological evidence offered in support of linguistic hypotheses, and linguistic reconstructions used to explain the distribution of the archaeological evidence. But this is not exactl y the case. The summary presented here has been necessarily brief, and cannot adequately convey the full subtlety and depth of the pertinent linguistic and archaeological evidence. But hopefully, what emerges from the foregoing discussion is a realizati on of the parallel distributions of a distinctive Neolithic material culture on one hand, and the Austronesian languages on the other. Where we find one, we also find evidence for the other. This is the
90 purest and simplest definition of a correlation. W hat is compelling in the argument is that these distributions of language and material culture have been established using analytical methods that proceed in opposite directions from their fundamental data to their primary conclusions. Historical linguist s work from the structure and range of present day languages backwards through time to recreate the origins and dispersal of the proto languages of prehistory. Archaeologists, although looking backwards at past cultures, nevertheless work forward through time, charting cultural innovation and evolution and the growth of social complexity over time. That these lines of evidence are thus so convergent in their ultimate conclusions speaks to the scientific merit of the hypothesis they underpin. There is also a great deal more to the argument of the connections between the spread of languages and agriculture in Southeast Asia, as we have not yet considered the role of the Austroasiatic language family in the cultural and biological history of the region. As w as noted above, the Austroasiatic family includes the Khmer, Vietnamese and Mon languages, as well as the Munda language of eastern India. As there are no Austronesian languages on the Southeast Asian mainland other than Cham (which most authorities agree was introduced to the area by more recent immigrants from island Southeast Asia, probably during the first millennium BC), our understanding of the spread of the Austro Tai languages sheds little light on the population history in this core area of Southe ast Asia (Blust 1993b, cited in Higham 1996). However, Blust (1993a, 1993b, 1996), Reid (1994), Higham (1996), and others (Bellwood 1997) have taken up the hypothesis first put forth by Schmidt (1906) that the Austroasiatic and Austro Tai language familie s are both descended from an even higher order superfamily called Austric. As Higham (1996:112) explains, The relationship between the Austroasiatic and Austro Tai languages is a matter of considerable relevance to the prehistorians. If, for example, it could be shown that
91 they are related, then we could seek a common origin and early population expansion in association with rice cultivation in Southeast Asia and eastern India. Conversely, The assumption than Austroasiatic and Austronesian are separate would support more than one transition to rice cultivation and Neolithic expansion, perhaps one in the Yangzi Valley and the other in tropical Southeast Asia (Higham 1996:110). The explication of the Austric hypothesis proceeds in much the same fashion as the argument for the Austronesian expansion, but it is necessarily focused on mainland, rather than island, Southeast Asia. The Austroasiatic languages are divided into two main subfamilies, Mon Khmer (which includes Vietnamese) and Munda. The presence of numerous cognates pestle and mortar, huskin proto language of people living in a Neolithic rice agricultural context. Based on this reconstruction, it is likely that the various Austroasiatic languages diverged from this ancestral proto language beginning around 400 0 years ago (Zide and Zide 1976; Diffloth 1991; Higham 1996; Higham and Thosarat 1998). The homeland of proto Austroasiatic has, like that of proto Austro Tai, been placed in southern China, owing to the presence of Austroasiatic loanwords in the historic and modern dialects of southern China. These loanwords includes words for various animals, including tiger dog fish and crab as well as words for shaman, child, son, ivory, to die and river (Higham 1996:115; Benedict 1942; Hashimoto 1972; Norman and M ei 1976; Norman 1985). Interestingly, some of the dialects in which these loanwords appear are themselves centered in Fujian and Guangdong provinces, noted above as the probable jumping off point for the Austronesian colonization of Taiwan. Like Austro Ta i, the existence of the Austric language superfamily is supported by linguistic evidence of structural and lexical similarities between the Austroasiatic and
92 Austronesian languages. These include a number of the seemingly omnipresent cognates for various facets of rice agriculture (Reid 1994; Higham and Thosarat 1998; Higham 1996, 2002). And as with the Austronesian expansion, it is hypothesized that the incised banded, dentate stamped Ban Kao style pottery found at Neolithic archaeological sites througho ut mainland Southeast Asia is evidence of populations of rice agriculturalists expanding from southern China. But here on the mainland, these migrating agriculturalists carried with them the Austroasiatic languages of Mon, Khmer and Vietnamese, such that when the Austronesian speaking agriculturalists spreading westward across Indonesia reached the Malay peninsula and the Vietnamese coast, they found these areas already inhabited by Austroasiatic speaking peoples (Higham 1996, 2002; Bellwood 1993, 1997). Lastly, just as the Austronesian expansion was constrained by the geography of island Southeast Asia, so too was the Austroasiatic expansion constrained by the geography of mainland Southeast Asia. These peoples appear to have settled preferentially along the inland valley floors and flood plains of the many rivers of the mainland, including the Chao Phraya, Mun, Chi, Mekong and Red Rivers. As was noted in the previous chapter, this represents a largely novel settlement pattern among the peoples of this r egion and was probably driven by the ecological and environmental demands of wet rice agriculture (Higham 1996). Thus, we are confronted with archaeological and linguistic evidence of the separate dispersals of three groups of early Neolithic peoples, from a posited common ancestral homeland in southern China, to various regions of South and Southeast Asia. All of these peoples possessed knowledge of rice agriculture, and the majority of them shared a distinctive dentate stamped pottery tradition. We have clear ideas of the paths of their migrations: the Munda, down the Brahmaputra river to eastern India; the Mon Khmer, down the Chao Phraya, Mekong
93 and Red Rivers to the heart of mainland Southeast Asia; and the Austronesians, across the islands of Taiwan, the Philippines and Indonesia to the whole of the Pacific Basin. The most parsimonious interpretation of this body of evidence is a single origin for these peoples, their languages, and their agricultural economy a single origin in southern China, follow ed by radiating migrations down the spoke like rivers of mainland Southeast Asia and across the stepping stone archipelagoes of island Southeast Asia and the Pacific (Blust 1993a, 1996; Bellwood 1993, 1997; Howells 1997; Higham 1996, 2002; Higham and Thosa rat 1998; Diamond and Bellwood 2003; Matsumura and Hudson 2005). This connection between the spread of agriculture and the spread of language is more than mere correlation: there is a clear causal relationship between the advent of agriculture and the i nitiation of vast population migrations. This relationship has been documented in association with the rise of agricultural populations all over the world and is grounded in mechanisms that are familiar and pervasive throughout the whole of anthropology. Quite simply, this causal relationship is attributed to the population growth that is an inevitable result of the adoption of a successful agricultural subsistence strategy (Barth 1952; Renfrew 1987; Renfew and Bahn 1996; Higham 1996; Bellwood 1997; Diamo nd 1999; Diamond and Bellwood 2003; Matsumura and Hudson 2005). Hunter gatherer populations, by their very nature, maintain a state of long term equilibrium with their environment wherein population densities are held in check by the distribution and dens ity of available food resources. The tightness and immediacy of this balancing feedback loop is such that hunter gatherer populations will rarely approach, much less exceed, the carrying capacity of their territories, all else being equal. However, once the difficulties of the transition to agriculture are surmounted, fully fledged agricultural societies universally exhibit rapid population growth, owing to the increases in birthrate and life
94 expectancy and decreases in infant and child mortality afforded by food surpluses and a sedentary lifestyle (Renfrew 1987; Renfew and Bahn 1996; Matsumura and Hudson 2005). Such populations will quickly reach and then outstrip the carrying capacity of their lands, necessitating the expansion of agricultural settlemen ts into new territories. The magnitude of such population expansions is generally sufficient to completely overwhelm any hunter gatherer populations already resident in the area of expansion. In addition to the advantages of sheer numbers, expanding agr icultural populations also carry with them the additional assets of the more advanced material culture and greater social complexity permitted and encouraged by a sedentary agrarian lifestyle (Barth 1952; Diamond 1999; Diamond and Bellwood 2003). Thus, th e Austronesian and Austroasiatic speaking, southern mongoloid agriculturalists who spread from southern China into Southeast Asia, with their polished stone tools and sophisticated pottery tradition, easily and readily displaced the resident inland popula tions of australoid Hoabinhian hunter gatherers, and converted the closed, forested inlands of Southeast Asia into wide open wet rice paddies. We have seen that coastal hunter gatherer populations, such as that at Khok Phanom Di, who themselves had develo ped a considerable measure of material wealth and social complexity, were not so easily displaced (Bellwood 1997; Higham and Thosarat 1998; Higham 2002). Instead, the people of this site likely interacted with the immigrant agriculturalists for several ce nturies, even briefly adopting agriculture for a time while environmental conditions were favorable. This circumstance highlights an important consideration when reconstructing population migrations: from an archaeological perspective, the demographic sh ift from one population to another associated with such migrations often appears to have occurred instantaneously, heralded by the sudden, dramatic appearance of new artifact types and new patterns of behavior. In fact, the process is
95 quite gradual, subtl e and accumulative (Renfrew 1987; Matsumura and Hudson 2005). Cavalli Sforza (Cavalli Sforza et al. 1988:6006) has drawn an analogy with punctuational events in evolution (Eldredge and Gould 1972), wherein an abrupt, radical change in this case, the adven t of agriculture and ensuing rapid population expansions sets in motion a lengthy period of gradual yet profound transformation that establishes a new equilibrium among the affected populations(s). And what of the Thai? As it happens, their early history provides a suitable example of this type of apparent punctuational population change. As was noted above and in the previous chapter, inscriptions in Thai do not appear in Southeast Asia until the famous stela of King Ramakhamhaeng of Sukhothai, dated to AD 1292 and written in an adapted Khmer script (Higham and Thosarat 1998). By this time, however, the ur Thai kingdoms of Sukhothai and Lanna were already well enough established in northern Thailand and the Chao Phraya basin to have begun gnawing away at the western fringes of the Khmer empire; Ayutthaya was soon to follow. These literate and well organized polities did not just burst forth over night, and so the Thai must have been resident in the general area of northern Thailand for some time prior to this efflorescence. However, since the Mon kingdom of Dvaravati dominated the Chao Phraya basin as far north as modern Chiang Mai from at least the sixth century AD until its subjugation by Angkor in the tenth to eleventh centuries AD, we can also reason ably surmise that the Thai did not arrive in the basin until the last centuries of the first millennium AD, at the earliest (Benedict 1942; Higham and Thosarat 1998; Higham 1996). As speakers of a Daic language, the Thai must also have spread to the regio n from the ancestral homeland of Austro Tai in southern China.
96 Today, the greatest concentration and diversity of Daic languages other than Thai remains centered in Yunnan and adjacent provinces in southern China. This suggests that the majority of the Austroasiatic and Austronesian speaking cousins fanned out across Southeast Asia and Oceania (Benedict 1942; Rinehart 1981; Nakbunlung 1994; Bellwood 1997; Higham and Tho sarat 1998; Baker and Phongpaichit 2005). This is not to imply cultural stagnation amongst the Daic speakers, however, as archaeological evidence suggests that a state level society emerged in Yunnan as early as the first century AD, and eventually came t o extend its influence over most of southern China between the coast and the foothills of the Himalaya (Rinehart 1981; Baker and Phongpaichit 2005). It is now generally believed that this Tai civilization, known as the kingdom of Nanchao, was under consta nt pressure from the relentless imperial expansion of the northern Han Chinese. Over the course of several centuries, this pressure, along with the open hostilities it likely precipitated, led to a gradual southward movement of the Daic peoples, including the Thai, overland and along the river valleys into Southeast Asia. Here, they encountered the Mon Khmer and Cham speaking populations resident in the area since at least the mid third millennium BC. The southward drift of Daic peoples was doubtlessl y further hastened by the Mongol conquest of southern China in AD 1276 (Briggs 1945; Rinehart 1981; Nakbunlung 1994; Higham 1996; Baker and Phongpaichit 2005). This likely lead to a rapid secondary influx of large numbers of people into the nascent Thai k ingdoms of Sukhothai and Lanna, providing leaders there with the manpower they would need to eventually stand and challenge Angkor (Rinehart 1981). Unlike the Austronesian and Austroasiatic expansions, there is little to suggest that the emerging dominan ce of the Thai people and language in the Chao Phraya basin can be attributed
97 merely to numerical or technological superiority. The Mon and Khmer had been resident in Southeast Asia for several millennia prior to the arrival of the Thai, and had given ris e to sweeping, populous imperial civilizations of surpassing complexity and robusticity. The Thai shared with them a common economy centered around rice agriculture and bronze and iron metallurgy, but as recent immigrants to the area, they lacked the type of economic infrastructure and political machinery with which Angkor asserted and maintained its dominance over the region. Thus, the Thai initially would have had neither an advantage of men nor of matriel that would have allowed them to overrun the Mo n and Khmer. Instead, by all accounts, the Thai integrated themselves into the resident populations, and for a time dwelt as subjects of Angkor. This is reflected in the high proportion of Khmer and Mon loanwords in Thai, but the comparatively low propor tion of Thai loanwords in Khmer (Kaplan 1981; Baker and Phongpaichit 2005). Nevertheless, the Thai retained their distinct linguistic and cultural identity, which Bayard (1979) believes included a social standing as something of a warrior elite, their ma rtial skills honed over centuries of conflict with the Han Chinese and the Mongols. If this is true, then the Thai may have been regarded by the dispossessed and overtaxed populace of the outlying western fringe of the Khmer empire as an attractive altern ative for their allegiance. By aligning themselves with the Thai, adopting their language and culture, and strengthening the power base of Sukhothai and Lanna, these converts hastened both the downfall of Angkor and the spread of the Thai language across Southeast Asia (Briggs 1945; Rinehart 1981; Higham and Thosarat 1998; Higham 2002; Baker and Phongpaichit 2005). Thus, the initial spread of the Thai language can be attributed to the elite dominance model, in contrast to the initial spread of
98 Austronesia n and Austroasiatic languages, which is attributed to the numerical and technological superiority of these agricultural populations. Population Genetics As the forgoing discussion has hopefully made clear, the archaeological and linguistic i.e., the cultur al evidence in support of the Two Layer model of Southeast Asian population history is considerable. It would be an easy thing, however, to get caught up in the voluminous minutiae of linguistic reconstructions and archaeological excavations, and forget t hat at its core, the Two Layer model is a hypothesis regarding the biological origins of the modern populations of this region. It is a frustrating truth that linguistic cognates and pottery sherds will always be more abundant than ancient DNA samples and intact skeletal remains, and so the closer we get to the peoples in question, the more scant the evidence of their biological origins becomes. To date, onl y a handful of genetic studies 3 have been brought to bear directly on the question of population hi story in Southeast Asia, although an additional number have some indirect relevance. On a large scale, Cavalli Sforza and colleagues (Cavalli Sforza et al 1988; Cavalli Sforza 1991) have found that a linkage analysis of 120 allele frequencies f or 42 class ical genetic markers 4 in 42 worldwide populations generates a phylogenetic tree that closely corresponds to a reconstructed tree of linguistic phyla. In the genetic tree, Southeast Asians form the first major division within the Non African branch, opposi te of North Eurasians (including Europeans and 3 Xing et al. (2010:199) specifically site the populations of Southeast Asia as being particularly under represented, in terms both of previous genetic studies and of inclusion in the major genetic diversity databases, such as HapMap. 4 Classical genetic markers are derived from polymorphic loci that are expressed as serological and cytological characters, s uch as blood group antigens, serum antibodies, proteins, and enzymes. These markers include such familiar systems as ABO, Rh, and Diego blood types, the HLA complexes, PTC tasting ability, and the G6PD enzyme. Because of their ease of detection and analy sis from blood samples, such markers were used to study genetic relationships among populations prior to the direct genetic analyses permitted by the mapping of the mitochondrial and nuclear genomes.
99 Northeast Asian populations such as the Koreans, Japanese and Mongolians). Within the Southeast Asian division, separate groupings exist for Papua New Guineans/Australians, Pacific Islanders and mainland/isla nd Southeast Asians. Interestingly, this mainland/island Southeast Asian cluster includes southern Chinese populations. These genetic groupings are mirrored to a remarkable degree in the linguistic tree, with the Australian and Papuan languages as outlie rs to the Austronesian languages of island Southeast Asia and the Pacific and the Austroasiatic and Daic languages of the mainland, all of which are united under the Austric superfamily (Cavalli S forza et al. 1988:6003 6004). The southern Chinese populati ons are also something of linguistic outliers to the Austric cluster, as speakers of various dialects of Chinese, a Sino Tibetan language. However, this is likely a more recent linguistic tradition imposed upon these peoples by the expanding Han Chinese o f the last millennium BC and first millennium AD (Benedict 1942; Higham 1996; Howells 1997). At the time of its publication, Cavalli demonstration of the correlation between genetic and linguistic dive rgence times and the accord between these times and well dated events in the archaeological record. Since then, a number of other studies have looked to genetic data as a means of confirming hypotheses originally derived from archaeological and linguistic evidence. In a study similar in scope (121 alleles at 29 polymorphic loci) to that of Cavalli Sforza and colleagues (1988), but with different methodological approaches and analytical assumptions, Nei and Roychoudhury (1993) generated a phylogenetic tree which also groups together Southeast Asian and southern Chinese (Fujian province) populations, to the exclusion of Australian/Papuan populations on one hand and East Asian populations (Japanese, Tibetan, Korean) on the other. In a significant divergence from Cavalli
100 supercluster along with the Southeast Asians, Oceanians and Australo Melanesians, rather than with the European populations in a Eurasian cluster. Curiou sly, this tree grouped Polynesian and Micronesian populations into the Australo Melanesian branch, in direct opposition to the prevailing archaeological and linguistic reconstructions, which align them with southern mongoloid, Austronesian speaking populat ions (Bellwood 1997; Howells 1997). However, examination of the genetic distances (D A ) between the Poly/Micronesian populations and the Australo Melanesian populations (D A = 5.3 to 7.9) reveals them to be more than twice as great as the distances between the Poly/Micronesians and the Southeast Asians (D A = 2.1 to 3.8). This suggests that the grouping of Oceanic and Australo Melanesian populations may be an artifact of the clustering algorithm, rather than a reflection of true population relationships (Nei and Roychoudhury 1993:931, 934). Indeed, using different clustering parameters, Nei and Roychoudhury (1993) find the Southeast Asian Oceanic grouping is restored. Other studies of classical genetic markers have repeatedly underscored both the fundament al dissimilarity between the populations of Southeast Asia and Australia/Papua New Guinea, and the fundamental similarities among Southeast Asians, Polynesians and Southern Chinese (Barth 1952; Kirk 1982, 1986; Pietrusewsky 1994; Bellwood 1997; Omoto and S aitou 1997; Matsumura and Hudson 2005; Matsumura 2006). On a smaller scale, several studies of classical genetic markers in Thai populations have also demonstrated similarities to populations from southern China and island Southeast Asia (Simmons et al. 1 954; Fongsatikul 1997). In recent years, however, such studies of classical genetic markers have given way to population genetic studies based on direct sequencing of the mitochondrial and nuclear genomes. These studies permit the exploration and reconst ruction of population relationships with much greater precision and resolution. As with the classical genetic studies, they vary greatly in scope, from
101 broad considerations of populations world wide to targeted studies of specific polymorphic loci within circumscribed geographic or populational boundaries. In these studies too, mainland and island Southeast Asians usually group most closely with each other and with Oceanians, before they are joined in the general Asian cluster by East and Northeast Asian populations (Melton et al. 1995; Bellwood 1997; Matsumura and Hudson 2005; Lertrit et al 2008; Xing et al. 2010). In a study of more than 250,000 single nucleotide polymorphism (SNP) loci in 850 people from 40 worldwide populations, Xing and colleagues (2010), for example, found that this precise clustering pattern is repeated in 90 to 100% of 1000 bootstrapp ed iterations of the clustering algorithm. This very high replication rate speaks to the confidence with which this clustering pattern can be viewed as reflective of the true genetic relationships among these populations. In the same study, Xing et al. (2010:204 206) use the ADMIXTURE algorithm (Alexander et al. 2009) to estimate the major ancestry components of the individuals in their study sample, populatio et al. 2009:1655). Their analysis reveals that the ancestry of the individuals in their Thai sample is composed substantially of Indian/South Asi an, East Asian and Austronesian 5 components. This ancestry pattern is echoed as well in the ancestry of the Cambodian and Vietnamese samples, but these populations have a far smaller Indian/South Asian component to their ancestry than do the Thai. Such patterns of ancestry are precisely what would be expected for these population s based on reconstructions of their population history derived from archaeological and linguistic evidence. The Austronesian component, which accounts for approximately two thirds of the ancestry of the Thai and 5 Though left unnamed by Xing and colleagues, this compo nent dominates the ancestry of the Iban individuals in people of northern Borneo whose language belongs to the Austronesian family. Based o n this, I believe it is mongoloid populations in general.
102 Cambodian samples and one half of the ances try of the Vietnamese sample, is reflective of these Holocene. The Indian and East Asian component in the ancestry of the Thai speaks to the historical position of Thai land as the crossroads of Southeast Asia, through which the religious, political and mercantile emissaries of the great Indian and Chinese civilizations must have passed in order to spread their influence throughout Southeast Asia (Briggs 1945; Rinehart 19 81; Higham and Thosarat 1998; Baker and Phongpaichit 2005; Matsumura and Hudson 2005). The same is also true, albeit to a lesser degree, of the Cambodians, who during the Angkor period were the direct recipients of much of the cultural and material wealth brought to Southeast Asia by these Indian and Chinese scholars and merchants. The Vietnamese, on the other hand, though likewise descended from an ancestral southern mongoloid population, were much more heavily influenced by the Han Chinese, who conquere d the Red River valley early in the first century AD and who would maintain their hegemony over this Vietnamese heartland for the next millennium (Higham and Thosarat 1998; Stark and Allen 1998; Higham 2002). Unfortunately, Xing and colleagues (2010) did not include any populations from Australia or Melanesia in their study, so their results shed no light on the potential contributions of the preceding australoid Hoabinhian populations to the ancestry of modern Southeast Asians. If the immigrant southern mongoloid agriculturalists assimilated and intermixed with, rather than completely replacing, the resident hunter gatherer populations, as has been argued by many researchers (Bellwood 1997; Matsumura and Hudson 2005; Matsumura 2006), we should also see s ome traces of Australo Melanesian ancestry in modern Southeast Asians. Fortunately, some insight into this question is provided by several other studies. Firstly, analysis of Southeast Asian mtDNA sequences has demonstrated that the oldest haplogroup in Southeast Asia,
103 haplogroup M has a probable antiquity of 45 65,000 years (Schurr and Wallace 2002:437). This age is commensurate with the initial peopling of Southeast Asia by australoid populations, as reconstructed from the fossil record and from the l ikely dates of colonization of Australia by anatomically modern H. sapiens That this genetic signature was carried by the initial australoid colonizers of Southeast Asia is further supported by the observation that it is the dominant haplogroup in the hi ghland populations of Papua New Guinea. These are australoid populations who are rather genetically isolated due to the ruggedness and remoteness of their territory. Haplogroup M is also found at moderate frequencies (25 to 45%) in all Southeast Asian po pulations, suggesting that there is a near uniform, region wide substratum of australoid mtDNA sequences throughout the populations of Southeast Asia (Bellwood 1997; Schurr and Wallace 2002:437; Lertrit et al 2008:435). Additionally, haplogroup M attain s its greatest frequency in the aboriginal Negrito populations of Malaysia, who as noted above, are believed to be the direct descendants of the australoid Hoabinhian populations of Southeast Asia, but with varying degrees of genetic admixture with neighbo ring agricultural populations (Ballinger et al. 1992; Bellwood 1997; Schurr and Wallace 2002). Secondly, the mtDNA study (Oota et al. 2001) which revealed the Moh Khiew skeleton to be genetically closest to individuals from Papua New Guinea, in concordan ce with its posited australoid phenotypic characteristics (Matsumura and Pookajorn 2005; Matsumura 2006), also found that four skeletons from early Neolithic deposits (ca. 7500 to 8000 BP) at Sakai Cave in peninsular Thailand show nearly equal degrees of g enetic similarity to populations from Papua New Guinea (i.e., Australo Melanesians), Indonesia (i.e., Austronesians) and the Senoi agriculturalists of Malaysia. Although the modern Senoi are generally regarded in morphological terms as a Negrito (and ther efore australoid) population, they have clear genetic
104 links with Austroasiatic (Mon and Khmer) speaking populations (Saha et al. 1995; Bellwood 1993, 1997; Oota et al. 2001). It has been argued that the Senoi are the descendant population arising from a l arge degree of admixture between an ancestral Hoabinhian (australoid) hunter gatherer population and immigrant Neolithic (southern mongoloid) agricultural populations during the mid Holocene (Bellwood 1993, 1997). Again, if we are willing to accept as val id the results of these ancient DNA analyses, then the genetic similarity between the Sakai Cave specimens and the Senoi argues for a similar degree of admixture between the Hoabinhians and the southern mongoloid immigrants in at least one other early Neol ithic population of Southeast Asia. Other genetic studies have shed light on specific aspects of the southern mongoloid expansion. It has long been recognized by geneticists that Southeast Asian and Oceanic populations possesses a 9 base pair (bp) deletion between the Cytochrome Oxidase II (COII) and Lysine tRNA (tRNA Lys ) mitochondrial gene loci (Cann and Wilson 1983; Ballinger et al. 1992; Schurr and Wallace 2002). The frequency of this deletion within these populations ranges from approximately 2 0% in mainland Southeast Asia and eastern Indonesia, to approximately 40% in the Philippines and western Indonesia, to near fixation in Polynesians. The COII/ tRNA Lys 9 bp deletion occurs at much lower frequencies (7 to 15%) in East Asian populations, and is nearly absent (0 to 3%) in Australians and the highlanders of Papua New Guinea (Ballinger et al. 1992:144; Melton et al. 1995:407; Schurr and Wallace 2002). It has been estimated that this mutation arose as far back as approximately 60,000 years ago i n southern China (Redd et al. 1995; Ballinger et al. 1992; Melton et al. 1995; Fucharoen et al. 2001), and several researchers (Ballinger et al. 1992; Melton et al. 1995) have argued that both its near absence in Australo Melanesian populations and near fi xation in Polynesian populations are due to founder effects in
105 the colonizing populations of these regions. This would support the interpretation that the populations of Polynesia and Australo Melanesia are derived from separate lineages that arose some t ime after the origin of the mutation. The COII/ tRNA Lys 9 bp deletion also occurs at high frequencies (40 to 47%) in the aboriginal populations of Taiwan and the Philippines. As Taiwan is generally accepted as the aboriginal home of Austronesian speakers and the Philippines likewise as the first stage of their subsequent radiation throughout island Southeast Asia, it has been argued that the distribution of this gene throughout island Southeast Asia and Oceania is an index for the expansion of Austronesi an speakers beginning in the middle Holocene (Ballinger 1992; Melton et al 1995; Schurr and Wallace 2002). Further support for this interpretation comes from the evolution of three additional point mutations within the mtDNA control region that occur onl y in the presence of the COII/ tRNA Lys 9 bp deletion. These mutations, which consist of trans itions at bases 16217 (T hymine C ytosine, or T C ), 1 6247 ( Adenine Guanine, or A G), and 16261 (C T) (shorthanded et al. 1995:404), are found almost exclusively in Polynesian populations. This motif does, however, occur occasionally in eastern Indonesian populations, reinfo rcing the likelihood that this area served as the jumping off point for the settlement of Oceania (Hagelberg and Clegg 1993; Melton et al. populations of island and mainland S outheast Asia, precursor states of the motif are. The immediate precursor, which lacks the transition at the 16247 locus and therefore has the to moderate fre quencies (10 to 25%) in southern Chinese, western Indonesians (i.e., Borneans, Javans), Malaysians and Filipinos. Even more intriguing, the deep ancestral character state
106 ound in moderate to high frequencies throughout Southeast Asia. Of particular note, this character state is found in nearly 50% of southern Chinese but is completely absent both in Polynesian and in Australo Melanesian populations (Melton et al. 1995:410 411). The distribution of these three genetic mutation motifs gives the clear impression of an evolving character state being carried throughout Southeast Asia and the Pacific by the radiating migrations of an actively expanding population. In this case, that population is very likely the southern mongoloid, Austronesian speaking populations of rice agriculturalists who, archaeological and linguistic evidence suggest, moved first from southern China into Taiwan and the Philippines, before fragmenting into populations that would colonize nearly every island between Madagascar and Hawaii. The genetic traces of the island hopping Austronesian expansion thus established, other studies have turned their attention to the genetic history of the mainland Austroasi atic and Tai Kadai populations. Unfortunately, the reconstructions of the genetic history of these peoples, like the corresponding linguistic reconstructions, are complicated by the ebb and flow of the great civilizations of mainland Southeast Asia (Bellw ood 1997; Matsumura 2006; Lertrit et al. 2008). As Funan, Champa, Chenla, Dvaravati, Angkor, Sukhothai and Ayutthaya each waxed and waned in turn, their populations spread and contracted and intermingled with the resident populations of newly conquered te rritories, to a far greater extent than was possible among the inhabitants of the far flung islands of Oceania. Adding to the complexity are the unknown genetic contributions of the unnumbered Indian and Han Chinese travelers and traders who came to South east Asia to ply their spiritual and material wares (Briggs 1945; Bellwood 1997; Higham and Thosarat 1998; Higham 2002).
107 With regards to the Thai specifically, the question of biological descent from the ancient populations within their borders is a part icularly charged one. The Thai can, and do, clearly trace their direct descent back to the Kingdom of Sukhothai, whence came the first inscriptions repeatedly observed the founding of Sukhothai is a relatively late development in the population history of Southeast Asia. There is a tremendous amount of history and prehistory within the borders of modern Thailand prior to the Sukhothai period not least the civilization s of Dvaravati and Angkor that many Thai consider dear to their ethnic and cultural heritage. And this is to say nothing of the rich prehistoric cultures of Khok Phanom Di, Ban Chiang, Ban Kao and many others, of which the Thai are also, quite rightly, ra ther proud (Rinehart 1981; Nakbunlung 1994; Higham and Thosarat 1998; Baker and Phongpaichit 2005). In an effort to begin to address these questions, Lertrit and colleagues (2008) have undertaken what is perhaps the most comprehensive examination to date o f the genetic relationships among the prehistoric and modern populations of Thailand and other modern populations of East and Southeast Asia. Their study sample includes mtDNA sequence data from the Bronze Iron Age (1200 BC AD 400) skeletal population o f Ban Lum Khao and the Iron Age (400 BC AD 500 ) population from Noen U Loke 6 the first ever ancient DNA to be obtained from any archaeological population in Thailand as well as samples from modern populations of Austroasiatic (n = 3 populations), Sino Ti betan (Han Chinese populations from both northern and southern China; n = 9,) and Tai Kadai (including northern, northeastern and central Thai; n = 7) speaking peoples. This study is revealing in numerous ways. Firstly, moderate frequencies (20 to 61%) o f haplogroup M were observed throughout all the populations 6 Both sites are located in northeastern Thailand, in the upper reaches of the Mun River valley, a key settlement area during the Bronze and Iron Ages.
108 in the study (Lertrit et al 2008:435). This finding echoes the results of the genetic studies of island Southeast Asian and Polynesian that is, Austronesian populations (Schurr and Wallace 2002) and reinforces the interpretation that this haplogroup represents an australoid genetic substratum shared by all post Paleolithic Southeast Asian populations. Additionally with regards to haplogroup diversity, the Noen U Loke and Ban Lum Khao population s, along with the Tai Kadai, Austroasiatic and southern Han Chinese (from Yunnan and Guangdong provinces) populations share absent to low frequencies (0 to 12%) of haplogroup A (Lertrit et al. 2008:435). In contrast, northern Han Chinese and Tibetan popul ations display moderate frequencies (21 to 34%) of this haplogroup. Such a pattern indicates mutual genetic affinities among these four population groups, as would be expected under the hypothesis that both the Tai Kadai and Austroasiatic populations have their origins in southern China. Furthermore, it suggests broad southern mongoloid affinities for the prehistoric inhabitants of Ban Lum Khao and Noen U Loke and perhaps by extension for the other Bronze and Iron Age inhabitants of Southeast Asia as well These results clearly align this set of biological data with both the Austric linguistic hypothesis and the archaeological reconstructions of Neolithic southern mongoloid population migration patterns. In this light, it is intriguing that the southern Han Chinese populations should show such strong genetic similarities to the Austroasiatic and Tai Kadai populations clearly, the genetic traces of the ancient southern mongoloid populations of this region persist in the modern inhabitants, despite the seem ingly overwhelming cultural, linguistic and, by inference, biological influences of the historic Han Chinese. Yet the fact that haplogroup A is anywhere from two to ten times more common in northern Chinese populations than in southern Chinese, Austroasia tic or Tai Kadai populations speaks to the distinctness of these genetic lineages (Lertrit et al. 2008).
