Ancestry from Cranial Morphology and FORDISC

William White Howells was a Harvard physical anthropologist whose career spanned the mid-twentieth century. Between 1965 and 1980, he measured 82 standard craniometric variables on more than 2,500 adult skulls drawn from 28 (later expanded to more than 30) population samples from across the globe, including European, African, East Asian, South Asian, Polynesian, Native American, and Australian Aboriginal groups. The data were published in three landmark monographs: "Cranial Variation in Man" (1973), "Skull Shapes and the Map" (1989), and "Who's Who in Skulls" (1995). The entire dataset was made freely available for research download.

The 28 landmark-derived measurements Howells standardised include: maximum cranial length (glabella to opisthocranion), maximum cranial breadth (euryons), basion-bregma height, bizygomatic breadth (zygions), minimum frontal breadth (frontotemporales), biauricular breadth, orbital height, orbital breadth (dacryon to ectoconchion), nasal height (nasion to nasospinale), nasal breadth (alares), and several more around the palate, mastoid, and occipital regions. The precision of these measurements, each repeated to 0.1 mm using sliding and spreading calipers, is what gives the database statistical coherence.

Why does a database built between 1965 and 1980 remain the foundation of forensic ancestry assessment in the 2020s? Two reasons. First, Howells's documentation standards were extraordinary: each skull was drawn from a documented skeletal collection with known geographic provenance, and the measurements were published with individual data records, allowing others to verify or extend the analysis. Second, no comparably large, comparably well-documented, globally representative metric dataset has replaced it, though several extensions have been built from it.

The limitation is equally important. Howells's samples were drawn from skeletal collections assembled in the late nineteenth and early twentieth centuries. They represent population groups as they existed at that time. They do not capture the secular trend in cranial dimensions over the twentieth century, which has shifted mean values measurably in several populations. They do not represent the genetic and morphological diversity of twenty-first century migrant populations. And they over-represent European and East Asian groups relative to South Asian, Middle Eastern, and sub-Saharan African variation.

For South Asian casework specifically, the absence of a dedicated South Asian reference sample in the original Howells database is a practical problem. A skull from rural Maharashtra or West Bengal will not map cleanly onto either the "Hindu" (Howells's single South Asian sample, drawn from a pre-Partition Punjabi skeletal collection) or the East Asian reference groups. The classification result may be technically correct relative to the loaded samples while being misleading about actual population affinity.

FORDISC (Forensic Discriminant Computer program) was developed by Stephen Ousley and Richard Jantz at the University of Tennessee Knoxville (UTK). Version 1.0 was released in 1993. Version 2.0 (1996) added the Forensic Data Bank (FDB), a collection of modern documented skeletal cases predominantly from US medical examiner contexts. Version 3.0 (2005) and the current Version 3.1 added additional FDB samples and expanded the reference options. The software runs discriminant-function analysis on the user's measurements against user-selected reference groups and returns a classification result with associated posterior probabilities and a typicality probability.

The discriminant-function analysis that FORDISC performs is a multivariate statistical procedure. Given a set of measurements on an unknown skull, the software calculates which reference group's centroid the unknown falls closest to in multivariate measurement space, weighted by the covariance structure of each reference group. The output has three components. The classification is the reference group to which the unknown is assigned (e.g. "White Female", "Black Male", "Hispanic Male"). The posterior probability is the probability that the unknown belongs to the classified group, given that it must belong to one of the groups loaded for the analysis (ranging from 0 to 1.0). The typicality probability (also called the F-probability or the p-value for the chi-square distance) indicates whether the unknown is actually a typical member of any of the loaded groups; a low typicality probability (below 0.05) means the skull is not a good fit for any reference group in the analysis.

The FDB component of FORDISC draws on skeletal cases submitted from US medical examiner offices since the late 1980s. By the 2005 expansion, it contained over 1,000 documented modern cases. The Spradley 2008 work at Texas State University expanded the FDB to include a significantly larger and more ethnically diverse modern US sample, with particular improvements in the Hispanic and multiracial representation that had been under-sampled in earlier versions. This mattered because Hispanic individuals in the US frequently classify ambiguously between "White" and "Hispanic" reference groups in earlier FORDISC versions, reflecting the admixed genetic history of Latin American populations rather than any deficiency in the measurement.

