Classification Trends Among Modern Filipino Crania Using Fordisc 3.1

Forensic Anthropology Vol. 2, No. 4: 1–11 DOI: 10.5744/fa.2019.1005 RESEARCH ARTICLE Classification Trends among Contemporary Filipino Crania Using Fordisc 3.1 Matthew C. Goa,b* ● Ansley R. Jonesa ● Bridget F. B. Algee-Hewittc ● Beatrix Dudzikd ● Cris E. Hughesa,e ABSTRACT: Filipinos represent a significant contemporary demographic group globally, yet they are underrepresented in the forensic anthropological literature. Given the complex population history of the Philippines, it is important to ensure that traditional methods for assessing the biological profile are appropriate when applied to these peoples. Here we analyze the classification trends of a modern Filipino sample (n = 110) when using the Fordisc 3.1 (FD3) software. We hypothesize that Filipinos represent an admixed population drawn largely from Asian and marginally from European parental gene pools, such that FD3 will classify these individuals morphometrically into reference samples that reflect a range of European admixture, in quantities from small to large. Our results show the greatest classification into Asian reference groups (72.7%), followed by Hispanic (12.7%), Indigenous American (7.3%), African (4.5%), and European (2.7%) groups included in FD3. This general pattern did not change between males and females. Moreover, replacing the raw craniometric values with their shape variables did not significantly alter the trends already observed. These classification trends for Filipino crania provide useful information for casework interpretation in forensic laboratory practice. Our findings can help biological anthropologists to better understand the evolutionary, population historical, and statistical reasons for FD3-generated classifications. The results of our study indicate that ancestry estimation in forensic anthropology would benefit from population-focused research that gives consideration to histories of colonialism and periods of admixture. KEYWORDS: forensic anthropology, admixture, ancestry estimation, postcolonialism, Fordisc 3.1, Philippines Introduction The estimation of ancestral affiliation of unidentified forensic skeletal cases is an integral part of the identification process. Not only does ancestry offer an avenue for narrowing down putative identifications, but knowing ancestry also further calibrates other biological profile components, such as age, sex, and stature. Ancestry can also be one of the most challenging of these inferred parameters. From a statistical standpoint, the classifications are conditional on the assumption that reference data sets capture the range of pertinent human variation for any given case. In actual practice, many groups remain underrepresented or absent in these data sets, and, Department of Anthropology, University of Illinois at UrbanaChampaign, Urbana, IL 61801, USA b SNA International, supporting the Defense POW/MIA Accounting Agency, Central Identification Laboratory, Joint Base Pearl Harbor– Hickam, HI 96853, USA c Center for Comparative Studies in Race and Ethnicity, Stanford University, Stanford, CA 94305, USA d DeBusk College of Osteopathic Medicine, Lincoln Memorial University, Harrogate, TN 37752, USA e Carl R. Woese Institute for Genomic Biology, University of Illinois at Urbana- Champaign, Urbana, IL 61801, USA *Correspondence to: Matthew C. Go, Department of Anthropology, University of Illinois at Urbana- Champaign, 109 Davenport Hall, 607 South Mathews Avenue, Urbana, IL 61801, USA E-mail: [email protected] a Received 26 December 2018; Revised 30 December 2018; Accepted 18 January 2019 © 2019 University of Florida Press because reference materials are opportunistically acquired, even large samples are often limited in coverage, so that classification analyses must operate under the unrealistic expectation of broad regional homogeneity. The increasing ethnic diversity of the United States and the growth of transnational metropolises around the world necessitate a more inclusive approach to forensic anthropological case methods. Best-practice recommendations for forensic anthropologists caution practitioners against the use of reference samples that are not representative of the unidentified skeletal remains in question, whether in terms of sex cohort, biogeographic population, or time period (Scientific Working Group for Forensic Anthropology 2012). However, in the applied context of medicolegal casework, rarely is the case subject to forensic anthropological evaluation drawn from a closed population where “ancestry” is a priori known. When case ancestry is truly unknown, reference samples that may or may not be representative of the unknown’s population of origin must be used to provide an estimate of ancestry for the individual case. While it is unrealistic for any ancestry method to include every possible population, we contend that it is also unnecessary when appropriate statistical tools and adequate reference populations are available for ancestry inference, as in the case of Fordisc (Jantz & Ousley 1993, 2005, 2012; Ousley & Jantz 1996, 2012). Fordisc 3.1 (FD3) makes it possible for a broad audience of forensic anthropologists to apply discriminant function 2 analysis to craniometric data from unidentified skeletal cases for the allocation of population membership using an unparalleled collection of forensically relevant and globally sourced reference samples (Jantz & Ousley 1993, 2005, 2012; Ousley & Jantz 1996, 2012). Beyond providing hard classifications for forensic cases into one of the available reference populations, FD3 also captures broad continental ancestry (Asian, African, European, Indigenous American) variation— yielding results comparable to hard classifications generated from unsupervised approaches to population inference (Algee-Hewitt 2016). This information, while not immediately diagnostic, can be highly useful for understanding the general ancestry composition of the individual case in question, as already demonstrated by Hughes et al. (2018) for Latin American samples. Therefore, the methods implemented via FD3 can be applicable in unidentified death case scenarios for estimating continental ancestry, even when the true population of origin is not represented by the current aggregate of reference samples. We argue that the nature of the variation reflected by the FD3 reference samples can be used to reveal continental ancestry patterns that can be informative of population membership and, in turn, be of value to forensic anthropological casework investigations. Here we provide an example of the utility of FD3 for assessing general continental ancestry of Filipino individuals. While the Howells series from the Philippines is available in FD3, we consider only