AMERICAN JOURNAL OF PHYSICAL ANTHROPOLOGY 83:307-319 (1990) Efficiency of Cranial Bilateral Measurements in Separating Human Populations I. HERSHKOVITZ, B. RING, AND E. KOBYLIANSKY Department of Anatomy and Anthropology, Sackler Faculty of Medicine, Tel Aviv University, Israel KEY WORDS Isolates, Skull discrimination, Asymmetry ABSTRACT A set of 31 nonconventional paired cranial measurements, as well as six Conventional nonpaired measures, were Laken o n 266 skulls, representing two related populations: Bedouins of the Israeli Negev Desert and Bedouins of the Sinai. The data were subjected to univariate and discriminant analyses to determine the relative efficacy of paired vs. conventional measures in sorting individuals according to tribal and sex affiliation. It was found that paired measures have greater discriminatory power (87%) than conventional ones (47%) in terms of classifying individuals belonging to human isolates derived from a common ancestor and sharing similar environmental conditions. This greater discrimination attests to the value of the level of “developmental noise” (a measure provided by fluctuating asymmetry) in sorting human populations. Possible explanations are proffered for the above finding. As pointed out by Woo (1931) many years ago, “Racial differences were approached from the standpoint of appearance, in other words from the conception of portraiture. Anthropometricians endeavoured to give quantitative value to the differences that were obvious to them at first sight.” It is not surprising, then, that a large set of cranial measurements deemed to be most representative of skull morphology was proposed, and that these were used in intrapopulation and interpopulation analyses in anthropology for over a century (for early literature review, see Falk and Corruccini, 1981).In time, however, selection of the best discriminatory traits was im roved by using multivariate techni ues ( owells, 1969; Relethford, 1988).%actorand principal component analyses supported the prevailing notion (e.g., Young, 1957; Moss and Young, 1960) that the crania of humans and animals consist of a number of varying units that res ond in s ecial ways to a variety of externa forces owells, 1973; Key and Jantz, 1981; Cheverud, 1982). This is consistent with the results of twin studies (Nakata et al., 1974) that evinced the existence of several genetic and environmental components from the craniometric data. a (4 0 1990 WILEY-IXS, INC B The success of discriminant methods based on selected cranial morphological traits depends on the heritability level of the latter. Indeed, the belief in a strong genetic component of craniofacial traits encouraged physical anthro ologists to use them in drawing up racia taxonomies. However, numerous studies (e.g., Nakata et al., 1974; Susanne, 1977; Sharma, 1987) have shown that considerable variation can be found in the estimates of heritability for different craniofacial traits (e.g., head length = 0.90; head breadth = 0.72; head height = 0.66; estimates of heritability from twin studies, Sharma, 1987). Studies that used traditional measurements (Martin and Saller, 1957) usually failed to discriminate between skulls from closely related populations (Arensburg et al., 1987). In fact, only opulations remote in space and cline could e separated efficiently by traditional measurements (Howells, 1969; Falk and Corruccini, 1981). In the present paper, we adopted a different approach toward cranial discrimination P g Received May 23.1989, accepted January 4.1990 Address reprint requests to Dr I Hershkovitz, Dept of Anatomy and Anthropology, Sackler Faculty of Medicine, Tel Aviv University, Ramat Aviv, Israel 308 1. HERSHKOVITZ ET AL in that we focused on skull asymmetry rather than on general skull morphology. Rates of fluctuating asymmetry, defined as the random difference between both sides of a paired trait in an individual, are believed to reflect the magnitude of developmental stability of an organism, that is, tion to deviation from norma ontogeny (Livshits and Kobyliansky, 19871,and correlate with opulational homozygosity level (Kobylians y and Livshits, 1983). Hence we concluded that paired measurements may be .. used more effiaentlv to discriminate hetween high1 inbred human populations, even when t ey are derived from the same ethnic stock and live under similar environmental conditions. This conclusion was based on the assumption that, due to geographic and social isolation and a long inbreeding process, the human groups, in the course of time, would tend to become genetically divergent. The small size o f the groups and their strong emphasis on consan ous marriages will result in a s t e a Y E 1 crease in the homozygosity of indivi uals within each group. These homozygous individuals would be less able to buffer against developmental disturbances and hence will tend to deviate from perfect symmetr more than heterozygotes (Soule, 1979; KO yliansky and Livshits, 1983, 1985; Leary et al., 1984; Livshits and Kobyliansky, 1984; Palmer and Strobeck, 1986). It would follow, then, that individuals of different Bedouin tribes who vary (due to different inbreeding levels among the tribes! in their levels of developmental homeostasis will manifest ciifferent rates of cranial morphoiogicai fiuctuating asymmetry-a feature that would be of discriminatory value. The rate and magnitude of skull symmetry has never been used in sorting human skulls into group affiliations. Rather, the vast amount of research carried