The FORDISC reference groups available in Version 3.1 include: American White Female, American White Male, American Black Female, American Black Male, American Hispanic Female, American Hispanic Male, American Indian Female, American Indian Male, Vietnamese Male, Japanese Female, Japanese Male, Chinese Female, Chinese Male, and several others drawn from the Howells world sample. The analyst selects which groups to include in any given analysis. Including all groups simultaneously is not always the best strategy: if the case context (geographic location, associated artefacts, approximate time period) suggests a likely origin region, loading only the relevant reference groups improves classification accuracy by removing unrelated groups from the discriminant space.

The 28 measurements Howells standardised fall into four anatomical clusters: vault measurements (maximum cranial length, maximum cranial breadth, basion-bregma height, cranial base length, porion-bregma height), facial measurements (bizygomatic breadth, upper facial height, minimum frontal breadth, total facial height), orbital and nasal measurements (orbital height, orbital breadth, nasal height, nasal breadth, interorbital breadth), and occipital and mastoid measurements (foramen magnum length, foramen magnum breadth, mastoid height, biauricular breadth, occipital chord, parietal chord, frontal chord).

In casework, the skull is rarely complete. A case skull may present with a missing or damaged zygomatic arch (eliminating bizygomatic breadth), a fractured orbital region (eliminating orbital measurements), or a fragmented basicranium (eliminating mastoid and foramen magnum measurements). FORDISC accommodates missing measurements: the discriminant analysis simply operates in the reduced measurement space. The statistical power of the classification decreases with each missing variable, but even a five-variable analysis can yield a useful posterior probability.

The measurement precision required matters. Howells's own intra-observer error studies showed that most caliper measurements have a technical error of measurement (TEM) of less than 1.5 mm for trained observers. An inter-observer TEM of 2-3 mm at a given landmark is typically tolerable for discriminant-function purposes, but systematic bias (for example, consistently measuring glabella-to-opisthocranion as maximum length when the actual maximum is behind glabella) can shift a classification. The Buikstra-Ubelaker (1994) landmark standardisation, adopted by both FORDISC and the broader forensic anthropology community, provides the working definitions.

For the purposes of South Asian casework, the craniometric work of V.R. Bhasin and colleagues (notably Bhasin et al. 2011, "Genetic and Biochemical Polymorphisms and Their Application in Physical Anthropology and Forensic Science in South Asia") provides supplementary metric data on Indian population groups. Manjunath (2013) focused specifically on South Indian skeletal collections, providing morphometric data that can be used to contextualise FORDISC outputs. Neither dataset is incorporated directly into FORDISC 3.1's FDB, but both provide published reference values against which a FORDISC classification for a suspected South Asian case can be evaluated.

Morphoscopic ancestry traits are non-metric, visually assessed anatomical features that vary between population groups. Unlike measurements, they are assessed on ordinal scales (absent, slight, moderate, pronounced) or nominal scales (present or absent). They are not precise in the way that a caliper reading is precise, but they capture variation in anatomical form that metric landmarks do not fully reflect.

Joseph Hefner's 2009 paper in the Journal of Forensic Sciences systematised the trait inventory that had developed over decades of forensic anthropological practice, producing a validated list of eleven cranial morphoscopic traits with known inter-population frequency distributions.

The anterior nasal spine (ANS) is the bony projection at the base of the nasal aperture. It ranges from absent (0) through slightly projecting (1) to steeply projecting (3). European-ancestry individuals tend toward pronounced ANS; many African-ancestry individuals tend toward slight or absent. The malar tubercle is a bony protuberance on the inferior surface of the malar (zygomatic) bone. It is common in East Asian and some Indigenous American populations and relatively infrequent in European-ancestry groups. The nasal bone shape, assessed from the frontal view, varies from a pinched, narrow form common in European-ancestry populations to a broader, rounded form more common in African-ancestry populations; a tent-shaped intermediate is common in East Asian and South Asian groups.

The post-bregmatic depression is a slight depression of the frontal bone just posterior to bregma, common in East Asian and Pacific Island populations and infrequent in European-ancestry populations. The palate shape (as assessed on the hard palate outline) varies from narrow and parabolic (more common in European-ancestry) to wide and hyperbolic (more common in African-ancestry), with intermediate shapes in East Asian and South Asian groups. The inter-orbital breadth contributes to the overall nasal region impression: narrow inter-orbital distances are associated with European-ancestry populations, wider with African-ancestry and some East Asian groups.

