David Bulbeck
School of Archaeology and Anthropology, The Australian National University,
Canberra, Australia
Date of Document: December 2005
 

Introduction

Late Pleistocene human skulls are of immense interest in reconstructing the evolutionary history of our species Homo sapiens. In the 1980s and 1990s, Kamminga and Wright (1988), Wright (1992), Stringer (1994), Howells (1995), Powell and Neves (1999), and others used multivariate statistics to compare the measurements of late Pleistocene skulls, from China, Africa, Australia, and the Americas, against modern human populations as represented by the worldwide survey of William White Howells (see map at the bottom of this page). One general result that emerged, which was then interpreted as evidence for the "Out of Africa" theory on the origins of modern humans, is that the Pleistocene crania frequently did not find their closest affinity with Howells's recent populations in the same region. For instance, the landmark paper of Kamminga and Wright (1988) concluded that Zhoukoudien Upper Cave 1 could not be considered "Mongoloid" because its craniometrics are very different from those of recent populations in the region such as Chinese and Buriats. Therefore, one component of my The Contribution of South Asia to the Peopling of Australasia project (with Colin Groves) has involved testing Australian fossil crania with FORDISC 2.0 (Ousley and Jantz 1996), a computer program that facilitates multivariate statistical comparisons of individual crania with the 28 populations measured by Howells (1989).

However, the assumption that a skull should be classified with one of the populations of its region is not necessarily well-founded. When we enter all 21 measurements that FORDISC 2.0 uses in its module that compares the measured skull against Howells's populations, only 69.1% (males) and 70.2% (females) of the skulls in Howells's reference populations are correctly classified to their population by FORDISC. Roughly speaking, we may expect around 30% of skulls to be mis-classified by FORDISC. Of course, many of those "mis-classifications" would be between closely related populations, for instance North Japanese and South Japanese, so if we were only interested in the general affinities of a particular skull (e.g., whether it is of East Asian ancestry) we may expect a higher rate of correct classifications. On the other hand, there is no guarantee that populations other than those measured by Howells would be particularly similar in their craniometrics to a Howells population from the same region. For instance, Vietnamese crania might happen to be metrically more like Europeans or American Indians, or some other non-East Asian grouping, than are the East Asian populations measured by Howells, with the result that a substantial proportion might be classified by FORDISC as non-East Asian.

My methodology to deal with the "mis-classification" issue is to develop a FORDISC profile for a population by entering the measurements of large numbers of individual skulls. This methodology allows us to determine the range of FORDISC classifications that might be expected of a single, unknown skull if it were a member of that same population. My focus is on Australian crania, and accordingly I have tested 447 recent mainland Australian crania with FORDISC. With such a large database, we would be in a strong position to determine whether the suggested FORDISC affinity of an unknown skull is, or is not, consistent with a recent Australian affinity, based on how that affinity compares with the recent Australian profile. In particular, we can apply this methodology to an understanding of fossil Australian crania (individuals and samples), and to an understanding of how well-defined the Australian craniometric pattern is, by developing similar profiles for recent, non-Australian populations.

Methodological details

As already noted, a maximum of 21 measurements can be entered into FORDISC 2.0 for statistical comparison with Howells's 28 populations. These measurements are glabella-opisthocranion length, maximum biparietal breadth, basion-bregma height, nasion-basion length, nasion-prosthion length, biauricular breadth, nasion-bregma chord, bregma-lambda chord, lambda-opisthion chord, foramen magnum length, mastoid height, upper facial breadth, bizygomatic breadth, upper facial height, nasal height, nasal breadth, orbital height, orbital breadth, biorbital breadth, interorbital breadth, and external palate breadth (for details, see Ousley and Jantz 1996). In the case of most skulls, not all 21 measurements are available, because the recorder did not take that measurement and/or because the skull is incomplete. The criterion I employ here is to use the measurements for a skull when at least 50% (to the nearest percent) of the reference Howells specimens are correctly classified by FORDISC. This criterion allows some leeway for incomplete measurement sets, but also prevents reliance on the classification of a skull when only an inadequate menu of measurements (e.g., less than ten) is available. (In the discussion that follows, it should be understood that the FORDISC tests I refer to are those where the user selects all the available Howells populations for comparison - 28 in the case of males, and 26 in the case of females.)

The great majority of the cranial measurements that I have entered into FORDISC have been measured by other observers. This raises the issue of interobserver error, which I have addressed elsewhere (see Interobserver Comparability in FORDISC 2.0). In particular, the measurements of Peter Brown, Halina Milicerowa, Michael Pietrusewsky (excluding one problematic measurement), Daniel Rayner, and Pathmanathan Raghavan are shown to be suitable for entry into FORDISC. I have not specifically tested the measurements of Tsunehiko Hanihara with my proposed "means test", but it is obvious from the results for the individual specimens measured by Hanihara that his measurements are extremely suitable for entry into FORDISC. Data for the 447 recent Australian crania tested here come from Brown (n.d.), Milicerowa (1955) and Hanihara (unpublished - see Acknowledgments). Data for the 222 eastern Indonesian crania tested here were predominantly measured by Pietrusewsky (see Acknowledgments) with smaller numbers measured by Hanihara and by myself. The 92 Malay crania were measured predominantly by Daniel Rayner, a PhD student in my School, and the 185 Punjab crania were measured by Pathmanathan Raghavan, as part of "The Contribution of South Asia to the Peopling of Australasia" project.

One particularly useful feature of FORDISC is that it calculates, and displays, two types of probability that a specimen would have a craniometric affinity with any selected Howells population. These two probability types, typicality and posterior probabilities, convey different types of information (for more information and a more technical discussion, see Ousley and Jantz 1996). The typicality probability is calculated from variance-covariance matrices, and advises the user of the probability that a skull with those particular measurements would occur in the Howells population to which the skull is being compared. The posterior probability is calculated by linear discriminant analysis, on the assumption that the target skull must belong to one of the Howells populations, i.e., the sum of the posterior probabilities is one. Two hypothetical examples show how these probabilities convey different information. One extreme example is a skull which is craniometrically very similar to the centroids of many of the Howells populations, but not specifically close to any single Howells population. The typicality probabilities for this skull will be high for quite a few Howells populations, but all of its posterior probabilities may be 0.2 or less, if the data transformations that maximise the discrimination between the Howells populations do not construe that specimen's measurements as diagnostic of any particular Howells population. The other extreme example is a skull whose typicality probabilities are all 0.000, but which may have a posterior probability approaching 1.000 of being classified with a specific Howells population. In plain language, such a skull would have very unusual measurements, but those measurements could only be possibly found in the Howells population to which it is closest. Most skulls, of course, produce results that fall in between these extreme cases, but the interplay of typicality and posterior probabilities is very useful, as we shall see.