109 This same pattern of shared genetic affinities among the Ban Lum Khao, Noen U Loke, Tai Kadai, Austroasiatic and southern Chinese populations, to the exclusion of northern Han Chinese populations, is also seen with regards to haplogroup B which is nearly twice as common in the former populations as it is the northern Chinese. The relative frequency of haplogroup B in these populations is of particula r interest since other researchers have already linked its distribution with the Neolithic expansion of Austronesian populations, and both the 9 bp COII/ tRNA Lys background of this haplogroup (Ballinger et al. 1992; Melton et al. 1995; Fucharoen et al. 2001; Schurr and Wallace 2002). However, the age of this haplogroup has been estimated at anywhere between 25,000 and 60,000 years (Schurr and Wallace 2002), making it of suf ficient antiquity to be shared among the divergent Austronesian, Austroasiatic and Tai Kadai lineages. Fucharoen et al. (2001:122) have found the 9 bp deletion in 18 to 32% of Tai Kadai speaking populations in Thailand (including both the Thai and several ethnic minority groups); the majority of Polynesian motif. As was noted above, the distribution of this distinctive suite of the COII/ tRNA Lys and CAC CAT CGT genetic mutations associated with haplotype B gives the clear impression of an evolving character state being carried throughout Southeast Asia by the radiating migrations of an actively expanding population. Although initially proposed to ex plain only the distribution of Austronesian speaking populations, based on the foregoing genetic evidence, there is no reason not to expand the scope of this phenomenon to include the Austroasiatic and Tai Kadai speakers of mainland Southeast Asia as well. In short, then, this genetic signature is not merely the hallmark of the Austronesians, but possibly of all southern
110 mongoloid populations radiating throughout island and mainland Southeast Asia from the Neolithic onward. Further insights into the gene tic history and relationships of Tai Kadai and Austroasiatic populations are provided via examination of patterns of inter and intragroup haplotype diversity and genetic distances provided by Lertrit and colleagues (2008:435 437). Firstly, haplotype dive rsity (i.e., genetic variation) was found to be greater among the separate linguistic groups (Sino Tibetan, Austroasiatic and Tai Kadai) than within them, indicating that these populations are significantly distinct from one another. This distinctness is confirmed by cluster analysis, which places the Sino Tibetan populations on a separate branch from the clusters of Austroasiatic and Tai Kadai populations. This genetic distinctness of the Sino Tibetan group, on one hand, from the Austroasiatic and Tai Ka dai groups on the other is unsurprising given the linguistic distinctness and indeed, the clearly separate linguistic origins of these populations (Benedict 1942; Higham 1996, 2002; Bellwood 1997). In contrast, one might reasonably expect the Austroasiati c and Tai Kadai groups to show less separation, given their hypothesized shared Austric origins (Benedict 1942; Higham 1996, 2002; Bellwood 1997). However, it must be remembered that despite this shared ancestry, these groups constitute separate waves of immigrant populations spreading into Southeast Asia; their genetic patterns of intragroup homogeneity and intergroup diversity are therefore reflective of the founder effects that would have acted on each of these populations separately (Lertrit et al 200 8:438; Higham 1996, 2002; Bellwood 1997; Higham and Thosarat 1998; Baker and Phongpaichit 2005). The separate trajectories of the Austroasiatic and Tai Kadai populations are further revealed in the observations that 1) the Thai populations in the study h ave fewer haplotypes in common with Austroasiatic populations than they do with southern Chinese Sino Tibetan
111 populations; and 2) genetic distances between Austroasiatic populations are relatively greater than distances between the Tai Kadai populations in the study (Lertrit et al. 2008:435). Both of these observations reflect the earlier migration of Austroasiatic populations into Southeast Asia, which allowed them to accumulate in situ a greater degree of intrapopulational genetic diversity, as compared to later arriving Thai populations, who are less far removed from the founder effects induced by their more recent migration into Southeast Asia (Lertrit et al. 2008:438; Higham 1996, 2002; Bellwood 1997; Higham and Thosarat 1998; Baker and Phongpaichit 20 05). Thus geographically separated for the better part of four millennia, the Austroasiatic populations diverged from their Tai Kadai cousins, who gained or maintained closer genetic ties (i.e., more shared haplotypes and smaller genetic distances) to nei ghboring East Asian Sino Tibetan populations by virtue of their longer residence in southern China (Lertrit et al. 2008). Closer examination of the genetic relationships at the level of individual populations, rather than at the level of linguistic groups, is additionally revealing. Not surprisingly, the geographically and temporally proximate prehistoric skeletal populations from Ban Lum Khao and Noen U Loke are closest to each other in terms of genetic distance; they also show relatively close genetic di stances to geographically proximate modern Thai and Austroasiatic populations. However, under more robust principle components and cluster analyses, these prehistoric populations display much tighter associations with Austroasiatic populations, and are cl early separated from Thai populations (Lertrit et al 2008:437). Such Austroasiatic, rather than Tai Kadai, genetic affinities in the Bronze and Iron Age populations of Thailand are precisely what would be expected under the hypothesis, based on archaeolo gical and linguistic reconstructions, that mainland Southeast Asia was colonized during the Neolithic by intrusive populations of Austroasiatic speaking agricultural populations (Higham 1996, 2002; Bellwood 1997; Higham
112 and Thosarat 1998; Baker and Phongpa ichit 2005; Lertrit et al 2008). Similarly, genetic distances for the modern Thai populations in the study sample are closest first among the Thai populations themselves and secondly between the Thai and the southern Chinese populations from Yunnan and G uangdong provinces. As with all the other genetic relationships elucidated by Lertrit et al. (2008), these results are unsurprising and follow logically from the reconstructed population history of the Thai as derived from linguistic, archaeological and h istoric lines of evidence. They also reinforce the perceived distinctness of the Tai Kadai populations moving into Thailand from the late first or early second millennium AD on, compared to the Austroasiatic populations already resident there, despite the ir common Austric ancestry (Higham 1996, 2002; Bellwood 1997; Higham and Thosarat 1998; Baker and Phongpaichit 2005; Lertrit et al. 2008). In summary, then, it is clear that a considerable amount of genetic research into the biological affinities of the populations of Southeast Asia is in close agreement with the hypothesized population histories suggested by both the archaeological record and paleolinguistic reconstructions. Further, the genetic analyses, like the archaeological and linguist analyses, c learly lend strong support to the Two Layer model of Southeast Asian population origins. However, as illuminating as these genetic studies are, for the time being they remain altogether too limited in scope to stand alone as the definitive line of biologi cal evidence with which to fully resolve the driving questions of essential continuity versus discontinuity among the prehistoric and historic populations of Southeast Asia. Physical Anthropology In the absence of the requisite genotypic data, physical a nthropologists have turned to various forms of phenotypic data that can stand as reasonable surrogates. Thus, the population history of Southeast Asia has primarily been addressed via craniometric and odontometric
113 studies and analyses of non metric dental traits. Outside the archaeological literature, such studies represent the most voluminous body of knowledge that has been brought to bear on the debate between the Two Layer and Regional Continuity hypotheses. But other, more general observations pertai ning to patterns of human phenotypic variation have also been offered, particularly in support of the Two Layer model. For example, although frequently discussed in a manner which implies that they are physically homogenous, the modern populations of Sout heast Asia are, in fact, tremendously varied in terms of broad phenotypic characters such as skin color, hair color and form, eye shape, overall skeletal build and so forth. It has long been recognized that this variation has a clinal structure, with a gr adient between mongoloid and australoid phenotypes extending from the northwest to the southeast across the region. This has led various researchers all to the same general conclusion, that the most parsimonious explanation for this Mongoloid expansion into an Australo 1997:74 75; Barth 1952; Coon 1966; Howells 1997; Matsumura 2006; Pietrusewsky 2006). In the same vein, it has been noted that the degree of skin pigmentation in most Southeast Asian popula tions is much lighter than that of indigenous populations living at similar tropical latitudes (21N to 11S) elsewhere in the world (Bellwood 1997:75 76; Brace 2005; Matsumura and Hudson 2005). Although skin color is generally assumed to be under strong environmental selective pressure, it is also thought to be a slowly evolving character that has taken tens of millennia to attain its present range of variation in populations worldwide (Molnar 2002; Mielke et al. 2006). Reasoning from the latter to the f than expected skin color of Southeast Asian populations indicates a more recent expansion into the area from more temperate latitudes, likely in the north. Conversely, i f Southeast Asian populations had evolved largely in situ since the terminal
114 Pleistocene, as argued by the Regional Continuity model, they should reasonably be expected to have darker skin tones more comparable to the adjacent populations of indigenous Aus tralians, Papuans and Melanesians (who are known for certain to have evolved in situ since the terminal Pleistocene). These observations, although intriguing and contributing to our understanding of population variation in Southeast Asia, are nonetheless s omewhat anecdotal, and thus not fully persuasive. The same can be said of early craniometric studies, as these tend to be largely descriptive rather than analytical in nature, owing to the lack, at the time, of relevant comparative datasets and of the com putational power necessary to execute sophisticated multivariate statistical analyses (Howells 1973; Nakbunlung 1994; Pietrusewsky 2000). Yet such studies can still be informative. For example, Sangvichien (1966a, 1996b, et al. 1969) described the propor tions and morphology of skeletal remains from Ban Kao as similar to the modern southern mongoloid populations of Thailand (Bellwood 1993, 1997; Nakbunlung 1994; Pietrusewsky and Douglas 2002; Matsumura and Hudson 2005). Although Sangvichien (1966a) interp reted this similarity as evidence that the Thai had not migrated to their present territory from elsewhere in Asia (i.e., a determination of local population continuity), this conclusion was drawn without the benefit of much of the major archaeological fie ldwork in the region. In light of our present understanding of Southeast Asian prehistory, one would expect the Neolithic inhabitants of Ban Kao to have southern mongoloid phenotypes, as this population (and its distinctive pottery tradition) has been tie d so closely to the Austronesian and Austroasiatic expansions (Bellwood 1993, 1997; Higham 1996). Under the Two Layer model, the modern populations of Southeast Asia should be locally continuous with archaeological populations extending back to the Neolit hic, such as that from Ban Kao; that is, we should expect the discontinuity to occur further
115 back in time, between the respective australoid and southern mongoloid populations of the pre and post Neolithic epochs (Pietrusewsky and Douglas 2002; Matsumura and Hudson 2005; Matsumura 2006). Modern multivariate studies of cranial and dental morphology directed at resolving the question of Southeast Asian population origins have taken a range of approaches to the problem. The more general approaches have ten ded to focus broadly on samples of modern and recent historic populations from various locations within Asia, Australia and the Pacific, and have yielded correspondingly broad conclusions. A major contribution of such studies is that their results have te nded to be in agreement with, and in some cases have foreshadowed, the results of similar genetic studies, thus affirming the validity of craniometric and dental data for illustrating population relationships in the absence of genetic data. The results ha ve also tended to correspond directly with expected population relationships predicted by archaeological and linguistic data. Indeed, these studies have more or less consistently demonstrated: the fundamental division between Asian and Australo Melanesian populations, driven by the stark contrast between the mongoloid and australoid phenotypes; the inclusion of Polynesian populations within the general Asian cluster, and their tendency to cluster most closely with populations from Southeast Asia 7 in genera l and island Southeast Asia in specific; the tendency of East and Southeast Asian populations to form separate subclusters within the general Asian cluster, and the frequent (but not exclusive) alignment of populations from southern China with the populati ons of Southeast Asia; the general Australo Melanesian affinities of Southeast Asian aboriginal populations, such as the Andaman Islanders (Howells 1989; Brace and Hunt 1990; Brace et al 1991; Li et al. 1991; Brace and Tracer 1992; Bellwood 1997; Pietrusewsky 2000; Matsumura and Hudson 2005). 7 et al. 1991; Li et al. 1991) consistent, idiosyncratic findings of closer relationships between Polynesians and the Ainu and Jomon of Japan notwiths craniometric analyses are compelling, there is little archaeological, linguistic or genetic evidence to support a link between the prehistoric populations of Japan and the ancient or modern Polynesians.
116 A few more detailed studies have also included the occasional archaeological population, and although the sample sizes are usually small, the results of these stu dies have been additionally illuminating. For example, Li and colleagues (1991) found that a Chinese Neolithic sample from the Yangshao culture of the central Yellow River valley routinely fell in an intermediate location between the East and Southeast As ian subcluster. The authors reject the subsequent interpretation that this population possibly represents a common ancestor to the East and Southeast Asian populations, solely on the grounds of the sample size being too small to permit this population to be reliably placed by the clustering algorithm (Li et al. 1991:272). However, there is a kernel of archaeological truth attached to the craniometric classification of this population. Archaeological evidence suggests that the Yangshao culture was centere d around rice agriculture brought to the Yellow River valley in the fourth millennium BC (Glover and Higham 1996). As such, the Yangshao sample likely represents the northward expansion of a southern mongoloid population similar to those that spread south ward into mainland and island Southeast Asia. Although the Yangshao population would therefore not be ancestral to modern populations owed to their common descent from the Neolithic populations of the Yangzi Valley. This interpretation garners further support from an expanded analysis that included more populations but considered fewer craniometric variables, in which the Yangshao sample is joined in its intermedi ate position by Neolithic populations from the sites of Ban Chiang and Ban Na Di in Thailand (Li et al. 1991:275). In a similar study, Brace and colleagues (1991:254) found that Neolithic populations from both China and Thailand cluster most closely with modern Southeast Asian populations, affirming the southern mongoloid phenotypic affinities of these prehistoric populations.
117 As intriguing as these results are, such craniometric studies do not include archaeological population samples of sufficient time depth or antiquity to provide much more than general insights into the relationships between the prehistoric and modern populations of Southeast Asia. As was noted above, it is expected under the Two Layer model that the Neolithic populations of southern China and Southeast Asia should be highly similar to the modern populations of this region, because one half of the core hypothesis of this model is that the latter are the direct descendants of the former. If the Two Layer model is indeed true, this evi dence of population continuity from the Neolithic onwards should be juxtaposed against a distinct discontinuity between the Neolithic, historic and modern southern mongoloid populations of Southeast Asia on one hand and the australoid populations of the Pa leolithic on the other (Pietrusewsky and Douglas 2002; Matsumura and Hudson 2005; Matsumura 2006). The most persuasive multivariate studies of cranial and dental morphology directed at resolving the question of Southeast Asian population origins have soug ht to probe the available skeletal material specifically for evidence of just such a population discontinuity (Pietrusewsky and Douglas 2002; Matsumura and Hudson 2005; Pietrusewsky 2005; Matsumura 2006). Thus, the case for the Two Layer model is made most convincingly by Matsumura and Hudson (2005; Matsumura 2006) who have presented a methodologically robust multivariate analysis of craniometric, odontometric and non metric datasets. In addition to the standard principle components, factor, cluster and di scriminant function analyses presented in most craniometric studies, these researchers have used a variety of multidimensional scaling methods to further distill and explore the relationships of various skeletal populations in multivariate space. Their ar gument gains additional strength from the breadth and depth of their datasets, which, in addition to containing a wide spectrum of modern East and Southeast Asian,
118 Australian, Melanesian and Oceanic populations, also include a broad sampling of late Paleol ithic (i.e., early and middle Holocene) and Neolithic skeletal populations from a great diversity of sites in mainland and island Southeast Asia. This representation of archaeological skeletal populations is far greater in breadth and depth than has been presented in any other studies reviewed in the course of this dissertation. In each of their analyses, regardless of the particular statistical approach and whether based on cranial or dental data, the same patterns of population relationships were reveal ed, superimposed upon the ever present first order separation of Australo Melanesian and Asian population clusters, and the subdivision of this latter cluster into East and Southeast Asian subclusters. First, the Paleolithic archaeological samples from Sou theast Asia consistently clustered most closely with modern and archaeological Australians and Melanesians. This includes not only the late Pleistocene remains from Moh Khiew (Thailand) and the Hoabinhian period skeletons from Gua Gunung Runtuh, Gua Cha, Guar Kepah, (Malaysia), Mai Da Nuoc and Mai Da Dieu (Vietnam) as discussed above, but also the early to middle Holocene hunter gatherer populations from Indonesia, Malaysia, Laos and Vietnam (including samples from the late/post Hoabinhian cultures of Bac Son and Da But). Matsumura and Hudson (2005; Matsumura 2006) interpret these results as clear confirmation of the australoid morphological character of these early (i.e., pre Neolithic) indigenous populations of Southeast Asia. Curiously, different anal ytical iterations also repeatedly place the early Neolithic skeletal samples from the sites of Ban Kao, Ban Chiang and Non Nok Tha perhaps the three most important Neolithic sites in Thailand into this Australo Melanesian grouping. At first blush, this wo uld see to run contrary to the pattern of population relationships expected under the Two Layer model. However, it must be remembered that the cultural influences of expanding
119 agricultural populations would have arrived earlier and had a greater initial i mpact on the resident populations of the region than did the genetic influence of these immigrant populations (Matsumura 2006:52). Thus, although the earliest Neolithic populations of Southeast Asia had taken on the material and social trappings of an agr icultural lifestyle, a number of generations may have passed before they also began to manifest the southern mongoloid phenotype carried by the intrusive agriculturalists and currently characteristic of the region (Matsumura and Hudson 2005:205). The poss ibility of a slight phenotypic shift (corresponding to an underlying genetic shift) towards the southern mongoloid pattern among the early Neolithic populations of Thailand is suggested by multidimensional scaling analysis, which places these samples close r than any other australoid population to the zone of overlap between the Asian and Australo Melanesian spheres (Matsumura 2006:48). These analytical observations and subsequent interpretations strike a blow against critics of the Two Layer model who int erpret the model as insisting upon full replacement of the resident australoid hunter gatherer populations by immigrant southern mongoloid agriculturalists (Turner 1990; Hanihara 1992, 1993). Indeed, the major proponents of the Two Layer model (Bellwood 1 997; Matsumura and Hudson 2005; Matsumura 2006) argue for a much more complicated and subtle process of change, one which must allow for intermarriage, local evolution, and also for the important concept that expansion [of agricultural populations] involv ed more a change in the structure of a Mongoloid Australoid cline than a migration of uniform and distinct peoples from a remote area such as China (Bellwood 1997:82). The complexity of this process of demic change is underscored by the fact that in one of from Non Nok Tha again takes a position within the Australo Melanesian cl uster, while the sample from Ban Kao falls within the Asian cluster. This suggests varying levels of genetic
120 influence from East Asian populations among the early Neolithic peoples of Thailand, and presumably the rest of Southeast Asia as well (Matsumura and Hudson 2005:203; Matsumura 2006). Secondly, the positioning of other archaeological populations under cluster and scaling analyses merits consideration as well. As in other analyses, the Neolithic population sample from southern China falls clearly within the Southeast Asian, rather than East Asian, subcluster, reaffirming the phenotypic similarities between the modern inhabitants of Southeast Asia and their hypothesized Neolithic ancestors. Although Matsumura and Hudson (2005; Matsumura 2006) do n ot include any Bronze Age populations from Southeast Asia in their studies a serious shortcoming they do include the early Iron Age population of Dong Son, in the Red River valley of northern Vietnam, which dates to the early first millennium BC (Higham 20 02). This population falls clearly within the Southeast Asian subcluster, aligning most closely with the modern population of Vietnam, but also showing affinities for the modern population of Thailand as well as the Anyang and Yayoi Bronze Age populations of central China and Japan, respectively. This placement is significant because it suggests a terminus ante quem for the firm establishment of the southern mongoloid phenotype in the prehistoric populations of Southeast Asia. The rapidity of this pheno typical shift from australoid to mongoloid, occurring over the course of, at most, two millennia, is far too dramatic to be reasonably attributed to the in situ evolution of local populations in the absence of any external influence. Nor can it be ascribe d to i.e., reductions in tooth size, facial prognathism and overall cranial robusticity, as well as a trend towards increasing brachicephalization as proposed by Bulbeck (1982; cf. B ellwood 1997; Matsumura
121 and Hudson 2005). Such microevolutionary processes, although powerful in the overall development of modern human races since the Pleistocene, are inadequate to account for such a radical morphological change in so brief a period of time (Bellwood 1997; Jantz and Jantz 2000). 40,000 years across a range of temperate and tropical environments in Australia and New Guinea, [have] been insuffici ent to produce major phenotypic differences equal to those we see in the major Australo arguments for in situ population evolution and continuity extending back to the late Pleistocene in S outheast Asia very difficult to entertain. Lastly, and perhaps most significantly for resolving questions of population origins in Southeast Asia, is the behavior of the modern Southeast Asian population samples in Matsumura 006) analyses. As was noted above, under cluster analysis of odontometric data, the modern populations of Southeast Asia cluster more closely with the populations of East Asia than with Australo Melanesians, yet form their own distinct subcluster within t he larger Asian grouping. However, under multidimensional scaling analyses of both metric and non metric dental data, modern Southeast Asians tend to fall in intermediate positions between the East Asian populations on one hand and the Australo Melanesian populations on the other. This potential for revealing intermediate populations is cited by Matsumura and Hudson (2005:198; Matsumura 2006:49) as a major analytical advantage of multidimensional scaling analyses over cluster analyses, as the latter will always force intermediate populations to join one cluster or another, creating the perception of discrete populations groupings where a continuous spectrum of variation is present.
122 The intermediate positioning of Southeast Asian populations revealed unde r multivariate analyses can also be seen in the simple univariate population frequencies of certain non metric dental traits. For example, the frequency of incisor shoveling ranges from 75 to 95% in modern East Asians (Japanese, Northern Chinese, Mongolia ns, Okinawans), 40 to 55% in modern Australo Melanesians (Australians, New Britain and Loyalty Islanders, Andaman Islanders), and 50 to 75% in modern Southeast Asians (Thai, Lao, Vietnamese). Similar patterns obtain for and double rooted premolars (Matsumura and Hudson 2005:196 20l; Matsumura 2006:50). Interestingly, however, this intermediate position is not wholly uniform across analyses: in terms of metric dental traits and some non metric traits, the modern populati ons of Southeast Asia shift slightly closer to East Asians, while in terms of other non metric dental traits they are more closely aligned with Australo Melanesian populations. This intermediate positioning of modern Southeast Asian populations, along wi th their fluctuating affinities for the populations of East Asia and Australo Melanesia, highlights two important aspects of the singular pattern of human variation among these populations. First, it reinforces our understanding of the clinal nature of th is variation, with the populations of Southeast Asia positioned intermediately both geographically and morphologically between the northern, mongoloid and southern, australoid extremes of the continuum. Secondly, it underpins the perception of the general southern mongoloid phenotype of the modern Southeast Asian populations as a hybrid between that of the broad mongoloid and australoid phenotypes a perception that has long been held by many researchers (Bellwood 1997:74 75; Barth 1952; Coon 1966; Howells 1997; Matsumura 2006; Pietrusewsky 2006). As was noted previously, the clinal nature of population variation within the region is most simply explained by a model of demic expansion of the southern mongoloid Austroasiatic and Austronesian agricultural
123 pop ulations into an area already inhabited by australoid hunter gatherer populations, followed by varying degrees of admixture between these populations. If we accept that the expanding southern mongoloid populations intermixed with, rather than simply repla ced, these resident australoid populations, then the resulting morphological pattern of the descendant populations should reflect this superimposition of a mongoloid phenotype upon the australoid phenotype in other words, we should expect an intermediate, populations. And indeed, this is what Matsumura and Hudson (2005; Matsumura 2006), along with many others (Barth 1952; Coon 1966; Bellwood 1997; Howells 1997), argue is the central phenotypic character of moder n, southern mongoloid Southeast Asians: a unique morphological pattern that shows the strongest affinities to the East Asian populations that share the dominant southern mongoloid phenotype, but that also maintains traces of the underlying indigenous aust raloid phenotype. of dental and cranial remains are in full congruence with the conclusions drawn from the substantial bodies of research in Southeast Asian archaeology linguistics and population genetics. Collectively, these independent, yet intertwined, lines of evidence make a compelling argument for the Two Layer model of population history in Southeast Asia, and this is whence the model draws its strength. The co nvergent interpretations and conclusions derived from these varied fields, requiring few, if any, ad hoc justifications to explain inconsistencies in the data, speak to the near certain biological and historical veracity of Southeast Asian population histo ry as reconstructed under the Two Layer model. And yet. The question of population history in Southeast Asia would be a lesser area of scientific enquiry if all were consensus. Thus, in the latter half of the twentieth century, an
124 alternative hypothesis of regional continuity among the modern, historic and prehistoric populations of Southeast Asia was proposed expressly to challenge the dominant theoretical paradigm of expansion, migration and admixture among these populations (Bulbeck 1982; Turner 1987, 1990, 2006; Nakbunlung 1994; Matsumura and Hudson 2005; Pietrusewsky 2005, 2006; Matsumura 2006). The emergence of the Regional Continuity model of modern Southeast Asian population origins was but one facet of a larger reactionary movement in anthropolog y at the time, in which theoretical models were shifted away from invoking population migration and its implied correlate, population replacement as a means of demic change (Adams et al. 1978; Hutterer 1983; Cavalli Sforza et al 1988; Wolpoff and Caspari 1997; Turner 2006). In some cases, this was a general rejection of migration as an anthropological deus ex machina for explaining quantum shifts in social complexity, technological advancement or biological characteristics within an archaeological or cult ural sequence, without giving due consideration to other mechanisms of change. In the case of Southeast Asia, however, this shift was intended as a important center of cultural diffusion from which new innova Nakbunlung 1994; Stark and Allen 1998; Pietrusewsky and Douglas 2002; Matsumura and Hudson 2005). Surrounded and overshadowed by the enormous cultural complexes of India and especia lly China, Southeast Asia was long viewed merely as a passive recipient of, rather than an active participant in, the broad advances in human cultural, political, economic and biological development that have shaped the modern populations of Asia and the P acific. In the post colonial era, however, researchers sought to distance themselves from the inherent Indo centric and Sino centric biases of the previous eras, and to view the cultures of Southeast Asia as
125 independent and dynamic agents in the region. This change in perception was accelerated by the groundswell of archeological and historical research in Southeast Asia in the middle to late twentieth century, which clearly illuminated both the richness and depth of the historic and prehistoric civilizat ions here, as well as the breadth of their contributions to the development of the modern cultures and peoples of eastern Asia and the Pacific (Hutterer 1983; Nakbunlung 1994; Bellwood 1997; Stark and Allen 1998; Higham 2002; Pietrusewsky and Douglas 2002; Matsumura and Hudson 2005). Additional momentum in this era was provided by the emergent nationalisms of the modern states of Southeast Asia, which structured their identities and legitimated their rights to sovereign rule and territorial control via the ir perceived descent from the great ancient civilizations of the region (Rinehart 1981; Baker and Phongpaichit 2005). In light of these epistemological shifts, both scientific and political agendas were driven, implicitly and explicitly, by a desire to de monstrate direct, lineal continuity between the modern peoples of Southeast Asia and their historic and prehistoric predecessors. The primary proponent of the Regional Continuity model is Turner (1987, 1990, 2006), who has argued for local population con tinuity in Southeast Asia extending as far back as the late univariate distribution frequencies of 28 non metric dental traits in numerous modern and archaeologi cal Asian and Pacific populations, in which he has repeatedly observed two discrete population groupings. These population groupings correspond to two distinct suites of dental morphology, which Turner has designated Sundadonty and Sinodonty The former grouping, which is characterized by what Turner (1987, 1990, 2006) asserts is a simpler and therefore more primitive dental morphological pattern, contains all the modern and archaeological Southeast Asian populations in the analyses, including:
126 late to te rminal Pleistocene (including Hoabinhian) specimens from Niah, Tabon, Gua Cha and Guar Kepah, along with other contemporary specimens from Malaysia and the Indonesian archipelago; Mesolithic Neolithic specimens from Laos and Vietnam; Neolithic to Iron Age samples from Ban Chiang, Non Nok Tha and Don Khlong in Thailand; AD 500; Modern samples from mainland and island Southeast Asia, the Philippines and Polynesia (Turner 1987, 1990, 2006). Convers ely, the second grouping contains the more northerly populations of East Asia (i.e., the northern and southern Chinese, Japanese, Koreans, Mongolians and Asiatic Siberians, among others). This group is characterized by the Sinodont dental pattern: a more complex and therefore more derived morphology containing many of the features, such as small relative tooth size, and high frequencies of incisor winging and shovel shaped incisors, which we today tend to (Turner 1987, 1990, 2006). These basic clustering patterns are reiterated in additional analyses comprising different selections of populations, reaffirming for Turner the biological and evolutionary validity of the Sinodont and Sundadont dichotomy. Turn er (1987, 1990, 2006) derives several primary conclusions from these groupings. First, the tight clustering of modern and archaeological samples from Southeast Asia (i.e., the Sundadonts), to the exclusion of all populations from East Asia, indicates long term, deep time population continuity in both regions. This conclusion is emphasized by the observation that the sample from southern China clusters not with the Southeast Asian populations as expected under the Two Layer model, but with the other East A sian Sinodont populations. Additionally, as was alluded to above, Turner (1987, 1990, 2006) argues that the southern Sundadont pattern is the more primitive dental pattern, based on both its overall morphological simplicity compared to the
127 northern Sinodo nt pattern, and its similarity to late Paleolithic fossil dentitions from northeastern y, Turner argues, this morphological dichotomy could have evolved only if the Sundadont and Sinodont populations had diverged a long time ago and progressed along separate population history trajectories with minimal to no gene flow between them. This div ergence and subsequent isolation would have been the only means of preserving the regional continuity of the distinct East and Southeast Asian populations, thus allowing the development of the derived Sinodont complex in East Asia and the retention of the Sundadont complex in Southeast Asia (Turner 1987, 1990, 2006; Bellwood 1997). Based on the fossil and modern skeletal data, Turner (2006:455; Turner 1990; Turner et al. 2000) argues that the Sinodont complex may have emerged as early as 30,000 years ago a populations migrating across the Bering land bridge. The precise evolutionary mechanism by which Sinodonty emerged is unclear, but Turner (1987:311, 315; Turner 19 90) posits the action of genetic drift or founder effects as the most likely cause, while not completely ruling out selective pressures favoring more elaborate dental architecture in northern populations. Secondly, two subclusters are discernible within t he Sundadont populations: the first containing the modern populations of island Southeast Asia and the Philippines, plus Burma and Nepal; the second containing all the archaeological populations, along with the modern Thai and other mainland Southeast Asi ans. Turner (1987:308) admits that this clustering pattern does not follow as logically from geography as does the southern Sundadont/northern Sinodont division, but he argues that this split within the Sundadont cluster possibly reflects the greater hist oric influence of India and the Middle East (correlated with the spread of Buddhism and Islam) on the
128 modern populations of the first subcluster, while the archaeological populations of the second subcluster would be free of such influence. Alternately, i t is suggested that the populations of the first cluster reflect greater degrees of admixture with northern Sinodont populations. Lastly, Turner (1990, 2006) argues that there is a certain degree of similarity between Sundadonty and the dental morphology of modern Australo Melanesians, indicating that these years ago, and providing additional evidence of the antiquity and long term continuity of the Sundadont comp lex among the populations of Southeast Asia. If Sundadonty has its origins in the late Pleistocene, following the separation of the Southeast Asian and Australo Melanesian populations, Turner (1990, 2006) reasons, then there is no possible way that the Su ndadont phenotype observed in the modern Southeast Asians can be attributed to admixture between resident australoid and immigrant southern mongoloid populations dating only from the Neolithic onward. is little in his analyses or conclusions that cannot be explained with an equal or greater degree of parsimony by the Two Layer model. To begin with, the tight grouping of modern and archaeological Southeast Asian populations within the Sundadont cluster, and the lack of diachronic morphological change between archaeological and modern Southeast Asian populations observations that form the linchpin in are neither incompatible with, nor unexpected under, the Two Layer m primarily of Neolithic and Metal age populations from mainland Southeast Asia populations that under the Two Layer model should be morphologically continuous with the modern populat ions of Southeast Asia, as they all postdate the hypothesized Neolithic southern
129 mongoloid expansion. As was noted earlier, under the Two Layer model, the expected population discontinuity instead should occur between the pre and post Neolithic populatio ns of Southeast Asia, which should be characterized respectively by australoid and southern mongoloid phenotypic patterns (Bellwood 1997; Matsumura and Hudson 2005; Matsumura 2006). It is noteworthy that the few pre Neolithic populations included in Tur analyses are treated in a uniformly unsatisfactory analytical manner, one that seems certain to Neolithic specimens from Laos and Vietnam are lumped together with the far more numerous and far more populous post Neolithic samples from Thailand. Whereas on its own, this Mesolithic Neolithic sample might be quite revealing of population affinities at the critical hunter gatherer to agricultu ralist transition, instead it is subsumed within a temporally, culturally, archaeologically, and therefore biologically meaningless amalgam that Turner patterning in th is transitional population, along with any potential insights into population history post Neolithic populations. Similarly, Turner (1987:306 308) points t o the inclusion of the terminal Pleistocene specimens from Malaysia and Indonesia in the Sundadont cluster as evidence of the great temporal depth of the Sundadont complex and, therefore, of population continuity in Southeast Asia. Yet he fails to acknowl something of an outlier within his Sundadont cluster. This is a critical oversight, because this s have
130 long been argued to display the australoid, rather than the southern mongoloid, phenotype (Barth 1952; Trevor and Brothwell 1962; Cuong 1986; Bellwood 1993, 1997; Matsumura and Zuraina 1995, 1999; Matsumura and Hudson 2005; Matsumura and Pookajorn 2 005; Matsumura 2006). Indeed, Matsumura and Hudson (2005; Matsumura 2006) found that in terms of dental morphology, these same specimens are much closer to modern Australo Melanesian populations than to modern Southeast Asians. Thus, the inclusion of thi in the Sundadont cluster could just as easily be explained within the Two Layer model as a function of the greater similarity between these australoid remains and the intermediate/hybrid phenotype of the modern Southeast heavily biased towards the northern extremes of Eastern Asia, including four separate northern Chinese a nd eastern Siberian populations. As such, the Sinodont cluster is likewise heavily biased towards the morphological extremes of the northern mongoloid phenotype, which Howells (1997:205) argues is a highly specialized morphological suite that arose as an Small wonder, then, that the early specimens from Malaysia and Indonesia would cluster with other tropical populations, regardless of which population histor y model is correct. impression that he is struggling mightily to fit square data into a round theoretical hole. For example, the two subclusters of populations Turner o bserved within the Sundadont group one comprising the modern populations of island Southeast Asia, the Philippines, Burma and Nepal, and the other containing modern and archaeological populations from mainland Southeast Asia correspond almost perfectly to the Austronesian/Austroasiatic population dichotomy
131 posited from archaeological and linguistic reconstructions. But rather than acknowledging this correspondence, which would seem to bring much of his analysis in line with the Two Layer model, Turner (198 7, 1990) resorts to convoluted justifications involving historical population influences that are patently incorrect. It was noted previously that Turner ascribes the separation of these two subclusters to the disproportionate influence of either 1) India and the Middle East, or 2) northern Sinodont populations, on the populations of island Southeast Asia, the Philippines, Burma and Nepal. Yet both of these suggestions are highly nonsensical. There is absolutely no archaeological or historical support fo r a greater degree of Indian influence in the early civilizations of island Southeast Asia compared to those of mainland Southeast Asia in fact, quite the opposite is true (Bellwood 1997; Higham and Thosarat 1998; Higham 2002). Similarly, although Islam ha s had a profound effect on the modern and historic populations of Indonesia and Malaysia going as far back as the thirteenth century, this early influence never penetrated in any meaningful way into the Philippines or the islands of eastern Indonesia that in western Indonesia, where Islam gained its greatest footholds in Sumatra and Java dating back to the fifteenth century AD a large degree of this influence d ates only to the nineteenth century and is attributable throughout more to the flow of ideologies and commerce, rather than of people or genes, to the archipelago (Bellwood 1997; Winchester 2003). It therefore is likely of little bearing on the biological affinities of these modern populations, or on the question of regional continuity. It is equally absurd to suggest that the far flung populations of island Southeast Asia would show greater evidence of admixture with northern mongoloid populations than w ould the more geographically proximate populations of mainland Southeast Asia. It is Layer model that his southern Chinese
132 population sample is grouped with the Sinodonts rather than the Sundadonts, as this sample is composed substantially of modern and historic individuals from Canton, who likely are predominantly of Han Chinese descent and therefore should be expected to display a northern mongoloid/Sinodont phenotype. Collectively, these critique seriously undermined by a flawed rendering of Southeast Asian prehistory, and of the archaeological, historical and linguistic evidence germane to the debate of population origins in this region. Turner (1987, 1990) also shows a paradoxical regard for the influence of geography on the population origins debate that weakens his position, as he alternately emphasizes geographical proximity over known historic population relationships, or he disregar ds geography entirely in order to somehow explain away incongruent data and interpretations. Indeed, the whole issue of population continuity versus discontinuity within Southeast Asia is largely resolved if southern China is appropriately regarded as bio geographically continuous with, and modern political borders (Barth 1952; Bellwood 1993, 1997; Matsumura and Hudson 2005; Pietrusewsky 2006). The subtle distin ction between the modern and historic populations of this who are ethnically Chinese (i.e., Han Chinese) and the archaeological populations who inhabited sites that today are located in China is a critical point that often seems to be lost on the advocates of the Regional Continuity model. Turner (2006:456; 1987, 1990) exacerbates this error by incorrectly framing the Two elanesians [ sic ] modern populations of China, who are universally and correctly regarded as Northeast Asians
133 (i.e., northern mongoloids), and the archaeological populations of regions that would eventually become part of China, who are more appropriately regarded as Southeast Asians (i.e., southern mongoloids) (Howells 1997; Bellwood 1993, 1997; Matsumura and Hudson 2005). Such misinterpretations, whether accidental or intentional, inject interference into the scientific dialogue between the competing theoretical models and impede progress towards an objective, scientifically valid resolution of the underlying debate. At worst, Turner (1987, 1990, 2 006) foregoes evolutionary parsimony to construct an alternative interpretation of raw data and analytical results already in agreement with the dominant theoretical model of Southeast Asian population origins. Turner (1987, 1990) reasonably concludes tha t the Sinodont complex likely originated due to genetic drift or a founder effect among populations spreading northward from the Sundadont homeland in Southeast Asia. But then, in order to account for the long term continuity of the distinct Sundadont and Sinodont lineages, he relies on a thoroughly unsatisfactory scenario of tens of millennia of near absolute genetic isolation between ostensibly culturally similar populations, spread across an area that lacks any significant geographical or climatic barri ers to population movement and admixture, in a region that has been a thoroughfare for large scale population movements since the time of Homo erectus (Bellwood 1997; Higham and Thosarat 1998; Higham 2002; Baker and Phongpaichit 2005). And yet in nearly t he same breath, Turner (1990, 2006) argues that the Australo Melanesian and Southeast Asian populations who have been separated by thousands of miles for as long as 60,000 years retain sufficient similarities in dental morphology to be grouped together to the exclusion of the Sinodonts, without suggesting a suitable evolutionary mechanism by which this similarity was maintained.