out on skull asymmetry in humans and nonhumans has been focused mainly on the different functions of the right and left hemispheres of the brain or on the masticatory mechanism (e.g. Woo, 1931; Vig and Hewitt, 1974).The skull complex, however, consists of several paired and unpaired bones and therefore manifests characteristics that display some degree of asymmetry (such as size differences between homologous bones or between bilateral parts 1. The relative contribution o f genetic and environmental factors t o the variation in the predisposi- E K t level of fluctuating asymmetry is as yet unclear (Palmer and Strobeck, 1986; Livshits and Kobyliansky, 1988). Many reports have shown that environmental stresses can significantly elevate fluctuating asymmetry in humans as well as in animal species (e.g., Siege1 and Smookler, 1973; E es er et al., 1986). Estimates of fluctuating asymmetry heritability are usually found to be low and not statistically significant (e.g., Leamy, 19861, but recently Livshits and Kobyliansky (1988) found that the mean FA values of Several metric traits indeed reveal a statistically significant genetic component. The present paper describes a method for discriminating between related inbred populations by utilizing the asymmetrical nature of the skull. MATERIALS AND METHODS Two Bedouin populations, one from the Israeli Negev and the other from South Sinai, were studied. From a historical point of view, the Bedouins who today inhabit the Negev and South Sinai regions are only a fraction of the total Bedouin population of the Near East and Arabian Peninsula. They began to settle in the region of the Negev and Sinai between the 10th and 14th centuries C.E. (Sharon, 1977).The Bedouin tribe of the Negev included in the present study is within the social framework of the “Tayaha” tribal affiliation, whereas the studied Sinai tribes belong to the “Towara” tribal affiliation (Hershkovitz, 1985). These two groups of tribes are geographically and socially isolated and have no intermarital contacts. Aimost 97% of Sinai Bedouin males (Her siikovitz, 1985) and 86% of Negev males (Muhsam, 1966)find their mates within their own tribes. Marriages within the extended family, with preference for first-cousin matings, are very common among Sinai and Negev Bedouins, and may comprise more than 30% of the total (Hershkovitz, 1985).Unlike other Bedouin tribes of the Near East and Saudi Arabia, the Sinai and Negev tribes are small, the Negev tribes being generally larger than those of the Sinai (Muhsam, 1966; Hershkovitz, 1985). Concerning the origin of the material and its affiliation, the skeletal remains of the Negev Bedouins were recovered in ancient burial caves from the Roman-Byzantine periods near Kibbutz Lahav (Fig. l ). The reuse of ancient burial caves by Israeli Bedouins is quite common. From coins, beads, metal 309 SEPARATING HUMAN POPULATIONS bracelets, and other artifacts accompanying the osteological material, the skeletons were dated to the late 19th century. Historical documents (Muhsam, 1966; Sharon, 1977)as well as direct testimony of the present Bedouins of the region (Marx, 1974; Bailey, 1980) confirm the origin of the material and its affiliation to the Ramadin tribe (still living in this region). The skeletal material from the Sinai Desert came from two main sites, Wadi Qid and Wadi Slaf (Fig. 1). From the artifacts and testimonies of the Bedouins themselves, the skeletons are affiliated to the Muzeina and Awlad Said tribe.. (amnng the larger tribes of the Towara). According to the literature (Ben-David, 19781, there is documentation that a severe epidemic spread through these regions during the later 19th century, an account of which also a pears in many Bedouin folk stories ershkovitz, 1985). The existence of mass burials seems to sup ort this notion. Because the tribafaffiliation of both the Negev and Sinai skeletal sample could be determined, we were able, through Bedouin history books and documents, as well as other sources (Hershkovitz, 19851, to follow their tribal history almost until their time of (fi ______ ---I 7 Fig. 1. Locations of sites in the Negev and Sinai Deserts separation from a common mother population more than 600 years ago. In the course of the present investigation, cranial measurements were taken on 266 adult Bedouin skulls, of which 164 were from the Negev and 102 from the Sinai; the skulls are currently housed in the Department of Anatomy and Anthropology, the Sackler Faculty of Medicine, Tel Aviv University. All measurements were taken by one of the authors (B. Ring). The measured coefficient of repeatability was high (r = 0.981. The anthropometric criteria comprised 31 new paired measures (chor6s 2nd XCC) mr! six traditional single measures (Table I, TABLE I. List of variables used in the present study' Code Variable Measures of 1. PBC PBA 2. SBC SEA 3. PLC PLA 4. SLC SLA 5. ALE ALA 6. ABC ABA 7. NAC NAA 8. OAC OAA 9. PNSS 10. PSQ 11. SON the cranium2 = porion-bregma chord = porion bregma arc = sphenion-bregma chord -= sphenion-bregma arc = porion-lambda chord = porion-lambda arc = sphenion-lambda chord = sphenion-lambda arc = asterion-lambda chord = asterion-lambda arc = asterion-bregma chord = asterion-bregma arc = nasion-asterion chord = nasion-asterion arc = opisthion-asterion chord = opisthion-asterion arc = parietal notch-sphenosquamosal suture = porion-squamosal suture = supraorhital notch-midcoronal suture (lateral frontal arc) L \ * ...... . edbures of the 12. NEL = nasion-ectoconchion length 13. ORBH = orbital height 14. ORBW = orbital width 15. MZL = maximum zygomatic length 16. MZH = maximum zygomatic height 17. PZA = prosthion-zygomaxillare anterior chord 18. NSA = nasal spine-alare chord Measures of the base of the skull' 19. BML = basion-mastoid length 20. RSS = basion-sphenoid spine chord 21. SSF = sphenobasion-stylomastoid foramen chord 22. SBP = sphenobasion-pterion chord 23. FIM = foramen incisivum-tip of mastoid chord Conventional measures CIND = cephalic index FIND = facial index RIND = basal index (sphenobasion-Op/Po-Po) BIAB = biauricular breadth MAXL = maximum cranial length MAXR = maximum cranial breadth 'For location of measures, see Figures 1 a n d 2. 