To display a population's profile, the typicality and posterior probabilities will be presented according to their decimal percentile values. The distribution of these probabilities is strongly negatively skewed, therefore summarising them in terms of a mean and standard deviation (with the implicit assumption of a normal distribution) would be entirely inappropriate. Decimal percentile values allow us to gauge the proportion of the target population which is similar to any given Howells population. This provides a well-rounded indicator of which Howells population(s) resemble(s) the target population. A simpler indicator, such as the proportion of specimens in the target population to which one or the other Howells population is closest, would be more likely to leave the analyst with too narrow a view. However, I have only calculated the percentile values for those Howells populations which are the closest population to at least one of the skulls in the target population (using this criterion to exclude very dissimilar populations from the comparisons). Finally as a technical note, the sample size (n) used for calculating the percentile values will be larger for the Howells populations where both sexes are represented, and smaller for the two Howells populations (Filipinos and Anyang Chinese - see map at bottom of this web page) represented only by males.

Aboriginal Australians past and present

The results for recent mainland Australian Aborigines will be presented first, as the anchor for comparison. Twenty-two of the Howells populations are considered, on the basis of being the closest Howells population to at least one Australian skull. The typicality probabilities are listed by specimen in Recent Australian Typicality Probabilities, and the posterior probabilities in Recent Australian Posterior Probabilities. In these tables, the skulls measured by Brown (n.d.) are displayed first, followed by those published by Milicerowa (labelled Wroclaw R_), and finally those measured by Hanihara. Brown's (n.d.) sample is from the Murray Valley, while the skulls in Milicerowa (1955) are predominantly from the northern two-thirds of the Australian continent, and Hanihara's sample is continent-wide with a focus on the Adelaide region of South Australia. These Adelaide skulls are geographically close to the Lake Alexandrina sample (predominantly from the Swanport cemetery, the place name that I shall use) chosen by Howells (1973, 1989) to represent Australian Aborigines, and this geographical proximity is reflected in high resulting probabilities. Brown gives 18 of the 21 possible measurements (foramen magnum length, mastoid height and interorbital breadth are excluded), Milicerowa gives 19 (upper facial breadth and mastoid height are excluded), while Hanihara also gives 19 measurements (biorbital breadth and foramen magnum length are excluded).

Table 1. Recent Australian Typicality Probabilities for Closest Populations

Percentile	Australians	New Britain	Zulu	Tasmanians	Teita (Kenya)
Note. Australians are the closest population in 223 cases, New Britain in 67 cases, Zulu in 54 cases, Tasmanians in 39 cases, and Teita in 8 cases.
10th percentile	0.673	0.497	0.433	0.402	0.334
20th percentile	0.475	0.257	0.235	0.227	0.181
30th percentile	0.333	0.146	0.133	0.114	0.096
40th percentile	0.221	0.094	0.068	0.061	0.043
Median	0.139	0.058	0.037	0.031	0.021
60th percentile	0.078	0.027	0.018	0.015	0.010
70th percentile	0.037	0.010	0.008	0.005	0.003
80th percentile	0.009	0.003	0.001	0.001	0.001
90th percentile	0.001	0	0	0	0

 

Table 2. Recent Australian Posterior Probabilities for Closest Populations

Percentile	Australians	New Britain	Zulu	Tasmanians	Teita (Kenya)
Note. Australians are the closest population in 223 cases, New Britain in 67 cases, Zulu in 54 cases, Tasmanians in 39 cases, and Teita in 8 cases.
10th percentile	0.944	0.535	0.393	0.312	0.113
20th percentile	0.846	0.260	0.173	0.163	0.046
30th percentile	0.723	0.141	0.075	0.068	0.024
40th percentile	0.561	0.075	0.033	0.034	0.012
Median	0.391	0.039	0.016	0.017	0.005
60th percentile	0.231	0.017	0.008	0.007	0.003
70th percentile	0.097	0.008	0.003	0.002	0.001
80th percentile	0.048	0.004	0.001	0.001	0
90th percentile	0.011	0.001	0	0	0

As would be expected, Howells's Swanport Australians are by far and away the closest Howells population to the Australians measured by Brown, Milicerowa and Hanihara (Tables 1 and 2). This is obvious whether we refer to typicality or posterior probabilities, whether we refer to the 10th percentile or the 80th percentile or a percentile in between, whether we use as our criterion the depth of the percentile before reaching a zero probability, or whether we use as our criterion the number of skulls for which Swanport Australians are the closest Howells population. Further, the next four closest populations are New Britain Tolai, Zulu, Tasmanians and Teita, in that order, and again we can observe that order virtually regardless of which criterion (of those mentioned in the previous sentence) we employ. The large sample size (n = 447) yields a well-behaved data set with extremely regular statistical patterns (Tables 1 and 2). However, the results also show that it would be a mistake to expect FORDISC to necessarily classify an Australian skull as Australian, because less than half (223, or 49.9%) would be so classified, even though the Australian profile of these Australian skulls is absolutely unmistakeable.

Tables 3 to 6 give the percentile values of the typicality and posterior probabilities for the other 17 populations which are the closest Howells population to at least one Australian skull measured by Brown, Milicerowa or Hanihara. The quantum difference between these populations and those listed in Tables 1 and 2, in terms of their affinity with Australian skulls, is obvious (almost) regardless of which criterion is employed. Determining the order of similarity to Australian skulls amongst these unAustralian-like populations, however, depends heavily on the arbitary choice of the specific criterion used to infer similarity. This is because the majority of typicality and posterior probabilities evaluate to zero, which drastically reduces the sample size of feasible comparisons. Fortunately, it hardly matters which dissimilar populations may be more, or less, dissimilar from Australians. The important point is that, if the tested Australian sample is large enough, we can expect even these very dissimilar populations to occasionally be the closest Howells population to an Australian skull.