134 Melanesian dental complex are descended from an ance stral population in Southeast Asia, and his position that the defining features of Sinodonty and the Australo Melanesian complex were established by genetic drift and founder effects, are both probably true (Howells 1997). Yet the evolutionary scenario he has constructed to explain this pattern of descent, relying on a sole mechanism of divergence and subsequent continuity in isolation, is most certainly flawed. Matsumura and Hudson (2005, 2006) have argued that these dental complexes are neither as discr ete nor as static over time as Turner contends and as the continuity model demands. Indeed, despite the clarity groupings, his presentation of individual trait frequencies demonstrates a continuous, rather than bi modal, pattern of dental morphological variation among these populations. Although analysis of univariate trait frequencies is not the most powerful or elegant statistical approach to exploring populat ion relationships, such analyses do suggest intermediate morphological patterns that may be obscured by the rigid inclusion/exclusion parameters of most cluster analyses (Pietrusewsky 2000; Matsumura and Hudson 2005, 2006; Matsumura 2006). Within this con tinuous spectrum of variation, Turner (1987, 1990; Matsumura and Hudson 2006) acknowledges that his Sundadont populations often tend to fall in intermediate positions between the complex Sinodont morphological pattern and the simple Australo Melanesian mor phological pattern. Turner (1990:315) also acknowledges that there is evidence of some diachronic change in trait frequencies between prehistoric and modern Southeast Asians, with the later populations displaying a subtle shift towards the Sinodont patter n. These observations, admitted only in passing by Turner, are not in keeping with an evolutionary pattern of divergence followed by isolation, but are fully consistent with a pattern
135 of divergence followed by admixture in short, with the patterns of population history posited by the Two the biological evidence delivers us once again to the conclusion that the unique morphological pattern of the modern Southeast Asian s, intermediate between the phenotypic and geographic extremes of East Asia and Australo Melanesia, is most parsimoniously explained by the recombination of these divergent mongoloid and australoid populations in a region where they overlap. From archaeol ogical and linguistic evidence, we can be increasingly confident that this overlap was brought about by the southward migration of agricultural populations from southern China into the midst of the resident hunter gatherer populations of Southeast Asia (Be llwood 1997, Howells 1997, Higham and Thosarat 1998; Higham 1996, 2002; Matsumura and Hudson 2005, 2006; Matsumura 2006). torch of the Regional Continuity model usi (1994) dissertation is noteworthy as one of the few studies in existence that specifically address the question of the origins of the modern Thai people. In this study, modern and prehistoric Thai and Chines e samples are compared using a somewhat unconventional statistical approach based Thai are unlike both the archaeological Thai and archaeological Chinese sample s, leading to the conclusion that neither group is a likely ancestor for the modern Thai. However, Nakbunlung (1994:72, 106) notes that the modern Thai are less unlike the prehistoric Thai than they are unlike the Neolithic Chinese, and that therefore her data are generally in agreement with the Regional Continuity model.
136 interpretations, unfortunately, are undermined by several flawed assumptions. First, given the historic al and linguistic evidence that the ethnic Thai did not arrive in the Chao Phraya basin until the late first millennium AD at the earliest, it should come as no surprise that the modern led skeletal data from Khok Phanom Di (late hunter gatherers), Ban Kao (Neolithic), Non Nok Tha (Neolithic/Bronze Age), Ban Na Di (Bronze Age) and Tham Ongbah (early Iron Age) (Nakbunlung 1994:43). of pooled data from five populations: four from the Yellow River basin and only one small population (n = 9) from southern China. As such, this sample is hardly representative of the agricultural southern Chinese populations whence the Austroasiatic and Thai populations of post Neolithic mainland Southeast Asia are thought to be derived. These two archaeological samples are thus of little comparative value for exploring the population history of the modern Thai or testing the relative merits of the Two L ayer or Regional Continuity models with regards to this modern population. It must also be study underscore a challenge acknowledged by all craniometric r esearchers: that it is exceedingly difficult to obtain sufficient craniometric data from comparative populations that are geographically and temporally appropriate to the question at hand. Still, researchers must be cognizant of the interpretive limitati ons imposed by such incomplete or inadequate datasets, regardless of how unavoidable they may be. Most notable among the supporters of the Regional Continuity model, however, is Pietrusewsky (1981, 1984, 1990, 1992, 1994, 1996, 2000, 2005, 2006, 2008a,b; P ietrusewsky and Douglas 2002; Pietrusewsky et al. 1992, among many others), who, although less vociferous
137 than Turner, has been no less stalwart in his adherence to this theoretical framework over the last three decades. Through numerous methodologically and statistically robust craniometric studies, Pietrusewsky has concluded, like Turner, that the biological distances and resultant clustering patterns demonstrated among Asian, Australo Melanesian and Pacific populations are most consistent with the Regio voluminous, the majority of his studies differ from one another only in terms of the number of populations he has included in his analyses. While some studies are rather narrowly focused on specific countries or time periods looking, for instance, at the modern, historic and prehistoric populations of Japan (Pietrusewsky 1992, 1996), or at early Metal Age populations in Thailand (Pietrusewsky 1981, 1984; Pietrusewsky and Douglas 2002) oth er studies are remarkable in their breadth, incorporating as many as 70 modern and archaeological populations from a vast swath of Asia, Australia and Oceania. Through this collective body of work, Pietrusewsky has made two critical accomplishments above and beyond making his case for the Regional Continuity model: 1) he has compiled a comprehensive array of craniometric datasets from nearly every modern and archaeological population in Asia and the Pacific that is relevant to the question of population o rigins in this region (although his samples are primarily only males); and 2) his repeated analyses of similar sets of population samples, with their overwhelmingly similar outcomes, have established the precision and reliability of the clustering patterns detailed in these studies. 2006, 2008a,b; Pietrusewsky et al 1992) cluster analyses of Mahalanobis (D 2 ) distances repeatedly demonstrate that these modern and archa eological populations fall first into two large, now familiar superclusters, one for Asia and Polynesia and one for Australia and Melanesia.
138 Within the Asia/Polynesia supercluster, Pietrusewsky invariably finds a primary division between Polynesians and E ast/Southeast Asian populations. Within this latter cluster, he likewise consistently finds distinct subclusters of Japanese, Chinese and Southeast Asian populations, although, interestingly, the specific interpopulational relationships are rather fluid wi thin each these subclusters. These clustering patterns are also confirmed by stepwise multiple discriminant (canonical) function analyses, which generally show three discrete, well separated constellations of populations along the canonical axes: Austral o Melanesians, Polynesians/Micronesians and East/Southeast Asians (this last grouping frequently resolves into separate East and Southeast Asian constellations). Almost without exception, Pietrusewsky interprets these patterns as indicative of separate po pulation origins and long term continuity within each of these regions, congruent with the Regional Continuity model (Pietrusewsky 1990, 1992, 1994, 2000, 2005, 2006, 2008a,b; Pietrusewsky et al. 1992). 2005, 2008a,b; Pietrusewsky et al. 1992) craniometric analyses have revealed a host of population relationships that are clearly in line with conclusions from archaeological, linguistic, historic and genetic evidence, including: Connections between the mo dern populations of island Southeast Asia and Polynesia, consistent with the theory that the Polynesians have their ultimate origins among the Neolithic Austronesian populations of the Indonesian archipelago, rather than among the australoid populations of Melanesia (Benedict 1942; Ballinger et al. 1992; Melton et al. 1995; Higham 199 6; Bellwood 1997; Howells 1997). Close relationships between the populations of mainland and island Southeast Asia, indicating that these populations have a common origin at so me point in prehistory (Higham 1996, 2002; Higham and Thosarat 1998; Bel lwood 1993, 1997; Howells 1997). A clinal pattern of morphological variation within island Southeast Asia, ranging from populations of greater mongoloid affinities in the north and wes t to populations with more australoid affinities in the south and east, reflecting the respective spheres of influence of these two phenotypic complexes (Barth 1952; Coon 1966; Bellwood 1997; Howells 1997; Matsumura and Hudson 2005; Matsumura 2006).
139 Of gre atest interest for addressing the endogenous versus exogenous debate surrounding Southeast Asian population history, however, are the clustering patterns among the archaeological and modern populations of China and mainland Southeast Asia. Here, Pietrusew sky points to two recurring observations as definitive evidence of the direct, locally continuous, in situ development of the modern populations from their prehistoric antecedents (Pietrusewsky 1981, 1992, 1996, 2000, 2005, 2006, 2008a,b). First, the arch aeological populations from China and mainland Southeast Asia generally group most closely with geographically proximate modern populations. Thus, the Bronze Age (ca. fourteenth to eleventh century BC) Chinese sample from Anyang, in the eastern Yellow Riv er valley, clusters with modern Chinese and other East/Northeast Asian populations, including Koreans and Taiwanese; Similarly, samples from Khok Phanom Di, Ban Chiang and Non Nok Tha in Thailand group closely with one another, with Hoabinhian/early Neolit hic samples from Laos and Vietnam, and with modern Southeast Asian populations (Pietrusewsky 1981, 1984, 1992, 1994; 1996, 2006; importantly, the modern populations of East A sia (particularly those from China) and Southeast Asia always form separate clusters with little to no overlap, clearly indicating the separate origins of these regional populations with little to no admixture between them (Pietrusewsky 2005, 2006, 2008a,b ; Pietrusewsky et al. 1992). demonstrated high level of replicability, lend a considerable amount of weight to his arguments in support of the Regional Continuity mo del, which have rarely been challenged. As with much of science, however, what constitutes the most persuasive lines of evidence is generally a matter
140 the sepa ration between the East and Southeast Asia clusters. But it bears noting that this separation is not incompatible with the Two Layer model, which accepts that the post Neolithic populations of both China and Southeast Asia each followed separate developme ntal trajectories along the path towards literate state level civilizations. As such, we should expect to see some separation between the modern populations of these two areas, despite their common roots in the early agricultural populations of China. Ma tsumura and Hudson (2005, Matsumura 2006), on the other hand, take a different view the similarities between the Chinese and Southeast Asian subclusters, which consistently join with one another before either group clusters with any other Asian population group. These repeated demonstrations of strong affinities between the Chinese and Southeast Asian subclusters argue just as strongly for a significant historic al pattern of gene flow and biological relatedness between them, as would be suggested under the Two Layer model. Indeed, in light of the widely acknowledged close biological and cultural ties between Polynesia and (especially island) Southeast Asia the s eparation between which only dates back ca. 4000 years (Bellwood 1997; Howells 1997) we might reasonably expect the Polynesian and Southeast Asian subclusters to link up first. The fact that they do not argues for the great strength of the biological ties between East and Southeast Asians dating back over the same time period. It hardly needs mentioning at this point that this interval of four millennia also corresponds almost precisely to the post Neolithic era in Southeast Asia and the hypothesized time line for the migrations of agricultural populations from southern China into Southeast Asia. As such, the clustering patterns between East and Southeast Asian populations demonstrated in
141 L ayer model as easily as they can be said to support the Regional continuity model (Matsumura and Hudson 2005). Granted, the more distant clustering relationships between Polynesians and Southeast Asians could possibly be dismissed as evidence of a sharp di vergence between these populations created by profound genetic drift in Polynesians, as a result of the severity of the repeated founder effects imposed during their colonization of the Pacific. However, the strong genetic and linguistic similarities whic h exist between modern Polynesians and Southeast Asians despite these founder effects (Balinger et al. 1992; Melton et al 1995; Bellwood 1997) would seem to argue against this interpretation. Likewise, the close relationships between Chinese and Southeas t Asian populations could be attributed to recent admixture among the modern populations of these areas, which certainly has occurred but is unlikely to be sufficiently uniform in its extent across all of Southeast Asia to fully account for this clustering pattern. Accordance with the Two Layer model is a far more parsimonious and a far less ad hoc justification for the observed patterns of population relationships. Curiously, additional support for the Two f ound if, instead of stepping back from the data and examining clustering patterns at a regional level, we close in and look more narrowly at the clustering behavior of individual populations, rather than temporal or geographic groups of populations. Conve niently, a number of prehistoric (Pietrusewsky 1981, 1984, 1994, 2006; Pietrusewsky et al. 1992; Pietrusewsky and Douglas 2002), allowing us to evaluate the popu lation relationships between archaeological and modern populations that are central to the debate of Southeast Asian origins. In some of the earliest studies (Pietrusewsky 1981, 1984), cluster analyses of Mahalanobis (D 2 ) distances positioned a
14 2 small samp le (n = 10) from the Neolithic/Bronze Age site of Ban Chiang as outgroup to a general Asian cluster containing modern mainland and island Southeast Asian and East Asian s are positioned as the outgroup to all populations, including Ban Chiang. Direct examination of the distance matrix as well as the canonical plots from stepwise discriminant function analysis also reveals somewhat close affinities between the Ban Chiang population and the populations of island Southeast Asia. Although Pietrusewsky (1981:23) interprets these observations as indicative of population continuity back to the late Pleistocene, this clustering pattern also lends itself to several conclusions co nsistent with the Two Layer. First, while the distance matrix does align the Ban Chiang population with island Southeast Asia, the distances here are relatively large (5.75 to 13.8, with most between 7.0 to 9.0) compared to distances among the modern pop ulations of East and Southeast Asia (which range widely, but are typically on the order of 1.5 to 7.0) (Pietrusewsky 1981:11 12). This general modern Asian pop ulations (Pietrusewsky 1981:8). Collectively, these observations suggest that Laye r model, we would expect Neolithic populations such as Ban Chiang to display a unique phenotype reflecting the recent initiation of admixture between intrusive southern mongoloid and resident australoid populations. This would result in variable Neolithic populations that are similar to, yet noticeably distinct from, the modern populations in which morphological variation has been somewhat homogenized via continued admixture over the course of millennia.
143 1984) analyses is fully in agreement with the Two Layer model, as these populations in fact probably date to the Hoabinhian period and therefore are likely to display the australoid phenotype (Matsumura and Hudson 200 5; Pietrusewsky 2006). Their extreme separation from modern Asian populations, as reflected in their relatively large Mahalanobis distances and their position as the ultimate outgroup, underscores the morphological distinction between these australoid Pal eolithic populations and the modern mongoloid populations of Asia. Furthermore, their morphological separation from the Ban Chiang population is a demonstration of precisely the type of discontinuity between pre and post Neolithic populations predicted b y the Two Layer model. Similar clustering patterns, with Neolithic/Bronze Age populations from Thailand positioned as an outgroup to all modern Asian populations, are observed in several other studies as well (Pietrusewsky 1994, 2006). In each case, the archaeological populations join the Asian cluster immediately before the Polynesian cluster joins to create th e Asian/Oceanic supercluster. Such a clustering pattern further indicates a degree of close morphological, and by extension genetic, similarity among these populations that is at odds with the scenario of separation and divergence at the core of the Regional Continuity model. But even the rare exceptions to this pattern are illuminating: in one study (Pietrusewsky et al. 1992), a pooled sample o f skeletal specimens from Ban Chiang, Ban Na Di (Bronze/Iron Age) and Non Pa Kluay (Bronze Age/Iron Age) clustered not with Asian populations but with populations from Tasmania and Melanesia. Although it is unwise to draw too broad a conclusion from a sin gle, apparently idiosyncratic analytical result, this clustering pattern provides a tantalizing suggestion of some remnant degree of australoid morphology among the Metal Age populations of Thailand.
144 either of two competing hypotheses speaks to the intricacy of the debate over Southeast Asian origins, as well as the subtlety and complexity of the population relationships driving that debate. None of the forgoing discussion is to say that Pietrusewsky is mistaken in his interpretations of his own data, which are uniformly scientifically and logically sound. My point here is only to demonstrate that his data do not provide the iron clad support for the Regional Continuity model that he repeatedly argue s they do. But this leaves us no closer to resolving the debate in any sort of satisfactory fashion. So how do we pursue such a resolution? The archaeological and linguistic lines of evidence bearing on the question of Southeast Asian population origins are so well established and so widely accepted so unlikely to undergo significant revision that they provide few avenues for potential scientific advances in our understanding of this debate. Others will surely argue that genetic studies are the way forwa rd. As demonstrated above, it is true that populations genetics research has the potential to yield great insight into the population history of Southeast Asia. However, just as craniometric and odontological analyses have failed to resolve the matter on their own, it is equally unlikely that population genetics would be able to singlehandedly put the matter to rest. The first problem is that it is unlikely, although certainly not impossible, that we would be able to obtain sufficient DNA sequences from the salient archaeological populations in Southeast Asia to allow a truly definitive assessment of the affinities of these populations. The time, expense, curatorial objections to, and sheer difficulty of collecting, extracting and sequencing ancient DNA argues against this possibility. Although the possibility of reconstructing the time depth of specific mutations and other sequence data using molecular clocks and DNA sequences from modern populations does help the situation somewhat, it is less ideal t han being able to obtain samples
145 directly from the ancient populations themselves. Therefore, genetic analyses will likely contribute only a part, not the whole, of the answers we are seeking with regards to population history in Southeast Asia. At this p oint, it seems inevitable that a good deal of the remaining portion of those answers will have to come from craniometric studies, which have none of the temporal, financial, or technical difficulties of DNA, and only a very few of the social and philosophi cal difficulties. But repeated analyses of the same sets of populations, with only a few new populations added here and there, is unlikely to yield any new insights or to get us any closer to the answers we seek. A significant flaw of the existing cranio metric studies is that they frequently consider archaeological and modern populations somewhat monolithic, using a single, inappropriately small sample from a single locale as a proxy for the entire population. Thus, in many of the craniometric studies re individuals obtained from a single medical school in Bangkok. We have no idea how well this small sample reflects the true extent of craniometric variation in the Thai population as a whole, much less the extent of variation in the 80% of the population that does not live in Bangkok. (1981, 1984, 1994, 2006; Pietrusewsky et al. 1992; Pietrusewsky and Dou glas 2002) analyses consists of only 10 to 12 individuals, selected from among of 142 burials recovered from nearly two millennia of cultural deposits. How can such a sample be truly representative of this ancient population, which spans the interval from the Neolithic to very near the advent of literate civilization in Thailand? Little to no attention has been paid to the possibility of significant variation within temporally or geographically bounded populations, or the potential impact of such variation on
146 the reconstruction of population relationships. For example, is there significant morphological variation among the skeletal series from various Neolithic sites in Thailand? Any influx of agricultural populations from southern China at the dawn of th e Southeast Asian Neolithic was unlikely to be uniform in its distribution across mainland Southeast Asia, so there is no reason to expect the Neolithic populations of that region to be completely homogenous in terms of morphology. The various analyses of skeletal series from Ban Chiang, Non Nok Tha and Khok Phanom Di (Pietrusewsky 1981, 1984, 1994, 2006; Pietrusewsky et al. 1992; Pietrusewsky and Douglas 2002; Matsumura and Hudson 2005; Matsumura 2006) indicate different close relationships between each o f the individual Thai samples and various geographically proximate, non Thai populations, further suggesting that there may be significant intrapopulation diversity and regional geographic patterning within the ancient and modern Thai populations. Matsumura (2006) has argued that the way forward in resolving the debate between the Two Layer and Regional Continuity models of Southeast Asian population history does not lie in broader geographic sampling of populations. Rather, future sampling efforts must have greater temporal depth, yet be more geographically focused, because the nature of prehistoric and historic population interactions likely varied substantially from one time period to another and from location to location within the larger geogra phic contexts of Southeast Asia. This is particularly true for Thailand, because superimposed on the complex patterns of population interactions (i.e., locally varying degrees of admixture and replacement between indigenous and immigrant populations) posi ted by the Two Layer model, is an equally complex pattern of recent and historical population movements and interactions (Rinehart 1981; Baker and Phongpaichit crossroads of cultural and biological influences from prehistoric, historic and modern
147 populations all across central, eastern and southeastern Asia (Briggs 1945; Rinehart 1981; Bellwood 1997; Higham and Thosarat 1998; Higham 2002; Baker and Phongpaichit 2 005). The only way to truly grasp the nature and intensity of these varying influences on the modern Thai population, and our only hope of even beginning to unravel the complex, deep time population history of the Thai and the other peoples of Southeast A sia, is to sample these modern populations as thoroughly as we can at the local level, to build up our understanding of local patterns of intra and interpopulation morphological variation, and then to compare any such patterns to similar local patterns of variation in the salient archaeological populations
148 Figure 3 1 Selected archaeological sites in China ( Redrawn from original map data obtained from the CIA World FactBook (2010) and used with permi ssion. Site locations following Glover and Higham 1996; Higham 1996, 2 002 ). A. Anyang B. Baiyankun C. Fenshanbao D. Hemudu E. Pengtoushan
149 Figure 3 2. Relationships among the various language families of Southeast Asia.
150 CHAPTER 4 MATERIALS AND METHOD S The human features and countenance, although composed of but some ten parts or little more, are so fashioned that among so many thousands of men there are no two in existence who cannot be distinguished from one another. Pliny the Elder Natural History, VII, 8 (AD 77 79) Resolution of the debate over Southeast Asian population origins cannot be accomplished within the space of a single dissertation nor, indeed, within the space of a single career. What follows instead is an attempt to begin to address several persistent i ssues in craniometric studies of Southeast Asian populations, all of which have hampered our understanding of these populations in the contexts of both forensic anthropology and population history. The first of these issues is the exceedingly inadequate representation of Southeast Asian populations in craniometric databases used by forensic anthropologists in the identification of unknown skeletal remains. As such, most practitioners have only a vague sense of the scope and pattern of craniometric variat ion within these populations. Such deficiencies have a serious negative impact on forensic identification efforts in Thailand, in cities and regions of the United States with significant Southeast Asian communities, and by agencies such as the J oint P riso ner of CIL) that frequ ently encounter the remains of Southeast Asians in the course of their casework. Fortunately, this problem is easily remedied through more extensive sampling of modern populations; in the case of Thailand, robust skeletal collections are available at any of several medical schools throughout the country. In addition to advancing the cause of forensic identification, more extensive sampling of the modern Thai population allows us to explore patterns of intrapopulation craniometric variation on a much finer scale, and to consider the possible impact of such variation on the
151 reconstruction of population history through craniometric comparisons. It also allows u s to address the question of whether a sole, relatively small sample from a single anatomical or archaeological skeletal collection is an appropriate and accurate proxy for the Thai population as a whole. In short, gaining a greater understanding of patte rns of variation which exist within the modern Thai population, both between modern Thai males and females and among regional Thai subpopulations, is a matter of both practical and academic interest. The Samples In order to gain a greater understanding o f patterns of craniometric variation in the modern Thai, twenty four standard cranial measurements (Tabl e 4 1; Figures 4 1 to 4 3 ) (Martin and Saller 1956; Howells 1973; Moore Jansen et al. 1994) were collected on a total of 429 known individuals from the modern anatomical skeletal collections maintained by the Departments of Anatomy at Naresuan University 1 (NU), Chiang Mai University (CMU) and Khon Kaen University (KKU), Thailand (Table 4 2). The source collections are composed of individuals of known age for gross anatomy dissection; the majority of individuals have dates of death between 1992 and 2006, with a few individuals having dates of death as early as 1979. Likew ise, dates of birth range between 1899 and 1983, with the majority of individuals (70%) having been born between 1926 and 1956. These individuals represent local subpopulations in and around three of Kaen, respectively) and also broadly North and N ortheast, respectively (Figure 4 4 ) (Bunge 1981; Higham and Thosarat 1998; Baker 1 The skeletal collection at Naresuan University consists of primarily of skeletons loaned to the University from Siriraj Hospital, Mahidol University, Bangkok, along with a small number of individuals (n=8) on loan from the collections at Khon Kaen University. These latter have b een added to the sample of individuals measured from the KKU collection; the others are designated as the Naresuan University sample but represent, inasmuch as is possible, the population of Bangkok.
152 and Phongpaichit 2005). As su ch, these craniometric data are representative of a substantial portion of the total range of morphological variation within the modern Thai population, and therefore provide some insight into the degree of morphological variation among several regional su bpopulations within Thailand. All craniometric data were collected by this author between June 2006 and November 2009, using GPM analog sliding and spreading calipers, with measurements recorded to the nearest millimeter. Measurements were taken followin g the landmark and measurement definitions provided by Moore Jansen et al. (1994); these are reproduced in Appendix A for reference. In the total sample, males outnumber females by a ratio of 2.13 to 1; this approximate ratio holds for each of the region al subsamples, with the exception of Khon Kaen, in which specimens were measured in alternating male female order due to time constraints. In males, age at death ranged from 22 to 95 years, with a mean age at death of 65.58 years (standard deviation [s.d. ] = 13.46 years); in females, age at death ranged between 27 and 94 years, with a mean age at death of 64.35 years (s.d. = 14.08 years). Mean ages at death are not significantly different between males and females (assessed using an independent samples t test for the equality of group means; p = .41), or among regional subsamples (assessed using a one way analysis of variance [ANOVA] ; p = .83). The distribution of age at death is significantly skewed to the left, and fails the Shapiro Wilk test for normal ity ( p > 0.000). However, both the sex ratio and the age distribution are typical of anatomical collections (Hunt and Albanese 2004; Buikstra and Komar 2008; Komar and Grivas 2008). Statistical Methods The majority of statistical analyses used in this di ssertation were performed in SPSS Release Version 19.0.0 (SPSS, Inc. 2010). Some preliminary data cleaning and summary statistics were performed using Microsoft Office Excel 2007.