'Measured on both sides of the skull. 310 I. HERSHKOVITZ ET AL. Fig. 2. Paired measures taken on skull, lateral view. Explanation of numbers is given in Table 1. Fig. 3. Paired measures taken on skull, basilar view. Explanation of numbers is given in Table 1. Figs. 2, 3). The paired measures were selected in order to characterize the three main anatomic regions of the skull, as follows: calvarium, 19 measures; face, seven measures; and base of the skull, five measures. Of these 31 measurements, half were corresponding measurements on single homologous bones, and the rest spanned t,wo or more bones. For each skull, age and sex were determined independently by three experienced anthropologists. Where opinions diverged, the skull was reexamined and the majority decision accepted. All skulls were assigned into three age groups according to suture closure, as follows: 1)18-30 years (117 crania), all ectocranial sutures still open; 2) 30-45 years (106 crania), ectocranial sutures partially closed; and 3) 45+ years (43 crania), ectocranial sutures synostosed. The degree of agreement regarding assignment of the skulls into the above age groups was over 90%. We found the age distribution to be similar in both the Sinai and Negev samples. Determination of sex was based on size of the glabella and supraciliary eminence, inclination ofthe frontal bone, frontal bossing, shape of the orbit, thickening of the orbital 311 SEPARATING HUMAN POPULATIONS margin, and general gracility of the skull. Size of the mastoid process and the relief of the planum nuchale, which are usually used in sex determination, were ignored here because Bedouin women tend to resemble males in this respect, as they regularly carry heavy burdens on their heads. Sex ratio (maledfemales x 100)for the Negev sample was 102.5 and for the Sinai group 142.9. Agreement achieved in our sex classification was over 80%. It is noteworthy that anthropological studies on living Bedouin populations in the Negev (Muhsam, 1966)and Sinai (IIei SlikuVitL, ISS5) h a v e reveaied simiiar results regarding sex ratio. Statistics For each characteristic, the mean (X) and standard deviation (SD) were calculated. By analysis of variance with one factor interchange (ANOVA),the variability of the studied traits in two Bedouin groups was determined and compared separately for each sex (Nie et al., 1975). The subprogram Discriminant was used to generate canonical discriminant functions (Nie et al., 19751, which in turn allowed classification of individuals on the basis of their proximity to the various group centroids. We carried out three separate classifications: of male skulls according to tribal affiliation; of female skulls according to tribal affiliation; and of all skulls together according to tribal affiliation and sex. Individuals with any missing data were excluded from the analysis. RESULTS Although there were some differences in ANOVA results between sets ofright and left measures, we present here only the mean values and standard deviations for the measures taken on the right side of the skull (Table 2). The Sinai and Negev females differed significantly in six of 31 right side traits (SBC, SBA, NAC, NAA, SON, and SSF) and the males in five traits (SON,ALC, ABC, ABA, and SSF).Only two traits (SON and SSF) were significantly different in both the male and female intratribal comparisons. For left-side measures, the males differed significantly in nine traits (PBC, PLC, SON, ALC, ALA, ABC, ABA, ORBH, and SSF),but in females no significant discriminating left trait was found. Of the six conventional measures (Table 11, cephalic index and maximum cranial breadth showed sig- nificant differences among the females; facial index was significantly different among males (Table 2). Of the total 68 measures (31paired and 6 single) used in the discriminant analysis (mimimum F to enter = l),33 manifested significant discriminatory ower for the combined sample (Table 3, le t column). Conventional measures selected for cranial identification, such as cephalic index and basal index, manifested low Wilks lambda values (Table 3). The contribution of paired measures to Bedouin skull differentiation is evident from the results ofthe stepwise variable selection; i.e., in nine variables, both the measures on the right and left sides were selected by the stepwise procedure. Correlations of bilateral measures in order of decreasing intensity appear in Table 4. If we examine the composition of the 15 variables with the highest Wilks lambda values (Table 31, we find the following: 1) nine of the 15 variables are measures of individual bones; 2) none of the nonpaired variables or indices appear in the list of traits; 3) nine of the variables are measures taken on the right side and six on the left; 4) although about half of the 31 