Table 3. Recent Australian Typicality Probabilities for Fairly Close Populations

Percentile	Eskimos	Easter Island	Ainu	Norse	Dogon
Note. Eskimos are the closest population in 11 cases, Easter Island in 8 cases, Ainu in 7 cases, Norse in 5 cases, and Dogon in 3 cases.
10th percentile	0.080	0.077	0.132	0.123	0.127
20th percentile	0.023	0.020	0.049	0.043	0.040
30th percentile	0.007	0.006	0.019	0.015	0.018
40th percentile	0.002	0.002	0.009	0.007	0.008
Median	0.001	0.001	0.004	0.003	0.003
60th percentile	0	0	0.001	0.001	0.001
70th percentile	0	0	0.001	0	0
80th percentile	0	0	0	0	0

Table 4. Recent Australian Posterior Probabilities for Fairly Close Populations

Percentile	Eskimos	Easter Island	Ainu	Norse	Dogon
Note. Eskimos are the closest population in 11 cases, Easter Island in 8 cases, Ainu in 7 cases, Norse in 5 cases, and Dogon in 3 cases.
10th percentile	0.013	0.009	0.033	0.022	0.020
20th percentile	0.002	0.001	0.008	0.004	0.005
30th percentile	0.001	0	0.002	0.001	0.002
40th percentile	0	0	0.001	0.001	0.001
Median	0	0	0	0	0

Table 5. Recent Australian Typicality Probabilities for Dissimilar Populations

Percentile	San (Bush)	Guam	Atayal	Santa Cruz	Anyang Chinese	Zalavar
Note. San and Guam are the closest population in 4 cases each, Atayal in 3 cases, Santa Cruz and Anyang in 2 cases each, and Zalavar and the remainder in 1 case each.
10th percentile	0.101	0.050	0.077	0.078	0.036	0.107
20th percentile	0.028	0.016	0.024	0.025	0.008	0.036
30th percentile	0.010	0.006	0.008	0.009	0.002	0.014
40th percentile	0.004	0.003	0.003	0.004	0.001	0.007
Median	0.001	0.001	0.001	0.001	0	0.003
60th percentile	0	0.001	0	0	0	0.001
70th percentile	0	0	0	0	0	0
	South Japan	North Japan	Egypt	Hawaii	Andaman Islands	Peru
10th percentile	0.083	0.062	0.075	0.025	0.027	0.029
20th percentile	0.024	0.018	0.023	0.008	0.007	0.009
30th percentile	0.009	0.007	0.009	0.003	0.002	0.003
40th percentile	0.003	0.003	0.003	0.001	0.001	0.001
Median	0.001	0.001	0.002	0	0	0
60th percentile	0	0	0	0	0	0

Table 6. Recent Australian Posterior Probabilities for Dissimilar Populations

Percentile	San (Bush)	Guam	Atayal	Santa Cruz	Anyang Chinese	Zalavar
Note. San and Guam are the closest population in 4 cases each, Atayal in 3 cases, Santa Cruz and Anyang in 2 cases each, and Zalavar and the remainder in 1 case each.
10th percentile	0.013	0.006	0.005	0.006	0.003	0.014
20th percentile	0.003	0.001	0.001	0.001	0	0.004
30th percentile	0.001	0	0	0	0	0.001
40th percentile	0	0	0	0	0	0
	South Japan	North Japan	Egypt	Hawaii	Andaman Islands	Peru
10th percentile	0.009	0.004	0.010	0.002	0.001	0.001
20th percentile	0.002	0.001	0.002	0	0	0
30th percentile	0.001	0	0	0	0	0
40th percentile	0	0	0	0	0	0

How do ancient Australian crania fare on FORDISC? It is widely recognised that Australian skulls at the Pleistocene/Holocene boundary differ metrically from recent Australian Aborigines (e.g., Brown 1989), but can we detect this using FORDISC? There is now a quite reasonable sample available to investigate this question, beginning with the Liang Lemdubu fossil from Aru in eastern Indonesia, dating to a time when Aru was connected to the Pleistocene continent of Sahulland (Bulbeck 2005), and continuing with the southeastern Australian skulls dated to the general period between 12,000 and 7,000 years ago (see Bulbeck 2001). Many fossil skulls of interest, including all those from the Willandra Lakes and most from Kow Swamp, regrettably do not have enough measurements available to satisfy the criterion of causing the classification of at least 50% of the reference Howells specimens, and so cannot be included here. The measurements used for the southeast Australian skulls are from Brown (1989, n.d.) and Thorne (1975). Brown's measurements are used for Kow Swamp 5 as these yield higher typicality probabilities than Brown's, while Thorne's measurements are used for Kow Swamp 1 as Brown published a smaller set of measurements, too few for 50% of the reference Howells specimens to be classified. The FORDISC typicality and posterior probabilities are listed in Sahulland Probabilities, for all populations that are the closest Howells population to at least one ancient Sahulland fossil, and summarised in terms of their decimal percentile values in Tables 7 and 8.

Table 7. Typicality Probabilities for Sahulland Australian Skulls

Percentile	Tasmania	Tolai	Australia	Zulu	Guam	Easter	Ainu	Eskimo	Santa Cruz
Note. Tasmanians are the closest population in 12 cases, New Britain Tolai in 11 cases, Australians, Zulu, Guam and Easter Islanders in 3 cases each, and Ainu, Eskimos and Santa Cruz in 1 case each.
10th	0.115	0.081	0.054	0.054	0.038	0.010	0.016	0.007	0.004
20th	0.016	0.046	0.017	0.018	0.008	0.002	0.004	0.001	0.001
30th	0.005	0.006	0.004	0.003	0.003	0	0.001	0	0
40th	0.002	0.002	0.002	0.001	0.001	0	0	0	0
Median	0.001	0.001	0	0	0	0	0	0	0
60th	0	0	0	0	0	0	0	0	0

 

Table 8. Posterior Probabilities for Sahulland Australian Skulls

Percentile	Tasmania	Tolai	Australia	Zulu	Guam	Easter	Ainu	Eskimo	Santa Cruz
Note. Tasmanians are the closest population in 12 cases, New Britain Tolai in 11 cases, Australians, Zulu, Guam and Easter Islanders in 3 cases each, and Ainu, Eskimos and Santa Cruz in 1 case each.
10th	0.730	0.808	0.401	0.255	0.281	0.341	0.036	0.011	0.059
20th	0.556	0.446	0.214	0.081	0.056	0.003	0.015	0.004	0.002
30th	0.428	0.316	0.106	0.042	0.036	0.001	0.004	0.002	0
40th	0.221	0.220	0.062	0.026	0.007	0	0.001	0	0
Median	0.085	0.158	0.021	0.015	0.004	0	0.001	0	0
60th	0.036	0.042	0.014	0.003	0.002	0	0	0	0
70th	0.013	0.025	0.004	0.002	0.001	0	0	0	0
80th	0.004	0.008	0.003	0.001	0	0	0	0	0
90th	0	0.002	0	0.001	0	0	0	0	0