153 Basic Assumptions Craniometric data lend themselves to a variety of exploratory and analytical multivariate statistical methods. These various methods allow for both the description of patterns of variation within and between groups of known individuals, and the classification of unknown individuals into one of the consti tuent groups in an analysis. In general, these multivariate methods are underpinned by the paired assumptions that the variables in the analysis are characterized by normally distributed response values (i.e., measurement values) and homogenous variances among the different groups in the analysis (Kachigan 1991; Feldesman 1997 ; Field 2009 ). As there is no efficient way to test for multivariate normality, it is generally accepted that if the individual variables in an analysis are each normally distribute d, then the assumption of multivariate normality holds (Kachigan 1991; Feldesman 1997). Normality of the individual craniometric variables was tested using the Shapiro Wilk test (Shapiro and Wilk 1965) at p 3. As is clear from this table, the majority of cranial vault measurements are normally distributed (the exceptions generally being breadth measurements such as maximum cra nial breadth), while the majority of facial measurements have test values indicating significant deviations from normality. These cases of non normality are attributable to any of several potential causes. The first of these are simple limitations of the Shapiro Wilk test, which becomes more sensitive to small deviations from normality with increasing sample sizes. The sensitivity of the Shapiro Wilk test is also significantly affected by large numbers of ties in the dataset (i.e., numerous individuals i n the dataset with identical measurement values for a given variable). Ties are common in the present dataset due to the relatively narrow range of human cranial variation overall and the fact that all measurements were recorded to the nearest millimeter. Large sample sizes and large numbers of ties within the dataset may lead to artificial indications of non
154 normality under the Shapiro Wilk test, when in fact the sample values do not deviate from the normal distribution to an extent sufficient to affect robust multivariate an alytical methods (Field 2009). In addition to these immediate statistical methodology considerations, there are underlying evolutionary and developmental factors that are likely to have a significant impact on the distribution of resp onse values for craniometric variables. The functional matrix model of cranial development suggests that the growth and mature morphology of the skull is driven by the growth and development of neural tissues, oral and nasal passages, and special sense or gans such as the eyes, olfactory apparatus and glossopharyngeal structures (Moss 1973, 1997a c; Humphrey 1998). Accordingly, the intimate association between these soft tissue structures and the basicranium and portions of the facial skeleton places these cranial regions under significant, but independent, developmental and morphological constraints. As such, it is reasonable to expect a narrowed range of morpho logical and metric variation in features such as the orbits and cranial base (Moss 1973, 1997a c). Therefore, the distribution within a population of measurement values for these cranial features may be truncated at both the upper and lower extremes, resulting in a statistically significant deviation from univariate normality for these craniometric variables. It seems certain that this consideration would apply to all human populations, and as it does not seem to have posed a problem in previous craniometric studies, it is assumed it will not have a significant impact on the present study. Both of t hese sets of concerns regarding the normality of the dataset can be ameliorated somewhat by examining the magnitude of skewness (asymmetry) and kurtosis (peakedness) values for individual craniometric variables, as well as by examining the quantile ( Q Q ) p lots and histograms of response values for individual variables for their extent of deviation from the
155 normal curve. With regard to the present dataset, several observations can be made. The majority of instances of non normality appear to be due to posi tive kurtosis values, indicating a higher than normal peak to the distribution curve. This is a further indication that the individual with identical measureme nt values at, or very close to, the mean. Such tight grouping of response values around the mean is additionally suggestive of tight constraint on the range of variation for certain variables. Furthermore, the fact that the male sample had twice as many variables fail the test of normality than the female sample suggests that the validity of the test is unduly affected by size of the male sample relative to the female sample. This is reinforced by the observation that, with regard to many of the variabl es in which the male sample deviates significantly from normal but the female sample does not, the male sample actually has lower absolute skewness and kurtosis values than the female sample. This indicates relatively smaller deviations from normality in the male sample than in the female sample, despite the determination of significant non normality for males under the Shapiro Wilk test. In light of these considerations, it is believed the non normality of these variables is insufficient in magnitude to have a serious negative impact on the statistical analyses herein, which generally are robust to slight departures from normality. Homogeneity of variances ( homoscedasticity ) between male and female samples and was ). This method is appropriate for use with this dataset as it is relatively insensitive both to differences in sample size and deviations from normality (Ott and Longnecker 2001). Homoscedasticity was confirmed for all variables at except cranial base le p = .001 ( Table 4 4 ). For this variable, variance in the male sample (20.822) is nearly twice that in the female sample (11.810), indicating a narrower
156 range of variation in cranial base length in females. The ma le sample contains several outliers with extreme values for this variable, including one individual with a cranial base length of 88 mm and three with BNL measurements of 112 or 113 mm. However, removal of these individuals from the sample did not resolve the heteroscedasticity of this variable. An alternate explanation is that a developmental factor is at work. It is broadly accepted that females generally display a narrower range of morphological variation than males ( the hypothesis of female canalizat ion; Waddington 1942; Hamilton 1982; Stinson 1985; John Byrd, pers comm.) ; this effect may be especially profound with regards to the cranial base, which is intimately associated with a number of critical neural and endocrine structures, such as the crania l nerves, brainstem, pituitary gland and hypothalamus ( Moss 1973, 1997a c ) Data transformations and non parametric approaches Clearly, failure of a given dataset to meet the basic assumptions of normally distributed variables with equal variances introd uces significant doubt into the validity and reliability of the results any subsequent multivariate statistical analyses performed on that dataset. Due to the complexity of the methods involved, it can be difficult to assess to what degree the violations of these assumptions may be driving significant results in the multivariate analyses and masking or exaggerating the true patterns of variation within the dataset. To a certain degree, these doubts can be mitigated by a close understanding both of the limitations of the tests used to establish the the preceding discussion of the Shapiro Wilk and Levene tests (Field 2009) However, these considerations ar e generally (and correctly) regarded as inadequate for fully eliminating all and the veracity of the results of those analyses
157 Many methods exist for resolving these issue s of non normality and heteroscedasticity using various data transformations. These methods generally involve the global application of a mathematic operator, such as logarithms, square roots, or trigonometric functions (angular or arcsine transformations ), to the dataset which renders the variables into normal, homoscedastic distributions ( Conover and Imam 1981; Sokal and Rohlf 1995). Although these methods are quite powerful from an analytical standpoint, they can have a significant drawback for practi cal statistical applications (Conover and Imam 1981) such as the use of discriminant functions in forensic anthropology identifications. In this applied context, if an analyst wishes to apply a discriminant function derived from transformed data to a nov el unknown individual, the battery of craniometric data collected on that novel individual must also undergo the same transformation, thereby creating an extra computational step that 1) presents an additional opportunity for the introduction of error into the classification analysis; and 2) may be challenged in court as an undesirable or unwarranted manipulation of the skeletal data. Fortunately, there exists a simple r and analytically expedient means of evaluating the effects of non normality and heterosc edasticity on the outcomes of multivariate analyses, as well as for teasing out what violations of these assumptions are due to the limitations of the test methods themselves using rank order ing Rank order ing involve s sorting the re sponse values for eac h variable, across all groups, from smallest to largest and then assigning the smallest response value the rank of 1, the next highest the rank of two, the next highest the rank of 3, and so forth. Ties in the response values are assigned the mean values of their total ranks (i.e., four individuals with maximum cranial breadths (XCB) of 131 mm and ranked as 5 through 8 in the ordering would each be given the rank of 6.5). Rank ordering mitigates the non normality and heteroscedasticity of the underlying d ata, and reduces the disproportionate effects of outliers on
158 both of these characteristics, yet preserves a faithful representation of the central tendency and spread of the data (Ott and Longnecker 2001; Field 2009). Typically, rank ordered datasets are then analyzed using non parametric statistica l tests (e.g., the Mann Whitney test ) that are homologous to the more familiar parametric tests (e.g., t test), but which are freed from the constraints of assumed normality and heteroscedasticity H owever, Conover and Iman (1981) have shown that standard parametric tests, such as the t test, can be applied to rank ordered data and are fundamentally equivalent to non parametric tests in such applications, which they call rank transformations This is due to the preservation in the rank ordered dataset of faithful measures the central tendency and spread of the original data two criteria that are essential to parametric statistics. In rank transformations if the parametric tests of the rank or dered data return results that are essentially identical to the original parametric tests on the unranked data, then it can generally be assumed that the original parametric tests have not been unduly affected by violations of the assumptions of normality and homoscedasticity in the underlying dataset The rank transformation technique will be applied to the present dataset as warranted within the context of each of the analyses outlined in the following chapter. Multivariate Statistical Analyses Despite their long history as an integral part of the practice of physical anthropology, for many years the potential of craniometric data to explore, describe and utilize patterns of intra and interpopulation variation was severely constrained by the limitation s of available Pietrusewsky 2000, 2005). Such analyses as were possible often relied on a single measurement or on a single index, such as the cephalic index, which combined a few measurements in relatively simple fashion. Such simplistic analyses, unavoidable as they may have been,
159 necessarily led to simplistic interpretations of patterns of human variation and equally simplistic typological classification rubrics (Howells 1969; Gould 1996; Pietrusewsky 2000, 2005). Not until the development of several essential multivariate statistical methods including analysis of variance (ANOVA) (Fisher 1925), discriminant function analysis (DFA) (Fisher 1936; Rao 1948, 1952), Generalized Distance (D 2 ) (Mahalanobis 1936) in the 1920s to 1940s, and the subsequent advent of increasingly powerful computing systems, could craniometric analyses begin to mature int o the powerful analytical tool we recognize today (Pietrusewsky 2000:377; Howells 1973; Kachigan 1991; Feldesman 1997; Pietrusewsky 2005). As noted previously, multivariate statistical analyses can be descriptive or predictive, depending on the aims of t he analysis, and they have numerous practical and theoretical advantages for the interpretation of craniometric data, regardless of their specific approaches to data analysis (Feldesman 1997:86; Huberty 1984). First, serial pairwise statistical analyses o f suites of craniometric variables in multiple populations are, by their nature, lacking in statistical hypothesis) with each additional pairwise test ( Ott and Longnecker 2001; Field 2009). For example, if one were to compare the mean values for maximum cranial length (GOL) in three t tests for the equality of means, comparing populations 1 and 2, 1 and 3, and 2 and 3, with a level of significance of p = 0.05, the proba bility of making a Type I ( ) error 2 for each test is 5%, ( = 0.05) and the probability of not making a Type I error is therefore 95% (1 = 0.95) Because the tests are independent, the 2 In this example, the null hypothesis would be that the two populations being compared do not have significantly different mean values for maximum cranial length. A Type I error, falsely rejecting the null hypothesis, would entail the conclusion that thes e two populations do have significantly different mean values for maximum cranial length, when in actuality they do not
160 multiplicative rule of probability applies, such that the total probability of not making a Type I error for the three tests is actually 85.7% (0.95 x 0.95 x 0.95 = .857) and the probability of making a Type I error is therefore 14.3% ( = 1 .857 = 0 .143). This is far greater than the initial condition of = 0.05, and is well above even the most liberal accepted levels of significance at p = 0.1 (example adapted from Field 2009:348; Ott and Longnecker 2001). To counter the effects of this compoundi ng error in serial pairwise comparisons, the initial values must be set to a much lower value, creating a much more stringent threshold for concluding statistical significance. More importantly, such serial analyses have the undesired effect of decomposi ng an integrated whole (i.e., a skull and also a discrete individual) into a number of quasi independent parameters, the sum of which is much, much less than the whole. As a result, it is profoundly difficult to gain a true appreciation of the morphologic al character of the skull of a given individual, much less that of a population of individuals (Howells 1969, 1973; Pietrusewsky 2000). As Howells (1969:312) has eloquently stated, in univariate statistical analyses, the individual is sunk. Nor is a pop limitation of univariate statistics: there is no real vector, or profile, representing either individuals or populations. tistics are the statistics of individuals and populations (Pietrusewsky 2000:377; Howells 1969:312; Howells 1973). Multivariate statistical methods achieve this integration of separate measurements into whole individuals and collected populations by permi tting the analysis of many variables simultaneously, while accounting for both the nature and extent of the relationships among the variables (Howells 1969, 1973; Kachigan 1991; Pietrusewsky 2000). This is accomplished by various means depending upon the specific structure of a particular multivariate statistical method, but in general is the result of the reduction of dimensionality in a
161 dataset. To clarify what this means by way of an example: if we consider an individual cranium as a multidimensional object (which it clearly is), with each dimension defined by each one of the 24 craniometric variables in a dataset, then the given cranium can be described by its position in a 24 dimensional multivariate space. In this multivariate space, each craniomet ric variable corresponds to an individual coordinate axis of the space (x axis, y axis, z axis and so on), and the position of a given individual is determined by its measurement values for each of the metric variables, which act as Cartesian coordinates f or the dimensional axes. A population of 4) of positions within a particular region of this same 24 dimensional space, and different populations would occupy differen t regions of that space (Howells 1969, 1973; Zegura 1971; Corruccini 1973; Kachigan 1991; Feldesman 1997; Pietrusewsky 2000). Obviously, 24 dimensional space is exceedingly difficult to comprehend logically, and in fact is nearly equally difficult to manag e computationally. Thus, many multivariate statistical methods reduce this troublesome dimensionality by creating a smaller number of composite or known variously as linear combinations factors or variates based typically on patterns of covariation among the individual variables. When applied to craniometric data, these variates summarize the morphological character (i.e., the size and shape) of an individual, accentuate certain aspects of the patterns of morphological variation within and among populations, and reduce redundancy and noise in the data. This reduction in dimensionality is possible because craniometric measurements overlap somewhat in their information content, simply by virtue of the fact t hat they are all measuring some aspect of the
162 individual craniometr ic variables are intercorrelated For example, there is a high degree of intercorrelation between nasal height (NLB), which is the distance from nasion to nasospinale in the sagittal plane, and upper facial height (UFHT), which is the distance from nasion to prosthion i n the sagittal plane (Figure 4 1 ), as the latter measurement includes the same distance measured by the former. In multivariate analyses, the newly generated variates account and adjust for such intercorrelation, thereby optimizing the amou nt of information contained within. Thus, the craniometric dataset containing 24 measurements (and hence information in 24 dimensions) may be reduced to, say, 5 variates, all of which act as new coordinate axes that allow us to consider the individuals an d populations in a 5 dimensional space. A 5 dimensional space is still rather difficult to grasp, but is certainly much more intuitive and much less unwieldy than a 24 dimensional space (Howells 1969, 1973; Zegura 1971; Kachigan 1991; Feldesman 1997; Piet rusewsky 2000). An essential feature of these variates is that collectively they still capture the majority of the variation within and among the population samples under analysis, despite the reduced dimensionality. Variates generated by multivariate s tatistical analyses are often purposely structured to be uncorrelated or orthogonal with regards to one another. Each orthogonal variate with the first varia te explaining the greatest proportion of the variance, and each subsequent variate explaining a slightly smaller, yet still unique, component of the total variance. Thus, intra and interpopulational patterns of variation within the dataset are preserved, and in many analyses are emphasized. An additional advantage of multivariate statistical methods is that they permit not only the simultaneous analysis of multiple variables, but also the simultaneous analysis and comparison
163 of three or more groups or p opulations at a time. The results of such multigroup analyses are much more reliable and more statistically powerful than are pairwise comparisons of each population in turn. Furthermore, the economization of information contained in the orthogonal varia tes permits the identification of factors that are correlated with specific aspects of biological variation, such as differences in size or in specific shape parameters. All of this in turn allows the exploration and interpretation of complex patterns of morphological variation within and among populations in a manner that is morphologically holistic and therefore biologically meaningful, as well as statistically powerful and more logically intuitive (Kachigan 1991; Feldesman 1997; Pietrusewsky 2000). Fr om a practical standpoint, many multivariate statistical methods are also very robust to (i.e., unaffected by) deviations from normality and from homogeneity of variances among individual variables, to a much greater degree than univariate methods. This i s especially important in physical anthropology analyses, where skeletal samples can often be quite small and statistically irregular. For the current research, the gathered craniometric data from modern and archaeological populations in Thailand will be analyzed using the following methods: Analysis of variance Analysis of variance (ANOVA) is the statistical solution to the problem of compounding error rates that occur when performing serial pairwise statistical tests of three or more groups. It allows t he simultaneous testing of the equality of all group means while maintaining a fixed Type I error rate, as established by the significance level for the test set by the analyst, and is therefore the foundation for nearly all other multivariate statistical methods (Kachigan 1991; Feldesman 1997; Ott and Longnecker 2001; Field 2009). Without delving too deeply into the mathematical basis for ANOVA, suffice it to say that the approach is built on a statistical model that states that the total variance (the sp read or dispersion of data) within a dataset is equal to the
164 sum of the within group variance plus the among group variance. When used as a statistical test to compare populations, ANOVA tests the equality of group means by considering the ratio of among group variance to within group variance; this is known as the F ratio If the among group variance is approximately equal to the within group variance, the F ratio will be close to 1. This result is interpreted as indicating that the groups in the analys is share a common mean and variance, and therefore are not significantly different from one another. However, as the amount of among group variance increases relative to the amount of within group variance, so too does the F ratio, indicating that the gro ups have significantly different means and therefore are truly distinct from one another. Like all multivariate statistical methods, ANOVA is underpinned by the assumptions of multivariate normality and homogeneity of variances (homoscedasticity), but it i s also quite robust to substantial deviations from these conditions if sample sizes are sufficiently large (i.e., greater than 20) and approximately equal (Kachigan 1991; Feldesman 1997; Ott and Longnecker 2001; Field 2009). Discriminant function analysis Discriminant function analysis (DFA) is one of the most widely used multivariate statistical methods in physical anthropology. The broad appeal of DFA stems from its combination of descriptive and predictive capabilities in a single analysis, providing an efficient means both for expressing the patterns of variation among separate populations (or among groups within a population), and for assigning an unknown individual to one of these populations or groups (Zegura 1971; Huberty 1984; Kachigan1991; Feldesm an 1997). Discriminant function analysis was first applied to forensic anthropology identifications by Giles and Elliot in 1962 for purposes of race determination, and in 1963 for purposes of sex determination. The power of discriminant function analysis for exploring population relationships and illuminating subtle patterns of human intra
165 study Cranial Variation in Man (Howells 1973) and subsequent volumes (Howells 1989, 1995). Today, D FA finds continued application in a broad variety of analytical contexts, including questions germane to functional morphology and human evolution (Feldesman 1997), population history and global and regional patterns of human morphological variation (Pietr usewsky 2000, 2005, 2006, 2008a,b, etc; Zegura 1971), and forensic identification (Jantz and Ousley 2005). Discriminant function analysis was initially developed by Fisher (1936) and was enhanced by contributions from Hotelling (1931), Mahalanobis (1936) and Rao (1948, 1952). The central goal of DFA (also known as canonical variates analysis when applied to three or more groups, following Rao 1948, 1952) is to create orthogonal variates, or canonical axes which maximize the amount of among group varianc e and minimize the amount of within group variance for each population or group, to ensure maximal separation and minimal overlap of all groups in an analysis. These canonical axes define a transformed multivariate space in which each dimension or axis ac counts for some proportion of the total variance in the dataset. As was noted above, the first variate or canonical axis explains the greatest proportion of the total variance, with each subsequent variate/axis explaining a lesser portion of the remaining variance. Usually the number of variates or axes generated in a specific DFA is one less than the number of groups in the analysis. Mathematically, the creation of a small number of orthogonal variates or canonical axes achieves the desired reduction in dimensionality and allows the identification of the specific individual variables that are driving the separation between groups. Each canonical axis is magnitude of Kachigan 1991; Pietrusewsky et al. 1992; Feldesman 1997; Pietrusewsky 2000).
166 Once the groups are separated in the multivariate space created by the canonical axes, linear or quadr atic equations the actual discriminant functions themselves can be derived that clusters of individual points representing the different populations or groups. The line or curve described by the discriminant function is usually oriented obliquely to all the canonical axes, so that it captures a summary of all the variation among groups described by the canonical axes (Zegura 1971; Howells 1973; Kachigan 1991). Thes e linear or quadratic discriminant functions are also the means by which an unknown individual is classified as belonging to one of the craniometric values into the disc riminant function and determining which population swarm the unknown individual falls within or is closest to (Howells 1969, 1973; Kachigan 1991; Feldesman 1997; Pietrusewsky 2000). Discriminant functions can be constructed in one of two ways. The first is to include all the variables in a given dataset in the discriminant function, regardless of how informative an individual variable is with regard to the separation and discrimination among populations. A more powerful approach is called stepwise DFA in which only those variables which contribute the most to the separation among populations are included in the equation. Determining which variables make the greatest contributions can be achieved through either of two approaches: forward selection or b ackward elimination (Sokal and Rohlf 1995). In forward selected stepwise DFA, which is the approach used in this dissertation, variables are added into the equation one at a time through a sequence of iterative analyses, until the point of diminishing ret urns is reached where the inclusion of subsequent variables does not significantly contribute to the separation between groups or improve the discriminatory power of the equation. With the addition of each
167 new variable, the contributions of all of the var iables already included in the equation are re evaluated. If the addition of a new variable significantly decreases the contribution of a variable already in the equation below the level of significance, that lesser variable is then removed from the equat ion. Backward elimination stepwise methods take the opposite approach: all variables are considered in the initial analysis, and each subsequent iteration removes from the equation the variable that contributes the least to the discriminatory power of th e discriminant function. Stepwise approaches thus provide further insight into which specific variables are driving the separation between populations, and which variables therefore have the greatest discriminatory power and the greatest potential for fu ture investigation. They also help reduce the redundancy of data in the analysis by avoiding the inclusion of pairs of variables that are highly intercorrelated with one another, such that little additional information is gained by including both variable s (Kachigan 1991; Pietrusewsky 1996, 2000; Field 2009). Stepwise many variables are included in the discriminant function relative to the number of individuals in the sample. In instances of overfitting, the model (i.e., the discriminant function) is characterized by a greater degree of complexity than that found in the data, such that the model ends up explaining noise or error in the dataset due to the intercorr elation of the individual variables, rather than explaining only the underlying pattern of variation in the data (Feldesman 1997; Ott and Longnecker 2001). There are many statistical means of evaluating which variables make the greatest contribution to t he discrimination between populations in a forward selected stepwise analysis. One of the most common approaches, and the method used herein, relies on a statistic known as group variance to total varian ce also
168 for an individual variable or for the discriminant ANOVA, namely that the total variance in a dataset is the sum of the within group variance plus the among the F ratio used in ANOVA (i.e., the ratio of among group variance to within group variance), but in an inverse fashion. Ju st as a larger F ratio indicates that there is greater distinction between groups (the among group variance is significantly greater than the within group group variance is a smaller proporti on of the total variance, such that a greater proportion of the total variance is accounted for by the among unexplained variance there is, and the greater the proportion of the total variance that is explained by a specific variable or by the discriminant function as a whole. Since the general aim of discriminant function analysis is to maximize the among group variation and minimize within group variation to facilitate discrimi nation between the two groups, forward selection minimizes this ratio for each iterative step in building the function. Distance (D 2 ) (Mahalanobis 1936), which is distance between the centroids 3 of two the ratio of between group variance to within group variance in pairwise comparisons of all the populations in an analysis, Mahalanobis distances provide a single, objective, quantitative 3 That is, the center point of the swarm of points representing a population in multivariate space. The centroid does not indeed, will not correspond to a particular individual in that population. Rather it is a multivariate indicator of central tendency, comparable (but not exactly equivalent) to the mean in a univariate analysis.
169 indicator of the similarity or dissimilarity that exists between each of the different populat ions in an analysis (Pietrusewsky 1996:70; Howells 1973; Feldesman 1997; Pietrusewsky 1994, 2000, 2 test (1931) provides a test of significance for D 2 values, facilitating the identification and interpretation of biologically meaningfu l distances among populations. Mahalanobis distance from each gr discriminant function analysis as belonging to the population from which it has the smallest D 2 distance. Measures of significance can be applied to these individual to population centroid gen eralized distances to generate probabilistic statements, in the form of posterior and typicality probabilities, regarding the likelihood that the unknown individual belongs to the population into which they have been classified (Howells 1973; Albrecht 1992 ; Feldesman 1997; Pietrusewsky 2000; Jantz and Owsley 2001). Discriminant function analyses may also incorporate Bayesian statistical approaches in the form of informed prior probabilities which allow the analyst to weight the classification function ac cording to demographic characteristics of the dataset and/or the underlying population from which it was derived. For example, in discriminant analyses for classification by sex, prior probabilities are often set to 0.5 ( equal priors ) for both sexes as a default, but may be adjusted for analyses of strongly sex biased populations to reflect the actual ratio of males to females in the population. Thus, in a population in which males outnumber females at a ratio of 3:1, the prior probabilities could be set to 0.75 for males and 0.25 for females, to accommodate for the probability that an unknown individual from this population is more likely to be a male simply by chance. Similar informed prior probabilities can be applied to discriminant function analyses
170 for purposes of race assessment, to reflect the actual racial/ethnic composition of a given population As with all multivariate statistical methods, DFA carries with it the twin assumptions of multivariate normality and homogeneity of variances, but adds the additional assumption that all populations in the analysis have equal covariance matrices, an assumption which can be tested primarily an issue when using DFA for classification of unknown individuals, as the groups with to more frequent misclassification. However, this is generally not a significant concern unless the covariance matrices are extremely different; unequal covariance matrices between groups can also be accommodated for by using quadratic, rather than linear, discriminant functions (Zegura 1971; Kachigan 1991; Feldesman 1997; Jantz and Ousley 2005; Fie ld 2009). It is additionally fortunate for the present analysis that DFA is relatively robust to departures from normality due to kurtosis, rather than due to skew, which is the case for most of the craniometric variables under consideration here (if they are indeed significantly non normal, as questioned above) (Feldesman 1997:87 90). When using discriminant functions to classify unknown individuals, it is additionally assumed that unknown individual is reasonably derived from one of the reference populat ions from which the discriminant function was generated. Discriminant functions will always assign an unknown individual to one of the groups in the analysis, regardless of whether the individual actually belongs to any of those groups (Zegura 1971; Giles and Elliot 1963; Jantz and Ousley 2005). This is of little concern when using DFA for sex determination, as all individuals are either male or female, but is a persistent problem when the method is used for ancestry
171 determination, since many ancestry gro ups are underrepresented or wholly unrepresented in current databases of normative craniometric data. In this regard, it is incumbent upon the analyst to review not only the classification results, but also the posterior probability and typicality probabi lity as a means of assessing the validity of the classification. Final Considerations The complexity of these multivariate methods and their flexibility for application in a variety of analytical contexts means that each statistical approach presents the researcher with an array of operational parameters that can be adjusted to fine tune the analysis to the specific demands of the research question and/or the dataset at hand. As the present research is intended as an exploratory study of patterns of cran iometric variation in the modern Thai, rather than one which is focused on addressing specific hypotheses, the specific methodological approach to the statistical analysis of this dataset evolved with each subsequent analysis, in concert with the evolving understanding of the structure of that craniometric variation. Thus, it is not possible here to have provided more than the foregoing broad summary of the salient statistical methods employed in this research. Rather, the specific statistical an alytical steps followed in each individual analysis will be outlined in greater detail in the following chapter, within the framework of each specific exploratory analysis. Although this is an unconventional structure for a scientific paper, it is felt th at this will lend greater lucidity to both the structure of the analyses and the interpretation of the results.
172 Table 4 1. Standard cranial measurements used in this study. # Measurement name a Code b # Measurement name Code 1 Maximum cranial length GOL 13 Nasal height NLH 2 Maximum cranial breadth XCB 14 Nasal breadth NLB 3 Bizygomatic breadth ZYB 15 Orbital breadth OBB 4 Basion bregma height BBH 16 Orbital height OBH 5 Cranial base length BNL 17 Biorbital breadth EKB 6 Basion prosthion length BPL 18 Interorbital breadth DKB 7 Maxillo alveolar breadth MAB 19 Frontal chord FRC 8 Maxillo alveolar length MAL 20 Parietal chord PAC 9 Biauricular breadth AUB 21 Occipital chord OCC 10 Upper facial height UFHT 22 Foramen magnum length FOL 11 Minimum frontal breadth WFB 23 Foramen magnum breadth FOB 12 Upper facial breadth UFBR 24 Mastoid height MDH a,b Measurement names and codes are following FORDISC 3.0 (Jantz and Ousley 2005; Moore Jansen et al. 1994)
173 Figure 4 1. Anterior views of the skull, de tailing cranial measurements. Measurement n u mbers correspond to Table 4 1; s ee Appendix A for landmark and measurement definitions.
174 Figure 4 2. Left lateral view of the skull, de tailing cranial measurements. Measurement n umbers corre spond to Table 4 1; s ee Appendix A for landmark and measurement definitions.
175 Figure 4 3. Inferior view of the skull, de tailing cranial measurements. Measurement n um bers correspond to Table 4 1; s ee Appendix A for landmark and measurement definitions.
176 Table 4 2. Summary statistics of modern Thai population samples. Population Males n = Age range (mean) Females n = Age range (mean) Totals Chiang Mai 242 22 95 years 104 27 93 years 346 (65.87 years) (63.74 years) Khon Kaen 30 40 80 years 24 36 94 years 54 (61.47 years) (67.50 years) Naresuan a 20 N/A 9 N/A 29 N/A N/A Totals 292 137 429 a Age data not available for the Naresuan skeletal collection.
177 Figure 4 4 Vernacular sociopolitical regions of Thailand. (Redrawn from original map data obtained from the CIA World FactBook (2010) and used with permission. Regional boundaries following Bunge 1981; Higham and Thosarat 1998 ; Baker and Phongpaichit 2005 ).
178 Table 4 3. Summary of Shapiro Wilk tests of normality for individual craniometric variables. Variable Males Females Variable Males Females GOL Y a Y NLH N(.006) N(.001) XCB N(.003) b Y NLB N(.000) N(.004) ZYB Y Y OBB N(.000) N(.002) BBH Y Y OBH N(.000) N(.001) BNL Y Y EKB N(.013) Y BPL Y Y DKB N(.000) N(.003) MAB Y Y FRC Y(.041) Y MAL Y Y PAC Y Y AUB Y(.020) Y OCC N(.000) Y UFHT Y Y FOL N(.001) N(.001) WFB Y Y FOB N(.000) N(.003) UFBR Y(.021) Y MDH Y(.026) Y a Normally distributed at p 0.05, unless otherwise stated b p values for significant deviations from normality given in parentheses. Table 4 4 Summary of tests of normality for individual craniometric variables. Variable Levene Statistic df1 df2 Sig. Variable Levene Statistic df1 df2 Sig. GOL 3.817 1 424 .051 NLH 2.703 1 421 .101 XCB 1.774 1 406 .184 NLB .150 1 416 .699 ZYB .357 1 424 .550 OBB .014 1 427 .905 BBH .494 1 424 .483 OBH 3.280 1 427 .071 BNL 11.298 1 427 .001 EKB .770 1 427 .381 BPL 1.172 1 168 .281 DKB .308 1 427 .579 MAB .020 1 83 .888 FRC .004 1 425 .952 MAL 2.680 1 163 .104 PAC 1.888 1 423 .170 AUB .410 1 426 .522 OCC .195 1 423 .659 UFHT .005 1 154 .946 FOL 3.245 1 424 .072 WFB .829 1 421 .363 FOB .667 1 425 .415 UFBR 1.518 1 427 .219 MDH 3.141 1 424 .077
179 CHAPTER 5 RESULTS AND DISCUSSI ON The whole of science is nothing more than a refinement of everyday thinking. Albert Einstein Physics and Reality (1936) Numerous multivariate statistical analyses using the methods discussed in Chapter 4 were performed in order to explore patterns of morphological variation between the sexes and among regional subpopulations within the modern Thai population. These analyse s were performed in an iterative fashion in order to ascertain both the analytical parameters and the craniometric variables that provide the greatest illumination of this morphological variation; each of these will be detailed in turn below. As all of th e multivariate statistical methods employed in this study are, at their core, rooted in the differences in mean craniometric dimensions between the sexes and among regional subpopulations, these data are provided in Tables 5 1 and 5 2. Discriminant Functio n Analysis By Sex Discriminant Function 1: All Variables Patterns of sexual dimorphism in the modern Thai were first explored using a linear discriminant function analysis incorporating all 24 craniometric variables and applied to the total sampl e of 42 9 individuals (292 males/ 137 females). However, due to missing measurement data for several variables in a number of individuals, this initial analysis included only 38 males and 32 females who had recorded values for all 24 craniometric variables. Never test for the equality of the covariance matrices for the male and female samples failed to reject the null hypothesis of equal matrices at p = 0.426, indicating that this reduced sample still meets the basic assumptions for the appropriate use of linear discriminant function analysis. This is particularly reassuring in light of the possible non normality of many of the individual craniometric variables, as discussed in Chapter 4, and confirms the likelihood that the Shapiro
180 Wilk test is be ing overly sensitive to small, non significant deviations from univariate normality due to the relatively large total sample size. Because the male and female samples included in this specific analysis are nearly equal in size, prior probabilities were le ft at the default value of 0.5 for each sex. An independent samples t test for the equality of means was calculated for each of the 24 craniometric variables prior to DFA, as a preliminary means of probing which variables differ significantly between mal es and females (Table 5 3). This t test revealed significant differences at p < 0.01 1 between the male and female means for each craniometric variable except maxillo alveolar length, which was significant at p < 0.05. Despite the non homogeneity of varia nces p t test is still significant at p < 0.001 under analytical parameters that do not assume equality of variances. Likewise, independent samples t test s executed on the rank ordered data for each of the variables shown in Table 4 3 to be non normally distributed at p < 0.05 (nominally, XCB, AUB, UFBR, NLH, NLB, OBB, OBH, EKB, DKB, FRC, OCC, FOL, FOB, MDH) also shows sign ificant differences at p < 0.01 betwee n the male and female mean ranks for each of these variables (Table 5 4) Because the results of this rank transformed t test mirror the results of the t test as applied to the unranked data, it can be concluded that t he non normality and heteroscedasticity detecte d in the dataset are likely of an insufficient 1 It may be questioned at this point whether a Bonferron i correction (or similar) is required here to adjust the pairwise t tests are not being used here for hypothesis testing i.e., to establish whether craniometric differences ex ist between males and females as this is already known a priori Nor is it being argued here that these craniometric variables are collectively significantly different between males and females at p < 0.05. Rather these univariate t tests are simply be ing used as a preliminary tool for data exploration, to provide a general preview of which individual craniometric variables may be emphasized by the subsequent DFA to generate a classification function for sex determination. As the results of the t tests are not being considered in a summary fashion for hypothesis testing, the compounding increase in the pairwise error rate is of little concern here, and it is felt that no Bonferroni correction is needed for this or subsequent univariate tests which are u sed for the same purpose of general data exploration.