Additional Hefner traits include: inferior nasal aperture morphology (guttered, rounded, or sharp sill at the lower aperture border), nasal aperture width, nasal overgrowth (the presence of a nasal overhang or nasal bridge), supranasal suture (the persistence of the suture between the two nasal bones superior to their junction), and zygomaxillary suture shape.

Morphoscopic scoring is subject to inter-observer variability at a level that metrical analysis is not. Intra-class correlation coefficients (ICCs) for Hefner's traits between trained observers range from 0.60 to 0.85 in published reliability studies, with some traits (malar tubercle, post-bregmatic depression) showing ICCs nearer the lower bound. This is not a reason to exclude morphoscopic traits from a report, but it is a reason to report them with explicit acknowledgment of the inter-observer error literature and to use them as corroborative evidence rather than as the sole basis for a classification.

The nasal overgrowth and supranasal suture traits show relatively high inter-observer agreement (ICC above 0.80) and are worth prioritising when measurement data are limited by fragmentation.

The core issue is straightforward. FORDISC's discriminant functions remain the operational standard; population affinity is assessed alongside sex estimation from the skull and mandible and stature estimation as part of the full biological profile. FORDISC's discriminant functions are derived from reference groups that are predominantly North American and European in their post-1980s FDB component, supplemented by Howells's global samples. The FDB groups reflect the population diversity of US medical examiner caseloads in the 1980s to 2000s: predominantly White American, Black American, and Hispanic American, with smaller Indigenous American, East Asian, and Vietnamese samples. A skull from a population group not represented in the loaded reference samples will be forced into the nearest available category.

A South Asian skull from India, Pakistan, Bangladesh, or Sri Lanka presents a predictable calibration problem. The Howells "Hindu" sample (drawn from Punjab-region collections) is the closest reference, but it is a small sample (approximately 60 individuals) and represents a geographically narrow slice of the subcontinent's morphological variation. When South Asian skulls are run through FORDISC against the standard FDB groups, they not infrequently classify as "White" or as "Asian" (East Asian), depending on which measurements are most complete. This is not a random misclassification: South Asian cranial morphology is genuinely intermediate between several Howells reference groups, reflecting the subcontinent's position at the crossroads of human dispersal routes.

The practical response in South Asian forensic contexts is to use the available population-specific metric data as a calibration check. Mukherjee and Bandyopadhyay (1955) and the subsequent Indian anthropological metric literature provide craniometric data on Bengali, Punjabi, South Indian, and Tribal Indian skeletal series. Pan (1924) provided early skeletal metric data on South Indian populations. If a case skull's measurements place it within the 90 per cent confidence interval of one of these South Asian reference series, the FORDISC classification should be reported alongside that context, not in isolation.

East African populations present a similar calibration challenge. Hennessy and Stringer (2002), working with Sub-Saharan African skeletal collections at the Natural History Museum London, provided craniometric data that substantially expands on Howells's three African samples (Teita, Dogon, Zulu). East African populations, including Somali, Ethiopian, and East African Bantu groups, show cranial morphology that overlaps with both Howells's "African" groups and, at some variables, with South Asian and European groups. Classification accuracy for East African unknowns in FORDISC can be poor without loading an appropriate East African reference.

For indigenous populations in the Americas and Australia, the calibration problem is even more acute. The Howells samples for Indigenous American and Australian Aboriginal groups were drawn from nineteenth-century skeletal collections that are now subject to repatriation under NAGPRA (US) and the Aboriginal and Torres Strait Islander Heritage Protection Act (Australia). Many of these collections are no longer available for research augmentation. A FORDISC classification of an Indigenous American unknown as "White" (a common misclassification when the reference sample fails to capture the skull's actual affinity group) can have real consequences for the subsequent investigative direction.

The Spradley 2008 expansion of the FDB at Texas State addressed some of these limitations by adding modern documented skeletal samples that better represent recent US Hispanic and multi-ethnic diversity. However, the expansion did not add South Asian, South-East Asian (outside Vietnamese), or Sub-Saharan African samples in the proportions needed for high-confidence classification from those origin regions.

A structured ancestry-estimation workflow begins before any software is opened. The analyst documents available skeletal elements, notes any taphonomic alteration to measurement landmarks (porosity, animal gnawing, root etching, burning), records the context of recovery (geolocation, associated artefacts, estimated depositional period), and identifies which of the Howells variables can be reliably measured on the specific case skull.