The Sahulland fossil crania clearly have a different FORDISC profile from that of recent Australian crania. Tasmanians and Tolai, not Australians, are the closest Howells populations to the Sahulland skulls, even if it would be a toss-up as to whether Tasmanians or Tolai were the closest. There is a simple explanation for this finding; during the late Pleistocene, Tasmania and New Britain (via New Guinea) were connected to mainland Australia, and accordingly all their inhabitants constituted a single interbreeding population. There are other differences too, which are more subtle. The typicality probabilities for the Sahulland skulls clearly tend to be lower than is the case with recent Australian skulls. The ancient Sahulland skulls are not only distinctive from recent southwest Pacific populations, but indeed more distinctive than recent Australians are from other recent populations around the world. (This finding cannot be attributed to interobserver error, because Peter Brown's measurements - which constitute the majority of those used here - are highly suitable for use with FORDISC; see Interobserver Comparability in FORDISC 2.0). Finally, a possibly surprising result is that "Mongoloid" populations (Guam, Easter Islanders, Ainu, Eskimos and Santa Cruz) provide a substantial proportion (9/39, or 23.1%) of the closest "hits" to the Sahulland crania. All of these populations occur amongst the closest populations to recent Australian crania, but at a much lower rate (Tables 3 to 6). It may come as a surprise to think of Sahulland crania as more "Mongoloid" than recent Australian crania, but in terms of craniometrics this would appear to be the case.

The Roonka cemetery is of interest as its burials span the entire Holocene, and so form a chronological bridge between the ancient Sahulland and recent Australian crania. Roonka is located in the Murraylands upstream from Lake Alexandrina, close to Howells's reference Australian population. Most of the cranial measurements used here were provided by Hanihara (unpublished), with data for a few specimens coming from Pietrusewsky (1979) in those cases where his measurements provide higher typicality probabilities, or where Hanihara did not record the skull. Hanihara's sample also includes some Murraylands crania that are probably not from Roonka itself, but are included here as late Holocene Roonka crania - which would account for approximately half of the usable Roonka crania - are also recent Murraylands crania. Sex determination of the specimens is taken from Pretty et al. (1998), as is chronological age, apart from same refinements kindly provided by Donald Pate (pers. comm.). The FORDISC typicality and posterior probabilities are listed in Roonka Probabilities, and summarised in terms of decimal percentile values in Tables 9 to 12.

Table 9. Roonka/Murraylands Typicality Probabilities for Closest Populations

Percentile	Australians	New Britain	Tasmanians	Zulu	Teita (Kenya)
Note. Australians are the closest population in 19 cases, New Britain in 5 cases, Tasmanians in 3 cases, Zulu in 3 cases, and Teita in 2 cases.
10th percentile	0.493	0.332	0.322	0.461	0.464
20th percentile	0.334	0.161	0.213	0.141	0.124
30th percentile	0.282	0.082	0.042	0.043	0.078
40th percentile	0.180	0.048	0.022	0.032	0.048
Median	0.129	0.032	0.008	0.012	0.023
60th percentile	0.064	0.017	0.006	0.008	0.008
70th percentile	0.031	0.005	0.002	0.002	0.002
80th percentile	0.013	0.002	0.001	0.001	0
90th percentile	0.001	0	0	0	0

Table 10. Roonka/Murraylands Posterior Probabilities for Closest Populations

Percentile	Australians	New Britain	Tasmanians	Zulu	Teita (Kenya)
Note. Australians are the closest population in 19 cases, New Britain in 5 cases, Tasmanians in 3 cases, Zulu in 3 cases, and Teita in 2 cases.
10th percentile	0.915	0.574	0.222	0.288	0.187
20th percentile	0.863	0.108	0.105	0.089	0.109
30th percentile	0.774	0.046	0.017	0.025	0.046
40th percentile	0.680	0.019	0.006	0.013	0.030
Median	0.333	0.014	0.004	0.008	0.015
60th percentile	0.189	0.007	0.001	0.007	0.006
70th percentile	0.027	0.003	0.001	0.005	0.002
80th percentile	0.013	0.002	0	0.003	0.001
90th percentile	0.005	0	0	0	0

Table 11. Roonka/Murraylands Typicality Probabilities for Occasionally Close Populations

Percentile	Easter Island	Bush	Filipinos	Eskimos	Hawaii
Note. Easter Islanders are the closest population in 4 cases, Bush (San) in 3 cases, Filipinos in 2 cases, and Eskimos and Mokapu Hawaiians in 1 case each.
10th percentile	0.156	0.224	0.082	0.015	0.016
20th percentile	0.065	0.012	0.013	0.008	0.004
30th percentile	0.041	0.003	0.001	0.002	0.001
40th percentile	0.019	0.001	0	0.001	0.001
Median	0.010	0	0	0	0
60th percentile	0.003	0	0	0	0
70th percentile	0.001	0	0	0	0
80th percentile	0	0	0	0	0

Table 12. Roonka/Murraylands Posterior Probabilities for Occasionally Close Populations

Percentile	Easter Island	Bush	Filipinos	Eskimos	Hawaii
Note. Easter Islanders are the closest population in 4 cases, Bush (San) in 3 cases, Filipinos in 2 cases, and Eskimos and Mokapu Hawaiians in 1 case each.
10th percentile	0.341	0.053	0.020	0.001	0.001
20th percentile	0.087	0.003	0	0	0
30th percentile	0.027	0.001	0	0	0
40th percentile	0.011	0	0	0	0
Median	0.001	0	0	0	0
60th percentile	0.001	0	0	0	0
70th percentile	0	0	0	0	0

The Roonka profile closely echoes the recent Australian profile. Australians are clearly the most similar Howells population, followed by New Britain Tolai, Tasmanians, Zulu and East African Teita. Australians are the closest population to Roonka in 19/43 (39.5%) of cases. The typicality probabilities have a very similar percentile depth to those of recent Australians before they "zero out". There may be some subtle differences from the recent Australian profile, as in the relatively strong showing of Easter Islanders, and the unexpected result that Filipinos are the closest Howells population in two cases (whereas they were never the closest population to the recent Australian sample). Whether these discrepancies would represent Roonka as intermediate between the Sahulland and recent Australian profiles, and whether the Roonka skulls exhibit chronological change, are topics addressed elsewhere Roonka and FORDISC). For present purposes, it is sufficient to note that Roonka appears to demonstrate the recent Australian craniometric pattern, as indeed observed by other physical anthropologists (e.g., Brown 1989).