181 magnitude to have undesired effects on the results of the following parametric multivariate analyses. This also further supports the initial interpretation that the non normality of so many craniometric variables, as indicated by the Shapiro Wilk test, is due to the over sensitivity of this test to large sample sizes and large numbers of ties in the dataset. Last, t he se indication s of significant sexual dimorphis m for all craniometric variables in turn implies that the DFA should be successful in generating an accurate and reliable function for sex determination in unknown individuals. Similarly, alternate test of equality of group means for each craniometric variable (Table 5 5 ). Lambda is rooted in the same statistical model as ANOVA (total variance = among group variance + within group variance), which in turn underpins DFA, the results of this test may be more directly relevant than the t test for evaluating which variables will contribute significantly to group separation under DFA. This analysis revealed signi ficant differences at p < 0.05 between the male and female mean values for each of craniometric variables except orbital height (OBH), interorbital breadth (DKB), occipital chord (OCC) and foramen magnum breadth (FOB). As such, these four variables may be considered to display a low degree of sexual dimorphism and therefore are expected to contribute little to the separation between males and females. Interestingly, three of these four variables (OBB, DKB, FOB) are intimately associated with organs of spe cial sense (the eyes for OBB and olfactory apparatus for DKB) or significant neural structures (the proximal spinal cord and brainstem for FOB). Under the functional matrix hypothesis (Moss 1973, 1997a c) these craniometric dimensions are arguably under s ignificant developmental constraint, and might be less likely to display significant sexual dimorphism as a
182 different between males at females at p < 0.01, indicati ng that these variables are highly sexually dimorphic and should be significant components of the discriminant function. Review of the discriminant function structure matrix (Table 5 6 ), which lists in rank order the pooled within groups correlation bet ween each craniometric variable and the discriminant function, provides the clearest insight into the relative importance of each variable to the separation of male and female clusters in multivariate space (Feldesman 1997:93; Field 2009; SPSS, Inc 2010). As expected, the four variables with non (OBH, OCC, FOB, DKB) also have the lowest structure correlation coefficients, indicating that they provide little discriminatory power to the analysis. Structure coefficients for variables with significant ANOVA results are all positive and range from 0.511 to 0.193. This overall moderate range of correlation coefficients indicates that all these variables are contributing somewhat more or less equally to population separation. In other words, the separation between males and females is based on many moderately dimorphic craniometric variables, rather than only a few highly sexually dimorphic cranial dimensions. Curiously, according to the structure matrix the variable contributi ng the most to the separation between males and females is nasal height (NLH), which is a relatively small, somewhat specific cranial dimension overall. Typically, discriminant functions for sex determination are found to be driven by the larger, more gen eralized dimensions of the cranial vault and facial skeleton that contribute significantly to the overall size differences between males and females (Giles and Elliot 1963). In the current analysis, this can be seen in the relatively high correlations (se cond and third highest overall) for bizygomatic breadth (ZYB) and cranial base length (BNL), as well as the presence of maximum cranial length (GOL), basion
183 bregma height (BBH) and biauricular breadth (AUB) among the ten craniometric variables with the str ongest correlations. Mastoid height (MDH), which is considered a strongly dimorphic non metric trait (Buikstr a and Ubelaker 1994; Walker 2008), also has a relatively strong correlation in this analysis, indicating its significant metric discriminatory pow er between males and females. Because only two groups (males and females) are under consideration, only a single discriminant function was generated in the analysis and accounts for 100% of the variation between males and females. By multiplying the crani al measurements of an unknown individual by their respective coefficients from Table 5 7 and then adding the constant term, the discriminant function score for the unknown individual is obtained. This discriminant function score is then compared to the pr ovided sectioning point of 0.108, which is simply the midpoint between the respective mean discriminant function scores for males and females (Figure 5 1). If is classified as a male; if smaller than the sectioning point, the individual is classified as female. The classification accuracy of Discriminant Function 1 was tested using both resubstitution and leave one out cross validation (Table 5 8 ). Under resu bstitution nearly 90% (34/38) of males and 94% (30/32) of females were classified into their correct sex group, for an overall correct classification rate of 91%. However, estimates of classification accuracy based on resubstitution are artificially high as the same individuals which were used to create the discriminant function initially are also used to test its performance in effect, individuals are being tested against themselves, leading to inflated accuracy rates. In contrast, under leave one out cross validation, one individual is held out of the analysis and the discriminant function is generated from n 1 individuals in the sample; the held out individual is then classified into one
184 of the reference groups using the n 1 discriminant function. Th is process is repeated with each of the n individuals in the dataset, so that the discriminant function is iteratively tested against an Classification accuracy rates ar e almost always lower under leave one out cross validation than under resubstitution, but are a much more reliable indicator of the true predictive accuracy of the discriminant function (Feldesman 1997; Pietrusewsky 2000; Jantz and Ousley 2005). Using lea ve one out cross validation, 74% (28/38) of males and 81% (26/32) of females were correctly classified by Discriminant Function 1, for an overall accuracy rate of 77% (54/70 correctly classified by sex). This accuracy rate is still significantly greater t han chance (0.5), indicating the basic utility of this function for sex determination in modern Thai populations. Though the inclusion of all 24 craniometric variables in this discriminant allows us the opportunity to explore the overall pattern of sexual dimorphism in the cranial dimensions of the modern Thai, this is not necessarily the most powerful analysis for sex determination in this population. Indeed, the 77.1% correct classification rate, although considerably better than chance, is nonetheless s omewhat poor in comparison to the performance of other models developed on other populations. The discriminant functions for sex determination generated by Giles and Elliot (1963:63) for use on American Whites and Blacks, for example, had classification a ccuracy rates of 82 89%, appreciably better than the performance of this Thai specific discriminant function. Likewise, discriminant functions generated by FORDISC 3.0 (Jantz and Ousley 2005) for purposes of sex determination often have classification a ccuracy rates greater than 90%. This suggests that the Thai specific sex determination function can be considerably improved upon.
185 There are several potential causes for the lower classification rate in this Thai specific function. There is, of course, a lways the possibility that the modern Thai population is not strongly sexually dimorphic, and is less so than other populations. If this were the case, a greater degree of craniometric and morphological overlap between males and females in the modern Thai would translate to greater difficulties distinguishing between them using discriminant functions. Given the overall gracility of the Thai population and, indeed, of all Southeast Asian populations in general this is a very real possibility which must be considered in any attempt to develop metric or morphological standards for the determination of sex in unknown skeletal remains from this population. It is much more likely, however, that this poorer correct classification rate is simply due to the struct ure of this initial analysis. As was noted above, many individuals were removed from this DFA due to missing measurement values for one or more of the craniometric variables. This resulted in a reduction in sample size from 429 individuals t o only 70 ind ividuals (38 males /32 females), meaning that 84% of the total sample was excluded from the analysis. There is no doubt that the removal of these individuals also equates to the loss of a significant proportion of the total pattern of variation within the modern Thai sample. As such, it is possible that Discriminant Function 1 would have an even lower classification accuracy over a period of long term application to a number of unknown individuals from forensic contexts. Clearly, this is an undesirable si tuation. Fortunately, both of these concerns can be addressed through various adjustments of the analytical parameters for generating the discriminant function for sex determination. By increasing the number of individuals in the analysis and/or selectin g only those variables with the greatest discriminatory power, it should be possible to generate other discriminant functions with greater classification accuracy rates. However, a failure of these
186 adjustments to produce any significant increase in model performance for the sex determination function may be considered an indicator of a relatively lower degree of sexual dimorphism in the modern Thai population. Discriminant Function 2: 20 variables One way to increase the number of individuals included in the discriminant function analysis is simply to exclude those variables that are missing for large numbers of the individuals in the dataset. In the present study, the majority of the individuals a re missing values for basion prosthion length (BPL), maxillo alveolar length (MAL), maxillo alveolar breadth (MAB) and upper facial height (UFHT) ( see Table 5 1). These measurements all involve osteometric points that are positioned on the alveolar proces ses of the maxilla particularly the landmark prosthion and are therefore strongly affected by antemortem tooth loss and subsequent alveolar resorption, as well as by dental conditions such as periapical abscesses and advancing periodontal disease. Like mo st anatomical collections, the Thai skeletal collections from which the current sample is drawn are composed primarily of older adults who display varying degrees of dental pathology, antemortem tooth loss and accompanying alveolar resorption resulting in a large number of individuals with missing values for one or more of these four variables. By removing BPL, MAL, MAB and UFHT, the sample size is increased from 70 individuals to 385 individuals (262 males/ 123 females), representing approximately 90% of the original dataset. Examination of the structure matrix for Discriminant Function 1 ( see Table 5 6 ) shows that these variables contributed only moderately to the separation between males and females in the first discriminant function analysis. Therefore, the cost of removing these variables, in terms of lost morphological information, is far outweighed by the gain of information in terms of the improved representation of the overall pattern of variation within the Thai population that is provided by an enlarged sample size.
187 trices was performed to ensure the data meet the underlying assumptions of DFA. This test failed to reject the null hypothesis of equal covariance matrices at p = 0.153, indicating that this sa mple conforms to the assumption of equal covariance matrices in addition to the previous established assumptions of normality and homoscedasticity craniometric variables revealed significant differences between the male and female mean values for all 20 measurements inc luded in the analysis (Table 5 9 ), indicating that all variables are contributing to the separation between males and females. Of these, 19 variables were significantly different at p < 0.01; orbital height (OBH) was significant at p < 0.05. Review of the structure matrix (Table 5 10 ) finds that all variables have positive correlation coefficients ranging from 0.655 to 0.136. This range of correlation coefficients extends somewhat higher than the range of coefficients from the first analysis; indeed, six variables (ZYB, NLH, BNL, MDH, AUB, BBH) have correlation coefficients that are greater than the highest correlation in the first analysis (NLH correlation coefficient: 0.511; see Table 5 6 ). This indicates that in this second analysis there is a stronge r relationship between certain variables and the statistical separation between males and females; these variables therefore have a correspondingly greater influence in the discriminant function. This is an important practical consideration, because the more a given discriminant function is driven by a few highly correlated variables, the more susceptible it becomes to classification errors due to inaccurate values for those variables. For example, the true measurement value of bizygomatic breadth is oft en affected by antemortem or perimortem cranial trauma. If an analyst is not cognizant of this distortion and enters an inaccurate value into a discriminant function which is strongly driven by bizygomatic breadth, this small measurement
188 error can have a disproportionately large effect on the classification of an unknown individual, increasing the likelihood of a classification error. However, this is not a significant concern for Discriminant Function 2 due to the overall moderate magnitude of the correl ation coefficients. In general, the suite of variables contributing the most to the separation between males and females in Discriminant Function 2 is nearly identical to that in Discriminant Function 1, with only slight differences in the rank order of th e variables. In Discriminant Function 2, bizygomatic breadth (ZYB), nasal height (NLH), cranial base length (BNL), mastoid height (MDH), biauricular breadth (AUB), basion bregma height (BBH) and maximum cranial length (GOL) have the highest correlation co efficients in the structure matrix. As expected, Discriminant Function 2 is largely driven by measurements of overall cranial vault and facial size, as well as by measurements of the strongly sexually dimorphic mastoid processes. Also as in the first an alysis, interorbital breadth (DKB), occipital chord (OCC) and orbital height (OBH) have the lowest correlation coefficients, reinforcing the interpretation that that these craniometric dimension are minimally sexually dimorphic and therefore contribute lit tle to the discrimination between males and females. The unstandardized canonical coefficients, male and female group means, and sectioning point for Discriminant Function 2 are presented in Table 5 1 1 Under leave one out cross validation, 82.1% (215 / 262) of males and 83.7% (103 / 123) of females were classified into their correct sex group, for an overall correct classification rate of 82.6% (318 / 385 individuals correctly classified by sex) (Table 5 1 2 ; Figure 5 2). This represents a marked increa se over the accuracy rate for Discri minant Function 1, indicating a substantial improvement in model performance when the discriminant function is derived from a larger sample size. The accuracy rate for Discriminant Function 2 also falls within the range of correct sex classification rates for
189 discriminant functions based on American Whites and Blacks reported by Giles and Elliot (1963:63). This generally eliminates the concern that the modern Thai population displays a lower degree of sexual dimorphism than the American White and Black populations. Further, it indicates that the lower correct classification rate seen in Discriminant Function 1 was the result of the restricted pattern of craniometric variation in the small sample from which this function was derived. It also should be noted that these correct classification rates for Discriminant Function 2 were calculated using equal prior probabilities for males and females. However, since the male sample size is twice as large as the female sample siz e, classification rates were recalculated using informed prior probabilities based on sample size, of 0.681 for males and 0.319 for females. This led to a 7.6% increase in the correct classification rates for males, to 89.7%, but also resulted in a 10.3% decrease in the correct classification rates for females, to only 73.2%. Despite a slight increase (1.8%) in overall accuracy rate to 84.4%, the significant reduction in correct classification of females indicates that while the use of informed unequal p riors may be statistically justified, in practice it is unwarranted for this equation. Equal prior probabilities of 0.5 for males and females also much more accurately reflect the percentages of males (49.2%) and females (50.8%) in the modern Thai popula tion ( based on 2010 Thailand government census data, available at http://188.8.131.52/stat/pk/pk53/ pk_53.pdf accessed 25 June 2011), and thus are more appropriate for use when analyzing unknown skeletal remains originating from this population. Discrim inant Function 3: Forward Selected Stepwise Analysis An alternate and generally more powerful approach to DFA is to use a stepwise selection process to generate a discriminant function incorporating only those variables which collectively account for the majority of separation between the groups under analysis. A discriminant
190 function generated through stepwise selection or elimination of variables can be loosely regarded as an optimal solution to the competing analytical demands for the highest possible classification accuracy using the fewest possible variables to account for most of the separation between groups. Therefore, a third discriminant function for sex determination in the modern Thai population was generated via forward selected stepwise analysis selection criterion. In order to maximize the sample size at the outset of the analysis, basion prosthion length (BPL), maxillo alveolar length (MAL), maxillo alveolar breadth (MAB) and upper facial height (UFHT) were again excluded from the analysis due to large numbers of missing values. Thus, the forward selected stepwise analysis started from the same initial conditions as Discriminant Function Analysis 2. A total of eight variables were selected by the forward s tepwise analysis: bizygomatic breadth (ZYB), nasal height (NLH), mastoid height (MDH), cranial base length (BNL), biorbital breadth (EKB), biauricular breadth (AUB), frontal chord (FRC) and minimum frontal breadth (WFB ) (Table 5 1 3 ). This suite of variab les is generally similar to the craniometric dimensions with the highest correlation coefficients for Discriminant Functions 1 and 2 ( see Tables 5 6 and 5 10 ), a level of internal consistency that would seem to confirm the general validity of all the discr iminant functions generated in this study. Interestingly, biorbital breadth, frontal chord and minimum frontal breadth all had relatively low correlations with Discriminant Functions 1 and 2, so their prominence in the stepwise function is intriguing. Th eir inclusion, however, does preserve the general balance between cranial vault and facial dimensions present in the full measurement battery, and is a clear indication that both regions of the cranium are significantly sexually dimorphic. It is additiona lly interesting to note that breadths of the vault and face
191 appear to be more sexually dimorphic than are length measurements, as this equation includes four breadth measurements but only two lengths. This echoes the pattern seen in the postcranial skelet on, where epiphyseal breadths and other measurements in the transverse plane are more diagnostic for sex determination than measurements in the sagittal plane (Stewart 1979; France ; Spradley and Jantz 2011 ). The unstandardized canonical coefficients, male and female group means, and sectioning point for Discriminant Function 3 are presented in Table 5 1 4 Under leave one out cross validation, 82.0% (228 / 278) of males and 84.7% (111 / 131) of females were classified into their correct sex group, for an overall correct classification rate of 82.9% (339 / 409 individuals 2 correctly classified by sex ; Table 5 1 5 Figure 5 3). These correct classification rates for the stepwise selected discriminant function are approximately equa l to those for Discriminant Function 2, which was based on 20 craniometric variables. Although the stepwise selection of craniometric variables did not lead to a significant improvement in model performance, Discriminant Function 3 has the considerable ad vantage of only requiring eight measurements for a high level of accuracy in sex determination of unknown remains. As such, Discriminant Function 3 has greater utility in the analysis of remains from forensic contexts, as it can be readily applied to cran ia in which trauma or postmortem damage preclude measurement of the 20 variables necessary for Discriminant Function 2 but which retain the eight stepwise selected variables 2 The male and female sample sizes in the cross validation sample represent the individuals with recorded measurements for all eight stepwise selected variables, hence the difference here from the ini tial analytical sample size of 385 individuals 20 of 24 cranial measurements.
192 Discriminant Function Analysis By Regional Subpopulation Intrapopulation var iation in the modern Thai was also explored using linear discriminant function analysis, in order to assess the nature and extent of any craniometric differences among the three regional subpopulations from Bangkok (Naresuan University collection), Chiang Mai (Chiang Mai University collection) and Khon Kaen (Khon Kaen University collection). This exploration was undertaken on the premise that a ny significant differences that exist among these regional subpopulations may be reflective of semi unique populat ion histories, traceable to relative differences in the degree of genetic influence of various prehistoric and historic non Thai populations, such as the Mon or Khmer. As with the discriminant function analyses performed for purposes of sex determination, patterns of intrapopulation variation were addressed beginning with broad approaches including the maximum possible number of variables and progressing to more focused more powerful approaches relying on forward stepwise selection of the most diagnostic variables. Though it might be more efficient, and certainly more expedient, to proceed directly with the stepwise ana lysis, it is felt that there are nevertheless some important insights into patterns of morphological variation that are revealed by the no n stepwise approaches. From the outset, analyses of intrapopulation variation in the present sample we re somewhat more challenging than analyses of sexual dimorphism. Because of the obvious and demonstrated craniometric differences between males and fema les, discriminant function analyses of regional subpopulations were separated by sex. The sample sizes for each of the regional subpopulations are already rather unequal ( see Table 4 2), with the Chiang Mai sample roughly six times larger than the Khon Kaen sample and 12 times larger than the sample from Naresuan. However, when the data are parsed by sex the differences in sample size become even more dramatic and some, such as the Nare suan females (n = 9), drop well below reasonable
193 lower limits for sample size in DFA (Feldesman 1997). These small, markedly unequal sample sizes cause further problems for meeting the underlying assumptions of DFA, including univariate and multivariate n ormality of the craniometric variables and the homogeneity of variances and of covariance matrices. Failure of the data to meet these analyses causes problems throughout all phases of data analysis but is particularly problematic for formal tests of signi ficance, which can become somewhat unpredictable and decidedly unreliable as a result. Though data transformations and removing individuals from larger groups to make sample sizes more equal can remedy some of these issues, they also result in a certain l oss of information and statistical power, and may further complicate interpretation of results (Kachigan 1991; Feldesman 1997; Field 2009). Despite the likelihood that these critical assumptions are untenable in this instance, it is still a worthwhile in itial exercise to see how the data behave as a whole under DFA, as long as the interpretation of the results does not overreach the limits of the analysis under these non i deal conditions. This initial exercise also forms an important baseline for evaluat ing how later adjustments to the dataset have affected the outcomes of the analyses. The three regional subpopulations were first assessed for univariate normality of the individual craniometric variables using the Shapiro Wilk test ( see Table 5 2). Non normally distributed variables for the Chiang Mai male sample includes XCB, AUB, UFBR, NLH, NLB, OBB, OBH, EKB, DKB, OCC, FOL, FOB and MDH; non normally distributed variables for the Chiang Mai female sample include NLH, NLB, OBB, OBH, EKB, DKB, FOL and FO B. This pattern of non normality matches almost exactly that for the pooled male and female samples discussed in Chapter 4 (see Table 4 3) As Chiang Mai males and females are the largest of the six population*sex samples, many of the same statistical in fluences on the reliability of the Shapiro
194 Wilk test discussed in Chapter 4 obtain in this analysis as well. These influences include large sample sizes and large numbers of ties within the dataset leading to oversensitivity of the Shapiro Wilk test to on ly slight deviations from univariate norm ality. This oversensitivity is further suggested by the fact that far fewer variables display significant deviations from normality in the much smaller Naresuan and Khon Kaen male and female samples. Non normalit y for MAB in the Khon Kaen sample is certainly the result of only five individuals (3 males, 2 females) having recorded values for this variable; this is of little concern as this variable will be excluded from later analyses due to a high number of indivi duals with missing measurement values for this variable overall. Other non normalities, such as foramen magnum length (FOL) in the Khon Kaen male sample and biauricular breadth (AUB) in the Khon Kaen female sample, are due to flat (platykurtic) distributi ons with a high number of ties. The remaining deviations from normality are due primarily to skewed distributions. In general, non normality is of greater concern when DFA is being used for classification purposes (again, a greater issue where formal tes ts of significance are concerned), than when it is used merely for exploratory purposes, as here (Feldesman 1997). As such, it is believed that these suspected non normalities do not represent an insurmountable impediment to further analysis The regiona l Thai subpopulation samples were also tested for homogeneity of variances was noted above, this method is appropriate for use with this dataset as it is relatively insensitive both to differences in sample size and deviations from normality (Ott and Longnecker 2001). In males, homogeneity of variances was confirmed for all variables except nasal breadth (NLB) and p = 0.031 and p = 0.025, respectively. In
195 females, homogeneity of variances among the regional subpopulations was confirmed for all p = 0.029. Prior to DFA, a one way ANOVA was also calc ulated for each of the 24 craniometric variables, as a preliminary means of exploring which variables differ significantly among the regional Thai subpopulations (Table 5 1 6 ). As with the t tests comparing males and females above the se individual ANOVAs are not being considered in a summary fashion for purposes of hypothesis testing, and it is felt that a Bonferroni correction for the pairwise error rates is not needed in this application T h ese ANOVA s revealed significant differences at p < 0.05 among group means only for cranial base length (BNL), basion prosthion length (BPL), minimum frontal breadth (WFB) and foramen magnum breadth (FOB). With the exception of foramen magnum breadth (FOB), these variables meet all assumptions of normality and homog eneity of variance, so the corresponding significant ANOVA results are generally considered valid and reliable. However, given the disparity in sample sizes for the different populations, these significant differences may not actually be meaningful in bio logical terms or even within the statistical context of the DFA. In fact, the overwhelming number of non significant ANOVA results, as well as their magnitude (with p values as high as 0.899), suggest that there is little to no meaningful craniometric dif ferentiation among the three regional subpopulations. This is reinforced by the ANOVA results for female groups, in which none of the craniometric variables were significantly different amo ng the regional subpopulations ( p = 0.200 to 0.955 ; full data not shown ). Unfortunately, ANOVA is one of the few parametric tests that cannot be readily applied to rank ordered data, as the F ratio at the core of ANOVA lacks robustness when calculated on data of this type (Conover and Iman 1981:128). Therefore a rank transformed ANOVA cannot be
196 executed here to further probe the potential impact of the violations of normality and heteroscedasticity on the outcomes of the subsequent multivariate analyses, as with the rank transformed t tests described above. Instead, the Kruskal Wallis test, a non parametric procedure that is roughly homologous with, but less powerful than, ANOVA (Field 2009), was applied to the rank ordered data for those variables shown in Table 5 2 to be significantly non normally distributed. The results of the Kruskal Wallis test mirror those of the ANOVA, with the test failing to reject the null hypothesis of no differences among the three Thai subpopulations, at significance values ranging from 0.1 to 0.864 for males and 0.85 to 0.904 for female s This similarity of results between the parametric and nonparametric tests suggests that the results of the ANOVA are generally valid and reliable in spite of any underlying violations of normality and/or heteroscedasticity Furthermore, since violations of these assumptions in non equal sample sizes typically reduce the statistical power 3 of ANOVA and bias the test towards a larger Type I error rate meaning that the critical p value would have to be set lower than the standard p < 0.05 t o maintain the stringency of the test the profound non significance of the majority of the ANOVA results suggests that we need exercise little concern over the possibility that the test is falsely indicating significant differences where none actually exis ts. In spite of this discouraging outlook, discriminant function analyses were performed for the multiple purposes of comparison with the ANOVA results and gaining the opportunity to examine the relationships of the population clusters in multivariate sp ace. The first DFA was conducted using 20 craniometric variables, with basion prosthion length (BPL), maxillo alveolar breadth (MAB), maxillo alveolar length (MAL) and upper facial height (UFHT) excluded, as before, due to large numbers of individuals wit h missing values for one or more of these four 3
197 variables. In males, this led to effective sample sizes of n = 216 from Chiang Mai, n = 29 from Khon Kaen, and n = 17 from Naresuan (total n = 262); in females, the working sample sizes were n = 93 (Chiang Ma i), n = 23 (Khon Kaen), and n = 7 (Naresuan), for a total sample size of n = 123. Although the small sample sizes for Khon Kaen and especially Naresuan compared to the number of variables in the analysis means that there is a very high risk of overfitting the function in this analysis, this concern will be addressed by a subsequent forward selected stepwise DFA. test for the homogeneity of covariance matrices among the three regional subpopulations. the test fails to reject the null hypothesis of equal matrices at p = 0.176; for the female Chiang Mai and Khon Kaen samples, the nul l hypothesis is rejected at only p = 0.048, as the covariances for the Chiang Mai sample are somewhat greater than those for the Khon Kaen sample. Field (2009:604) and Tabachnick and Fidell (2007) indicate that in analyses where the differences in sample size and covariances are concordant (i.e., larger samples have larger covariances), the probability values for multivariate tests of significance will be more conservative and therefore less likely to indicate significance when this is not the case. When examining patterns of intrapopulation variation, falsely concluding the existence of significant distinctions among subpopulations when no such differences exist would seem to be a more serious error than falsely concluding the opposite. This conservativi sm as a result of the unequal covariances is therefore deemed reasonable and tolerable with regards to the aims of present analysis. In any case, the results of the initial DFA clearly and consistently point to the absence of any significant differences
198 Lambda to test the equality of group means demonstrates that the only variables to display significant differences among the three subpopulations are basion nasion length (BNL) among male s ( p = 0.033), and frontal chord (FRC) among females ( p = 0.038). These results largely echo those of the ANOVA, both in the overwhelming non significance of the results as a whole, and in the magnitude of that non significance, with p values as high as p = 0.818 among males and p = 1.0 in females. In addition to its utility in assessing the discriminatory power of individual variables, a value can also be calculated as a test of significance for the discriminatory power of the canonical var significant at p = .262, indicating that the two discriminant functions 4 generated here do not achieve any meaningful differentiation of the three subpopulations in the analysis. For mbda values for the discriminant functions generated above for sex determination were all significant at p < 0.001 a clear indication of the strong discriminatory power of these functions. Further review of the structure matrices for the two discrimina nt functions finds only low to moderate correlations between the individual craniometric variables and the discriminant function; these range from 0.418 to 0.435 for males and 0.383 to .0481 for females (Table 5 1 7 ). These low correlations are an additional indication that the subpopulations are not well differentiated by any variables on either canonical axis This is confirmed by examination the canonical scatter plots for males and females (Figures 5 4 and 5 5 ), which show only a slight separat ion of group centroids along the canonical axes and a near total overlap in scatter among the three subpopulations. Lastly (and thoroughly unsurprising at this point), the results of leave one out cross validation show that the overall correct classificat ion rate for these discriminant 4 As three groups are included in this analysis, two discriminant f unctions are needed to separate them on horizontal (Function 1) and vertical (Function 2) axes.
199 functions is 45.4% for males and 46.3% for females, both of which are only slightly greater than chance (0.33 ) (Tables 5 1 8 and 5 1 9 ) An attempt at generating a more powerful discriminant function using forward selection stepwise analysis, again Although the analysis did generate functions which achieved statistically significant separation among the three regional Thai subpopulations, these functions ar e of little practical utility as they each only contain a single variable. For males, the stepwise analysis selected basion nasion length (BNL) as the discriminating variable. Selection of this measurement is unsurprising, given that it was identified as differing significantly among the three subpopulations by both the initial females, the stepwise analysis selected frontal chord (FRC) as the discriminating variable, which among the subpopulations. Both of these variables have confirmed normal distributions and homogeneity of variances among the three subpopulations, and both the male and female stepwise discriminant function analyses had non equal population covariances. Therefore, all the underlying assumptions of DFA are met in this case, and the outcomes of the tests of significance can be taken at face value, that these variables truly do differ significantly among the regional subpopulations. However, statistical significance does not necessarily equate to biological significance. In discriminant functions for sex determination, where the statistical separation of populations by sex can be traced to real, biologically meaningful differences in absolute size and relative proportions between males and females. Likewise, in discriminant functions analyses for race determination, the separation among different ancestry groups is a function of subtle yet specific
200 patterns of shape (and to a lesser degree, size) differences that persist over large numbers of subpopulations across broad geographic areas, and even across broad time interv als. These patterns were established through a variety of evolutionary mechanisms (e.g., genetic drift, founder effects, selective pressure s ) and can be traced back to the distinct population histories of different ancestry groups since at least the termi nal Pleistocene or earliest Holocene (Howells 1973, 1989, 1995, 1997; Pietrusewsky 2000; Roseman 2004; Jantz and Ousley 2005). In contrast to these established patterns of sexual dimorphism and racial variation, there is no readily apparent, biologically meaningful explanation as to why two measurements, cranial base length in males and frontal chord in females and these two measu rements alone should differ significantly among regional subpopulations of the modern Thai. The population differences in these single craniometric dimensions cannot be related to general patterns of broad morphological variation among the populations of Southeast Asia, or even of Asia as a whole. And there is the additional consideration that while the complexity of craniometric variation among modern human populations is ultimately a function of our complex evolutionary history, it is also more proximat ely a function of the complexity of the polygenic basis of craniometric traits in general. By the very nature of their polygenic inheritance, individual craniometric dimens ions do not exist in a vacuum, and the meaningful shape and size differences betwee n sexes and among ancestry groups are driven by coherent patterns of variation across numerous craniometric variables, believed to be underpinned by coherent (if poorly understood) patterns of variation in numerous genes (Howells 1973, 1989, 1995, 1997; Pi etrusewsky 1994, 1999, 2000; Jantz and Ousley 2005; Matsumura 2006). Therefore, if we are using patterns of craniometric variation as a surrogate for patterns of genetic variation, we should not expect meaningful differences between populations to be reve aled in terms of a single craniometric variable.
201 Though a single measurement can be highly informative in some circumstances (e.g., use of the humeral head diameter for purposes of sex determination), in an analysis of this scope, seeking meaningful patte rns of intrapopulation variation to illuminate broader questions of population history, a single craniometric variable has so little real information content as to be practically useless, regardless of its statistical significance 5 In this case, the for ward selected stepwise discriminant functions are more informative for the variables that they do not i nclude than they are for the single variables that they do select. As has been noted stepwise DFA is generally considered the most powerful approach to developing functions which create meaningful separation between populations in an analysis (Kachigan 1991; Pietrusewsky 2000 ; Jantz and Ousley 2005 ). The fact that even this approach was unable to detect a suite of variables with which to differentiate, in any meaningful way, among regional subpopulations of the modern Thai speaks more clearly than any of the prior analyses to the overall similarity of these subpopulations. As if to underscore this, the leave one out cross validation of the stepwise disc riminant functions had total correct classification rates among the three regional subpopulations of only 25 .0 % for males and 24.8% for females these rates, obvious ly, are substantially lower than correct classifications due simply to chance, and reiterate the absence of any meaningful separation between the three regional Thai subpopulations. Collectively, all of these results for the various aspects of discriminant function analysis are so resoundingly unequivocal as to leave little doubt to the conclusio n that there is an overall lack of differentiation among the regional Thai subpopulations even in light of the concerns about the overall normality and homo scedasticity of the dataset and their impacts on the analysis. However, a single analysis, no matt er how convincing, generally does not a robust conclusion 5 And this, of course, is to say nothing of the scientific and methodological unsoundness of purposefully calculating a discriminant function to separate popu lations on the basis of a single variable.
202 make. In order to address the lingering possibility that the compounded effects of the unequal sample sizes and various suspected and confirmed violations of the underlying assumptions of DFA have rendered the above analyses totally invalid and unreliable and in spite of the rank transformations and non parametric tests which suggest this is not the case a final approach to the discriminant function analysis of intrapopulation variation in the moder n Thai was undertaken. Because the majority of these structural concerns stem from the extremely disproportionate size of the Chiang Mai sample relative to the Khon Kaen and Naresuan samples, the former was parsed into smaller datasets comparable in size to the se latter, and the forward selected stepwise discriminant function analysis was rerun using these smaller datasets. Only the stepwise DFA approach was used in order to eliminate the probability of overfitting the function to these small sample size s. Additionally, due to the unmanageably small size of the Naresuan female sample, this final analytical approach was applied only to the male samples. This procedure is somewhat similar in principle, though certainly not in scale, to the much more robu st bootstrap methods of resampling. In bootstrapping, statistics such as the means and standard deviations of individual variables are calculated on a large number of random subsamples drawn from within the total sample in an iterative fashion; depending on the dataset and the variables, a typical bootstrap might involve 1000 to 10,000 sampling iterations. The means and standard deviations (and their sampling distributions) as calculated over such large numbers of bootstrapped samples 1) can be used to re solve issues of unequal sample sizes and underlying non normalities and heteroscedasticities; and 2) have been shown to be highly accurate estimates of the true population parameters of central tendency and dispersion, from which formal tests of significan ce can be calculated (Sokal and Rohlf 1995; Field 2009). However, bootstrapping can be computationally and interpretively demanding, and with regards
203 to the present study, it is reserved as an avenue for future expansion of the analysis of intrapopulation variation within the modern Thai. For this final analysis, the Chiang Mai male sample was culled to create four datasets of 30 individuals each; these were designated Culled Samples (CS) 1 through 4. Individuals were sorted by accession number and those with missing measurement values were deleted from the initial dataset, leaving 216 individuals in the sample. These 216 individuals represent the same set of individuals used in all of the above analyses once the BPL, MAB, MAL and UFHT variables were excl uded. From this pool, individuals were selected for inclusion in the culled samples using a random number table (provided by Ott and Longnecker 2001) to pick 120 numbers between 1 and 216. The 120 i ndividuals in the culled samples represent 56% of the or iginal dataset, which should reduce sampling bias concerns, and should also allow any patterns observed across the analyses of these reduced datasets to be reasonably generalized to the modern Thai dataset as a whole. Unfortunately, the culled samples di d not fully resolve the issues of non normality and heterogeneity of variances. The non normal distributions of maximum cranial breadth (XCB) and orbital height (OBH) previously observed in the Naresuan sample and noted above ( see Table 5 2) persist in th is analysis, as this sample has not changed. Additionally, at least one and as many as four non normally distributed variables were revealed in the culled Chiang Mai samples by the Shapiro Wilk test (Table 5 20 ). These variables differed by culled dataset, and all but maximum cranial length (GOL; CS 3) and basion bregma height (BBH; CS 4) had been previously noted for the complete Chiang Mai dataset. The novel non normality of these two variables is likely due t o sampling error introduced during the creation of the culled samples.