paired measures manifest correlation coefficients higher than 0.800, of the group of 15 variables with the highest Wilks lambda, 67% manifest a correlation coefficient lower than 0.800, which indicates a clear tendency toward the selection of traits with higher FA levels; 5) nine of the 15 variables are measures of the calvarium, three are of the face and three of the base of the skull: and 6) two pairs of right-left measures a pear amorg the first 15 variables (OAC ancf ABC). Analysis of the first 15 measures with the highest FA level (0.379-0.592), calculated as 1-r2,(Table 4),shows that nine are measures of the calvarium, four are of the base of the skull, and only two of the face, thus reiterating the well known fact that the face is the most symmetrical part of the skull (Woo, 1931). In the analysis, three functions were obtained, the first two accounting for 90% of the between-group variance. Function 1 explains 51% of the between-group variance and Function I1 39%. The pooled within-group correlations between canonical discriminant functions and discriminating variables are given for all three functions in Table 5. Function I has its highest correlation with the right asterion- fp 312 1. HERSHKOVITZ ET AL, T A B L E 2. Basic descriptive statistics for two Bedouin populations, b y sex Group' ___ N 4 80 76 57 35 1 2 3 4 78 77 56 35 I 2 3 4 '1'1 1 2 3 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 X RPBC 124.4 121.5 125.6 121.1 RPLC 115.2 112.0 115.9 112.5 RALC 84.2" SD N 4.7 7.8 5.1 4.0 79 76 53 33 5.0 5.5 4.8 3.7 79 78 56 36 b.1 78 83.0 4.2 86.2* 5.0 83.0 4.0 RNAC 78 147.2 4.5 75 140.6" 4.6 52 148.1 5.7 32 143.0* 4.7 RPNSS 75 65.4 4.4 76 62.2 4.2 52 65.3 4.9 35 62.6 3.8 RORBH 78 32.9 2.2 2.2 78 32.9 56 32.3 1.8 37 32.6 2.4 RPZA _______________ 74 54.8 3.0 73 53.6 3.3 55 54.6 3.5 35 53.5 3.9 RSSF 81 50.2* 2.4 80 48.9* 2.3 55 51.9* 3.1 2.3 36 50.5" 80 56 35 76 55 34 79 79 52 33 76 74 52 35 80 79 56 37 77 79 54 37 71 76 50 34 CIND 1 2 3 4 1 2 3 4 79 73.5 2.8 77 74.2* 2.4 55 72.5 2.8 37 72.9* 2.7 ~ _ _ _ BIAB _ _ _ _ _ _ _ 81 100.5 4.5 79 96.5 4.7 57 100.7 5.2 38 97.3 4.4 79 78 57 36 80 79 55 39 X RPBA 153.7 1,50.7 154.2 148.4 RPLA 139.4 135.9 140.3 136.3 KALA 94.0 92.9 95.5 92.3 RNAA 182.4 174.2* 183.8 177.3" RPSQ 49.3 47.5 49.0 46.7 RORRW 39.7 38.9 39.8 39.4 RNSA 16.0 14.9 15.8 14.8 RSBP 71.2 69.0 70.6 68.5 FIND 73.1* 72.2 70.1* 70.4 MAXL 183.8 178.2 184.7 178.7 SD N 6.9 7.2 6.4 5.1 71 73 53 33 6.3 5.0 5.4 5.0 70 73 52 32 5.1 6.0 5.0 , 76 53 33 5.9 6.2 7.1 6.0 78 72 53 36 4.7 4.1 4.2 3.4 79 80 58 35 1.9 2.0 1.8 1.4 61 61 48 26 1.4 1.7 1.3 1.4 73 72 53 31 3.3 3.4 4.2 3.7 70 72 50 29 4.7 4.7 4.7 4.8 77 75 50 36 6.2 5.5 5.9 5.8 80 78 57 37 6.2 nt X RSRC 94.0 92.3* 93.7 89.9* RSLC 140.5 138.0 140.8 135.8 RABC 134.3" 130.0 137.2* 131.6 ROAC 61.4 60.0 62.0 60.2 RSON 103.5* 99.9* 100.2* 96.2* RMZL 53.3 50.1 52.3 49.9 RBML 50.9 48.7 51.8 49.3 RFlM 106.0 102.5 106.8 103.6 BIND 65.7 65.7 65.1 65.6 MAXB 134.8 132.2* 134.0 130.1* SD N ~. 5.0 5.3 4.8 4.3 6.0 7.1 5.7 5.1 71 75 54 35 ___ 70 73 52 34 5.1 4.3 4.9 4.3 78 79 54 34 2.7 2.9 2.5 2.8 78 75 53 36 6.7 6.2 6.4 4.8 80 80 57 38 3.1 2.9 3.6 3.7 73 74 57 35 2.4 2.5 2.6 2.1 69 67 54 37 X RSRA 110.7 109.0* 111.1 105.9' RSLA 173.2 169.4 172.8 167.4 RARA 164.2' 159.3 167.5* 160.4 ROAA 68.5 66.6 68.6 66.5 RNEL 50.7 49.2 51.2 49.4 RMZH 47.3 45.0 46.6 45.4 RBSS 39.3 37.5 39.2 37.8 _-SD 6.4 6.2 6.9 6.3 8.1 7.0 7.3 6.0 6.9 6.3 6.3 5.3 3.5 3.6 3.5 3.2 1.9 2.1 1.9 1.9 2.9 2.5 2.9 2.9 2.0 2.2 2.7 1.9 4.7 4.6 5.2 4.6 4.5 4.6 4.4 3.4 5.7 4.9 5.3 4.4 *Significantly different a t the level of P 10.05. between the same sex a t different sites. 1 = Negev males; 2 = Negev females; 3 = Sinai males; 4 = Sinai females. See 'Fable 1 for description of trait, abbreviations. ' bregrna chord and the left asterion-bregma arc, and Function I1 with maximum cranial length and right nasion-asterion arc. Function I11 has low correlations with most variables (Table 5). The Mahalanobis distances between group centroids are significantly different for all comparisons (P < 0.01). The group centroid values are presented in Table 6 for the three functions. For Function I the polar groups are Sinai males and Negev females, with values of 2.15 and - 1.78, respectively; 313 SEPARATING HUMAN POPULATIONS TABLE 3. Stepwise variable selection: summary table For sex and tribal affiliation Entered Wilks traits lambda ______________ ,581 RARC ,459 RSON .422 RNSA ,402 RSSF .363 1x33 LOAC ,302 LNAC ,302 RBML ,276 ,264 ROAC .257 LALA .ZSY XfZA 230 RSLC .214 RMZH LAX ,188 LABA ,175 RNAL ,167 .158 LSSF RIND .I45 ,141 RNAC RMZL .I31 .I33 RPLC .127 RBSS RORBW .124 CIND .120 LNAL ,117 .111 RABA .106 LALC RPLA .lo3 LPLA .098 .091 LORBH L.NAA ,087 ,081 RORBH Males for tribal affiliation Entered Wilks traits lambda -Females - _ _ _ _ _ _for _ _tribal _ _ _ _affiliation ___~ LABC RSON RPZA LOAC MAXBR RABA LPZA T,NSA LPNAP RBSS LSKA LBML RSBC LOAA LABA ROAA LSBA LSBC LSLC CIND ISON RMZL LNAA _ _ . ,756 ,600 .455 ,435 ,416 ,404 ,390 ,381 ,365 ,360 ,346 .338 ,328 .313 -307 ,301 ,293 ,289 ,280 .269 Entered traits Wilks lambda ---_ RSSF ,901 ,615 543 RNAA RNAI, LBSS LNAC MAXBR LSBP LALA LOAC ,494 .443 ,416 ,390 .363 ,343 RRSS 21 Q LSBA RABA LBML RABC RMZH 1,ORBI-I FIND RORBW RPST ,291 .266 ,232 223 ,213 ,205 .195 ,188 ,181 .271 .263 ,253 TABLE 4. Correlations of bilateral measures in order of intensity and fluctuating asymmetry level for 31 pairs of craniofacial traits Measure PBC SON PNAP NAC FIM ORBH PSQ SLC NEL MZL PLA NAA ALC PLC MZH 'Calculated as