Craniometric Profiles of non-Australian populations

The Australian craniometric profile appears to have prevailed across the continent throughout most of the Holocene, based on our above results. It would be interesting to document how quickly this profile cuts out when we move out of Australia. It is widely recognised that eastern Indonesia is a sharp zone of transition between "Australoid" populations to the south and east, and non-Australoid populations to the north and west. Accordingly, eastern Indonesian crania should exhibit a mixed FORDISC profile, with muted parallels to the Australian profile. As regards populations northwest of eastern Indonesia, the Australian craniometric pattern should be simpy unrecognisable. Malay and Punjab Indian cranial measurements can be used to test this latter expectation. Twenty measurements are available for the majority of eastern Indonesian crania (Michael Pietrusewsky's interorbital breadth is excluded), and the full set of 21 measurements is available for the Malay and Punjab crania. The original FORDISC results, for all populations which on at least one occasion are the closest Howells populations to a skull in the studied population, can be accessed through the mini-table below. The results are summarised in Tables 13 to 30, following a brief discussion of their main implications.

Typicality and Posterior Probability Data for non-Australian Crania

Eastern Indonesia		Malays		Punjab
Typicality	Posterior	Typicality	Posterior	Typicality	Posterior

The Australian craniometric pattern tapers off rapidly as we move northwest from Australia. It is detectable in eastern Indonesia as the most salient of the region's two major strands, the other strand comprising "southern Mongoloid" populations adjacent to eastern Indonesia, along with Andaman Islanders (Tables 13 and 16). When we consider the Malay and Punjab results, however, any southwest Pacific affinities are demoted to a secondary status (Tables 19 to 29). The Malay skulls have a clear "Mongoloid" profile, while the Punjab skulls have a complex profile in which Andaman Islanders, and African and European populations, show the highest frequencies of "hits". In both cases, Howells's southwest Pacific populations are at best occasionally close to any Malay or Punjab skull.

The typicality and posterior probabilities show some revealing contrasts. In the Australian results (Tables 1 to 6), the Australians' posterior probabilities re Howells's Swanport Australians tend to be higher than their typicality probabilities, but compared to all other Howells populations the typicality probabilities tend to be higher than the posterior probabilities. In effect, there is a tendency for Australian skulls to resemble skulls from many regions of the world, yet to still be statistically definable as distinctly Australian. This pattern, whereby the posterior probabilities exceed the typicality probabilities for the most similar population(s), but the typicality probabilities equal or exceed the posterior probabilities for less similar Howells populations, is also apparent for eastern Indonesians (Tasmanians being the single closest population) and Malays (Filipinos, Hawaiians and Buriats being the three closest populations). With the Punjab skulls, however, the pattern breaks down (Tables 25 to 30). Virtually invariably, the posterior probabilities exceed the typicality probabilities, both in their depth of non-zero values and their value for any population at any percentile level. That is, the Punjab skulls do not have an appropriate analogue in the Howells database, hence skulls are routinely classified with a population that the skull does not resemble. The lack of a suitable analogue for Punjab skulls in the Howells database is also apparent from the complex composition of the most similar populations (Tables 25 and 28).

Which percentile benchmarks would be most appropriate for displaying results? Medians have the advantage of being a conventional statistical measure of the central tendency. However, with our strongly skewed data sets, the median might be an insensitive indicator; for instance, the median typicality probabilities for eastern Indonesians all fall between 0.000 and 0.006, while the median typicality probabilities for Punjabis all evaluate to zero. While these homogeneously low values provide useful information in demonstrating the shallowness of these probabilities' non-zero values, the fortieth percentile may be an even better indicator. The quantum difference at the fortieth percentile between Australians (highest typicality probability 0.221), Malays (highest typicality probability 0.036), eastern Indonesians (highest typicality probability 0.017), and Punjabis (highest typicality probability 0.001), is very clear.

Another useful benchmark is the tenth percentile value, especially in how it compares with the fortieth percentile value. For instance, looking at eastern Indonesian posterior probabilities (Table 16), we observe that the tenth percentile value compared to Tasmanians is 0.750, which plummets to 0.040 by the fortieth percentile; a similar pattern is apparent for the posterior probabilities of Buriats for Malays (0.516 to 0.002) and he posterior probabilities of Andamanese for Punjabis (0.740 tp 0.009). The implications would be the same in all three cases; a distinct minority of the skulls have rather extreme measurements that classify them with a particular Howells population (Tasmanians for eastern Indonesians, Buriats for Malays, Andamanese for Punjabis), but these skulls could not be confused with the Howells population in question (as shown by the low, accompanying typicality probabilities), and they do not reflect a pattern which applies to their general population. Contrast these cases with the clearly Australian pattern of the tested Australian crania (typicality and posterior probabilities all high at both the tenth and fortieth percentiles). Accordingly, examination of the typicality and posterior probabilities at strategic benchmark values can provide far more contextualised information than a simple consideration of the number of "hits" with a particular Howells population.