those reference data that are forensically relevant, including temporally appropriate given concerns for secular change in the cranium (Jantz 2001; Weisensee & Jantz 2011; Wescott & Jantz 2005), and are believed to be similar in their ancestral makeup. To this latter point, Algee-Hewitt et al. (2018b) have observed differences in the mean quantities of trihybrid ancestry between the Howells and Hanihara samples and the contemporary individuals from the Manila North Cemetery (Go et al. 2017) studied here. Accordingly, we choose only from the individuals sourced from the Forensic Anthropology Data Bank (FADB) (Jantz & Moore-Jansen 1988). The contemporary subset of Asians included in FD3 contains only people of Chinese, Japanese, and Vietnamese origins or descent. By establishing the classification trends for our Filipino cases into the continental categories comprising the FADB groups in FD3, we can evaluate accuracy (defined here as the assignment of an individual case to one of the reference populations that would make up a larger continental-level Asian reference sample) and rates of error (defined here as the classification to any one of the populations that would not fall into this macrogeographic Asian reference sample) for inferred ancestry. We employ these contextual definitions of accuracy and rates of error to reflect how the FD3 outcomes would potentially lead investigators to make inferences of inclusion and exclusion based on continental ancestry. For example, if the FD3 group classification “accurately” associated with Fordisc 3.1 Classification Trends among Filipino Crania Asian continental ancestry, then the Philippines (along with other Asian populations) would be included as a potential source population for the unknown case. In contrast, if FD3 classified a case into a non-Asian group, the practitioner would likely exclude the Philippines (along with other Asian populations) as a potential source population for the unknown case. This study does not aim to provide general recommendations for the use of FD3 in ancestry estimation. Because of the complex population history of the Philippines, and thus the potential for highly heterogeneous cranial morphologies among present-day Filipinos, it is helpful to determine if FD3 classifications of Filipino crania are both consistent and sufficiently intuitive, such that they signify to the investigator possible Asian ancestry and imply that the Philippines should not be excluded as a possible source population. For this study, we posit that (1) the majority of the Filipino crania will be assigned into FD3 groups with East Asian and Southeast Asian ancestry, and (2) given Philippine colonial history specific to the Manila-based study sample, some proportion of these individuals will be alternatively allocated into groups with a limited proportion of European ancestry (e.g., classified as Hispanic). Misclassification of Hispanics using FD3 has previously been demonstrated and is likely associated with both first peopling and later Western colonial histories of Latin America (Dudzik & Jantz 2016; Hughes et al. 2018). The results of this study will provide a better understanding of classification and general ancestry estimation trends of FD3 for crania representing populations not explicitly captured by the available reference samples. Asians (defined as persons with ancestral origins in the Far East, Southeast Asia, or the Indian subcontinent) remain an understudied group despite making up 5.6% of the population recorded in the 2010 U.S. census (Hoeffel et al. 2012) and 60% of the global population. The increasing importance of Asians in U.S. demographics is easily demonstrated by the rapid shift in their share of the U.S. population at large. Within a decade from 2000 to 2010 the Asian population in every state except Hawai’i grew by at least 30%; 57% of Hawai’i’s population was composed of Asians by 2010 (Hoeffel et al. 2012). Although Asians represent the fastest growing racial group in the US, Filipinos in particular have received little to no attention in forensic anthropological literature (see Go 2018). This is further surprising given that Filipinos are the third-largest Asian demographic group in the United States. More than 3.4 million Americans report as having some degree of Filipino ancestry (24.4% of Asians in the United States), with more than 2.5 million identifying as solely Filipino (United States Census Bureau 2010). Filipinos are the most populous Asian group in most of the western half of the United States, including Alaska and Hawai’i. The Philippines also represents the third- and fifth-largest source country for documented and undocumented immigrants, Go et al. respectively (Baker & Rytina 2013, 2014), with 52% of Filipinos in the United States being foreign-born (United States Census Bureau 2010). Thinking more broadly, these trends are consistent, as in Canada, Filipinos rank first in number of permanent residents by source country (Citizenship and Immigration Canada 2015). The Philippines has experienced a unique Western colonization history relative to other Asian countries. The Philippines experienced over four centuries of consecutive colonial rule under Spain (1521–1898) and then the United States (1898–1946), 250 years of which saw regular trade routes between Latin America via the Manila-Acapulco galleon trade (1565–1815). Historical documents suggest intermarriages between Filipino indios, Latin Americans, Spanish, and Chinese were encouraged during Spanish colonization (De Mas 1843), although these pairings were likely most common in the capital and other major posts (Phelan 2011). During U.S. rule, intermarriages between Filipinos and American Whites in the Philippines continued (Winkelmann 2017). The flow of European genes into the archipelago was undoubtedly male biased owing to gendered colonial activities of subjugation via religious conversion, state exertion, and military expansion, or what has been called “bachelor colonization” (Molnar 2017). This history most certainly encouraged gene flow, among other microevolutionary processes that may not be immediately tractable from the morphological study of presently available Asian skeletal samples. Research exploring nuclear, mitochondrial, and Y- chromosome diversity have shown Filipinos to possess unique genetic histories relative to the surrounding region, not only reflecting initial colonization of the Asia-Pacific region, but also several waves of migration thereafter, including the postcolonial period (e.g., Banda et al. 2015; Bugawan et al. 1994; Delfin et al. 2011, 2014; Tabbada et al. 2010). Apart from anecdotal claims (Delfin 2015:450; Delfin et al. 2014:236; Howells 1989:110; Potter et al. 1981:34), no study has explicitly evaluated the