l-? Correlation coefficient r ,928 ,891 ,886 ,885 382 379 ,866 ,861 FA level' Measure .139 SLA SRP ABC ALA ORBW SSF PLA BSS NSA OAA ,207 215 ,217 ,222 ,228 ,250 ,259 ,859 ,850 ,262 ,849 ,842 ,834 ,832 ,279 ,291 ,802 .278 .305 ,308 .357 SRC .OAC PRA SBA ABA BML Correlation coefficient r ,797 .788 .786 .774 ,770 ,764 ,761 ,755 .747 .740 ,735 ,707 .701 ,667 .662 ,639 F A level' .365 .379 ,381 .401 ,407 .417 .421 .430 .442 ,453 .460 .501 509 ,555 ,562 ,592 314 1. HERSHKOVITZ ET AL. TABLE 5. Pooled withira-group correlations between canonical discriminant functions and discriminating variables Function TI I Variable RARC LABC RABA RNAA RNAC LABA LPLC RBML RPLC RSSF LBML MAXI, LSSF LPBA BIAB 537 515 429 417 388 189 188 191 316 301 ,259 .I17 ,214 137 ,385 ,379 343 330 1% n37 325 32 1 ,275 ,256 ,236 303 320 ,148 ,178 ,183 111 - 046 - 279 - 003 138 134 ---.I87 .058 .I44 058 "2 - 051 - 027 .022 ,065 -.004 TABLE 6. Group centroid values of four Bedouin groups Group Function I Function I1 Function 111 1 Negev males 2 Sinai males 3 Negev females 4 Sinai females -0.35 2.15 -1.78 -0.13 1.72 -0.13 -0.43 -2.26 0.51 -0,42 -0.67 1.02 Function I1 Variable I RPBA RPLA LPLA RFIM LPBC LALC LOAC LFIM LSRA HPCC RALC RALA LPSQ ROAC LOAA 236 229 072 030 123 112 - 054 -.049 212 141 146 110 099 .049 -.067 228 222 22 1 ,216 ,207 207 198 - 111 ,028 .om .I66 .. 164 :?f; < 074 - ($13 - 119 - 114 lJLi 186 185 ,178 ,172 .I70 075 052 ,169 ,037 -.069 ----,006 --.I52 .010 for Function 11, Sinai females and Negev males are extreme and for Function I11 Sinai and Negev Classification of the skulls according to sex and tribal affiliation (Table 7) shows all groups to have approximately similar numbers of misclassifications, ranging from 7% to 17%. The average percentage of correct TABLE 7. Classification of four Bedouin groups into their tribal and sex affiliations' _ _Actual _ _ _ _group _~~ 1 Xegev i d e s No. of cases _. ~-~ 1 41 34 (s2.9%1,) 2 Sinai males 39 3 Negev females 38 4 Sinai females 21 2 (5.1%)) 3 (7.9%) 0 (0%) -. Predicted group ____..._ --membership 3 2 _____ 2 3 (4.9%) 36 (92.3%) 0 (owl) 1 (4.8%)) (12.2%) 2 (2.6%) 33 ._____.. -. 4 0 (0%) 0 (0%) 2 (5.3%) 18 (85.7%) (86.8%) 2 (9.5%) 'Percent of grouped cases correctly classified = 87.058. TABLE 8. Classification.of skulls of two Bedouin groups into their tribal affiliation, males only' Actual group No. of cases 1 Negev males 41 2 Sinai males 40 _ TABLE 9. Classification of skulls of two Bedouin groups into their tribal affiliation,females only' Predicted group _membership _ ~ ____ 1 2 Actual group ._______ 38 (92.7%) 1 (2.5%) 'Percent of grouped cases correctly classified = 95.06% 3 1 Negev females (73%) 2 Sinai females 39 (97.5%) No.of cases 47 Predicted group membership 1 2 42 (89.4%) 22 I (4.%) 'Percent of grouped cases correctly classified 91.03% 5 (t0.6'%) 21 (95.5%) SEPARATING HUMAN POPULATIONS classification is very high, being 87.05%. Of the total 17.1% misclassified Negev male skulls, most (12.2%)were erroneously attributed to the Negev female group, and only 4.9% were misclassified as Sinai males. The Negev females present a similar picture, with most of the misclassified skulls attributed to the Negev male group. As for the Sinai males, approximately half of the 13.2% misclassified were assigned to the Negev males group (7.9%) and half to the Sinai females (5.3%).Most of the 14.3% misclassified Sinai females were attributed to the Simi males (9.5%> and the zest (4.8%) LO tile Negev females. The predicted tribal memberships for male and female samples separately (Tables 8 and 9, respectively) indicate a very high level of correct classification for both groups, with that for males being somewhat higher than for females. As can be seen in Table 3, prediction of group affiliation, by sex, is generally based on different measures. Most measures that highly discriminate between the Negev and Sinai females pertain to the base of the skull and the calvarium, whereas those discriminating between the Negev and Sinai male skulls are from all three regions. When using a higher criterion for variable selection in discriminant analysis (with minimum F ratio to enter into the function equal to 41, correct classification of skulls according to sex and tribal affiliation (with only four variables in the function) is greatly reduced (53.1%).However, it is still greater than that achieved by the five conventional variables used by Arensburg et al. (1987) of $74, When sex is determined first, successful classification into tribal origin (with only four to six variables in the function) remains relatively high: 82.4% for males and 76.5% for females. This clearly indicates the discriminatory power of our sets of bilateral measures. In order to evaluate properly the validity of the classification results obtained for a given number of individuals (266) and given number of descriptors (661, we determined the percentage of correct classifications due merely to chance. In anthropology, discriminant functions are usually generated and used without reference to the level of chance classification; random results are assigned the value of 50% correct classification (in the case of two groups), and any results greater than 50% are usually considered to be due to the infor- 315 mation contained within the descriptors (measurements). However, for a given number of individuals, the probability of fortuitously obtaining 100%correct classification increases as the number of features (d) increases from 1to the number of individuals in the