Table 13. Eastern Indonesian Typicality Probabilities for Closest Populations

Percentile	Tasmania	Filipinos	Tolai	Hawaii	Guam	Andaman Islands	Australia
Note. Tasmanians are the closest population in 51 cases, Filipinos in 12 cases (under-represented, as only male comparisons are possible), Tolai and Hawaiians in 18 cases each, Guam in 16 cases, Andaman Islanders in 15 cases, and Australians in 11 cases.
10th percentile	0.177	0.383	0.153	0.130	0.124	0.320	0.071
20th percentile	0.078	0.084	0.055	0.046	0.044	0.060	0.022
30th percentile	0.043	0.029	0.025	0.020	0.023	0.020	0.009
40th percentile	0.017	0.010	0.011	0.006	0.006	0.005	0.004
Median	0.006	0.003	0.004	0.002	0.003	0.002	0.001
60th percentile	0.002	0.001	0.001	0.001	0.001	0	0
70th percentile	0.001	0	0	0	0	0	0
80th percentile	0	0	0	0	0	0	0

Table 14. Eastern Indonesian Typicality Probabilities for Occasionally Close Populations

Percentile	Atayal	Santa Cruz	Peru	Moriori	Dogon	Zulu	Teita
Note. Atayal are the closest population in 9 cases, Santa Cruz in 7 cases, and the others in 6 cases each.
10th percentile	0.205	0.112	0.157	0.065	0.092	0.091	0.081
20th percentile	0.085	0.036	0.040	0.020	0.027	0.024	0.027
30th percentile	0.026	0.010	0.013	0.008	0.007	0.007	0.009
40th percentile	0.008	0.004	0.005	0.003	0.002	0.003	0.003
Median	0.002	0.001	0.002	0.001	0.001	0.001	0.001
60th percentile	0.001	0	0	0	0	0	0
70th percentile	0	0	0	0	0	0	0

Table 15. Eastern Indonesian Typicality Probabilities for Dissimilar Populations

Percentile	Zalavar	Arikara	Hainan Chinese	Eskimos	Egypt	Easter Island
Note. Zalavar and Arikara are the closest population in 5 cases each, Hainan Chinese and Eskimos in 4 cases each, Egyptians, Easter Islanders and Berg in 3 cases each, South Japan, Norse and North Japan in 2 cases each, and Ainu and San in 1 case each.
10th percentile	0.161	0.100	0.217	0.030	0.140	0.036
20th percentile	0.063	0.019	0.034	0.013	0.039	0.008
30th percentile	0.007	0.007	0.010	0.004	0.015	0.002
40th percentile	0.007	0.003	0.003	0.001	0.004	0.001
Median	0.003	0.001	0.001	0	0.001	0
60th percentile	0.001	0	0	0	0	0
70th percentile	0	0	0	0	0	0
	Berg	South Japan	Norse	North Japan	Ainu	Bush (San)
10th percentile	0.118	0.218	0.089	0.195	0.057	0.018
20th percentile	0.019	0.067	0.042	0.040	0.022	0.002
30th percentile	0.005	0.017	0.014	0.009	0.006	0.001
40th percentile	0.001	0.006	0.005	0.003	0.002	0
Median	0	0.002	0.001	0.001	0.001	0
60th percentile	0	0.001	0	0	0	0
70th percentile	0	0	0	0	0	0

 

Table 16. Eastern Indonesian Posterior Probabilities for Closest Populations

Percentile	Tasmania	Filipinos	Tolai	Hawaii	Guam	Andaman Islands	Australia
Note. Tasmanians are the closest population in 51 cases, Filipinos in 12 cases (under-represented, as only male comparisons are possible), Tolai and Hawaiians in 18 cases each, Guam in 16 cases, Andaman Islanders in 15 cases, and Australians in 11 cases.
10th percentile	0.750	0.254	0.242	0.212	0.193	0.182	0.151
20th percentile	0.460	0.090	0.075	0.060	0.061	0.046	0.036
30th percentile	0.139	0.026	0.028	0.015	0.026	0.010	0.009
40th percentile	0.040	0.011	0.011	0.006	0.010	0.004	0.003
Median	0.017	0.004	0.003	0.002	0.004	0.001	0.001
60th percentile	0.004	0.002	0.001	0.001	0.001	0	0
70th percentile	0.002	0.001	0	0	0	0	0
80th percentile	0	0	0	0	0	0	0

Table 17. Eastern Indonesian Posterior Probabilities for Occasionally Close Populations

Percentile	Atayal	Santa Cruz	Peru	Moriori	Dogon	Zulu	Teita
Note. Atayal are the closest population in 9 cases, Santa Cruz in 7 cases, and the others in 6 cases each.
10th percentile	0.083	0.073	0.063	0.033	0.031	0.027	0.021
20th percentile	0.028	0.014	0.021	0.006	0.005	0.006	0.006
30th percentile	0.012	0.003	0.008	0.002	0.001	0.001	0.002
40th percentile	0.005	0.001	0.003	0.001	0	0.001	0.001
Median	0.002	0	0.001	0	0	0	0
60th percentile	0.001	0	0	0	0	0	0
70th percentile	0	0	0	0	0	0	0

Table 18. Eastern Indonesian Posterior Probabilities for Dissimilar Populations

Percentile	Zalavar	Arikara	Hainan Chinese	Eskimos	Egypt	Easter Island
Note. Zalavar and Arikara are the closest population in 5 cases each, Hainan Chinese and Eskimos in 4 cases each, Egyptians, Easter Islanders and Berg in 3 cases each, South Japan, Norse and North Japan in 2 cases each, and Ainu and San in 1 case each.
10th percentile	0.050	0.043	0.039	0.025	0.055	0.009
20th percentile	0.020	0.011	0.010	0.004	0.013	0.001
30th percentile	0.010	0.004	0.003	0.001	0.003	0
40th percentile	0.006	0.002	0.001	0	0.001	0
Median	0.003	0.001	0	0	0.001	0
60th percentile	0.001	0	0	0	0	0
70th percentile	0	0	0	0	0	0
	Berg	South Japan	Norse	North Japan	Ainu	Bush (San)
10th percentile	0.024	0.073	0.051	0.020	0.017	0.005
20th percentile	0.006	0.028	0.010	0.007	0.005	0
30th percentile	0.002	0.007	0.004	0.002	0.002	0
40th percentile	0.001	0.003	0.002	0.001	0.001	0
Median	0	0.001	0.001	0	0	0
60th percentile	0	0	0	0	0	0

 

Table 19. Malay Typicality Probabilities for Closest Populations

Percentile	Filipinos	Hawaii	Buriats	Guam	Hainan Chinese	Atayal
Note. Filipinos are the closest population in 16 cases (all male), Hawaiians in 15 cases, Buriats in 12 cases, Guam in 5 cases, and Hainan Chinese and Atayal in 4 cases each.
10th percentile	0.330	0.312	0.110	0.331	0.312	0.206
20th percentile	0.170	0.121	0.030	0.095	0.087	0.072
30th percentile	0.070	0.035	0.016	0.040	0.052	0.023
40th percentile	0.036	0.024	0.007	0.011	0.024	0.011
Median	0.015	0.007	0.001	0.005	0.011	0.004
60th percentile	0.005	0.001	0	0.001	0.003	0.001
70th percentile	0.002	0	0	0	0.001	0
80th percentile	0.001	0	0	0	0	0
90th percentile	0	0	0	0	0	0