degree of European genetic introgression in postcolonial Philippine populations. One study found that only 3.57% (1/28) of their small Filipino sample possessed a European Y-chromosome haplotype (Capelli et al. 2001). Using STRUCTURE to analyze a set of ancestry informative markers, another study by Yang et al. (2005) found overwhelming correspondence between predicted ancestry and self-identification within their entire Asian subgroup of 80 Koreans, Japanese, Chinese, and Filipinos. Only 2 out of 26 Filipinos showed large European contributions. Similar results were found by a larger study (total N = 103,006, Filipino n = 1,708), as they remark that “for self-reported Filipinos, a substantial proportion” of the ~10% exhibiting Asian-European admixture who self-reported as Asian have “modest levels of European genetic ancestry reflecting older admixture” (Banda et al. 2015:1293). It is worth noting that the latter two studies 3 sampled Filipinos living in California, while Capelli et al. (2001) do not provide more details on their sample apart from that they are “from the Philippines.” Conversely, a genetic admixture study by Rodriguez-Rodriguez et al. (2018) reported nearly one-third of people sampled from Guerrero, Mexico, particularly from the city of Acapulco, had greater than 5% and up to 10% Asian ancestry, which they found was most closely related to populations from the Philippines and Indonesia. Most recently, Algee-Hewitt et al. (2018b) report, using craniometrically derived continental ancestry proportions, that present-day Filipinos are notably admixed, as they carry about 20% less Asian ancestry than the mean quantity (90%) estimated for the other Asian groups, representing specifically peoples from Vietnam, Thailand, China (Hong Kong), Japan, and Korea, included in their study. Certainly, the degree of admixture across Philippine populations is likely highly varied across regional, temporal, and social lines of difference. Census counts demonstrate that the level of Spanish immigration to the colonial Philippines did not reach such heights as those with colonial Mexico (Barrows 1905:478; Phelan 2011), even into the American period (Table 1). However, the United States actually increased its military interests in the country even after granting independence in 1946. When U.S. military bases in the Philippines permanently closed in 1992 and their troops withdrew, an estimated 50,000-plus infants, children, and adolescents sired by U.S. soldiers were left orphaned and impoverished (Kutschera TABLE 1—Census Data on Race in the Phillipines during U.S. Colonization. Race* Brown Males Females Yellow Males Females White Males Females Black Males Females Mixed Males Females Unreported Total Males Females 1903 1918 1939 6,914,880 3,435,848 3,479,032 42,097 41,071 1,026 14,271 11,450 2,821 1,019 767 252 15,419 7,516 7,903 9,386,826 4,692,426 4,694,400 50,826 47,296 3,530 12,390 8,592 3,798 7,623 4,029 3,594 34,663 17,974 16,689 6,987,686 3,496,652 3,491,034 9,492,328 4,770,317 4,722,011 15,758,637 7,905,222 7,853,415 141,811 107,093 34,718 19,300 11,112 8,188 29,157 15,511 13,646 50,519 25,868 24,651 879 16,000,303 8,065,281 7,935,022 Note: data from the United States Bureau of the Census (1905), Census Office of the Philippine Islands (1920), and Commonwealth of the Philippine Commission of the Census (1941). *Racial terminology reflects those used in the original census, where brown refers to “Malay” Filipinos, yellow refers to East Asians such as Chinese and Japanese, white refers to Europeans, and black includes both Negritos and Africans. 4 et al. 2012). The estimate of these “biracial” Filipino “Amerasians” grows to 250,000 when adults and second-generation progeny are included (Kutschera & Caputi 2012). These postinstallation cities have continued informally sanctioned military prostitution systems today in the form of sex tourism hotspots catering to white men (Chapman 2017; Kutschera et al. 2015). Filipinos were the first historic Asian group to immigrate to the Americas, at first by escaping servitude aboard Spanish ships in 16th-century California (Cordova 1983), and then later by establishing the first Asian settlement in 18th-century Louisiana (Espina 1988). U.S. rule had only accelerated the transport of Filipinos to the United States (Espiritu 2003). In the early 1900s, sakadas (Filipino farmers) were exported to Hawai’i to work in sugar plantations, and later they were exported to California for similar agricultural needs. Government-sponsored pensionados were also sent to the United States during this time to be schooled in U.S. history and government. The U.S. occupation and onset of numerous 20th-century global conflicts also saw many Filipino men conscripted into U.S. military service. Filipinos immigrating to the United States during this period also faced strict anti-miscegenation laws (Baldoz 2004). Later capitalist booms, such as oil in the Gulf States in the 1970s, Asian Tiger economies in the 1980s, and health care and information technology industries in the 1990s, increased demand for cheap domestic and manual labor. Filipinos were encouraged by the government to pursue this demand abroad and provide a form of foreign remittance for the country. The tradition of deploying such Overseas Filipino Workers en masse remains strong today (Rodriguez 2010). The consequences of a unique colonial history and the strong Filipino presence within the United States and around the world at the present time necessitate a better grasp of Filipino skeletal variation for forensic anthropological investigations. To begin to understand such variation, we use, for the first time, the FD3 software as a tool to evaluate ancestry estimates relative to Filipino skeletal remains. Appreciation of these results will further the call for practicing forensic anthropologists to more fully comprehend the biological and statistical motivations for cranial (mis)classification trends in order to give the appropriate weight to or interpretation of the software’s output when assigning ancestry to unidentified remains. Materials and Methods The current test sample consists of mostly identified adult Filipino crania curated by the Archaeological Studies Program, University of the Philippines Diliman (Go et al. 2017). These individuals were accessioned from a large public cemetery in Manila, having been exhumed from low-cost niche Fordisc 3.1 Classification Trends among Filipino Crania tombs with unpaid burial maintenance fees. They represent cases that remain unclaimed by next of kin. The earliest individual birth year is 1911; the majority of individuals in this sample died in 2010 and 2011. Age at death ranged from 20 to 88 years, with an average of 52 years. All craniometric data were recorded by