study (N) (Stouch and Jurs 1985a,b). These classifications, although correct, are due only to artifacts of mathematics governing the process of generating linear discriminant functions (LDF). They are not due to any relationship between the individuals, and the resulting LDF will have no predictive ability beyond random guessing (Stouch and Jurs, 1985a,b). Until recently, it was accepted that if the number of descriptors is kept below one third the number of individuals used, the probability of complete separation due to chance could be kept low. The ratio of NIda.3 was accepted as a minimum requirement (Stuper and Jurs, 1976; Varmuza, 1980). More recent studies (Stouch and Jurs, 1985a,b), however, have shown that, a t that ratio, random classification results ranged around 90%. Even one descriptor for every ten individuals would yield random correct classification of about 75%. It was also found that unequal group sizes serves to increase correct random classification and that for any one value of the ratio N/d, the percentage correctly classified is the same regardless of the number of individuals used in the study. The N/d ratio in our study was four prior t o the stepwise procedure, after which the ratio increased to six (in certain runs). Because a posteriori reduction in the number Q€ variables (i.e.. using only those vtlriabies thet work best) does not solve this particular problem, the best way to show that our discriminant functions are reliable discriminant tools, despite the low N/d ratio, is by showing that they will work equally on an independent but otherwise similar sample. Unfortunately, there are no such samples available in our country on which to perform complementary studies. Nevertheless, we strongly believe, for the following reasons, that the proposed method has great potential. 1. The theoretic background (relationship between level and pattern of asymmetry and population genetic structure) of the proposed method is well established in the literature. 2. An important problem in pattern recognition is the selection of the features used to 316 I. HERSHKOVITZ ET AL. characterize an object or, in this case, the skull. In the first and most important step, anthropologic and genetic knowledge must be, and was, used to select those measurements that hopefully are related t o the classification problem. After the preliminary preselection of features, we further selected, using the mathematical method, for the best features for a given classification problem (see F = l , F=4 method). To verify that our traits properly serve the goals for which they were intended (i.e., discrimination between closely related tribes), we used these same traits and same samples in a discrimination for which they were not designated and that lacks the justification of any practical or theoretic genetic background, i.e., discrimination between two arbitrary age groups. Because we determined the age of each skull, we split the total sample into two age groups (below and above 30 years of age) and used the same descriptors for age classification. The stepwise procedure (F=4) found very few variables (d=2) that could serve such a purpose, and the average classification success was, of course, much lower than that obtained in our study for tribal classification under the same conditions (82.4%vs. 66.1%). Moreover, when we used 60 asymmetry parameters (30 of fluctuating asymmetry and 30 of directional asymmetry), they still showed a high level of success (77.9%) in discriminating the skulls by tribes, although the results were somewhat lower than obtained using metric traits. 3. Although both parametric and nonparametric methods are used for the developmeat of discriminants (asing different statistical parameters), the N/d requirement value in the literature was obtained from a nonparametric method; ours is parametric. 4. Skulls that were misclassified by the functions were categorized generally within their own ethnic group (314 cases; see Table 7); i.e., of the 17.1%wrongly classified Negev male skulls, 12.2% were classified into the Negev females class, and only 4.5% into the Sinai males group. 5. The minimum requirement of variableto-pattern ratio (N/d) is 3 (based on Varmuza’s 1980 textbook, on which all later studies are based, and on the study of Stuper and Jurs (1976),and our ratio is 4. Recent studies (Stouch and Jurs, 1985a; Jurs, 1986)propose to increase the ratio to 10. In physical anthropology studies, such ratios are rarely achieved, as the skeletal samples are almost always too small. Surely this does not mean that we must stop using discriminate functions, but that we must be careful in evaluating the results. In addition to all of the above, the “Monte Carlo” techniques, which is a simulation method employed by Stouch and Jurs (1985a,b) to establish reliable criteria, were used to simulate a study similar to that presented here, involving 266 individuals and 66 descriptors. This study was formulated to determine i f there vcrere substsntia! changes in classifications provided by “real data” as opposed to those of “idealized data.” Six different runs were performed. Parametric discriminant functions were used under all circumstances (F=l or 4; class = 2 or 4). The results obtained by the simulation runs were much lower compared with those with “real data”: for two-group classifications (when F=l), 95.1% vs. 70.9%; and for fourgroup classifications, 87.0% vs. 50.7%. In sum, the 87.4% successful classifications in our study could well