204 foramen magnum breadth (FOB), and occipital chord (OCC) in CS 2, CS 3, and CS 4 respectivel y. The heterogeneous variances for PAC had been previously observed in the original dataset, but the varying pattern of heteroscedasticity across the four culled samples is again likely due to sampling error. Despite this, the collective results of the st epwise DFAs of the four culled samples are generally in accordance with the analyses of the modern Thai dataset in its entirety. As in all previous analyses, cranial base length (BNL) was found to differ significantly among the regional Thai subpopulation s in each of the four culled samples, based on the calculation of ality of group means ( see Table 5 20 ). This test identified additional significant differences among the three subpopulations in bizygomatic breadth (ZYB) in CS 2, and minimum frontal breadth (WFB) and occipital chord (OCC) in CS 4. Of these, however, only minimum frontal breadth had been previously identified (by a one way ANOVA applied to the entire dataset; see Table 5 1 6 ) as differing signific antly amo ng the subpopulations. These results are echoed in the variables selected by the forward stepwise analysis for inclusion in the discriminant functions. Cranial base length (BNL) was selected in all four stepwise analyses, while the functions generated fo r CS 4 also included minimum frontal breadth (WFB) and occipital chord (OCC). Curiously, the stepwise analysis of CS 2 also selected interorbital breadth (DKB), rather than bizygomatic breadth (ZYB), for inclusion in the functions, even though DKB was not populations (at p = 0.2 ). Review of the scatter plot on the canonical variates for the CS 2 functions (not shown) reveals that the individual data points are aligned in regular rows s uggesting there is an underlying issue with the data, such as too small a sample size
205 The discriminant functions generated for each of the culled datasets all have significant indicating that the stepwise discriminant functions do differentiate amongst the three regional subpopulations to a statistically significant degree. Raised again, however, is the concern over the distinction between statistical significance and biologica l meaning. As in the stepwise analysis of the entire Thai dataset, the discriminant functions for CS 1 and CS 3 contain only a single variable, cranial base length (BNL), which is, as noted above, an inadequate basis for arguments for a meaningful pattern of morphological variation and differentiation among the regional Thai subpopulations correlated with aspects of population history. If there were to be biologically meaningful differences among these three regional subpopulations, related to local diff erences in the genetic influence of various prehistoric and historic non Thai populations, we should expect a pattern of variation across several cranial dimensions simultaneously (Howells 1973, 1989, 1995; Pietrusewsky 2000, 2006; Matsumura 2006). Of part icular yet cautious interest, therefore, were the 3 variable discriminant functions p < 0.001, appear from a purely statistical standpoint to have discriminatory power comparable to that of the sex determination functions discussed above. Additionally this model has a 53.2% correct classificat ion rate under cross validation; this is somewhat greater than chance and is the highest correct classification rate seen in any of the DFA s addre ssing variation among the regional subpopulation s At this point, then, it would be an easy thing to get carried away by the seeming potential of this model, as it appears superficially to have several analytical strengths. Aside from being the mos t substantial stepwise function generated in these intrapopulation analyses, the three variables contained in this model (minimum frontal breadth, occipital chord and cranial
206 base length) could be argued to approximate a reasonable minimum selection of crani al vault dimensions (i.e., breadth, height, length) with which to capture something of the essential shape and/or size differences which might exist among populations. This could in turn suggest that th e population separation generated by this model may b e biologically meaningful as well as statistically significant. The discriminant functions appear to achieve slight separation of Chiang Mai and Naresuan from Khon Kaen on the first canonical axis and Chiang Mai and Khon Kaen from Naresuan on the second a xis (Figure 5 6 ). Examination of the function coefficients indicates that higher scores on the first axis are driven by large measurements for WFB and small measurements for OCC, while higher scores on the second axis are driven by large measurements for BNL and small measurements for WFB (Table 5 2 1 ). However, that it is clearly of little real analytical value or power. First, there is marked overlap among the scatters for eac h of the regional subpopulations, a strong sign that this model is not as powerful in achieving separation among the regional subpopulations as its statistical significance would imply In order to verify the actual discriminatory potential of this model, a discriminant function analysis using only minimum frontal breadth (WFB) occipital chord (OCC) and cranial base length (BNL) was executed on the entire modern Thai male sample (n = 284, after removing individuals with missing values for these three vari ables ). The covariance matrices for p = 0.863); subpopulations ( p = 0.022). However, leave one out cross validation reveals an overall correct classification rate of only 37% only marginally better than chance (0.33). Thus, despite the initial indications of statistical robusticity and the faintest possibility of biol ogical validity, this
207 discriminant function model does not appear to have any real utility for distinguishing among regional subpopulations when applied to the actual modern Thai sample as a whole. Viewed collectively and at a distance, these discriminant function analyses of the regional Thai subpopulations appear rather unpredictable and inconsistent in their results, with different suites of variables emphasized by the discriminant functions in each analysis (see Tables 5 1 7 and 5 20 ), wide fluctuations in the correct classification rates for each of the subpopulations (see Tables 5 1 8 to 5 20 and 5 2 2 ), and a frustrating juxtaposition between the high statistical significance of some of the functions and their overall poor performance in classifying indi viduals into their correct subpopulation. These observations give the impression that, as its analytical power severely undermined by the conflict betwe design to mathematically identify and emphasize morphological distinctions among population samples and its application, in this instance, to a set of subpopulations where no such distinctions actually exist. If there were a true pattern of mean ingful differences among these regional Thai sub populations, we should expect it to have emerged over the course of these several analytical approaches. That it does not, that no larger pattern of population segregation can be detected acro ss these multiple analyses save for the repeated identification of a single variable, cranial base length, that varies significantly among the subpopulations (but is of dubious physical anthropological significance) serves as something of a final confirmat ion of the craniometric homogeneity of the modern Thai population as represented in the current dataset.
208 Table 5 1. Means, standard deviations, and number of missing cases, for 24 cranial measurements for the modern Thai sample, pooled by sex. Males (n = 292) Females (n = 137) Variable Mean s.d. # Missing Mean s.d. # Missing GOL 171.77 6.71 3 165.15 5. 60 0 XCB 144.51 5.31 13 140.72 5.54 8 ZYB 133.56 4.94 1 126.85 4.67 2 BBH 138.26 4.8 5 3 133.12 5.1 8 0 BNL 99.64 4.56 0 94.52 3.4 4 0 BPL 97.87 4. 70 183 94.39 5.4 3 76 MAB 66.29 3.25 244 63.24 3.34 100 MAL 54.10 2.94 187 52.82 3.69 77 AUB 126.56 5.00 1 121.09 4.67 0 UFHT 72.47 4.13 191 68.31 4.05 82 WFB 93.12 4.67 4 90.73 4.35 2 UFBR 105.36 4.05 0 101.91 4.33 0 NLH 53.02 3.07 5 49.20 2.71 1 NLB 27.19 1.92 9 26.27 1.81 2 OBB 40.19 1.79 0 38.80 1.81 0 OBH 34.53 2.00 0 33.98 1.73 0 EKB 98.57 3.80 0 95.64 3.96 0 DKB 21.28 2.24 0 20.57 2.29 0 FRC 112.59 4.42 2 108.33 4.26 0 PAC 108.35 5.85 4 106.18 6.33 0 OCC 97.61 5.96 4 95.55 5.96 0 FOL 34.89 2.30 1 33.41 1.99 2 FOB 29.89 2.05 0 28.73 1.93 2 MDH 28.41 3.24 2 24.84 2.84 1
209 Table 5 2. Means and standard deviations for 24 cranial measurements for the modern Thai sample by sex and population Chiang Mai (n = 242 M / 104 F) Khon Kaen (n = 30 M / 24 F) Naresuan (n = 20 M / 9 F) Variable M ales s.d. Females s.d. M ales s.d. Females s.d. M ales s.d. Females s.d. GOL 171.75 6.91 164.99 5.82 172.57 4.98 165.54 5.01 170.72 6.74 165.89 4.76 XCB 144.32 a 5.47 140.52 5.84 145.73 3.93 140.92 4.49 144.88 5.15 142.63 4.84 ZYB 133.40 4.87 126.95 4.47 135.43 5.06 126.54 4.96 132.58 5.23 126.43 7.02 BBH 138.36 4.84 133.11 5.28 138.83 4.04 132.50 4.43 136.06 5.78 134.89 5.99 BNL 99.52 4.47 94.43 3.49 101.50 4.41 95.21 3.01 98.25 5.29 93.67 3.94 BPL 98.22 4.62 94.48 5.47 98.90 4.68 94.00 6.60 93.70 3.59 93.75 4.79 MAB 66.22 3.37 63.41 3.35 67.33 1.16 64.50 2.12 66.25 3.40 60.67 3.51 MAL 54.28 2.86 52.89 3.82 54.71 3.35 53.50 2.65 52.27 2.97 50.67 1.53 AUB 126.51 5.05 121.33 4.77 127.67 4.47 120.17 3.88 125.37 5.04 120.78 5.61 UFHT 72.79 4.21 68.54 4.01 71.33 2.73 68.33 2.08 70.64 3.83 65.50 5.45 WFB 92.84 4.63 90.74 4.10 95.07 4.55 91.08 5.36 93.55 4.82 89.78 4.49 UFBR 105.19 3.97 101.88 4.16 106.73 4.01 101.88 4.86 105.35 4.87 102.44 5.34 NLH 53.00 3.05 49.09 2.68 53.27 3.01 49.33 2.55 52.95 3.53 50.11 3.55 NLB 27.14 1.82 26.25 1.79 27.55 2.15 26.29 1.63 27.35 2.64 26.56 2.60 OBB 40.15 1.80 38.91 1.68 40.67 1.61 38.21 1.53 39.95 1.88 39.11 3.41 OBH 34.55 1.98 33.98 1.77 34.37 2.03 34.04 1.78 34.45 2.31 33.78 1.20 EKB 98.40 3.68 95.77 3.87 99.83 3.72 95.00 3.62 98.60 5.00 95.78 5.89 DKB 21.22 2.21 20.56 2.31 21.40 2.31 20.96 2.27 21.85 2.52 19.67 2.00 FRC 112.43 4.52 108.33 4.18 113.73 3.49 107.38 4.28 112.79 4.38 110.89 4.60 PAC 108.34 6.01 105.86 6.59 109.69 3.85 107.12 5.34 107.32 5.05 107.33 5.81 OCC 97.76 5.90 95.69 6.04 95.97 5.48 94.62 6.45 98.32 7.16 96.44 3.36 FOL 34.79 2.28 33.30 1.97 35.50 2.15 33.75 1.89 35.25 2.77 33.78 2.54 FOB 29.76 1.98 28.72 2.02 30.70 2.34 28.92 1.56 30.25 2.20 28.44 1.81 MDH 28.48 3.22 24.92 2.94 28.53 3.59 24.61 2.41 27.40 2.93 24.44 2.88 a Red values indicate the variable fails the Shapiro Wilk test for normality at p < 0.05
210 Table 5 3. Independent samples t tests for equality of group means, by sex, for 24 craniometric variables. Levene's Test t test for Equality of Means F Sig. t Degrees of Freedom (df) Sig. (2 tailed) Mean Difference Std. Error Difference 95% Confidence Interval (C.I.) of the Difference Lower Upper GOL Equal variances assumed 3.817 .051 10.011 424 .000 6.622 .662 5.322 7.922 XCB Equal variances assumed 1.774 .184 6.601 406 .000 3.784 .573 2.657 4.911 ZYB Equal variances assumed .357 .550 13.258 424 .000 6.708 .506 5.714 7.703 BBH Equal variances assumed .494 .483 10.012 424 .000 5.146 .514 4.136 6.157 BNL Equal variances assumed 11.298 .001 11.674 427 .000 5.122 .439 4.260 5.985 Equal variances not assumed 12.906 344.034 .000 5.122 .397 4.342 5.903 BPL Equal variances assumed 1.172 .281 4.375 168 .000 3.478 .795 1.909 5.048 MAB Equal variances assumed .020 .888 4.239 83 .000 3.048 .719 1.618 4.479 MAL Equal variances assumed 2.680 .104 2.446 163 .015 1.279 .523 .247 2.311 AUB Equal variances assumed .410 .522 10.777 426 .000 5.469 .507 4.472 6.467 UFHT Equal variances assumed .005 .946 6.042 154 .000 4.156 .688 2.797 5.515 WFB Equal variances assumed .829 .363 5.005 421 .000 2.385 .476 1.448 3.321 UFBR Equal variances assumed 1.518 .219 8.026 427 .000 3.444 .429 2.600 4.287 NLH Equal variances assumed 2.703 .101 12.423 421 .000 3.826 .308 3.221 4.431 NLB Equal variances assumed .150 .699 4.663 416 .000 .920 .197 .532 1.308 OBB Equal variances assumed .014 .905 7.436 427 .000 1.385 .186 1.019 1.752 OBH Equal variances assumed 3.280 .071 2.759 427 .006 .549 .199 .158 .941 EKB Equal variances assumed .770 .381 7.348 427 .000 2.930 .399 2.146 3.714 DKB Equal variances assumed .308 .579 3.063 427 .002 .715 .233 .256 1.174 FRC Equal variances assumed .004 .952 9.391 425 .000 4.258 .453 3.367 5.149 PAC Equal variances assumed 2.225 .137 3.608 422 .000 2.232 .619 1.016 3.449 OCC Equal variances assumed .195 .659 3.325 423 .001 2.056 .618 .841 3.272
211 Table 5 3 Continued. Levene's Test t test for Equality of Means F Sig. t df Sig. (2 tailed) Mean Difference Std. Error Difference 95% C.I. of the Difference Lower Upper FOL Equal variances assumed 3.245 .072 6.423 424 .000 1.479 .230 1.026 1.931 FOB Equal variances assumed .667 .415 5.507 425 .000 1.154 .209 .742 1.565 MDH Equal variances assumed 3.141 .077 11.028 424 .000 3.572 .324 2.935 4.209 Table 5 4. Rank transformed i ndependent samples t tests for equality of group means, by sex, for 14 craniometric variables. Levene's Test t test for Equality of Means F Sig. t df Sig. (2 tailed) Mean Difference Std. Error Difference 95% C.I. of the Difference Lower Upper XCB .628 .428 6.477 406 .000 77.392 11.948 53.904 100.880 AUB 2.368 .125 10.760 426 .000 122.242 11.361 99.911 144.573 UFBR .607 .436 7.906 427 .000 94.682 11.976 71.142 118.222 NLH 9.628 .002 12.798 421 .000 137.839 10.770 116.669 159.010 NLB .085 .771 4.625 416 .000 56.308 12.176 32.374 80.242 OBB .354 .552 7.380 427 .000 88.076 11.934 64.618 111.533 OBH 4.026 .045 3.132 427 .002 39.266 12.538 14.623 63.909 EKB .110 .741 7.115 427 .000 86.204 12.115 62.391 110.017 DKB 1.445 .230 3.172 427 .002 39.909 12.582 15.179 64.639 FRC .917 .339 9.098 425 .000 106.385 11.693 83.402 129.368 OCC .138 .710 2.870 423 .004 36.219 12.621 11.411 61.028 FOL 3.126 .078 6.726 424 .000 81.350 12.095 57.578 105.123 FOB 1.573 .210 5.472 425 .000 67.164 12.274 43.039 91.290 MDH 7.987 .005 11.372 424 .000 127.017 11.169 105.063 148.971
212 Table 5 5 Discriminant Function Analysis (DFA) 1. Wilks' Lambda F value df1 df2 p = GOL .780 19.201 1 68 .000 XCB .915 6.290 1 68 .015 ZYB .732 24.842 1 68 .000 BBH .805 16.481 1 68 .000 BNL .757 21.865 1 68 .000 BPL .831 13.870 1 68 .000 MAB .813 15.663 1 68 .000 MAL .876 9.627 1 68 .003 AUB .815 15.392 1 68 .000 UFHT .776 19.679 1 68 .000 WFB .928 5.265 1 68 .025 UFBR .872 10.018 1 68 .002 NLH .703 28.760 1 68 .000 NLB .937 4.596 1 68 .036 OBB .857 11.361 1 68 .001 OBH .964 2.575 1 68 .113 EKB .935 4.747 1 68 .033 DKB 1.000 .016 1 68 .901 FRC .897 7.802 1 68 .007 PAC .943 4.079 1 68 .047 OCC .964 2.546 1 68 .115 FOL .931 5.006 1 68 .029 FOB .974 1.791 1 68 .185 MDH .768 20.575 1 68 .000 Table 5 6 Structure matrix for Discriminant Function 1. Variable Correlation coefficient Variable Correlation coefficient NLH .511 MAL .296 ZYB .475 FRC .266 BNL .446 XCB .239 MDH .433 WFB .219 UFHT .423 FOL .213 GOL .418 EKB .208 BBH .387 NLB .204 MAB .377 PAC .193 AUB .374 OBH .153 BPL .355 OCC .152 OBB .321 FOB .128 UFBR .302 DKB .012
213 Table 5 7 Unstandardized canonical coefficients for Discriminant Function 1. Variable Coefficient Variable Coefficient GOL .060 NLH .165 XCB .036 NLB .224 Male group mean: 1.150 ZYB .143 OBB .109 Female group mean: 1.366 BBH .099 OBH .117 BNL .122 EKB .278 Sectioning point: BPL .099 DKB .175 Females < 0.108 < Males MAB .044 FRC .009 MAL .264 PAC .047 AUB .023 OCC .089 UFHT .047 FOL .034 WFB .055 FOB .094 UFBR .151 MDH .127 (Constant) 31.270 Figure 5 1. Frequency distribution of male and female discriminant function scores for Discriminant Function 1.
214 Table 5 8 Classification results for Discriminant Function 1. S ex Predicted Group Membership Total M ales F emales Resubstitution Count M 34 4 38 F 2 30 32 % M 89.5 10.5 100.0 F 6. 2 93.8 100.0 Cross validated Count M 28 10 38 F 6 26 32 % M 73.7 26.3 100.0 F 18. 7 81.3 100.0 91.4% of resubstituted grouped cases correctly classified. 77.1% of cross validated grouped cases correctly classified. Table 5 9 Discriminant Function Analysis 2. Wilks' Lambda F value df1 df2 p = GOL .815 86.719 1 383 .000 XCB .907 39.199 1 383 .000 ZYB .722 147.723 1 383 .000 BBH .808 90.922 1 383 .000 BNL .761 120.148 1 383 .000 AUB .807 91.483 1 383 .000 WFB .950 20.322 1 383 .000 UFBR .873 55.492 1 383 .000 NLH .735 138.023 1 383 .000 NLB .949 20.419 1 383 .000 OBB .891 47.092 1 383 .000 OBH .984 6.370 1 383 .012 EKB .892 46.483 1 383 .000 DKB .976 9.373 1 383 .002 FRC .831 78.057 1 383 .000 PAC .973 10.580 1 383 .001 OCC .979 8.357 1 383 .004 FOL .906 39.862 1 383 .000 FOB .935 26.477 1 383 .000 MDH .776 110.644 1 383 .000
215 Table 5 10 Structure matrix for Discriminant Function 2. Variable Correlation coefficient Variable Correlation coefficient ZYB .655 EKB .367 NLH .633 FOL .340 BNL .591 XCB .337 MDH .567 FOB .277 AUB .515 NLB .243 BBH .514 WFB .243 GOL .502 PAC .175 FRC .476 DKB .165 UFBR .401 OCC .156 OBB .370 OBH .136 Table 5 1 1 Unstandardized canonical coefficients for Discriminant F unction 2. Variable Coefficient Variable Coefficient GOL .023 OBB .225 XCB .002 OBH .052 Male group mean: 0.648 ZYB .183 EKB .208 Female group mean: 1.380 BBH .030 DKB .139 BNL .064 FRC .022 Sectioning point: AUB .070 PAC .004 Females < 0.366 < Males WFB .063 OCC .029 UFBR .008 FOL .016 NLH .119 FOB .005 NLB .028 MDH .126 GOL .023 (Constant) 22.677 Table 5 1 2 Classification results for Discriminant Function 2. S ex Predicted Group Membership Total M ales F emales Resubstitution Count M 220 42 262 F 17 106 123 % M 84.0 16.0 100.0 F 13.8 86.2 100.0 Cross validated Count M 215 47 262 F 20 103 123 % M 82.1 17.9 100.0 F 16.3 83.7 100.0 84.7% of original grouped cases correctly classified. 82.6% of cross validated grouped cases correctly classified
216 Figure 5 2 Frequency distribution of male and female discriminant function sc ores for Discriminant Function 2 Table 5 1 3 Stepwise selection matrix for variables included in Discriminant Function 3. Step Entered Lambda a df1 df2 df3 F ratio b df1 df2 Sig. 1 ZYB .722 1 1 383.000 147.723 1 383.000 .000 2 NLH .651 2 1 383.000 102.507 2 382.000 .000 3 MDH .609 3 1 383.000 81.556 3 381.000 .000 4 BNL .586 4 1 383.000 67.053 4 380.000 .000 5 EKB .567 5 1 383.000 57.863 5 379.000 .000 6 AUB .559 6 1 383.000 49.795 6 378.000 .000 7 FRC .551 7 1 383.000 43.911 7 377.000 .000 8 WFB .544 8 1 383.000 39.387 8 376.000 .000 a At each step, the variable that minimizes the overall Wilks' Lambda is entered. b Pairwise group comparisons
217 Table 5 1 4 Unstandardized canonical coefficients for Discriminant Function 3. Variable Coefficient ZYB .183 BNL .101 Male group mean: 0.626 AUB .075 Female group mean: 1.333 WFB .050 NLH .100 Sectioning point: EKB .091 Females < 0.354 < Males FRC .048 MDH .126 (Constant) 24.996 Table 5 1 5 Classification results for Discriminant Function 3. S ex Predicted Group Membership Total M ales F emales Resubstitution Count M 230 48 278 F 17 114 131 % M 82.7 17.3 100.0 F 13.0 87.0 100.0 Cross validated Count M 228 50 278 F 20 111 131 % M 82.0 18.0 100.0 F 15.3 84.7 100.0 84.1% of original grouped cases correctly classified. 82.9% of cross validated grouped cases correctly classified
218 Figure 5 3. Frequency distribution of male and female discriminant function scores for Discriminant Function 3.
219 Table 5 1 6 One way analysis of variance (ANOVA) results comparing from three regional Thai subpopulations, males only. Variable Sum of Squares df Mean square F Sig. GOL Between Groups 38.929 2 19.465 .430 .651 Within Groups 12944.538 286 45.261 Total 12983.467 288 XCB Between Groups 55.714 2 27.857 .987 .374 Within Groups 7786.028 276 28.210 Total 7841.742 278 ZYB Between Groups 129.385 2 64.693 2.678 .070 Within Groups 6958.312 288 24.161 Total 7087.698 290 BBH Between Groups 99.592 2 49.796 2.137 .120 Within Groups 6664.422 286 23.302 Total 6764.014 288 BNL Between Groups 145.642 2 72.821 3.559 .030 Within Groups 5913.601 289 20.462 Total 6059.243 291 BPL Between Groups 195.696 2 97.848 4.739 .011 Within Groups 2188.506 106 20.646 Total 2384.202 108 MAB Between Groups 3.476 2 1.738 .159 .854 Within Groups 492.441 45 10.943 Total 495.917 47 MAL Between Groups 42.058 2 21.029 2.503 .087 Within Groups 856.990 102 8.402 Total 899.048 104 AUB Between Groups 64.264 2 32.132 1.288 .277 Within Groups 7183.551 288 24.943 Total 7247.814 290 UFHT Between Groups 53.107 2 26.554 1.571 .213 Within Groups 1656.022 98 16.898 Total 1709.129 100 WFB Between Groups 136.560 2 68.280 3.183 .043 Within Groups 6113.426 285 21.451 Total 6249.986 287 UFBR Between Groups 63.910 2 31.955 1.960 .143 Within Groups 4711.049 289 16.301 Total 4774.959 291
220 Table 5 1 6 C ontinued. Variable Sum of Squares df Mean square F Sig. NLH Between Groups 2.013 2 1.006 .106 .899 Within Groups 2692.817 284 9.482 Total 2694.829 286 NLB Between Groups 4.965 2 2.482 .671 .512 Within Groups 1035.346 280 3.698 Total 1040.311 282 OBB Between Groups 8.379 2 4.190 1.307 .272 Within Groups 926.261 289 3.205 Total 934.640 291 OBH Between Groups 1.063 2 .531 .131 .877 Within Groups 1167.718 289 4.041 Total 1168.781 291 EKB Between Groups 54.483 2 27.241 1.899 .152 Within Groups 4145.281 289 14.344 Total 4199.764 291 DKB Between Groups 7.707 2 3.854 .768 .465 Within Groups 1449.700 289 5.016 Total 1457.408 291 FRC Between Groups 46.341 2 23.171 1.185 .307 Within Groups 5610.004 287 19.547 Total 5656.345 289 PAC Between Groups 71.443 2 35.721 1.072 .344 Within Groups 9463.860 284 33.323 Total 9535.303 286 OCC Between Groups 95.967 2 47.983 1.356 .259 Within Groups 10088.478 285 35.398 Total 10184.444 287 FOL Between Groups 16.240 2 8.120 1.535 .217 Within Groups 1523.457 288 5.290 Total 1539.698 290 FOB Between Groups 26.605 2 13.302 3.207 .042 Within Groups 1198.666 289 4.148 Total 1225.271 291 MDH Between Groups 22.006 2 11.003 1.050 .351 Within Groups 3008.163 287 10.481 Total 3030.169 289
221 Table 5 17 Structure matrices for discriminant function analysis of regional subpopulations, by sex, based on 20 craniometric variables. Function Function Males 1 2 Females 1 2 FOB .435 .016 OBB .383 .085 WFB .395 .026 AUB .212 .131 ZYB .345 .262 EKB .211 .126 FOL .316 .104 BBH .159 .053 UFBR .305 .031 OCC .122 .046 FRC .303 .014 FOL .121 .028 EKB .297 .008 ZYB .084 .081 OCC .272 .171 UFBR .040 .037 OBB .239 .118 OBH .030 .010 XCB .226 .042 FRC .291 .481 NLB .198 .110 DKB .116 .252 GOL .114 .109 BNL .185 .244 OBH .112 .045 GOL .057 .230 BBH .049 .418 XCB .043 .213 BNL .368 .384 FOB .065 .188 MDH .018 .348 PAC .130 .155 AUB .194 .281 NLH .034 .108 DKB .040 .276 MDH .087 .101 NLH .112 .142 WFB .022 .058 PAC .067 .122 NLB .004 .006 largest absolute correlation between each variable and any discriminant function.
222 Figure 5 4. Scatter plot of discriminant function scores for males, by regional subpopulation for discriminant function analysis using 20 craniometric variables
223 Figure 5 5. Scatter plot of discriminant function scores for females, by regional subpopulation for discriminant function analysis using 20 craniometric variables Table 5 1 8 Classification results for males by regional subpopulation, from discriminant function analysis using 20 variables. POP Predicted Group Membership Total CM KK NU Resubstitution Count CM 118 58 40 216 KK 5 18 6 29 NU 5 2 10 17 % CM 54.6 26.9 18.5 100.0 KK 17.2 62.1 20.7 100.0 NU 29.4 11.8 58.8 100.0 Cross validated a Count CM 102 66 48 216 KK 10 13 6 29 NU 8 5 4 17 % CM 47.2 30.6 22.2 100.0 KK 34.5 44.8 20.7 100.0 NU 47.1 29.4 23.5 100.0 55.7% of original grouped cases correctly classified. 45.4% of cross validated grouped cases correctly classified.
224 Table 5 1 9 Classification results for fe males by regional subpopulation, from discriminant function analysis using 20 variables. POP Predicted Group Membership Total CM KK NU Original Count CM 54 20 19 93 KK 7 15 1 23 NU 1 1 5 7 % CM 58.1 21.5 20.4 100.0 KK 30.4 65.2 4.3 100.0 NU 14.3 14.3 71.4 100.0 Cross validated a Count CM 44 25 24 93 KK 11 11 1 23 NU 4 1 2 7 % CM 47.3 26.9 25.8 100.0 KK 47.8 47.8 4.3 100.0 NU 57.1 14.3 28.6 100.0 60.2% of original grouped cases correctly classified. 46.3% of cross validated grouped cases correctly classified. Table 5 20 Summary of stepwise discriminant function analyse s of Culled Samples (CS) 1 through 4 (males only). CS 1 CS 2 CS 3 CS 4 Non normal variables: Chiang Mai FOL NLB EKB MDH GOL XCB OBB BBH OBH DKB OCC Khon Kaen XCB OBH XCB OBH XCB OBH XCB OBH Naresuan ----p -PAC FOB OCC p BNL BNL ZYB BNL BNL WFB OCC Variable selected: (in sequence) BNL BNL DKB BNL WFB OCC BNL (test of model) p = 0.036 p = 0.008 p = 0.008 p = 0.001 Cross Validation (% correct) 42.5 52.5 42.5 53.2
225 Figure 5 6 Scatter plot of discriminant function scores for males in Culled Sample 4, by regional subpopulation, for stepwise discriminant function analysis selecting 3 variables. Table 5 2 1 Unstandardized canon ical coefficients for stepwise discriminant function analysis of Culled Sample 4. Function 1 Function 2 Variable Coefficient Coefficient Function 1 Function 2 BNL .012 .241 WFB .177 .170 CM group centroid : 0.501 0.226 OCC .128 .039 KK group centroid : 0.605 0.132 (Constant) 5.222 4.543 NU group centroid : 0.111 0.617
226 Table 5 22 Cross validated classification results for stepwise discriminant function analyse s of Culled Samples 1 through 4. POP Predicted Group Membership Total CM KK NU Culled Sample 1 Count CM 4 12 14 30 KK 3 18 9 30 NU 2 6 12 20 % CM 13.3 40.0 46.7 100.0 KK 10.0 60.0 30.0 100.0 NU 10.0 30.0 60.0 100.0 Culled Sample 2 Count CM 13 9 8 30 KK 6 17 7 30 NU 4 4 12 20 % CM 43.3 30.0 26.7 100.0 KK 20.0 56.7 23.3 100.0 NU 20.0 20.0 60.0 100.0 Culled Sample 3 Count CM 3 14 13 30 KK 2 21 7 30 NU 2 8 10 20 % CM 10.0 46.7 43.3 100.0 KK 6.7 70.0 23.3 100.0 NU 10.0 40.0 50.0 100.0 Culled Sample 4 Count CM 15 7 8 30 KK 4 19 7 30 NU 4 7 8 19 % CM 50.0 23.3 26.7 100.0 KK 13.3 63.3 23.3 100.0 NU 21.1 36.8 42.1 100.0
227 CHAPTER 6 SUMMARY AND CONCLUSIONS Know then thyself, presume not God to scan; The proper study of Mankind is Man. A being darkly wise, and rudely great: With too much knowledge for the Sceptic side, With too much weakness In doub t his Mind or Body to prefer, Alexander Pope An Essay on Man (1733 34) In physical anthropology, the comparison of proximate and ultimate levels of analysis is a familiar construct that pervades nearly all scientific enquiry within the field, and though the analogy is imperfect, this dissertation is no exception. The foregoing researc h was initially motivated by c dataset for the modern Thai population, from which to generate population specific standards for determination of the biological profile in unknown skeletal remains from this population. This objective has, in considerable measure, been accomplished. A dataset of over 400 individuals has been compiled, and from this, discriminant functions have been generated that can be used immediately by forensic anthropologists in Thailand for the determination of sex in unknown remains within their country. The ev entual addition of these data to the FORDISC forensic database will introduce the first Southeast Asian female sample (and only the second sample of Asian females overall) to the database, and will permit all forensic anthropologists to apply this dataset and its representation of a small, yet important portion of global human craniometric diversity to their own casework, improving our ability to correctly identify individuals who fall nded future research will investigate whether these functions can also be validly generalized and applied to other
228 population history, they may be suitably representative of a much greater degree of the total span of human craniometric variation in the collective Southeast Asian populations than is currently appreciated. Additional impetus for this research was derived from the broader questions regarding the ultimate or igins of the Thai and the other modern populations of Southeast Asia. For here, in Southeast Asia, we find a compelling distillation of one of the essential debates in all of anthropology: what roles, and of what magnitude, have the in situ evolution of endogenous populations and the migratory influences of ex ogenous populations played in the development and structure of the populations of a region as we know them today. To have any hope of advancing towards a resolution of this debate, we must delve dee per into the available skeletal material, and consider the salient populations in greater detail than we have previously, to include considerations of patterns of regional and temporal variation possibly extant within them. In this regard, this dissertati on has also achieved a measure of success, albeit of somewhat lesser direct impact than the forensic application s noted above, but hopefully still of some significance in its implications for our understanding of Southeast Asian population history. As has been noted in previous chapters, the hypothesis that significant morphological differences may exist among regional subpopulations within the modern Thai population emerges proximately from first hand observations of the phenotypic diversity extant within the living population of Thailand, and is reinforced by the perceptions of this diversity expressed by the living Thais themselves. An understanding of the archaeology and linguistics of Southeast Asia leads almost ineluctably to the hypothetical corollar y that this perceived diversity, if it is in fact real, should be deeply rooted in the intricacies of Southeast Asian history and prehistory, in the procession of human civilizations and the peregrinations of human populations across the
229 landscape. The pl ausibility of this corollary is in turn reinforced by the varied cultural identities which exist within the modern Thai population, and by the differences in dialect and custom iopolitical regions geography and its modern and ancient history (Kaplan 1981; Rinehart 1981; Higham and Thosarat 1998; Baker and Phongpaichit 2005). Thus, the popul ations in the North and Northeast of Thailand are reported to look more bridges and noticeable epicanthic folds, echoing the proximity of these regions to, a nd the accordingly greater influence of, the peoples of East Asia. Chiang Mai and the surrounding regions of Northern Thailand look to the ancient kingdom of Lanna, which itself had its capital at Chiang Mai, as their legendary antecessor. The pride of t his place, first of the Tai kingdoms in the ur Thailand, always beyond the reach of great Angkor, is evoked by the prominent display throughout the region of the unique Lanna script and architectural style (Figures 6 1 and 6 2) (Kaplan 1981; Rinehart 1981; Higham and Thosarat 1998; Baker and Phongpaichit 2005). Khon Kaen is located on the Khorat Plateau in Northeastern Thailand, in the heart of what is perhaps the most culturally distinctive region of the country, Isan. Here, unique styles of food, dress, music and custom combine with a singular, predominantly Lao influenced dialect (also known as Isan) to create a regional culture which is quite distinct from and quite proud of its distinction from mainstream Thai culture. But t he Khorat Plateau ever the thoroughfare for the movements of peoples and the expansion of empires is home not just to the Isan culture (which shares ancient, common Daic roots with the northern Tai), but also to peoples who trace the paths of their descent to the ancient Austroasia tic Khmer of Angkor. Central Thailand,
230 birthplace of the modern kingdom in all its incarnations beginning with Sukhothai and ending with Bangkok, reborn from the ashes of Ayutthaya and for so long the point of entry for so much of the external influences from the great civilizations of China and India, is home to a cosmopolitan population of intermediate phenotype, neither wholly Asian nor wholly Indic in character, with perceived darker skin, stronger noses and sharper features than their northern counter parts. The populations of Southern Thailand, not addressed in the present study due to the social and political upheaval that precluded travel to the region, are known to be heavily influenced, culturally and biologically, by their Muslim Malay neighbors and the other Austronesian peoples of adjacent regions of island Southeast Asia (Kaplan 1981; Rinehart 1981; Higham and Thosarat 1998; Baker and Phongpaichit 2005). With such seemingly direct correspondences between the known ebbs and tides of peoples, cu ltures and languages, into and across Thailand over the course of millennia, and the human perceive meaningful phenotypic distinctions within their own populatio n becomes almost intuitively obvious. From there, the plausibility of real, discernable skeletal morphological differences among the subpopulations seems equally intuitive. And yet, as the results of the foregoing analyses have made abundantly clear, no such morphological distinctions can be detected. Given the analytical power of discriminant function analyses for teasing apart subtle patterns of human morphological variation (Howells 1973), and the breadth of the skeletal sample under analysis here, th ese findings of the absence of any clear patterns of differentiation among the regional subpopulations of the modern Thai cannot go unexplained. There is, of course, always the possibility that such patterns of differentiation never existed to begin with and that the population of Thailand has always been relatively homogenous. This
231 is, of course, a thoroughly unsatisfactory answer. To suggest that a region characterized by such complexity and reticulation in its population history has been craniometri cally homogenous throughout that entire history is to disregard all that we understand about the patterning of human craniometric diversity, along with all that we know about the prehistoric and historic development of the populations of Southeast Asia (Ke nyhurcz et al. 2010) It is simply untenable to argue that, all else being equal, the myriad peoples the Mon, Khmer, Cham, Tai, Lao, Vietnamese, Chinese, Indian, Malay, the unnamed prehistoric populations of Austronesians and Austroasiatics, and the Australo Melanesian hunter gatherers before them who have contributed, in greater or lesser degrees, to the pool of modern Thai genetic, linguistic and cultural diversity, have left no corresponding pattern of discernable craniometric diversity within this population. Something is clearly missing. As was stated at the outset of this dissertation, to understand the biological history of the Thai population it is essential that we first understand the cultural history of the people. For herein lies the most plausible explanation for the pattern (or perhaps lack of patterning) of craniometric variation observed in the modern Thai. According to historical accounts, the battles for political and cultural supremacy that roiled among the early Tai kingdoms of La nna, Sukhothai and Ayutthaya, and between Ayutthaya and Angkor in the fourteenth and fifteenth centuries, were generally accompanied by the displacement of large segments the defeated f as prisoners and slaves of the conquering armies. This is particularly true of the sacking of Ayutthaya by the Burmese in AD 1559 and again in AD 1767, the latter of which reduced the entire city to rubble and left it completely abandoned. A succession of similar attacks on the great Lanna capital of Chiang Mai
232 fr om Siam in the early 1800s, a similar fate had befallen most of the towns and cities once under Over the last quarter of the eighteenth century and the first of the nineteenth century, as the Thai fought to contain the Burmese and reclaim their lands, areas that had formerly been only loosely tributary to Ayutthaya such as Lanna and the Khorat fronti er were brought under the firm control of the new Siamese capital at Bangkok. But reclaiming their devastated cities was only the first step in rebuilding what would in short order become the modern nation of Thailand. Siam also needed to reclaim its peo ple, driven off and carried away during the sieges. To do north and east and engaged in their own wars of aggression against the Lao and Khmer conquering the former in 1828 and the latter in 1833. After each conquest, thousands of people Thai and non Thai, commoners and elites alike were again forcibly displaced from the annexed territories and resettled throughout Siam, to rebuild its cities and replant its f ields (Baker and Phongpaichit 2005:26 29). As Baker and Phongpaichit (2005:29) summarize, In the 1770s a 1 armies captured many thousands of Lanna Yuan, Lao Wiang, Lao Phuan, Black Tai, and Khmer. The southern expeditions brought back se veral thousand Malays. Possibly 30,000 40,000 Mons voluntarily migrated to Siam. In the early 1800s, the Bangkok and Lanna troops went further north to seize Khoen, Lu, and Shan. After the 1827 war against Vientiane [the Lao capital], over 150,000 were captured and some 50,000 marched down to the Chao Phraya basin. In the 1830s, the Bangkok armies made six expeditions into the Lao regions, depopulating the left bank of the Mekong, and bringing back Lao Phuan from the Plain of Jars, Tai Dam from Sipsongc hutai, Khmer, and Vietnamese. 1 Taksin was the first King of Siam in the post Ayutthaya era, who established an interim capital at Thonburi, on the west bank of the Chao Phraya river. He was deposed in a coup in 1782, whereupon the capital wa s shifted east across the river to Bangkok, and the Chakri Dynasty whence all subsequent kings, up to an including the present King of Thailand, Bhumipol Adulyadej (Rama IX) was established.