MCG. Measurements are among those employed by FD3, and their collection followed the most recent definitions used by the FADB (Langley et al. 2016). In cases of bilateral variables, the leftside measurement was used, substituting with the right side if the left side was absent. Two-sample t-tests assuming unequal variances showed no significant differences between left and right sides for every bilateral variable included in the study. A small number of fragmented crania were reassembled, in which case only reliable interlandmark distances were recorded. When a landmark was absent, associated measurements were not recorded. We explore FD3 classification trends within the context and traditional methodology of actual casework, whereby users are cognizant of its recommended guidelines that guard against model overfitting, especially with an unknown case. In order to avoid such overfitting, which arises from the inclusion of too many measurements with respect to the minimum group sample size, a maximum of nine standard cranial measurements were chosen via forward stepwise variable selection using Wilks’ lambda (Table 2). For the purposes of this article, we employ a common practice of limiting the number of variables used to 3m ≤ n, where m is the number of variables and n is the smallest group sample size (Huberty 1994). The most recent updates to the Fordisc help file relax this requirement to n – 1 variables (Ousley 2012:83). Given our current evaluation of a population not represented within the FADB groups, we follow the more conservative 3m ≤ n rule. Individual test cases with two or more missing variables out of the nine were omitted from analysis. Multivariate outliers flagged by FD3 and those individuals with two or more univariate outliers (less than or greater than three times the TABLE 2—Standard Cranial Measurements Used, Their Abbreviations, and Univariate Statistics for the Filipino Study Sample. Males (n = 69) Measurement Maximum cranial length Maximum cranial breadth Bizygomatic breadth Basion-bregma height Biauricular breadth Upper facial height Bimaxillary breadth Nasal breadth Orbital breadth Females (n = 41) Abbreviation Mean SD Mean SD GOL 175.07 6.95 167.46 7.19 XCB 141.10 5.74 136.34 5.08 ZYB BBH AUB UFHT ZMB NLB OBB 131.34 136.94 123.03 66.54 95.49 26.66 38.01 4.51 5.24 4.18 4.55 5.38 1.89 2.00 124.55 131.33 118.12 63.80 93.40 25.95 36.68 5.24 4.80 4.35 4.79 4.45 1.63 2.13 Go et al. 5 standard deviation) were also omitted. When an individual only had one outlying variable, it was run through FD3 with the outlying variable omitted. This resulted in a final Filipino crania sample size of 41 females (PHF) and 69 males (PHM); their univariate descriptive statistics are shown in Table 2. Each individual was run through FD3 (Version 3.1.314), and of the software-generated output, both the assigned membership to one among the available reference groups and the associated probabilities were recorded. The 13 FD3 reference groups originate from the FADB and include individuals with biogeographic ancestral ties to Europe, Africa, the Americas, and Asia (Table 3). More information about the provenience of each of these reference samples can be found in the Fordisc help file (Ousley 2012). Sexes of the Filipino test crania were treated as unknown, and therefore all FD3 reference groups were used for each case regardless of sex. We opted not to focus sex-specific categories for each Filipino test case to reflect the most conservative casework scenario where sex may be indeterminate. Therefore, the results of this study may differ if sex-specific analyses were performed in FD3. However, because so few female Asian reference samples are available, it may prove beneficial to include both male and female reference samples in the FD3 analyses of female Filipino test cases. Aside from regular interlandmark distances, shape-transformed values, for which the effect of size was reduced (Darroch & Mosimann 1985; Rosas & Bastir 2002), were also run through FD3 for each individual to account for scaling differences related to sex. Because FD3 assigns group membership to one of the reference groups included in the analysis, two probability measures are provided for evaluation. These values should be assessed simultaneously with the classification choice in order to gauge the strength of the classification (Ousley & TABLE 3—Fordisc Reference Groups Used in This Study and Their Abbreviations, Grouped into Broad Continental Ancestry Categories. Reference Group Abbreviation N African American Black Females American Black Males BF BM 34 60 European American White Females American White Males WF WM 165 340 Asian Japanese Females Japanese Males Chinese Males Vietnamese Males JF JM CHM VM 123 194 74 48 Indigenous American American Indian Females American Indian Males Guatemalan Males AF AM GTM 26 49 70 Hispanic Hispanic Females Hispanic Males HF HM 35 165 Jantz 2012). Posterior probabilities are measures of membership in each of the reference populations and, as proportions, must sum to one. As they are relative to the groups included in the function, they assume that the unknown belongs to one of these groups. Typicality probabilities “represent how likely an unknown belongs to a particular group, based on the average variability of all the groups in the analysis. Absolute distances are evaluated, rather than relative distances as in calculating [posterior probabilities]” (Ousley 2012:23). FD3 produces three measures of typicalities based on the F distribution, chisquare distribution, or ranked distances. This study uses the F distribution, which takes into account both the Mahalanobis distances and group sample sizes for each case. Recently, Konigsberg and Frankenberg (2018) have evaluated FD3’s calculation of typicalities from the F distribution, providing an alternative and what they state is the more appropriate equation. We use FD3-generated typicalities here, as this article focuses on the software’s outputs specifically. Lastly, the FADB craniometric data set was downloaded via the “Save Analyzed Data” option and combined with the Philippine sample in order to run a canonical variate analysis (CVA) using shape-transformed measurements in the software JMP® 10.0.0. Multivariate and univariate outliers as determined above were excluded. CVA was used in order to visualize group relationships and centroid trends in twodimensional space. Results For the overall model, the maximum total leave- one- out cross-validation rate acquired using nine variables was 50.6% for untransformed measurements and 40.2% for shapetransformed measurements. Generally, decreasing the number of variables used or removing the effects of size decreased the total cross-validation rate of the discriminant function, as expected (Ousley 2012). Of the results generated by FD3, this study focuses on the first and second population classification choice identified by the program (Tables 4 and 5), as well as the associated posterior and typicality (F distribution) probabilities for each case (Table 6 and Fig. 1). Regardless of sex, the majority of individuals classified into an Asian group (PHM = 72.5%, PHF = 73.2%), then the second-most- common group being Hispanic (PHM = 10.1%, PHF = 17.1%), third-most into an Indigenous American group (PHM = 7.2%, PHF = 7.3%), fourth-most as African (PHM = 5.8%, PHF = 2.4%), and least into a European group (PHM = 4.3%, PHF = 0.0%). Furthermore, over half of the 29 individuals (51.9%) who did not first classify as Asian had an Asian group as their second classification choice. Hispanics (HF and HM) and Indigenous Americans (GTM, AF, and AM) were also generally the next most common first or second choices after Asians. 