Table 20. Malay Typicality Probabilities for Occasionally Close Populations

Percentile	Ainu	Tasmania	Santa Cruz	Arikara	Peru	Andaman Islands	Berg
Note. Ainu and Tasmanians are the closest population in 5 cases each, Santa Cruz in 4 cases, and the others in 3 cases each.
10th percentile	0.130	0.120	0.151	0.209	0.108	0.159	0.165
20th percentile	0.033	0.014	0.029	0.056	0.050	0.051	0.057
30th percentile	0.004	0.008	0.008	0.020	0.010	0.010	0.019
40th percentile	0.001	0.002	0.003	0.009	0.006	0.003	0.003
Median	0	0.001	0.001	0.002	0.003	0.001	0.002
60th percentile	0	0	0	0.001	0	0	0
70th percentile	0	0	0	0	0	0	0

Table 21. Malay Typicality Probabilities for Dissimilar Populations

Percentile	Zalavar	Moriori	Egypt	Anyang Chinese	Dogon	Zulu	New Britain
Note. Zalavar, Moriori and Egyptians are the closest population in 2 cases each, and the others in 1 case each.
10th percentile	0.113	0.100	0.065	0.051	0.110	0.097	0.066
20th percentile	0.043	0.025	0.010	0.014	0.036	0.022	0.015
30th percentile	0.016	0.008	0.005	0.008	0.007	0.004	0.002
40th percentile	0.005	0.004	0.002	0.004	0.002	0.001	0
Median	0.003	0.003	0.001	0.002	0	0	0
60th percentile	0.001	0	0	0	0	0	0
70th percentile	0	0	0	0	0	0	0

 

Table 22. Malay Posterior Probabilities for Closest Populations

Percentile	Filipinos	Hawaii	Buriats	Guam	Hainan Chinese	Atayal
Note. Filipinos are the closest population in 16 cases (all male), Hawaiians in 15 cases, Buriats in 12 cases, Guam in 5 cases, and Hainan Chinese and Atayal in 4 cases each.
10th percentile	0.687	0.623	0.516	0.202	0.176	0.069
20th percentile	0.528	0.197	0.103	0.091	0.100	0.035
30th percentile	0.404	0.044	0.012	0.053	0.050	0.013
40th percentile	0.278	0.036	0.002	0.028	0.034	0.005
Median	0.104	0.013	0	0.011	0.023	0.003
60th percentile	0.067	0.003	0	0.001	0.010	0.001
70th percentile	0.021	0.001	0	0	0.003	0
80th percentile	0.014	0	0	0	0.001	0
90th percentile	0.001	0	0	0	0	0

Table 23. Malay Posterior Probabilities for Occasionally Close Populations

Percentile	Ainu	Tasmania	Santa Cruz	Arikara	Peru	Andaman Islands	Berg
Note. Ainu and Tasmanians are the closest population in 5 cases each, Santa Cruz in 4 cases, and the others in 3 cases each.
10th percentile	0.027	0.047	0.059	0.119	0.055	0.045	0.208
20th percentile	0.009	0.016	0.007	0.029	0.011	0.018	0.038
30th percentile	0.001	0.002	0.002	0.006	0.003	0.005	0.014
40th percentile	0	0	0	0.003	0.002	0.002	0.003
Median	0	0	0	0.001	0	0	0
60th percentile	0	0	0	0	0	0	0

Table 24. Malay Posterior Probabilities for Dissimilar Populations

Percentile	Zalavar	Moriori	Egypt	Anyang Chinese	Dogon	Zulu	New Britain
Note. Zalavar, Moriori and Egyptians are the closest population in 2 cases each, and the others in 1 case each.
10th percentile	0.058	0.018	0.009	0.017	0.024	0.014	0.008
20th percentile	0.017	0.011	0.001	0.006	0.007	0.003	0.002
30th percentile	0.008	0.004	0	0.003	0.001	0	0
40th percentile	0.003	0.001	0	0.002	0	0	0
Median	0.001	0	0	0	0	0	0
60th percentile	0	0	0	0	0	0	0

 

Table 25. Punjab Typicality Probabilities for Closest Populations

Percentile	Andaman Islands	Egyptians	Norse	Bush (San)	Teita	Zalavar
Note. Andaman Islanders are the closest population in 29 cases, Egyptians in 19 cases, Norse and San in 15 cases each, Teita in 11 cases and Zalavar in 10 cases.
10th percentile	0.042	0.047	0.029	0.015	0.013	0.040
20th percentile	0.009	0.013	0.009	0.002	0.004	0.015
30th percentile	0.002	0.004	0.002	0.001	0.001	0.004
40th percentile	0	0.001	0	0	0.001	0.001
Median	0	0	0	0	0	0

Table 26. Punjab Typicality Probabilities for Occasionally Close Populations

Percentile	Atayal	Tolai	Australia	South Japan	Tasmania	Zulu	Eskimos
Note. Atayal are the closest population in 9 cases, Tolai in 8 cases, Australians in 7 cases, and the others in 6 cases each.
10th percentile	0.019	0.014	0.008	0.030	0.010	0.023	0.003
20th percentile	0.006	0.003	0.002	0.006	0.002	0.003	0.001
30th percentile	0.002	0.001	0	0.002	0	0.001	0
40th percentile	0.001	0	0	0.001	0	0	0
Median	0	0	0	0	0	0	0

Table 27. Punjab Typicality Probabilities for Dissimilar Populations

Percentile	Hawaii	Peru	Guam	Berg	Filipinos	North Japan
Note. Hawaiians are the closest population in 6 cases, Peruvians in 4 cases, Guam and Berg in 3 cases each, Filipinos, North Japan, Santa Cruz, Anyang Chinese, Dogon and Moriori in 2 cases each, and Hainan Chinese and Ainu in 1 case each.
10th percentile	0.006	0.010	0.005	0.003	0.022	0.018
20th percentile	0.001	0.002	0.001	0	0.003	0.003
30th percentile	0	0	0	0	0.001	0.002
40th percentile	0	0	0	0	0	0
	Santa Cruz	Anyang Chinese	Dogon	Moriori	Hainan Chinese	Ainu
10th percentile	0.008	0.011	0.009	0.005	0.011	0.008
20th percentile	0.001	0.002	0.001	0.001	0.002	0.002
30th percentile	0	0	0	0	0.001	0
40th percentile	0	0	0	0	0	0