233 In this tumult of war and displacement, any remaining threads tying the local subpopulations of Siam (and adjacent regions of Southeast Asia) to their immediate prehistoric and historic antecessors would almost certainly have been broken. While many displaced Thais surely returned to rebuild and resettle the cities of their ancestry, thus reestablishing some semblance of regional ethnic and biological continuity with previous eras, the population of Siam was, in the main, rema ties and the great shuffle of displaced and replaced populations across the Chao Phraya ba sin and Khorat Plateau therefore must also have dramatically altered the structure of human biological variation within Thailand. The post reconstruction population of Siam likely contained a greater range of individual morphological diversity, due to the forced influx of thousands of non Thai, but this diversity must have been less patterned along regional and historic population lines, therefore making the modern population more homogenous overall. Throughout the nineteenth and twentieth centuries, thi s physical homogeneity would have been sustained and reinforced and in all likelihood, deepened by further population movements, first out of the urban areas and forested lands and into the rural provinces along the rapidly advancing modern agricultural fr ontier, and then later, back to the cities during the recent era of industrialization and urbanization (Baker and Phongpaichit 2005). Yet throughout nearly y, areas such as Isan and Lanna have still cleaved tightly to their unique regional ethnic and linguistic identities. It is hard to say to what extent these regional cultural complexes were disrupted by the earlier population upheavals, but like the popul ation of Thailand itself, these cultural identities must to some extent have been reclaimed, reconstructed, and reimagined by
234 the newly resettled populations following the ouster of the Burmese. In becoming the modern nation of Thailand, these regional cu ltural identities would be alternately celebrated as a measure of the cosmopolitan breadth of Siam or subdued in order to present to the world a singular Thai nation unified in language, custom and heritage (Baker and Phongpaichit 2005). Today, Thailand has achieved a delicate balance between both aspects of its character, acknowledging the richness of the manifold cultural influences and identities that shape Thai society as a whole, all the while asserting the overarching unity of a single Thai nation. And both aspects are of great import to the collective self image of the modern Thai. On one hand, internal unity has always been the means by which the Thai have resisted external pressure, whether from conquest, colonialism or globalization. Not surpr isingly, it is this same desire for unity within the modern Thai society that also motivates their desire for a direct connection to, through continuous descent from, the ancient populations of Thailand. On the other hand, the ural, linguistic, spiritual and biological heritages are the crucial ties that bind them to the other great cultures of southern and eastern Asia, and root them in their tion of and esteem for their diverse heritage is what drives their recognition of patterns of internal cultural and physical variation structured by internal regional boundaries. This per ception of suites of regionally distinctive characteristics within t he modern Thai population acts to reify the people s sense of deep time connections to the myriad historic and prehistoric cultures of mainland Southeast Asia. Given a greater understanding of these historical and cultural considerations in the populatio n history of the modern Thai, we can at last return to the pursuit of a corresponding understanding of the biological aspects of that population history. The critical finding we seek to
235 understand is that despite the colloquial assertions to the contrary, there appears to be no pattern of phenotypic distinctions within the modern population of Thailand or at least, none that we can describe craniometrically which can be correlated with the historic ethnic and linguistic distinctions among the internal subr egions of Thailand. By now, the reasons for this should be clear. The homogenizing effects of the deliberate upheaval and forced migrations of tens of thousands of dispossessed people, inflicted over three quarters of a century of intermittent warfare ac ross the whole of Southeast Asia, represents such a fundamental and profound disruption in the historic trajectories of peoples and cultures and of the historic connecti ons between peoples and places, that any correlation that might once have existed betw een them has long since been erased. Regardless of the cultural perceptions which exist among the Thai today, regional craniometric differentiation t o have evolved within the Thai population in the post population, though attributed largely to differences in phenotype, is in truth likely driven much more by the observ ation of cultural and ethnic factors, such as differences in dress, accent, dialect and manner, which were more easily preserved and which have more readily persisted in spite of the population upheaval of the eighteenth and nineteenth centuries. The new yet still incomplete, understanding of the patterns of morphological variation within the modern Thai population that has been developed within this dissertation leads to several other important considerations. First and foremost, the skeletal populatio ns from Chiang Mai, Khon Kaen and Bangkok, in addition to representing the populations of those cities and surrounding areas, can now be more comfortably accepted as representative of the modern Thai
236 population in general. 2 The genuine and justifiable con cern that the labyrinthine subtleties of Southeast Asian population history have created craniometrically distinct subpopulations within the mode rn Thai is somewhat ameliorated, as is the concern over the impact of such subpopulations on our ability to ult imately resolve the central question of essential continuity versus discontinuity between the pre and post Neolithic populations of Southeast Asia. Had the regional subpopulations examined in this study been found to differ significantly in terms of cran ial morphology, this finding would compel us to consider them as quasi independent populations in our broader studies of biological distance among the varied populations of Asia and the Pacific. But this is not so, and w hile the small sample sizes (ca. N many broad scale craniometric population studies still must necessarily fall far short of capturing a true picture of the full spectrum of craniometric variation within the Thai population, we are reassured that such samples can nevertheles s reasonably stand as a measure of the Thai populati on as a whole, and not just as the sum of one of its parts. Yet as reassuring as this should be, this finding also highlights the critical importance of approaching craniometric population studies with a thorough understanding of the subtleties of the individual histories of each population under consideration. Recognizing that certain populations speak related langu ages, or that they share ethnic, cultural or genetic characteristics inherited from a common ancestral civilization, are only the starting points for disentangling population relationships in a region so humanly complex as Southeast Asia. Historic events such as the Burmese conquest and the Siamese wars of reclamation add yet anothe r layer of complexity to the picture of population history in Thailand, and we underestimate the ripple 2 Ethnic minorities such as the Hill Tribes still notwithstanding of course.
237 need to delve deeper into our own anthropological know ledge of the forces which shape populations, to look not just at the ebb and tide of ancient civilizations, but also into the much more recent diasporas wrought by warfare, changing environmental conditions, political unrest. In the case of Thailand, an u nderstanding of historical events in Siam in the eighteenth and populations of the modern nation as long term reside nts in their present locations. Y et not a si ngle one of the craniometric studies cit ed in Chapter 3 even mentions these events, much less gives any consideration to their potential impact on our understanding of population history in Southeast Asia (though such events and their impact on Southeast Asian population prehistory are broadly alluded to by Kenyhercz et al. 2010). It was noted previously that Matsumura (2006) has argued for a more temporally robust and geographically focused approach to craniometric data sampling as the way forward in reso lving the central question of endogenous versus exogenous origins for the modern populations of Southeast Asia. This argument is supported, indirectly, by the results of this dissertation. Analysis of a tightly concentrated geographic and ethnic populati on, both as a whole and broken down into regional subpopulations, has found a pattern of morphological variation which runs counter to that which might be expected based on extrapolation from the prehistoric and early historic archaeological and linguistic data, but which is in broad accord with expectations produced by an understanding o f the previously unacknowledged recent historical events shaping that population. These results have assuaged some of the analytical concerns regarding the validity of usi ng a small pool of craniometric data derived from a single anatomical collection as representative of the entire population as a whole, but it has also raised other concerns about the thoroughness of our understanding of the history of the populations
238 unde greater attention to the recent historic sociopolitical factors relevant to the development of the specific geographic and ethnic populations we are analyzing. We are a lready quite adept at this when it comes to our own population for example, applying our knowledge of eighteenth to twentieth century American history to understand the composition of the Te rry or Hamann Todd collections. B ut it becomes especially importa nt when dealing with non Western populations, whose recent historic trajectories may not be so readily apparent to researchers venturing into new territories, or seeking new insight into familiar ones. My suspicion is that many of the na gging questions an d incongruences which cause so much distraction and misdirection in so many craniometric population studies will resolve themselves if we only understood a little more about the populations involved.
239 Figure 6 1 Sign in front of the International Cente r Hostel at Chiang Mai University, with text in the Thai (top), Lanna (middle), and Roman scripts. Photograph by the author.
240 Figure 6 2 Great Stupa of Wat Suandok, Chiang Mai ; an example of Lanna style religious architecture Photograph by the author
241 APPENDIX A DEFINITIONS OF CRANI OMETRIC LANDMARKS AN D MEASURMENTS For reference, the craniometric landmark and measurement definitions followed during data collection for this dissertation are reproduced here, with minor edits, as provided by Moore Jansen et al. (1994). Citations provided within each definition identify the original source of the landmark or measurement, as cited by Moore Jansen et al (1994). Definitions of Cranial Landmarks Alare (al) The most laterally positioned point on the anterior margin of the nasal aperture. This point should be marked on bot h t he right and left sides of the nasal aperture (after Bass 1971:60; Howells 1973:176). Alveolon (alv) The point where the midline of the palate is intersected by a straight tangent connecting the posterior borders of the alveolar crests. This point is used in the measurement of maxilla alveolar length and determined in practice, as the point where the mid sagittal plane intersects a wire needle placed against the posterior margins of the alveolar processes of the maxilla (Martin and Saller 1956:451). Auriculare (au) A point on the lateral aspect of the root of the zygomatic process at the deepest incurvature, wherever it may be. Basion (ba) The point where the anterior margin of the foramen magnum is intersected by the mid sagittal plane. The point i s located on the inner border of the anterior margin of the foramen magnum direction opposite of opisthion (o). In height measurements of the braincase, basion is positioned somewhat farther onto the underside of the margin of the foramen magnum (hypobasi on), so that the observer may distinguish between an inferior and a posterior basion for reasons of convenience and technical demands (Martin and Saller 1956:446). Bregma (b) The point where the sagittal and coronal sutures meet. In those cases where the most anterior segment of the sagittal suture deflects to one side, the point of the junction of the two sutures must be projected. Bregma is skulls with total obliteration of the sutures. In the latter case it may be possible to see existing traces of the sutures by slightly moistening the area (Martin and Saller 1956:444)
242 Dacryon (d) The point on the medial border of the orbit at which the frontal, lacrimal, and maxilla intersect. In other words, dacryon lies at the intersection of the lacrimomaxillary suture and the frontal bone. There is often a small foramen at this point (Martin and Saller 1956:450). Ectoconchion (ec) The intersection of the most anterior surface of the lateral border of the orbit and a line bisecting the orbit along its long axis. To mark ectoconchion, move a toothpick or other thin, straight instrument up and down, keeping it parallel to the superior orbital border, until you divide the eye orbit into two equal halves. Mark the point on the anterior orbital margin with a pencil (Howells 1973:168) Ectomolare (ecm) The most lateral point on the lateral surface of the alveolar crest. This point is generally positioned on the alveolar margin of the section maxillary molar. Euryon (eu) The most laterally positioned point on the side of the braincase. Euryon always falls on either the parietal bone or on the upper portion of the temporal bone and may be determined only by measuring maximum cranial breadth. The area of the root of the zygomatic arch, the supramastoid crest, and the entire adjacent region above the external auditory meatus, which sometimes exhibit excessive symmetrical lateral expansion, should be avoide d when determining the position of euryon (Martin and Saller 1956:447). Frontomalare The most laterally positioned point on the frontomalar suture temporale (fmt) (Martin and Saller 1956:451). Frontotemporale (ft) A point located generally forward and inward on the superior temporal line directly above the zygomatic process of the frontal bone (Martin and Saller 1956:451). Glabella (g) The most frontwardly projecting point in the mid sagittal plane at the lower margin of the frontal bone, which lies above the nasal root and between the superciliary arches. The point of glabella is depressed between the confining bony ridges, and is often delineated superiorly by a shallow gutter or a transversely running indentation on the surface of the frontal bon e (Martin and Saller 1956:442 443). Lambda (l) The point where the two branches of the lambdoidal suture meet with the sagittal suture. The determination of this point is uncertain in cases with strongly serrated sutures, as well as cases where sutures ar e totally obliterated. Locating lambda may be further complicated in crania with wormian or sutural bones at the apex of the occipital squama. In such cases the general direction of the two branches of the lambdoidal sutures is determined and two straig ht lines are projected along the branches of the
243 suture placing lambda at the point where these lines meet with one another and with the sagittal suture (Martin and Saller 1956:444) Nasion (n) The point of intersection of the nasofrontal suture and the mid sagittal plane. Nasion corresponds to the nasal root (Martin and Saller 1956:448) Nasospinale (ns) The lowest point on the inferior margin of the nasal aperture as projected in the mid sagittal plane. In crania with slight to moderate development of the anterior nasal spine, this point is easily determined by connecting the lowest point on the inferior margin of the nasal aperture right and left of the nasal spine. Nasospinale is located wherever this line is intersected by the mid sagittal plane at the base of the nasal spine. If the nasal spine is well developed, mark the point of nasospinale on the lateral wall of the projecting nasal spine. However, if the nasal spine is at or below the line connecting the lowest point on the inferior margins of th e aperture, nasospinale is found on the upper margin of the nasal spine (Martin and Saller 1956:448) Opisthocranion (op) The most posteriorly protruding point on the back of the braincase, located in the mid sagittal plane. Opisthocranion almost always fa lls on the superior squama of the occipital bone, and only occasionally on the external occipital protuberance. Opisthocranion can generally be established while obtaining the measurement of maximum cranial length. However, in some cases where the superi or squama forms a partial sphere with glabella as its midpoint, opisthocranion cannot be determined in this manner. In such cases, each point on the partial sphere represents the maximum distance and opisthocranion is located arbitrarily approximately at a point in the middle of the spheric segment (Martin and Saller 1956:445). Opisthion (o) The point at which the mid sagittal plane intersects the posterior margin of the foramen magnum. Opisthion is located on the inner border of the posterior margin of t he foramen magnum facing basion. Prosthion (pr) The most anterior point on the alveolar border of the maxilla between the central incisors in the mid sagittal plane. Note that in measuring basion prosthion length and palate length, prosthion is not posit ioned on the inferior margin of the advanced bony parts between the incisors, but is more anteriorly located on the anterior surface of the alveolar process. In measuring upper facial height, however, prosthion is located on the inferior tip of the alveol ar process. In cases of a defective or resorbed alveolar process, determination of prosthion becomes uncertain or impossible, and upper facial height cannot be measured (Martin and Saller 1956:449).
244 Zygion (zy) The most laterally positioned point on the z ygomatic arches. The position of zygion is defined from the measurement of bizygomatic breadth (Martin and Saller 1956:450) Definitions of Cranial Measurements Maximum cranial length (GOL; g op) The distance of glabella from opisthocranion in the midsagit tal plane measured in a straight line. Maximum cranial breadth (XCB; eu eu) The maximum width of the skull perpendicular to the midsagittal plane wherever it is located, with the exception of the inferior temporal line and the immediate area surrounding the latter. Bizygomatic breadth (ZYB; zy zy) The direct distance between both zygia located at the most lateral points of the zygomatic arches. Basion bregma height (BBH; ba b) The direct distance from the lowest point on the anterior margin of the foramen magnum, basion, to bregma. Cranial base length (BNL; ba n) The direct distance from basion to nasion. Basion prosthion length (BPL; ba pr) The direct distance from basion to prosthion. Maxillo aveolar breadth (MAB; ecm ecm) The maximum breadth across the alveolar borders of the maxilla, measured on the lateral surfaces at the location of the second maxillary molars. Maxillo aveolar length (MAL; pr alv) The direct distance from prosthion to alveolon. Biauricular breadth (AUB; au au) The least exterior breadth across the roots of the zygomatic processes, wherever found. Upper facial height (UFHT; n pr) The direct distance from nasion to prosthion. Minimum frontal breadth (WFB; ft ft) The direct distance between the two frontotemporale. Upper facial breadth (UFBR; fmt fmt) The direct distance between the two frontomalare. Nasal height (NLH; n ns) The direct distance from nasion to nasospinale.
245 Nasal breadth (NLB; al al) The maximum breadth of the nasal aperture. Orbital breadth (OBB; d ec) The laterally sloping distance from dacryon to ectoconchion. Orbital height (OBH) The direct distance between the superior and inferior orbital margins, perpendicular to orbital breadth. Biorbital breadth (EKB; ec ec) The direct distance between left and right ectoconchion. Interorbital breadth (DKB; d d) The direct distance between left and right dacryon. Frontal chord (FRC; n b) The direct distance from nasion to bregma, taken in the midsagittal plane. Parietal chord (PAC; b l) The direct distance from bregma to lambda, taken in the midsagittal plane. Occipital chord (OCC; l o) The direct distance from lambda to opisthion, taken in the midsagittal plane. Foramen magnum length (FOL; ba o) The direct distance of basion from opisthion. Foramen magnum breadth (FOB) The distance between the lateral margins of the foramen magnum at the point of greatest lateral curvature. Mastoid length (MDH) The projection of the mastoid process below and perpendicular to the Frankfurt Horizontal in the vertical plane.
246 LIST OF REFERENCES Adams WY, Van Gerven DP, Levy RS. 1978. The retreat from migrationism. Annu Rev Anthropol 7:483 532. Albrecht GH. 1992. Assessing the affinities of fossils using canonical variates and generalized distances. Hum Evol 7:49 69. Alexander DH, Novembre J, Lange K. 2009. Fast model based estimation of ancestry in unrelated individuals. Genome Res 19:1655 1664. Anderson DD. 1990. Lang Rongrien rockshelter: a Pleistocene Early Holocene archaeological site from Krabi, southwestern Thailand. Philadelphia: University of Pennsylvania Monograph No. 71. Baker CJ, Phongpaichit P. 2005. A history of Thailand. Cambridge: Cambridge University Press. Ballinger SW, Schurr TG, Torroni A, Gan YY, Hodge JA, Hassan K, Chen KH, Wallace DC. 1992. Southeast Asian mitochondrial DNA analysis reveals genetic continuity of ancient mongoloid migrations. Gene tics 130:139 159. Barth F. 1952. The southern Mongoloid migration. Man 52:5 8. Bass WM. 1971. Human osteology: A laboratory and field manual of the human skeleton. Columbia (MO): Missouri Archaeological Society. Baughan B, Demirjian A. 1978. Sexu al dimorphism in the growth of the cranium. Am J Phys Anthropol 49:383 390. Bayard DT. 1972. Excavation at Non Nok Tha, Northeastern Thailand, 1968. Asian Perspect 13:109 143. Bayard DT. 1979. Comment on the linguistic history of mainland Southeast A sia. In Smith RB, Watson W, editors. Early south east Asia. Oxford: Oxford University Press. p 278 280. Gombrich R, editors. The world of Buddhism. London: Thames and Hudson. p 147 158. Bellwood PS. 1993. Cultural and biological differentiation in peninsular Malaysia: The last 10,000 years. Asian Perspect 32:37 60. Bellwood PS. 1996. Phylogeny and reticulation in prehistory. Antiquity 70:881 890. Bell wood PS. 1997. Prehistory of the Indo Malaysian archipelago, revised edition. Honolulu: University of Hawaii Press. Benedict PK. 1942. Thai, Kadai and Indonesian: A new alignment in Southeastern Asia. Am Anthropol 44:576 601.
247 Benedict PK. 1975. A ustro Thai language and culture. New Haven, CT: HRAF Press. Blust RA. 1984 1985. The Austronesian homeland: A linguistic perspective. Asian Perspect 26:45 67. Blust RA. 1993a. The Austronesian settlement of mainland Southeast Asia. Unpublished pap er read at the second annual meeting of the Southeast Asian Linguistic Society. Blust RA. 1993b. Beyond the Austronesian homeland: the Austric hypothesis and its implications for archaeology. Unpublished paper read at a meeting on Austroasiatic langu ages, Philadelphia. Blust RA. 1996. Beyond the Austronesian homeland: The Austric hypothesis and its implications for archaeology. In Goodenough WH, editor. Prehistoric settlement of the Pacific. Philadelphia: American Philosophical Society. p 117 140. letter word: The genesis of the concept. Oxford: Oxford University Press. Brace CL, Hunt KD. 1990. A nonracial craniofacial perspective on human variation: A(ustralia) to Z(uni). Am J Phys Anthropol 82:341 36 0. Brace CL, Tracer DP. 1992. Craniofacial c ontinuity and c hange: A c omparison of l ate P leistocene an d recent Europe and Esia. In Akazawa T Aoki K, Kimura T, editors. The e volution and dispersal of modern h umans in Asia. Tokyo : Hokusen sha. p 439 472. Brace CL, Tracer DP, Hunt KD. 1991. Human craniofacial form and the evidence for the peopling of the Pacific. Bull Indo Pac Pre Hi 11:247 269. Briggs LP. 1945. Dvaravati, the most ancient kingdom of Siam. J Am Oriental Soc 65(2):98 107. Brow n P. 2000. The first Australians: The debate continues. Australasian Science May:28 31. Bryson, B. 1990. The mother tongue: English and how it got that way. New York: HarperCollins. Buikstra JE, Frankenberg SR, Konigsberg LW. 1990. Skeletal biological distance studies in American physical anthropology: Recent trends. Am J Phys Anthropol 82: 1 7. Buikstra JE, Komar DA. 2008. Forensic a nthropology : Contemporary t heory and p ractice. Oxford : Oxford University Press. Bu ikstra JE, Ubelaker DH (editors) 1994. Standards for d ata c ollection from h uman s keletal r emains. Fayetteville, AR : Arkansas Archaeological Survey. Bulbeck D. 1982. A re evaluation of possible evolutionary processes in Southeast Asia since the late Pleistocene. Bull Indo Pac Pre Hi 3:1 21.
248 Bunge FM (editor). 1981. Thailand: A country study. Washington, D.C.: American University Foreign Area Studies. Bunnag J. 1984. The way of the monk and the way of the world: Buddhism in Thailand, Laos and Cambodia. In Bechert H, Gombrich R, editors. The world of Buddhism. London: Thames and Hudson. p 159 170. Cann RL, Wilson AC. 1983. Length mutations in human mitochondrial DNA. Genetics 104:699 711. Carson EA. 2006. Maximum likelihood estimation of human craniometric heritabilities. Am J Phys Anthropol 131: 169 180. Cavalli Sforza LL, Piazza A, Menozzi P, Mountain J. 1988. Reconstruction of human evolution: Bringing together genetic, archaeological, and linguistic data. Proc Natl Acad Sci USA 85:6002 6006. Cavalli Sforza, LL 1991. Genes, peoples and languages. Sci Am 265: 104 110. Central Intelligence Agency. 2010. The world factbook: Thailand. Electronic document, https://www.cia.gov/library/publications/the world factbook/geos/th.html Accessed 6 November 2010. Chang KC. 1969. Fengpitou, Tapenkeng and the prehistory of Taiwan. New Haven: Yale University Publications in Anthropology 73. Chang KC. 1970. Prehistoric archaeology of Taiwan. Asian Perspect 13:59 78. Chang KC. 1977. A new prehistoric ceramic style in the southeastern coastal area of China. Asian Perspect 20:17 9 182. Chang KC. 1986. The archaeology of ancient China, 4 th ed. New Haven: Yale University Press. Chang TT. 1976. The rice cultures. Phil Trans Roy Soc London B 275:143 155. Chappell J. 1994. Upper Quaternary sea levels, coral terraces, oxygen is otopes and deep sea temperatures. J Geogr 103 :828 840. Chappell J, Thom BG. 1977. Sea levels and coasts. In Allen J, Golson J, Jones R, editors. Sunda and Sahul: Prehistoric studies in Southeast Asia, Melanesia and Australia. p 275 292. Cheverud J M. 1988. A comparison of genetic and phenotypic correlations. Evolution 42:958 968.
249 Ciochon R, Long Vu The, Larick R, Gonzalez L, Grun R, de Vos J, Yonge C, Taylor L, Yoshida H, Reagan M. 1996. Dated co occurrence of Homo erectus and Gigantopithecus f rom Tham Khuyen Cave, Vietnam. Proc Natl Acad Sci 93:3016 3020. Mmoirs du Service XIII:1. Comas J. 1960. Manual of physical a nthropology. Springfield (IL): Charles C Thomas Conover WJ, Iman RL. Rank transformations as a bridge between parametric and non parametric statistics. Am Stat 35:124 129. Coon CS. 1966. The living races of man. London: Jonathan Cape. Corruccini RS. 1973. Size and shape in si milarity coefficients based on metric characters. Am J Phys Anthropol 38:743 754. Cuong NL. 1986. Two early Hoabinhian crania from Thanh Hoa province, Vietnam. Z Morth Anthropol 77:11 17. Dawkins R. 1976. The selfish gene. Oxford: Oxford University Press. Diamond J. 1999. Guns, germs, and steel: The fates of human societies. New York: WW Norton. Diamond J, Bellwood P. 2003. Farmers and their languages: The first expansions. Science 300:597 603. Diffloth G. 1991. Austro Asiatic languages. Encyclopedia Brittanic Macropaedia. 22:719 722. Dizon E, Detroit F, Semah F, Falgueres C, Hameau S, Ronquillo W, Cabanis E. 2002. Notes on the morphology and age of the Tabon Cave fossil Homo sapiens Curr Anthropol 43:660 666. Djuric M, Dj onic D, Nikolic S, Popovic D, Marinkovic J. 2007. Evaluation of the Suchey Brooks method for aging skeletons in the Balkans. J Forensic Sci 52: 21 23. Douglas MT. 1996 Paleopathology in Human Skeletal Remains from the Pre Metal, Bronze, and Iron Ages, Northeastern Thailand: Non Nok Tha database 1 Craniometric measurements. Unpublished PhD dissertation, Department of Anthropology, University of Hawaii, Honolulu. Elect ronic document, http://seasia.museum.upenn.edu/skeletal/ (web ver 1.0). Accessed 18 June 2011). Dubois E. 1922. The proto Australian fossil man of Wadjak, Java. Koninklijke Akademie van Wetenschapp en te Amsterdam B23:1013 1051. Eldredge N, Gould SJ. 1972. Punctuated equilibria: An alternative to phyletic gradualism. In Schopf TJM, editor. Models in Paleobiology. San Francisco: Freeman. p 82 115.
250 Feldesman MR. Bridging the chasm: Demystifyi ng some statistical methods used in biological anthropology. In Boaz NT, Wolfe LD, editors. Biological anthropology: The state of the science. Corvallis (OR): IIHER Publications, Oregon State University Press. p 73 100. Field A. 2009. Discovering s tatistics using SPSS, 3 rd ed. London: SAGE Publications. Fisher RA. 1925. Statistical methods for research workers. Edinburgh: Oliver and Boyd. Fisher RA. 1936. The use of multiple measurements in taxonomic problems. Ann Eugenic 7:179 188. Fongsat ikul L, Nantachit N, Kamtorn N, Leetrakool N. 1997. HLA gene frequencies of Northern Thais. J Med Assoc Thailand 80:S38 S42. France DL. 1983. Sexual dimorphism in the human humerus. Unpublished PhD dissertation, Department of Anthropology, University of Colorado, Boulder. Fucharoen G, Fucharoen S, Horai S. 2001. Mitochondrial DNA polymorphisms in Thailand. J Hum Genet 46:115 125. Giles E, Elliot O. 1962. Race identification from cranial me asurements. J Forensic Sci 7 :147 157. Giles E, Elliot O. 1963. Sex determination by discriminant function analysis of crania. Am J Phys Anthropol 21 :53 68. Glover IC. 1996. The southern Silk Road: Archaeological evidence for early trade between India and Southeast Asia. In Srisuchat A, editor. Ancient t rades and cultural contacts in Southeast Asia. Bangkok: National Culture Commission. p 57 94. Glover IC, Higham CFW. 1996. New evidence for early rice cultivation in South, Southeast and East Asia. In Harris DR, editor. The origins and spread of agr iculture and pastoralism in Eurasia: Crops, fields, flocks and herds. p 413 441. Gorman CF. 1969. Hoabinhian: a pebble tool complex with early plant associations in Southeast Asia. Science 163:671 673. Gorman CF. 1971. The Hoabinhian and after: Su bsistence patterns in Southeast Asia during the late Pleistocene and early Recent periods. World Archaeol 2:300 320. Gorman CF. 1972. Excavations at Spirit Cave, north Thailand: Some interim impressions. Asian Perspect 13:79 107. Gorman CF, Charoenwon gsa P. 1976. Ban Chiang: A mosaic of impressions from the first two years. Expedition 18:14 26. Gould SJ. 1996. The m ismeasure of m an. New York : W.W. Norton.