be real and based on the actual differential asymmetry level and not due merely to chance. Furthermore, the possibility of high random correct classifications does not preclude the possibility of actual clustering of the skulls. The difference of more than 20% between “real” and idealized data provides evidence for a true tribal-asymmetry relationship, which may not be covered by appropriate random variables that would contain no information pertaining to biological asymmetry. Any numerical technique has limitation3. JVs believe that this should not discourage us, but rather oblige us to use them carefully. DISCUSSION Relatively little research has been carried out on the craniofacial characteristics of the Negev and South Sinai Bedouins, most of it being on the population level (Arensburg, 1973; Henke and Disi, 1981; Hershkovitz et al., 1983; Arensburg et al., 1987; Brown and Smith, 1988). The Bedouin skull has been described as being generally dolichocranic in males and mesocranic in females; facial height was usually low in males but medium in females. All studies mention marked sexual dimorphism in cranial size and shape. Anthropological studies on skulls as well as on living subjects revealed only small morphological differences among the tribes. Hence it was generally concluded that 317 SEPARATING H L W POPULATIONS the Bedouins of southern Israel, Jordan, and the Sinai, Egypt, are a relatively homogeneous population (Vallois, 1937; Arensburg, 1973; Milanesi and Petrucci, 1980; Hershkovitz et al., 1983). This explains why predictions of correct group affiliation based on conventional morphological measures yield poor results (Arensburg et al., 1987) compared with those obtained for human populations that are remote in time and space (Howells, 1973; Falk and Corruccini, 1981). The paired craniometric traits used in the present study enhanced our ability to identify Bedouin skulls by tribe-sex aftiliation, almost to the level attained by Howells (1973) for different ethnic populations (93.2% correctly classified males and 91.9% females). Arensburg et al., (19871, using several conventional measures (e.g., nasal height, cranial length, minimum frontal breadth, nasal breadth, cranial breadth) in discriminant analysis of the same Bedouin populations, could successfully identify by tribe and sex only 47.4%of skulls, as opposed to 87% in the present study. The isolation of and the ongoing inbreeding process within the Bedouin tribes, accompanied from time to time by a sharp decline in population size (as a result of epidemics, drought, etc.), produce different outcomes of the genetic-environmental interaction effect. As a result, the pattern and level of asymmetry will differ from tribe to tribe. The importance of asymmetry traits in discriminating skulls of closely related populations can be appreciated from the following findings: 1) the small contribution of facia! measures to the discrhinatinn hetween Bedouin skulls, with paired facial measures generally displaying the closest associations (probably because of their high heritability) and hence manifesting the least asymmetry (as already shown by Woo, 1931); 2) the reduced contribution of paired bilateral measures toward the discrimination between female Bedouin skulls of different tribes-only one pair of bilateral measures was selected here, as opposed to four for male skull discrimination and nine for the prediction of both sex and tribal affiliation; assumedly, females are better buffered against environmental stress (Hall and Macknair, 1972; Relethford et al., 1980; Stinson, 1985; Rudan et al., 1986) and therefore tend to manifest less asymmetry than males (Stinson, 1985); 3) the large number of paired measures selected by the stepwise procedure when both sex and tribal affiliation are predicted. Undoubtedly, the success of the measures used here should be partly ascribed t o the fact that half of these measures do not span more than one bone, which is highly advantageous from both a genetic and a functional standpoint. Kraus and coworkers (19591, in an article entitled “Heredity and the Craniofacial Complex,’’ claimed that “. . . the singlest type of traits-morphologic aspect of single bones-is the best indicator of the control of hereditary factors in the craniofacia1 complex.” hisewhere, 1Moss and Young (1960)have argued convincingly that each of the cranial bones responds to different functional demands, and therefore measurements like cranial length and width include several functionally independent parts of the skull. More recently, Key and Jantz (1981) maintained that the cranium consists of a number of independently varying units, which may respond in unique ways to a variety of external forces. From all of the above, we may conclude that generally the most commonly used conventional measures (e.g., cranial length and breadth, head circumference, facial breadth) are sums of several inde endent variables and should therefore ehave as any other com osite trait; that is, their variance shou d be greater than that of each of their components separately (for a detailed explanation, see Kobyliansky and Micle, 1986).Moreover, because the variation in size of each bone depends on intrinsic (genetic) and extrinsic (environmental) factors, biological isolation and the ensuing ~ ~ n ~ a nin ~Redouip i ~ i t ~ tribes may also contribute to the variability of conventional composite traits. Hence, not surprisingly, the discriminatory power of most conventional measures that span several