6 Fordisc 3.1 Classification Trends among Filipino Crania TABLE 4—FD3 Classification Counts by First Then Second Choice. Filipino Males (n = 69) Untransformed Asian Hispanic Indigenous American African European Filipino Females (n = 41) Shape-Transformed Untransformed Shape-Transformed 1st Choice 2nd Choice 1st Choice 2nd Choice 1st Choice 2nd Choice VM JF CHM GTM JM HM AM JM VM JF GTM CHM VM JF WM HF JM VM GTM 6 5 4 1 1 1 6 5 1 1 7 2 2 1 3 2 1 1 VM 20 22 12 8 7 4 1 3 3 3 1 1 1 6 2 1 3 2 2 1 1 JF JF JF CHM GTM AM JM VM CHM HF GTM BF CHM JF WM JM JF VM GTM WM VM 7 CHM 1 HF VM GTM BF JM CHM JF CHM GTM AF JM VM JM WF HM AF GTM 2 1 1 1 1 1 JF HM AF GTM GTM BM WM 2 1 1 1 1 1 1 HF 7 JF VM AF HM JF CHM GTM CHM BM VM AF 3 1 2 1 GTM 3 1 1 1 BF 1 1 18 CHM 13 JM 12 JF 7 HM 4 HF 3 GTM 4 AF 1 BF 2 BM 2 WM 3 JM 9 CHM 9 HF 5 HM 3 2 1 1 1 GTM 4 AM 3 JF BM CHM HM 1 1 1 1 BM 2 BF 1 JF WM BM CHM HM BM 1 1 1 WM 1 BM 1st Choice 2nd Choice VM 16 JF 9 JM 3 CHM 3 JF CHM HF AF GTM JM VM CHM HF BF CHM JF JM 7 5 2 1 1 3 2 2 1 1 2 1 3 5 1 1 HF 5 JF VM 4 1 JF HM BF 1 1 1 GTM AM 2 1 HM AF 2 1 HF 1 BM 1 AM 1 WF 1 HF 1 8 6 3 3 1 1 4 1 1 1 1 TABLE 5—FD3 Classification Percentages (and Counts) Based on the First Choice. Males (n = 69) Asian Hispanic Indigenous American African European Females (n = 41) Sexes Pooled (N = 110) Untransformed Transformed Untransformed Transformed Untransformed Transformed 72.5% (50) 10.1% (7) 7.2% (5) 5.8% (4) 4.3% (3) 72.5% (50) 11.6% (8) 10.1% (7) 4.3% (4) 1.4% (1) 73.2% (30) 17.1% (7) 7.3% (3) 2.4% (1) 0.0% (0) 75.6% (31) 12.2% (5) 7.3% (3) 2.4% (1) 2.4% (1) 72.7% (80) 12.7% (14) 7.3% (8) 4.5% (5) 2.7% (3) 73.6% (81) 11.8% (13) 9.1% (10) 4.5% (5) 1.8% (2) Excluding the effects of size, shape-transformed measurements did not significantly alter the classification trends for either sex. Looking only within those individuals that classified as Asian and using untransformed values, Filipino males most commonly classified into the three Asian male reference groups (36.0% into VM, 26.0% into CHM, and 24.0% into JM) and least into JF (14.0%), while Filipino females most commonly classified into the sole Asian female reference group (73.3% into JF) and then into VM (23.3%) and CHM (3.3%). When using shape-transformed values, males and females now follow a shared trend, mostly classifying as VM (PHM = 40.0%, PHF = 51.6%), then JF (PHM = 24.0%, PHF = 29.0%), and equally into JM (PHM = 18.0%, PHF = 9.7%) and CHM (PHM = 18.0%, PHF = 9.7%). Using untransformed values, the median posterior probability was 0.42 (PHM = 0.38; PHF = 0.45) and the median Go et al. 7 TABLE 6—Median Posterior Probabilities (PP), Typicality Probabilities (TP), and Leave- One- Out Cross-Validation Rates (CV) per First Choice Classification Group. Males Untransformed VM CHM JM JF HM HF GTM AF AM BF BM WF WM Females Shape-Transformed Untransformed Shape-Transformed PP TP CV PP TP CV PP TP CV PP TP CV 0.468 0.427 0.385 0.355 0.175 0.411 0.308 0.437 0.564 0.510 0.565 0.724 0.532 0.714 0.336 0.458 0.347 0.581 0.441 0.423 0.451 0.375 0.648 0.667 0.454 0.352 0.371 0.325 0.244 0.257 0.410 0.488 0.923 0.780 0.644 0.658 0.846 0.378 0.541 0.149 0.256 0.459 0.444 0.167 0.620 0.383 0.332 0.384 0.565 0.612 0.500 0.501 0.436 0.676 0.521 0.372 0.231 0.329 0.506 0.975 0.705 0.455 0.490 0.027 0.410 0.691 0.602 0.304 0.854 0.450 0.167 0.535 0.298 0.345 0.837 0.622 0.139 0.401 0.307 0.331 0.510 0.192 0.002 0.604 0.700 0.971 0.562 0.328 0.629 0.410 0.451 0.055 0.469 0.943 0.549 0.246 0.420 0.306 0.514 0.261 0.689 0.687 0.579 0.825 0.507 0.458 0.124 0.848 0.229 0.271 0.152 FIG. 1—Distribution of posterior probability (PP) and typicality probability (TP) values for the Filipino test sample. typicality probability was 0.50 (PHM = 0.51, PHF = 0.48). Only seven (6.4%) of the test cases had posterior probabilities greater than 0.70, four classifying as JF and one each into VM, GTM, and WM, all of which had typicalities greater than 0.05. However, there is no required threshold for posterior probabilities, because they are relative to each included reference group (Ousley & Jantz 2012). Greater than 90% (100/110) of the Filipino test cases had typicality probabilities that exceeded the value of 0.05 (or 5%)—the threshold adopted here to signify questionable membership or measurement error (see Ousley 2012; Ousley & Jantz 2012). Using shape-transformed values, the median posterior probability was 0.36 (PHM = 0.36, PHF = 0.36) and the median typicality probability was 0.61 (PHM = 0.59, PHF = 0.64). Nine cases (8.2%) had posteriors greater than 0.70, eight classifying as VM, three of which with typicalities less than 0.05, and one as BM. As with regular measurements, greater than 90% (101/110) had typicalities exceeding 0.05. The Mahalanobis distance matrix and plot generated from the CVA are found in Table 7 and Figure 2, respectively. The first two canonical variables explain a cumulative total of 68.8% of the variation (43.8% and 25.0%, respectively). 