 

Table 28. Punjab Posterior Probabilities for Closest Populations

Percentile	Andaman Islands	Egyptians	Norse	Bush (San)	Teita	Zalavar
Note. Andaman Islanders are the closest population in 29 cases, Egyptians in 19 cases, Norse and San in 15 cases each, Teita in 11 cases and Zalavar in 10 cases.
10th percentile	0.740	0.314	0.232	0.273	0.171	0.181
20th percentile	0.176	0.156	0.080	0.045	0.069	0.111
30th percentile	0.034	0.081	0.033	0.010	0.034	0.049
40th percentile	0.009	0.035	0.014	0.003	0.011	0.029
Median	0.002	0.015	0.006	0.001	0.003	0.013
60th percentile	0.001	0.006	0.002	0	0.001	0.003
70th percentile	0	0.002	0.001	0	0	0.001
80th percentile	0	0.001	0	0	0	0
90th percentile	0	0	0	0	0	0

Table 29. Punjab Posterior Probabilities for Occasionally Close Populations

Percentile	Atayal	Tolai	Australia	South Japan	Tasmania	Zulu	Eskimos
Note. Atayal are the closest population in 9 cases, Tolai in 8 cases, Australians in 7 cases, and the others in 6 cases each.
10th percentile	0.136	0.079	0.045	0.114	0.098	0.097	0.023
20th percentile	0.040	0.013	0.017	0.041	0.012	0.021	0.007
30th percentile	0.018	0.004	0.003	0.013	0.003	0.008	0.001
40th percentile	0.009	0.001	0.001	0.005	0.001	0.003	0
Median	0.003	0	0	0.002	0	0.001	0
60th percentile	0.001	0	0	0.001	0	0	0
60th percentile	0	0	0	0	0	0	0

Table 30. Punjab Posterior Probabilities for Dissimilar Populations

Percentile	Hawaii	Peru	Guam	Berg	Filipinos	North Japan
Note. Hawaiians are the closest population in 6 cases, Peruvians in 4 cases, Guam and Berg in 3 cases each, Filipinos, North Japan, Santa Cruz, Anyang Chinese, Dogon and Moriori in 2 cases each, and Hainan Chinese and Ainu in 1 case each.
10th percentile	0.015	0.018	0.031	0.008	0.023	0.041
20th percentile	0.003	0.004	0.004	0.001	0.009	0.013
30th percentile	0.001	0.001	0.001	0	0.004	0.002
40th percentile	0	0	0	0	0.001	0.001
Median	0	0	0	0	0.001	0
60th percentile	0	0	0	0	0	0
	Santa Cruz	Anyang Chinese	Dogon	Moriori	Hainan Chinese	Ainu
10th percentile	0.013	0.010	0.011	0.021	0.019	0.017
20th percentile	0.002	0.002	0.002	0.003	0.006	0.004
30th percentile	0.001	0.001	0.001	0.001	0.002	0.001
40th percentile	0	0	0	0	0.001	0
Median	0	0	0	0	0	0

Conclusion

FORDISC analysis of a large sample of recent Australian crania (n = 447) reveals a distinctly Ausralian craniometric profile. This is not especially apparent in the number of correct classifications, as just under fifty percent were classified by FORDISC as Australian. Instead, it is apparent from the typicality and posterior probabilities, which display a huge gulf between Lake Alexandrina Australians and any other Howells population. A sample of crania from the Roonka cemetery and other Murraylands locations, whose antiquity reaches back to the early Holocene, also clearly displays a recent Australian craniometric profile, even though only forty percent would be classified as Australian. When we consider ancient Sahulland crania (Late Pleistocene to early Holocene, almost entirely from southeastern Australia), however, their craniometric profile is quite different. Tasmanian and New Britain affinities are more pronounced than recent Australian parallels. These Sahulland fossils are better characterised as southwest Pacific ("Australoid") than specifically Australian in their craniometrics.

Three large, non-Australian samples were also analysed through FORDISC to compare their profile with the recent Australian craniometric profile. Malay skulls were similar to Australians in the sense that the analysis displayed a clear affinity, specifically, their affinity with East Asian and Polynesian/Micronesian crania. Filipinos would be the single closest Howells population to Malays, but, probably because Howells had measured so many "Mongoloid" populations (see map below), the Malays' affinities to specific Howells populations were quite widely dispersed. Eastern Indonesian skulls were revealed by FORDISC to be thoroughly intermediate between the Australian and Malay craniometric profiles. Six of the seven closest populations were either southwest Pacific populations or Malay-like "Mongoloid" populations (Andaman Islanders were the seventh of these populations). Tasmanians are the single closest Howells population to eastern Indonesians, as revealed more convincingly by the posterior than the typicality probability values. Finally, Punjab crania displayed a complex craniometric profile in which the most similar Howells populations - none of which are particularly similar - reach from the Andaman Islands to Hungary and south again to Cape Horn. The Howells database has no close analogues for Punjab crania, as reflected in the ambiguous FORDISC results.

Users of FORDISC should be wary of simple, literal readings of the results. Recent populations show considerable overlap in their craniometrics, to the extent that a high proportion of "wrong" classifications is fully to be expected. A reference population should be characterised in terms of a profile, based on a large sample size, which incorporates not only the pattern of FORDISC "hits" but also the population's typicality and posterior probabilities (preferably summarised with reference to the tenth and fortieth percentile values). On this basis, it has been possible to show that ancient Sahulland crania are not specifically Australian but are of distinctly southwest Pacific affinity.

Acknowledgments

My sincere thanks go to Professors Tsunehiko Hanihara and to Michael Pietrusewsky for their benevolence in sending me their unpublished cranial measurements for analysis. Dr Pathmanathan Raghavan, and Daniel Rayner and Adam Lauer, have allowed me to use the craniometric data that they collected as part of the The Contribution of South Asia to the Peopling of Australasia project. Associate Professor Donald Pate provided me his recent estimates of the chronological ages of the Roonka burials. Professor Maciej Henneberg kindly provided me with a copy of Milicerowa (1955).