Measurements are taken in duplicate with a reference digital sliding caliper and a Todd-type spreading caliper for biauricular and bizygomatic dimensions. Where two measurements of the same variable differ by more than 1.0 mm, a third measurement is taken and the median of three is recorded. The Buikstra-Ubelaker 1994 landmark definitions are used throughout.

The reference group selection for the FORDISC analysis should reflect case context. For a case recovered from India: include the Howells "Hindu" sample, the Howells East Asian samples (Japanese, Chinese), and Howells's three African samples as minimum coverage; if the FDB is also loaded, include Hispanic American and White American groups to complete the comparison space. For a case from sub-Saharan Africa: load all three Howells African samples, supplemented by any published East African craniometric data the analyst has access to; include East Asian and European comparison groups.

After FORDISC analysis, the analyst inspects three outputs in sequence. First, the typicality probability for the classified group: if it falls below 0.05, the skull is not a good fit for any loaded group and the analysis should be reported as inconclusive pending additional reference data. Second, the posterior probability: a value above 0.90 supports a strong affinity statement; a value between 0.60 and 0.89 supports a moderate affinity statement; below 0.60, the classification should be reported with explicit uncertainty. Third, the posterior probability of the next-best group: a difference of less than 0.15 between the first and second groups suggests genuinely ambiguous affinity and should be flagged.

Morphoscopic scoring follows the metric analysis. The traits are scored blind to the FORDISC result to avoid confirmation bias. Concordance between morphoscopic and metric classifications strengthens the report's foundation. Discordance triggers re-examination of both the measurements and the trait scores.

1. Pre-analysis documentation
   Record skeletal element inventory, taphonomic condition at each measurement landmark, case context (recovery location, associated evidence, estimated period). Establish which Howells variables are measurable.
2. Metric measurement
   Measure all available Howells variables in duplicate. Record to 0.1 mm. Use Buikstra-Ubelaker landmark definitions. Take medians where replicates diverge more than 1.0 mm.
3. Reference group selection
   Choose reference groups appropriate to case context. Do not load all groups by default. Include the most probable affinity groups plus three to four comparison groups from different world regions.
4. FORDISC analysis
   Run discriminant-function analysis. Record classification, posterior probability (classified group), typicality probability, and posterior probability of all loaded groups.
5. Morphoscopic scoring
   Score Hefner's eleven traits independently of the FORDISC result. Note which traits are assessable and which are obscured by damage. Record on a standardised trait score sheet.
6. Concordance evaluation
   Compare metric classification with morphoscopic trait profile. Note concordance or discordance. Where discordant, re-examine measurements and trait scores before drafting the affinity statement.
7. Population-affinity narrative
   Draft the report statement using posterior probability language: 'cranial morphology shows highest metric affinity to [group] with a posterior probability of [value]; typicality probability [value] indicates the skull is [a/not a] typical member of this group.' Do not use racial typology terms (Caucasoid, Mongoloid, Negroid).

The Howells world sample reflects the state of global physical-anthropological skeletal collection as of the mid-twentieth century, not an internationally agreed taxonomy of human groups. Each sample carries a geographic or institutional label that summarises the collection's provenance, not a complete description of the population.

The "Zulu" sample comprises individuals from a Zulu-speaking South African collection assembled in the early twentieth century. The "Dogon" sample comes from a West African Mali collection. The "Teita" sample comes from a Kenyan collection. Together they represent three distinct sub-Saharan African groups with meaningful morphological differences. Treating them as interchangeable "African" representatives in an analysis of a case from Lagos or Nairobi would be a significant calibration error.

The "Hindu" sample is a single North Indian (Punjab-region) collection. Treating it as representative of the entire Indian subcontinent suppresses real morphological variation between North Indian, South Indian, Northeast Indian, and Tribal Indian populations. For casework in South India, the Pan (1924) South Indian data and the Manjunath (2013) South Indian skeletal series provide more relevant references than the Howells "Hindu" sample.

The East Asian samples (Japanese, Chinese) are drawn from Japanese and Chinese skeletal collections. Both are well-represented (each over 130 individuals) and have been heavily validated. They are not, however, directly applicable to casework involving South-East Asian origins (Thai, Vietnamese, Cambodian, Filipino), where the available FORDISC reference is limited to the FDB Vietnamese sample.