8 Fordisc 3.1 Classification Trends among Filipino Crania TABLE 7—Mahalanobis Distance Matrix of Philippine and FD3 Reference Groups Based on Canonical Variate Analysis and Shape-Transformed Values. AF AM BF BM CHM GTM HF HM JF JM PHF PHM VM WF WM AF 0 3.09 7.06 7.68 7.88 3.37 3.30 3.71 5.57 6.43 9.90 10.52 7.80 11.78 9.47 AM BF BM CHM GTM HF HM JF JM PHF PHM VM WF WM 0 13.53 11.07 10.09 4.21 9.28 6.74 9.81 8.30 12.29 10.39 10.86 18.32 13.27 0 1.52 7.95 6.39 2.73 3.93 4.12 7.07 9.36 10.82 9.32 3.68 2.31 0 6.67 5.04 5.56 3.54 4.65 5.48 10.99 10.62 9.98 7.89 3.46 0 2.48 7.08 3.37 1.12 0.49 4.99 3.83 1.54 14.93 10.61 0 4.10 0.90 2.62 2.25 6.37 5.30 3.51 11.75 7.42 0 2.94 3.35 6.43 6.47 8.60 5.61 4.23 4.61 0 2.60 2.89 8.67 8.22 4.84 8.25 4.58 0 1.17 3.56 3.89 1.49 9.48 7.26 0 6.20 4.56 2.64 13.62 9.05 0 1.04 2.71 12.98 12.66 0 3.36 14.99 12.54 0 14.27 12.58 0 1.76 0 FIG. 2—Plot of group centroids based on the first two canonical variables and using shape-transformed craniometric values. Discussion Apart from visual methods for ancestry estimation, such as the use of macromorphoscopic traits (Hefner 2009; Rhine 1990), traditional craniometrics, as interlandmark distances, have been widely used to classify individuals into groups. Discriminant function analysis, which underlies the Fordisc software, has played a prominent role in the multivariate treatment of craniometric variables since the 1960s (Birkby 1966; Giles 1966; Giles & Elliot 1962). Owing to the nature of the statistical framework for this now-standard approach to ancestry estimation, classification success is tied to the assumption that the true population of origin for the unknown case is represented by the reference samples presently available in the FD3 software. In the forensic anthropology context, complete population representation is unrealistic on a worldwide scale. Therefore, it is useful instead to gain insight into the classification trends when populations not included in the reference samples are classified with discriminant function analysis. Do they adhere to likely classifications given their known population history? This study explores trends for Filipinos using FD3, noting that no Filipino population is currently represented among the program’s reference groups from the FADB. The majority of Filipino crania tested here classified into Asian groups regardless of sex. Among these Asian groups, Vietnamese Male was the most common classification when only incorporating shape differences, which is also concordant with the expectation of classification based on geographic proximity to the Philippines. When using shape and size, Filipino males classified most as Vietnamese Male and females as Japanese Female, indicating that sexual dimorphism is an important factor when considering population affinity. A negligible sex bias was observed. Both male and female Filipino crania were assigned Asian ancestry more than 70% of the time. After Asians, most individuals classified as Hispanic, and then Indigenous Americans to a lesser degree—a pattern likely owing to the shared Native American ancestry of those peoples who make up the social category of “Hispanic” and the weighted representation of Latinos of Mexican origin among the FADB Hispanics (Algee-Hewitt 2017a). Classification into the White Male or White Female group was the least likely result regardless of sex or measurement transformation, perhaps indicating that the majority of the potential Asian-European admixture present in this sample is captured through the Hispanic classification, as hypothesized. The high percentage of “typical” Filipinos suggests the FD3 reference samples collectively capture the cranial variation present within the test sample. Low to moderate posterior Go et al. probabilities for any given case also indicate that these posteriors are distributed across multiple reference samples, and no one reference group adequately represents Filipino variation. Typicalities greater than 0.05 and low posteriors could reflect the admixture represented in the cranial variation in that these individuals are falling into the region of overlap between multiple FD3 reference samples, similar to what is commonly seen with Hispanic individuals submitted to FD3 (Dudzik & Jantz 2016; Hughes et al. 2018; Spradley et al. 2008) and what unsupervised clustering has revealed for similarly admixed groups (Algee-Hewitt 2016, 2017a, 2017b; Algee-Hewitt et al. 2018a). Examining the CVA plot (Fig. 2), an oblique gradient from the top left to bottom right shows all Asian groups, then Indigenous Americans, then Hispanics, then American Blacks, and finally American Whites. The second canonical variate (accounting for 15.7% of the variation) assists in discriminating among the six Asian samples (VM, CHM, JM, JF, PHM, PHF). On this axis, we see a gradient that corresponds with latitudinal proximity of the Asian samples, with Japanes and Chinese samples more closely associated midplot, while the Filipinos are plotted in the upper quadrant and closest to the Vietnamese sample. Several limitations were imposed on this study’s design. First, only the program’s first— and to a lesser extent second— classification choice was considered here, but the distribution of posterior probabilities across the top three to four group may also be informative of potential admixture. Second, in a conservative approach, sexes and possible ancestries of the test cases were treated as unknown, and therefore every available group within the FADB was used for each case. Selection of only the relevant sex-specific reference groups per test case may produce different classification trends, but due to the results, particularly for females, it is unlikely that excluding male reference samples (and the majority of Asian reference groups with them) would improve outcomes. Third, eight or nine variables were input into the program depending on the completeness of each cranium. Classification trends may shift with the inclusion of more craniometric variables, most likely resulting in an increase in what are already robust correct classification rates for the Filipino sample. Fourth, reference groups were limited to those made available in the program sourced from the FADB. However, FD3 has the option of including 20th-century samples from the Howells database, which includes Filipino males that died before World War II. We did not include the Howells data, as they do not represent contemporaneous (i.e., forensically significant) groups in keeping with published statements on best practice for ancestry estimation (Scientific Working Group for Forensic Anthropology 2012). Finally, other, more nuanced analytical approaches to exploring ancestry, admixture, postcolonialism, and Filipino craniometrics 9 are needed. Work done by Algee-Hewitt et al. (2018b) using mixture analysis has revealed, in a model-bound but unsupervised way, that Filipinos are considerably more admixed compared to other Asian populations, with observable differences in admixture proportions between Philippine samples with differing provenience, including the Howells Philippine data set. Conclusion In our sample of 110 individuals, nearly three-fourths of the cases would have led to a conclusion of Asian ancestry using FD3 when their assignment is based on the first classification choice alone. The estimated ancestry for the remaining cases would have yielded potentially misleading identifications as, most notably, Hispanic, but also Indigenous American. Although there are reasonable population history explanations for these misclassifications, in a truly unknown casework context the incorporation of such prior expectations into the