251 Hagelberg ES, Clegg JB. 1993. Genetic polymorphisms in prehistoric Pacific islanders det ermined by analysis of ancient bone DNA. Proc R Soc Lond B 252:163 170. Hamilton ME. 1982. Sexual dimorphism in skeletal samples. In Hall RL, editor. Sexual dimorphism in Homo sapiens. New York: Praeger. p 107 163. Hanihara T. 1992. Dental and cranial affinities among populations of East Asia and the Pacific: The basic populations in East Asia, IV Am J Phys Anthropol 88:163 182. Hanihara T. 1993. Population p rehistory o f East Asia and the Pacific as v iewed f rom c raniofacial m orphology : The b asic p opulations in East Asia, VII. Am J Phys Anthropol 91:173 187. Hanihara T. 1996. Comparison of c raniofacial f eatures of m ajor h uman g roups. Am J Phys Anthropol 99:389 412. Hanihara, T. 2006. Interpretation of craniofacial variation and diversificati on of East and Southeast Asian s. In Oxenham M, Tayles N, editors. Bioarchaeology of Southeast Asia. Cambridge: Cambridge University Press. p 91 107. Hashimoto OY. 1972. Studies in Yue dialects 1: Phonology of Cantonese. Cambridge: Cambridge University Press. Ha Van Tan. 1980. Nouvelles recherch prehistoriques et protohistoriques au Viet Nam. 54. Ha Van Tan. 1985. Prehistoric pottery in Viet Nam and its relationships with Southeast Asia. Asian Perspect 26:135 146. Ha Van Tan. 1997. The Hoabinhian and before. Bull Indo Pac Pre Hi 16 :35 41. Higham CFW. 1996. Archaeology and linguistics in Southeast Asia: Implications of the Austric hypothesis. Bull Indo Pac Pre Hi 14:110 118. H igham CFW. 2002. Early cultures of mainland Southeast Asia. London: River Books. Higham CFW, Bannanurag R (editors). 1990. The Excavation of Khok Phanom Di a prehistoric site in central Thailand. Volume 1: The excavation, chronology, and human bur ials. London: Society of Antiquaries. Higham CFW, Bannanurag R (editors). 1991. The Excavation of Khok Phanom Di a prehistoric site in central Thailand. Volume 2: The biological remains. London: Society of Antiquaries. Higham CFW, Kijngam A (ed itor). 1984. Prehistoric investigations in Northeast Thailand. Oxford: BAR International Series.
252 richest habitat. Forth Worth (TX): Harcourt Brace. Higham CFW, T hosarat R. 1998. Prehistoric Thailand: From early settlement to Sukhothai. London: River Books. Hoang Xuan Chinh, Nguyen Ngoc Bich. 1978. The excavation of the archaeological site of Phung Nguyen [in Vietnamese]. Hanoi: Nha Xuat Ban Khoa Hoc Xa H oi. 378. Hotelling T. 1933. Analysis of a complex of statistical variables into principle components. J Ed Psych. 24:417 441, 498 520. Howells WW. 1973. Cranial v ariation i n m an. Cambridge: Papers of the Peabody Museum of Archaeology and Ethnology. Vol. 67. Howells WW. 1989. Skull s hapes and the m ap. Cambridge: Papers of the Peabody Museum of Archaeology and Ethnology. Vol. 79. w ho in s kulls. Cambridge: Papers of the Peabody Museum of Archaeology and Ethnology. Vol. 82. Howells WW. 1997. Getting here: The story of human evolution. Washington, D.C.: Compass Press. a nthropometry. Philadelphia : Wistar Institute of Anatomy and Biology. Huberty CJ. 1984. Issues in the use and interpretation of discriminant analysis. Psych Bull 95:156 171. Hudson MJ. 1999. Ruins of identity : ethnogenesis in the Japanese I slands. Honolulu: University of Hawaii Press Humphrey LT. 1998. Growth patterns in the modern human skeleton. Am J Phys Anthropol 105: 57 72. Hunt DR, Albanese J. 2004. History and demographic composition of the Robert J. Terry Anatomical Collection. Am J Phys Anthropol 127:406 417. Hutterer KL. 1983. The natural and cultural history of Southeast Asian agriculture. Anthropos 78: 169 212. Hutterer KL. 1988. The prehistory of the Asian rainforests. In Denslow JS, Padoch C, editors. People of the tropical rain forest. Berkeley: Univers ity of California Press. p 63 72. can MY, Loth SR, King CA, Shihai D, Yoshino M. 1998. Sexual dimorphism in the humerus: a comparative analysis of Chinese, Japanese and Thais. Forensic Sci Int 98: 17 29.
253 Jacob T. 1967. Some problems pertaining to the racial history of the Indonesian region. Unpublished PhD Diss ertation, University of Utrecht, The Netherlands. Jantz RL. 2004. Metric description of Hispanic skeletons : a preliminary analysis. Proceedings of the American Academy of Forensic Sciences [AAFS]. 10: 295. Jantz RL, Jantz LM. 2000. Secular change in craniofacial morphology. Am J Hum Biol 12:327 338. Jantz RL Ousley SD 2005. FORDISC 3.0: Personal c omputer f orensic d iscriminant f unctions. Knoxville : University of Tennessee. Jantz RL, Owsley DW. 2001. Variation among early North American crania. Am J Phys Anthropol 114:146 155. Kachigan SK. 1991. Multivariate statistical analysis: A conceptual introduction. New York: Radius Press. Kamminga J, Wright RVS. 1988. The U pper Cave at Zhoukoudian and the origins of the mongoloids. J Hum Evol 17:739 767. Kaplan I. 1981. The society and its environment. In Bunge FM, editor. Thailand: A country study. Washington, D.C.: American University Foreign Area Studies. p 49 11 8. Kenyhercz MW, Pietrusewsky M, Damann F, Ousley SD. 2010. Craniometric variation within Southeast Asia. Proceedings of the American Academy of Forensic Sciences [AAFS]. 16: 395. King CA, MY, Loth SR. 1998. Metric and comparative analysis of se xual dimorphism in the Thai femur. J Forensic S ci 43: 954 958. King CA. 1999. Sex determination from Thai tibiae. Homo 50: 107 117. Kirk RL. 1982. Microevolution and migration in the Pacific. In Proceedings of the VI International Congress of Human Gen etics, Jerusalem. Human genetics, Part A: The unfolding genome. New York: Liss. p 215 225. Kirk RL. 1986. Human genetic diversity in Southeast Asia and the western Pacific. In Roberts DF and de Stefano GF, editors. Genetic variation and its maint enance. Cambridge: Cambridge University Press. p 111 133. Klein RG. 1999. The human career : Human biological and cultural origins. Chicago : University of Chicago Press. Komar DA, Grivas C. 2008. Manufactured populations : What do contemporary re ference skeletal collections represent? A comparative study using the Maxwell Museum Documented Collection. Am J Phys Anthropol 137:224 233.
254 Konigsberg LW, Ousley SD. 1995. Multivariate quantitative genetics of anthropometric traits from the Boas data. Hum Biol 67:481 498. Kramer A. 1993. Human taxonomic diversity in the Pleistocene: Does Homo erectus represent multiple hominid species? Am J Phys Anthropol 91:161 171. Krigbaum J, Datan I. 1999. The Deep Skull of Niah. Borneo 5:14 17. Krogman WM, Charles C Thomas. Lai Van Toi. 1999. The initial inhabitants of Co Loa [in Vietnamese]. Khao Co Hoc. 1998(3):39 54. Leong SH. 1991. Jenderam Hilir and the mid Holocene p rehistory of the west coast plain of peninsular Malaysia. Bull Indo Pac Pre Hi 10:150 160. Lertrit P, Poolsuwan S, Thosarat R, Sanpachudayan T, Boonyarit H, Chinpaisal C, Suktitipat B. 2008. Genetic history of Southeast Asian populations as revealed by ancient and modern human mitochondrial DNA analysis. Am J Phys Anthropol 137:425 440. Levene H. 1960. Robust tests for equality of variances. I n Olkin I, Ghurye SG, Hoeffding W, Madow WG, Mann HB, editors. Contributions to probability and s tatistics: Essays in h onor of Harold Hotelling Palo Alto (CA): Stanford University Press. p 278 292. Li Y, Brace CL, Gao Q, Tracer DP. 1991. Dimensions of the face in Asia in the perspective of Geography and Prehistory. Am J Phys Anthropol 85:269 279. Loofs W Charoen site, Lopburi Province, Thailand. Bull Indo Pac Pre Hi 16:199 211. de la Loubre S. 1693. A new historical relation of the Kingdom of Siam. London: Thomas H orne. Macintosh, NWG. 1978. The Tabon Cave mandible. Archaeol Phys Anthropol Ocean 13:143 159. Mahalanobis PC. 1936. On the generalized distance in statistics. Proc Nat Inst Sci Calcutta 12: 49 55. Martin R, Saller K. 1956. Lehrbuch der Anthropologie 3 rd ed Volume 3. Stuttgart : Gustav Fischer Verlag. Mason GM. 1991. The molluscan remains. In Higham CFW, Bannanurag R, editors. The excavation of Khok Phanom Di a prehistoric site in central Thailand. Vol. 2: The biological remains. p 249 319.
255 Mason GM. 1996. The micromolluscs. In Thompson GB, editor. The excavation of Khok Phanom Di a prehistoric site in central Thailand. Vol. 4: Subsistence and environment: the botanical evidence. Matsumura H. 2006. The population history of Southeast Asia viewed from morphometric analyses of human skeletal and dental remain s. In Oxenham M, Tayles N, editors. Bioarchaeology of Southeast Asia. Cambridge : Cambridge University Press. p 33 58. Matsumura H, Hudson M. 2005. Dental perspect ives on the population history of Southeast Asia. Am J Phys Anthropol 127:182 209. reply to Turner. Am J Phys Anthropol 130:458 461. Matsumura H, Pookajorn S. 2 005. A morphometric analysis of the l ate Pleistocene human skeleton from the Moh Khiew Cave in Thailand. Homo 56: 93 118. Matsumura H, Zuraina M. 1995. Metrical analysis of the dentition of Perak man from Gua Gunung Runtuh in Malaysia. Bull Natl Sci Mu s Tokyo 21(Series D):1 10. Matsumura H, Zuraina M. 1999. Metric analyses of an Early Holocene human skeleton from Gua Gunung Runtuh, Malaysia. Am J Phys Anthropol 109:327 340. McCorriston J, Hole F. 1991. The ecology of seasonal stress and the origins of agriculture in the Near East. Am Anthropol 93:46 69. Melton T, Peterson R, Redd AJ, Saha N, Sofro ASM, Martinson J, Stoneking M. 1995. Polynesian genetic affinities with Southeast Asian populations as identified by mtDNA analysis. Am J Hum Genet 57 :403 414. Mielke JH. Konigsberg LW, Relethford JH. 2006. Human biological variation. Oxford: Oxford University Press. Mijsberg WA. 1940. On a Neolithic Paleo Melanesian lower jaw found in a kitchen midden at Gua Kepah, province Wellesley, Straits Se ttlements. In: Proceedings of the 3 rd Congress of Prehistorians of the Far East, Singapore. p 100 118. Molnar S. 2002. Human variation: Races, types, and ethnic groups, 5 th ed. Upper Saddle River (NJ): Prentice Hall. Moore Jansen PM, Ousley SD, Jantz RL. 1994. Data c ollection p rocedures for f orensic s keletal material, 3 rd ed. Knoxville : University of Tennessee Forensic Anthropology Series. Moss ML. 1997a. The functional matrix hypothesis revisited. 1. The role of mechanotransduction. Am J Orth od Dentofac 112:8 11. Moss ML. 1997b. The functional matrix hypothesis revisited. 2. The role of an osseous connected cellular network. Am J Orthod Dentofac 112:221 226.
256 Moss ML. 1997c. The functional matrix hypothesis revisited. 3. The genomic thes is. Am J Orthod Dentofac 112:338 342. Moss ML. 1973. A functional cranial analysis of primate craniofacial growth. Symp IV Intl Cong Primatol. Vol 3: Craniofacial biology of Primates. Basel, Karger. p 191 208. Nakbunlung, S. 1994. Origins and b iolog ical a ffinities of the m odern Thai p opulation : An o steological p erspective. Unpublished PhD Dissertation, Department of Anthropology, University of Illinois, Champaign Urbana. Natapintu S. 1991. Archaeometallurgical studies in the Khao Wong Prachan Valley, Central Thailand. Bull Indo Pac Pre Hi 11:153 159. Norman J. 1985. A note on the Chinese duodenary cycle. In Thurgood G, Matisoff JA, Bradley D, editors. Linguistics of the Sino Tibetan area: The state of the art. Canberra: Australian National University. p 85 89. Norman J, Mei T. 1976. The Austroasiatics in ancient South China: Some lexical evidence. Monumenta Serica 32:274 301. Nei M, Roychoudhury AK. 1993. E volutionary relationships of human populations on a global scale. Molecular Biology and Evolution 10:927 943. 446 Oota H, Kurosaki K, Pookajorrn S, Ishida T, Ueda S. 2001. Genetic study of Paleolithic and Neolithic Southeast Asians. Hum Biol 73:225 231. Unpublished Mast Zealand. Unpublished PhD dissertation, Department of Anthropology, University of Otago, New Ze aland. Ott RL, Longnecker M. 2001. An introduction to statistical methods and data analysis, 5 th ed. Pacific Grove (CA): Duxbury. Oxenham M, Tayles N (editors) 2006. Bioarchaeology of Southeast Asia. Cambridge : Cambridge University Press. Pietrusew sky M. 1981. Cranial variation in early metal age Thailand and Southeast Asia studied by multivariate procedures. Homo 25: 1 26. Pietrusewsky M. 1984. Pioneers on the Khorat Plateau: The prehistoric inhabitants of Ban Chiang. J Hong Kong Archaeol Soc 10:90 106.
257 Pietrusewsky M. 1990. Craniofacial variation in Australasian and Pacific populations. Am J Phys Anthropol 82(3): 319 340. Pietrusewsky M. 1992. Japan, Asia and the Pacific: A multivariate craniometric investigation. In Hanihara K (editor) : International symposium on Japanese as a member of the Asian and Pacific p opulations. Kyoto: International Research Ce nter for Japanese Studies. p 9 52. Pietrusewsky M. 1994. Pacific Asian relationships: A physical anthropological perspective. Oceanic Linguist 33(2):407 429. Pietrusewsky M. 199 6. Multivariate c raniometric investigations of Japanese, Asians and Pacific Islanders. In Omoto K (editor ): Interdisciplinary perspectives on the o rigins of the Japanese. Kyoto: International Researc h Center for Japanese Studies. p 65 104. Pietrusewsky M. 2000. Metric a nalysis of s keletal r emains : Methods and a pplications. In Katzenberg MA, Saunders SR, editors. Biological a nthropology of the h uman s keleton. New York : Wiley Liss. p 375 415. Pie trusewsky M. 2005. The physical anthropology of the Pacific, East Asia, and Southeast Asia. In Blench R Sagart L Sanchez Mazas A, editors. The peopling of East Asia. New York: Routledge Curzon. p 201 229. Pietrusewsky M. 2006. A multivariate cran iometric study of the prehistoric and modern inhabitants of Southeast Asia, East Asia, and surrounding regions : A human kaleidoscope? In Oxenham M, Tayles N, editors. Bioarchaeology of Southeast Asia. Cambridge : Cambridge University Press. p 59 90. Pi etrusewsky M. 2008a. Craniometric variation in Southeast Asia and neighboring regions: A multivariate analysis of cranial measurements. Human Evol 23:49 86. Pietrusewsky M. 2008b. The modern inhabitants of island Southeast Asia: A craniometric perspe ctive. In Indriati E, editor. Proceedings of the international seminar on Southeast Asian paleoanthropology: Recent advances on Southeast Asian paleoanthropology and archaeology. Yogyakarta, Indonesia: Gadjah Mada University Faculty of Medicine. p 18 5 201. Pietrusewsky M, Douglas MT. 2002 a Ban Chiang, a p rehistoric v illage s ite in Northeast Thailand I : The h uman s keletal remains. Thai Archaeology Monograph Series Volume 1. Philadelphia: University of Pennsylvania Museum of Archaeology and Anthropology. Pietrusewsky M, Douglas MT. 2002 b Ban Chiang, a p rehistoric v illage s ite in Northeast Thailand I : The h uman s keletal remains, Appendix E, Database E1 Craniometric mea surements. Thai Archaeology Monograph Series Volume 1. Philadelphia: University of Pennsylvania Museum of Archaeology and Anthropology. Electronic document, http://seasia.museum.upenn.edu/skeletal/ (web ver 1.1). Accessed 18 June 2011).
258 Pietrusewsky M, Li Y, Shao X, Quyen NQ. 1992. Modern and n ear m odern p opulations of Asia and the Pacific : A m ultivariate c raniometric i nterpretation. In Akazawa T, Aoki K, Kimura T, editors. The e volution and d i spersal of m odern h umans in Asia. Tokyo : Hokusen sha. p 531 558. Pigott VC, Weiss AD, Natapintu S. 1997. The archaeology of copper production: Excavations in the Khao Wong Prachan Valley, central Thailand. In Ciarla R, Rispoli F, editors. Southeast 119 157. Pope GG. 1985. Evidence of early Pleistocene hominid activity from Lampang, northern Thailand. Bull Indo Pac Pre Hi 6:2 9. Pope GG, Frayer DW, Liangcharoen M, Kulas ing P, Nakabanlang S. 1978. Palaeoanthropological investigations of the Thai American expedition to northern Thailand (1978 1980): An interim report. Asian Perspect 21:147 163. Pope GG, Barr S, MacDonald A, Nakabanlang S. 1986. Earliest radiometrically dated artifacts from Southeast Asia. Curr Anthropol 27:275 278. Pookajorn S. 1990. Hoabinhian cave excavations in Ban Kao District, West Thailand. In Glover I, Glover E, edit ors. Southeast Asian archaeology 1986. Oxford: BAR International Series. p 11 28. Pookajorn S. 1991. Preliminary report of excavations at Moh Khiew Cave, Krabi Province, Sakai Cave, Trang Province and ethnoarchaeological research of hunter gatherer g roup, so Pookajorn S. 1994. Final Report of Excavations at Moh Khiew Cave, Krabi Province, Sakai Cave Trang Province and ethnoarcheological research of hunter gatherer gr oup, so called Rao CR. 1948. The utilization of multiple measurements in problems of biological classification. J Roy Stat Soc B10:159 203. Rao CR. 1952. Advanced statistical methods in biometrical research. New York: John Wiley. Rankin DR, : the addition of missing Asian data to aid racial determinations in forensic casework. Proceedings of the American Acade my of Forensic Sciences [AAFS]. 10: 288 289. Redd AJ, Takezake N, Sherry ST, McGarvey S, Sofro ASM, Stoneking M. 1995. Evolutionary history of the COII/tRNA Lys intergenic 9 bp deletion in human mitochondrial DNAs from the Pacific. Mol Biol Evol 12:604 6 15. Reid LA. 1994. Morphological evidence for Austric. Oceanic Linguistics 33:323 344.
259 Relethford JH. 1984. Morphological size and shape variation among local human populations. J Hum Evol 13:191 194. Relethford JH. 1994. Craniometric variation amo ng modern human populations. Am J Phys Anthropol 95 : 53 62. Relethford JH. 2002. Apportionment of global human genetic diversity based on craniometrics and skin color. Am J Phys Anthropol 118: 393 398. Relethford JH, Harpending HC. 1994. Craniometric variation, genetic theory, and modern human origins. Am J Phys Anthropol 95: 249 270. Renfrew C. 1987. Archaeology and linguistics. Cambridge: Cambridge University Press. Renfrew C, Bahn P. 1996. Archaeology: Theories, methods and practice. New York: Thames and Hudson. Reynolds TEG. 1992. Exca vations at Banyan Valley Cave, n orthern Thailand: A r eport on the 1972 s eason Asian Perspect 31:77 97. de Ribadeneyra, FM. 1601. Historia de las Islas del Archipielago y Reynos de la Grand China. Ba rcelona: G. Graells. Rinehart R. 1981. The historical setting. In Bunge FM, editor. Thailand: A country study. Washington, D.C.: American University Foreign Area Studies. p 1 47. Rispoli F. 1992. Preliminary report on the pottery of Tha Khae, Lo pburi, central Thailand. In Glover I, editor. Southeast Asian archaeology 1990. Hull (UK): University of Hull Centre for Southeast Asian Studies. p 129 142. Roseman CC. 2004. Detecting interregionally diversifying natural selection on modern human c ranial form by using matched molecular and morphometric data. Proc Natl Acad Sci USA 101:12824 12829. Roseman CC, Weaver TD. 2004. Multivariate apportionment of global human craniometric diversity. Am J Phys Anthropol 125: 257 263. Saidin M. 2006. Bu kit Bunuh, Lenggong, Malaysia: New evidence of late Pleistocene culture in Malaysia and Southeast Asia. In Bacus EA, Glover IC, Pigott VC, editors. Uncovering 65. Saha N, Mak JW, Tay SH. 1995. Popula tion genetic study among the Orang Asli (Semai Senoi) of Malayasia. Hum Biol 76:37 57. Sangvichien S. 1966a. Neolithic skeletons from Ban Kao, Thailand, and the problem of Thai origins. Curr Anthropol 7:234 235.
260 Sangvichien S. 1966b. A preliminary re port on non metrical characteristics of Neolithic skeletons found at Ban Kao, Kanchanaburi. J Siam Soc 54:1 8. Sangvichien S. Sirigaroon P, Jorgensen JB, Jacob T. 1969. Archaeological excavations in Thailand. Volume 3, part 2: The prehistoric Thai sk eletons. Copenhagen: Munksgaard. Sarich VM. 1997. Race and language in prehistory. In Clark GA, Willermet CM, editors. Conceptual issues in modern human origins research. New York: Aldine de Gruyer. p 392 410. Sauer CO. 1952. Agricultural origi ns and dispersals. New York: American Geographical Society. Schaefer MC Black SM. 2005. Comparison of ages of epiphyseal union in North American and Bosnian skeletal material. J Forensic Sci 50: 777 784. Schmidt W. 1906. Die Mon Khmer Vlker: ein B indeglied zwischen Vlkern Zentralasiens und Austronesiens. Braunschweig. Schurr TG, Wallace DC. 2002. Mitochondrial DNA diversity in Southeast Asian populations. Hum Biol 74:431 452. Shapiro SS, Wilk MB. 1965. An analysis of variance test for normal ity (complete samples). Biometricka 52:591 611 Sieveking GG. 1954. Excavations at Gua Cha, Kelantan, 1954. Federation Museums Journal 1 2:75 143. Sieveking GG. 1987. Gua Cha in the light of recent research. In Shuhaimi NH, editor. Kelantan Zaman A wal. Kota Bharu: Muzium Negeri Kelantan. p 73 86. Simmons RT, Graydon JJ, Sringam S. 1954. A blood group genetical survey in Thais, Bangkok. Am J Phys Anthropol 12:407 412. Sjvold T. 1984. A report on the heritiability of some cranial measurements and non metric traits. In van Vark GN, Howells WW, editors. Multivariate statistical methods in physical anthropology. Dorderecht: D. Reidel Publishing Co. p 223 246. Slice DE, : Implications for forensic identification. Proceedings of the American Academy of Forensic Sciences [AAFS] 10: 280 281. Sokal RR, Rohlf FJ. 1995. Biometry, 3 rd edition. New York: W.H. Freeman and Co. Solheim WG. 1968. Early bronze in Northeastern Thailand. Curr Anthropol 9:59 62. Solheim WG. 1972. An earlier agricultural revolution. Sci Am 206:34 41.
261 Srensen P. 1972. The Neolithic cultures of Thailand (and north Malaysia) and their Lungshanoid relationships. In Barnard N, editor. Early Chinese art and it s possible influence in the Pacific Basin, Volume 2. p 459 506. Srensen P. 1973. Prehistoric iron implements from Thailand. Asian Perspect 16:134 173. Srensen P. 1979. The Ongbah cave and its fifth drum. In Smith RB and Watson W, editors. Early S outheast Asia. p 78 97. Srensen P. 1988. The kettledrums from Ongbah cave, Kanchanaburi Province. In Srensen P, editor. Archaeological excavations in Thailand: Surface finds and minor excavations. Copenhagen: Scandinavian Institute of Asian Studies, Occasional Paper No. 1. p 95 156. Srensen P, Hatting T. 1967. Archaeological investigations in Thailand. Vol. II: Ban Kao, Part 1: The archaeological materials from the burials. Copenhagen: Munksgaard. Spielman RS, Smouse PE. 1976. Multivariate classification of human populations I: Allocation of Yanomama Indians to villages. Am J Hum Gen 28:317 331. Spradley MK. 2006. Biological a nthropological a spects of the African D iaspora: Geographic o rigins, s ecular t rends, and p lastic versu s g enetic i nfluences u tilizing c raniometric d ata. Unpublished PhD dissertation, Department of Anthropology, University o f Tennessee, Knoxville. Spradley MK, Jantz RL. 2011. Sex estimation in forensic anthropology: Skull versus postcranial elements. J Forensic Sci 56:289 296. Spriggs M. 1989. The dating of the island Southeast Asian Neolithic: an attempt at chronometr ic hygiene and linguistic correlation. Antiquity 63:587 613. SPSS, Inc. 2010. IBM SPSS statistics for Windows, release version 19.0.0. Chicago: SPSS Inc. Stark MT, Allen SJ. 1998. The transition to history in Southeast Asia: An introduction. Int J Hist Archaeol 2:163 174. Steadman DW, Haglund WD. 2005. The scope of anthropological contributions to human rights investigations. J Forensic Sci 50: 23 30. Stewart TD. 1979. Essentials of forensic anthropology. Springfield (IL): Charles C Thomas. Stinson S. 1985. Sex differences in environmental sensitivity during growth and development. Yrbk Phys Anthropol 28:123 147. Susa nne C. 1977. Heritability of anthropological c haracters. Hum Biol 49: 573 587. SWGANTH. 2011. The Scientific Working Group for Forensic Anthropology. Electronic document, http://www.swganth.org Accessed 13 August 2011.
262 Swisher CC, Curtis GH, Jacob T, Getty AG, Suprijo A, Widiasmoro. 1994. Age of the earliest known hominids in Java, Indonesia. Science 263:1118 1121. Tabachnick BG, Fidell LS. 2007. Using multivariate statistics, 5 th ed. Boston: Allyn & Bacon. Tayles NG. 1999. The excavation of Khok Phanom Di a prehistoric site in centr al Thailand. Volume 5: The people. London: Society of Antiquaries. Thompson GB. 1992. Archaeobotanical investigations at Khok Phanom Di, central Thailand. Unpublished PhD Dissertation, Department of Prehistory, Research School of Pacific Studies, Au stralian National University. Thompson GB. 1996. The excavation of Khok Phanom Di a prehistoric site in central Thailand. Volume 4: Subsistence and environment: The botanical evidence. London: Society of Antiquaries. Thorne A, Grn R, Mortimer G, Sp ooner NA, Simpson JJ, McCulloch M, Taylor L, Curnoe D. J Hum Evol 36:591 612. Trevor JC, Brothwell DR. 1962. The human remains of Mesolithic and Neolithic date from Gua Cha, Ke lantan. Fed Mus eum J 7 :6 22. Turner CG II. 1987. Late Pleistocene and Holocene population history of East Asia based on dental variation. Am J Phys Anthropol 73:305 321. Turner CG II. 1990. Major features of sundadonty and sinodonty including suggest ions about East Asian microevolution, population history, and late Pleistocene relationships with Australian Aborigines. Am J Phys Anthropol 82:295 317. Turner CG II. 2006. Dental morphology and the population history of the Pacific Rim and Basin: Comm entary on Hirofumi Matsumura and Mark J. Hudson. Am J Phys Anthropol 130:455 458. Turner CG II, Manabe Y, Hawkey DE. 2000. The Zhoukoudian Upper Cave dentition. Acta Anthropol Sin 19:253 268. Ubelaker DH. 2004. International research in forensic anth ropology. Proceedings of the American Academy of Forensic Sciences [AAFS] 10: 281. van Heekeren HR, Knuth E. 1967. Archaeological excavations in Thailand: Sai Yok. Copenhagen: Munksgaard. Van Vark GN, Dijkema J. 1988. Some notes on the origin of the Chinese people. Homo 143 148. Van Vark GN, Howells WW (editors). 1984. Multivariate statistics in physical anthropology. Dordrecht: Reidel Publishing.
263 von Koenigswald GHR. 1952. Gigantopithecus blacki von Koenigswald, a giant fossil hominid from t he Pleistocene of southern China. Anthropol Pap Am Mus Nat Hist 43:295 325. Waddington CH. 1942. Canalization of development and the inheritance of acquired characters. Nature 14:563 565. Walker PL. 2008. Sexing skulls using discriminant function analysis of visually assessed traits. Am J Phys Anthropol 136:39 50. Watanabe T. 1984. Origin and dispersal of rice in Asia. East Asian Cultural Studies 24:33 39. Weidenreich F. 1945. Giant early man from Java and South China. Anthropol Pap Am Mus N at Hist 40:1 134. White JC. 1982. Ban Chiang: The discovery of a lost Bronze Age. Philadelphia: University of Pennsylvania Press. White JC. 1995. Incorporating heterarchy into theory on socio political development: The case from Southeast Asia. In Ehrenreich RM, Crumley CL, Levy JE, editors. Heterarchy and the analysis of complex societies. Archaeological Papers of the American Anthropological Association, No. 6. Arlington (VA): American Anthropological Association. p 101 123. Winchester S. 2003 Krakatoa: The day the world exploded. New York: HarperCollins. Wolpoff M, Caspari R. 1997. Race and human evolution: A fatal attraction. Boulder: Westview Press. Wood WR. 1989. Ethnohistory and historical method. In Schiffer MB, editor. Archaeological Method and Theory, Vol. 2 Tucson: University of Arizona Press. p 81 109. Xing J, Watkins WS, Shlien A, Walker E, Huff CD Witherspoon DJ Zhang Y Simonson TS Weiss RB Schiffman JD, Malkin D, Woodward SR, Jorde LB. 2010. Toward a more uniform sampli ng of human genetic diversity: A survey of worldwide populat ions by high density genotyping. Genomics 96:199 210. Yen DE. 1977. Hoabinhian horticulture: The evidence and the questions from Northwest Thailand. In Allen J, Golson J, Jones R, editors. Sunda and Sahul: Prehistoric studies in Southeast Asia, Melanesia and Australia. London: Academic Press. p 567 99. Zegura SL. 1971. A multivariate analysis of the inter and intrapopulation variation exhibited by Eskimo crania. Unp ublished PhD Dissertation, Department of Anthropology, University of Wisconsin, Madison. Zide ARK, Zide NH. 1976. Proto Munda cultural vocabulary: Evidence for early agriculture. In Jenner PN, Thompson LC, Starosta S, editors. Austro Asiatic Studies, part II. Honolulu: Oceanic Linguistics Special Publication 13. p 1295 1334.
264 Zuraina M. 1994. The excavation of Perak man, an epi Paleolithic burial at Gua Gunung Runtuh. In Zuraina M, editor. The excavation of Gua Gunung Runtuh. Kuala Lumpur: Depa rtment of Museums and Antiquity.
265 BIOGRAPHICAL SKETCH Laurel Elizabeth Freas graduated summa cum laude with a Bachelor of Arts degree in anthropology and a rchaeology from Cornell University in 2001. After working as a park ranger for the National P ark Service at Walnut Canyon National Monument near Flagstaff, Arizona, Ms. Freas enrolled in the graduate program in the Department of Anthropology at the University of Florida. There, she served as a graduate analyst at the C. A. Pound Human Identificat ion Laboratory (CAPHIL). Upon earnin g her Master of Arts degree in a nthropology in 2005, Ms. doctoral program and was appointed Laboratory Manager of CAPHIL. While at the University of Florida Ms Freas also taught courses in human osteology, forensic anthropology and human a natomy. In 2009, Ms. Freas joined the scientific staff of the Joint P risoner of War /M issing I n A ction (POW/MIA) (JPAC CIL ), in Honolulu, Hawaii. At the JPAC CIL, Ms. Freas serves as a deploying forensic anthropologist and has led recovery missions to Laos and Cambodia in search of the remains of Americans missing and unaccounted icts. Ms. Freas is the recipient of a National Science Foundation Graduate Research Fellowship (2002 2005), the J. Lawrence Angel Student Paper Award (2004), the Forensic Science 2007), the Ellis R. Ke rley Scholarship (2007), a University of Florida Graduate Student Teaching Award (2007 2008) and the William R. Maples Scholarship (2009).