interpretive process may not be possible. Moreover, the generally low to moderate posterior probabilities even in cases of Asian classifications should cause the forensic anthropologist to question the reliability and, so, the utility of the results, and at the very least, to revisit the input data and the decisions made when running the analysis. In a real laboratory setting the analyst would likely opt to remove the most dissimilar group and rerun the analysis in a stepwise fashion in hopes of achieving a more satisfactory classification (Ousley & Jantz 2012). Recall that Filipinos are not currently represented as one of the reference groups. Furthermore, an actual “real case” laboratory assessment of ancestry would likely draw from multiple indicators in conjunction with craniometrics such as macromorphoscopic traits. This study does not aim to provide general recommendations for the use of FD3 in ancestry estimation. FD3 is used here specifically to provide information on the heterogeneity represented in modern Filipino cranial variation and the effect this diversity in morphology may have on correctly associating Philippine cases with continental Asian ancestry. Overall, Filipinos would likely and rightly classify as Asian, but with a small percentage classifying as Hispanic. These results warn against assumptions of group homogeneity for broad regional categories such as Asian, which itself represents multiple nations and ethnicities, each of which has undergone a unique history. Indeed, the variation in the Philippines is so complex that AlgeeHewitt et al. (2018b) have even found differences in admixture proportions by sample source, indicating that there is within-population structure that may be related to geography, ethnicity, and time. The classification trends presented here may help us better understand the evolutionary, population historical, and statistical reasons for FD3 results. They 10 also demonstrate how further research that gives due consideration to the effects of colonialism and admixture on ancestry estimation is warranted in forensic anthropology. References Algee-Hewitt BFB. Geographic substructure in craniometric estimates of admixture for contemporary American populations. American Journal of Physical Anthropology 2017a;164: 260–280. Algee-Hewitt BFB. Population inference from contemporary American craniometrics. American Journal of Physical Anthropology 2016;160:604– 624. Algee-Hewitt BFB. Temporal trends in craniometric estimates of admixture for a modern American samples. American Journal of Physical Anthropology 2017b;163:729–740. Algee-Hewitt BFB, Hughes CE, Anderson BE. Temporal, geographic, and identification trends in craniometric estimates of ancestry for persons of Latin American origin. Forensic Anthropology 2018a;1:4–17. Algee-Hewitt BFB, Hughes CE, Go MC, Dudzik B. An admixture approach to trihybrid ancestry variation in the Philippines with implications for forensic anthropology. Human Biology 2018b; Early View. doi: 10.13110/humanbiology.90.3.01. Baker B, Rytina N. Estimates of the lawful permanent resident population in the United States: January 2013. United States Department of Homeland Security; 2014. Baker B, Rytina N. Estimates of the unauthorized immigrant population residing in the United States: January 2012. United States Department of Homeland Security; 2013. Baldoz R. Valorizing racial boundaries: Hegemony and conflict in the racialization of Filipino migrant labour in the United States. Ethnic and Racial Studies 2004;27:969–986. Banda Y, Kvale MN, Hoffmann TJ, Hesselson SE, Ranatunga D, Tang H, et al. Characterizing race/ethnicity and genetic ancestry for 100,000 subjects in the Genetic Epidemiology Research on Adult Health and Aging (GERA) Cohort. Genetics 2015; 200:1285–1295. Barrows DP. History of the population. In: Census of the Philippine Islands. Volume I— Geography, History, and Population. Washington, DC: United States Bureau of the Census; 1905: 411– 491. Birkby WH. An evaluation of race and sex identification from cranial measurements. American Journal of Physical Anthropology 1966;24:21–27. Bugawan TL, Chang JD, Klitz W, Erlich HA. PCR/oligonucleotide probe typing of HLA class II alleles in a Filipino population reveals an unusual distribution of HLA haplotypes. American Journal of Human Genetics 1994;54: 331–340. Capelli C, Wilson JF, Richards M, Stumpf MPH, Gratrix F, Oppenheimer S, et al. A predominantly indigenous paternal heritage for the Austronesian-speaking peoples of insular Southeast Asia and Oceania. American Journal of Human Genetics 2001;68: 432– 443. Census Office of the Philippine Islands. Census of the Philippine Islands: Taken under the direction of the Philippine Legislature in the year 1918. Manila: Bureau of Printing; 1920. Chapman L. “Just being real”: A post-colonial critique on Amerasian engagement in Central Luzon’s sex industry. Asian Journal of Women’s Studies 2017;23:224–242. Citizenship and Immigration Canada. Facts and figures— Immigration overview: Permanent residents 2014. Ottawa, ON: Minister of Public Works and Government Services Canada; 2015. Fordisc 3.1 Classification Trends among Filipino Crania Commonwealth of the Philippines Commission of the Census. Census of the Philippines: 1939, Volume II. Manila: Bureau of Printing; 1941. Cordova F. Filipinos, Forgotten Asian Americans: A Pictorial Essay, 1763– circa 1963. Dubuque, IA: Kendall/Hunt Publishing Company; 1983. Darroch JN, Mosimann JE. Principal components of shape. Biometrika 1985;72:241–252. Delfin FC. The population history of the Philippines: A genetic overview. Philippine Studies: Historical and Ethnographic Viewpoints 2015;63:449– 476. Delfin F, Ko AM-S, Li M, Gunnarsdóttir ED, Tabbada KA, Salvador JM, et al. Complete mtDNA genomes of Filipino ethnolinguistic groups: A melting pot of recent and ancient lineages in the Asia-Pacific region. European Journal of Human Genetics 2014;22:228–237. Delfin F, Salvador JM, Calacal GC, Perdigon HB, Tabbada KA, Villamor LP, et al. The Y-chromosome landscape of the Philippines: Extensive heterogeneity and varying genetic affinities of Negrito and non-Negrito groups. European Journal of Human Genetics 2011;19:224–230. De Mas S. Informe sobre el estado de Las Islas Filipinas en 1842. Tomo III. Politica interior. Madrid: s.n.; 1843. Dudzik B, Jantz RL. Misclassifications of Hispanics using Fordisc 3.1: Comparing cranial morphology in Asian and Hispanic populations. Journal of Forensic Sciences 2016;61:1311–1318. Espina M. Filipinos in Louisiana. New Orleans: A.F. Laborde; 1988. Espiritu YL. Home Bound: Filipino American Lives across Cultures, Communities, and Countries. Berkeley: University of California Press; 2003. Giles E. Statistical techniques for sex and race determination: Some comments in defense. American Journal of Physical Anthropology 1966;25:85– 86. Giles E, Elliot O. Race identification from cranial measurements. Journal of Forensic Sciences 1962;7:147–157. Go MC. Appraising forensic anthropology in the Philippines: Current status and future directions. Forensic Science International 2018;288:329.e1–329.e9. Go MC, Lee AB, Santos JAD, Vesagas NMC, Crozier R. A newly assembled human skeletal reference collection of modern and identified Filipinos. Forensic Science International 2017;271: 128.e1–128.e5. Hefner JT. Cranial nonmetric variation and estimating ancestry. Journal of Forensic Sciences 2009;54:985–995. Hoeffel EM, Rastogi S, Kim MO, Shahid H. The Asian population: 2010. United States Census Bureau; 2012. Howells WW. Skull shapes and the map: Craniometric analyses in the dispersion of modern Homo. Papers of the Peabody Museum of Archaeology and Ethnology 79. Cambridge, MA: Harvard University; 1989. Huberty CJ. Applied Discriminant Analysis. New York: John Wiley & Sons; 1994. Hughes CE, Dudzik B, Algee-Hewitt BFB, Jones A, Anderson BE. Understanding (mis)classification trends of Hispanics in Fordisc 3.1: Incorporating cranial morphology, microgeographic origin, and admixture proportions for interpretation. Journal of Forensic Sciences 2018; Early View. doi: 10.1111/1556- 4029.13893. Jantz RL. Cranial change in Americans: 1850–1975. Journal of Forensic Sciences 2001;46:784–787. Jantz RL, Moore-Jansen PH. A data base for forensic anthropology: Structure, content and analysis. Department of Anthropology Report of Investigations No. 47. University of Tennessee, Knoxville; 1988. Jantz RL, Ousley SD. FORDISC 1.0: Personal computer forensic discriminant functions. University of Tennessee, Knoxville; 1993. Go et al. Jantz RL, Ousley SD. FORDISC 3.1: Personal computer forensic discriminant functions. University of Tennessee, Knoxville; 2005. Jantz RL, Ousley SD. Introduction to Fordisc 3. In: Tersigni-Tarrant MA, Shirley NR, eds. Forensic Anthropology: An Introduction. Boca Raton, FL: CRC Press; 2012:253–269. Konigsberg LW, Frankenberg SR. Typicality and predictive distributions in discriminant function analysis. Human Biology 2018;90:31– 44. Kutschera PC, Caputi MA. The case for categorization of military Filipino Amerasians as diaspora. In: Proceedings of the 9th International Conference on the Philippines, October 28–30, 2012; East Lansing, MI. Kutschera PC, Pelayo JMG III, Talamera-Sandico MG. The continuing conundrum of Southeast Asia’s 50,000 Filipino military “Amerasians.” International Proceedings of Economics Development and Research 2012;48:62– 67. Kutschera PC, Tesoro EC, Talamera-Sandico MG, Pelayo III JMG. Modern day second generation military Filipino Amerasians and ghosts of the U.S. military prostitution system in west central Luzon’s “AMO Amerasian Triangle.” International Journal of Social, Behavioral, Educational, Economic, Business and Industrial Engineering 2015;9:3356–3366. Langley NR, Meadows Jantz L, Ousley SD, Jantz RL, Milner G. Data Collection Procedures for Forensic Skeletal Material 2.0. Knoxville: University of Tennessee; 2016. Molnar NT. American Mestizos, the Philippines, and the Malleability of Race: 1898–1961. Columbia: University of Missouri Press; 2017. Ousley SD. Fordisc help file. Version 1.48. http://math.mercyhurst .edu /~sousley/Fordisc/Help/Fordisc3_Help.pdf. Created 2012. Accessed September 13, 2018. Ousley SD, Jantz RL. FORDISC 2.0: Personal computer forensic discriminant functions. University of Tennessee, Knoxville; 1996. Ousley SD, Jantz RL. Fordisc 3 and statistical methods for estimating sex and ancestry. In: Dirkmaat DC, ed. A Companion to Forensic Anthropology. West Sussex, UK: Wiley-Blackwell; 2012:311–329. Phelan JL. The Hispanization of the Philippines: Spanish Aims and Filipino Responses, 1565–1700. Madison: University of Wisconsin Press; 2011. Potter RHY, Alcazaren AB, Herbosa FM, Tomaneng J. Dimensional characteristics of the Filipino dentition. American Journal of Physical Anthropology 1981;55:33– 42. Rhine S. Non-metric skull racing. In: Gill GW, Rhine S, eds. Skeletal Attribution of Race: Methods for Forensic Anthropology. Albuquerque, NM: Maxwell Museum of Anthropological Papers, No. 4; 1990:9–20. 11 Rodriguez RB. Migrants of Export: How the Philippine State Brokers Labor to the World. Minneapolis: University of Minnesota Press; 2010. Rodriguez-Rodriguez JE, Blanco-Portillo J, Ioannidis A, MorenoEstrada A. Reconstructing admixture and migration dynamics of post-contact Mexico. In: Proceedings of the 87th Annual Meeting of the American Association of Physical Anthropologists, April 11–14, 2018; Austin, TX. Rosas A, Bastir M. Thin-plate spline analysis of allometry and sexual dimorphism in the human craniofacial complex. American Journal of Physical Anthropology 2002;117:236–245. Scientific Working Group for Forensic Anthropology. Statistical methods. https://www.nist.gov/sites/default /files/documents /2018/03/13/swganth_statistical_methods.pdf. Created 2012. Accessed November 1, 2018. Spradley MK, Jantz RL, Robinson A, Peccerelli F. Demographic change and forensic identification: Problems in metric identification of Hispanic skeletons. Journal of Forensic Sciences 2008;53:21–28. Tabbada KA, Trejaut J, Loo J-H, Chen Y-M, Lin M, Mirazon-Lahr M, et al. Philippine mitochondrial DNA diversity: A populated viaduct between Taiwan and Indonesia? Molecular Biology and Evolution 2010;27:21–31. United States Bureau of the Census. Census of the Philippine Islands: Taken under the Direction of the Philippine Commission in the Year 1903. Washington, DC: United States Bureau of the Census; 1905. United States Census Bureau. Race Reporting for the Asian Population by Selected Categories: 2010. 2010 Census Summary File 1. United States Census Bureau; 2010. Weisensee KE, Jantz RL. Secular changes in craniofacial morphology of the Portuguese using geometric morphometrics. American Journal of Physical Anthropology 2011;145(4):548–559. Wescott DJ, Jantz RL. Assessing craniofacial secular change in American Blacks and Whites using geometric morphometry. In: Slice DE, ed. Modern Morphometrics in Physical Anthropology. Boston: Springer; 2005:231–245. Winkelmann TO. Rethinking the sexual geography of American empire in the Philippines: Interracial intimacies in Mindanao and the Cordilleras, 1898–1921. In: Choy CC, Wu JT- C, eds. Gendering the Trans-Pacific World. Leiden, The Netherlands: Koninklijke Brill NV; 2017:39–76. Yang N, Li H, Criswell L, Gregersen PK, Alarcon-Riquelme ME, Kittles R, et al. Examination of ancestry and ethnic affiliation using highly informative diallelic DNA markers: Application to diverse and admixed populations and implications for clinical epidemiology and forensic medicine. Human Genetics 2005;118:382–392.