bones decreases when the populations are closely related. Falk and Corruccini (1981) claimed that the overall external shape of the skull is more important in evolutionary terms than its constituent parts. Their findings, however, do not necessarily contradict our own because their measures on isolated bones were confined to a small part of the skull (basioccipital area) and because they compared human populations from very distantly related racial stocks. The face, generally considered to be the most discriminatory part of the skull-on an individual, ethnic, or racial level-is relatively similar among the studied Bedouin g P 318 I. HERSHKOVITZ ET AL groups (Arensburg, 1973; Henke and Disi, 1981). This similarity may derive from the fact that many facial traits manifest a high heritability level (see, for example, Sharrna, 1987). Consequently, they will also be less prone to developmental disturbances and as a result will manifest a lower level of asymmetry. It is this characteristic that renders craniofacial traits less useful in discriminating Bedouin populations. phy-Part 11. Tel Aviv, Israel: Eretz, Ministry of Defence, pp. 855-859. Bailey C (1980) The Negev in the nineteenth century: Reconstructin history from Bedouin oral traditions. Asian African%tud 14:35-80. Ben-David J (1978) The Bedouin Tribes in Southern Sinai (in Hebrew). Jerusalem: Keshet. Brown T and Smith P (1988) Craniofacial morphology of two skeletal populations from Israel. Hum. Biol. 60:55-68. Cheverud JM (1982) Phenotypic, genetic and environmental morphological integration in the cranium. Evolution 36:499-516. SUMMARY AND CONCLUSIONS Falk D and Corruccini R (1981) Efficiency, of Franial versus dzntal mea nto h i - s ~ p a r a ~ i nnuman g !a SUZ, S E C C ~ Ssh!! S F ~disjiriininaiiitii vf populations. Am. J . Phys. Anthropol. 57:123-127. closely related human groups cannot be per- Hall RL and Macknair PL (1972) Multivariate analysis formed by gross observation alone, but it of anthropometric data and classification of British Columhian natives. Am. J. Phys. Anthropol. 37:401necessitates more sophisticated statistical 410. tools, through which variation in the size W and Disi AM (1981) The craniology of recent and shape of individual bones of the calvar- Henke Jordanian Bedouins. JMAA 1.217-248. ium and base of the skull, as well as in the Hershkovitz I (1985) The Effect of Mating Type on level of asymmetry, can be detected. Growth and Development of Children in South Sinai Bedouin Isolates. Ph.D. Thesis. Tel Aviv, Israel: Tel Paired measures have great advantage Aviv University. over conventional ones in discriminating I, Kohyliansky E, and Arenshurg B (1983) crania of closely related populations, basi- Hershkovitz Migration and biological isolation of human populacally because they are better measures of the tions in influencing range and variation of metric and genetic-environmental interaction effect. non-metric traits of the skull. Antropos 10:344-356. Besides information on size and shape of Howells WW (1969) The use of multivariate techniques in the study of skeletal populations. Am. J. Phys. skulls of the different Bedouin tribes, they Anthropol. 31:311-315. also provide data on an additional selection Howells WW (1973) Cranial variation in man. Papers of criterion, “developmental noise,” which has Peahody Museum 67:l-269. never been used previously. Jurs PC (1986) Pattern recognition used to investigate multivariate data in analytical chemistry. Science Measures taken on the calvarium and base 235: 1219-1224. of the skull are more effective in skull disP and Jantz RL (1981) A multivariate analysis of crimination than are those of the face. Dis- Key temporal changes in Arikara craniometrics: A methodcriminant functions, which contain mainly ological approach. Am. J. Phys. Anthropol. 55:247conventional measures of the face, will a 259. IOeser YA, Groeneveld HT, and Preston CB (29%) Flucpriori manifest low discriminatory power. tuating dental asymmetry as a ineasure of odontogrnic Because discriminant functions must be canalization in man. Am. J. Phys. Anthropol. 71:437developed independently for each popula444. tion, we suggest that the measures used Kohyliansky E and Livshits G (1983) Relationship beshould be bilateral in nature and should tween level of biochemical and morphologicalvariability in human populations. Ann. Hum. Genet. 47:215cover all three anatomic areas of the cra223. nium, with preference to the calvarium and Kohyliansky E and Livshits G (1985) Differential fertilbase of the skull. ity and morphological constitution of spouses. Z. MorIt is the relationship between population phol. Anthropol. 76:95-105. structure and correlations observed between Kohyliansky E and Micle S (1986) Finger pattern characteristics as determinants of total ridge count varifluctuating asymmetry and heterozygosity ability. Intern. J . Anthropol. 1t49-58. that probably enables the successful disBS, Wise WJ, and Frei RH (1959) Heredity and crimination of Bedouins into their tribal af- Kraus the craniofacial complex. Am. J. Orthodont. 45:172filiations. 217. Leamy L (1986)Directional selection and developmental LITERATURE CITED stability: Evidence from fluctuating asymmetry of morphometric characters in rats. Growth 49:8-18. Arenshurg B (1973)The People in the Land of Israel from the Epipaleolithic to Present Times, Ph.D. Thesis, Tel Leary RF, Allendorf FW, and Knudson RL (1984) SupeAviv, Israel: Tel Aviv University. rior developmental stability of heterozygotes of enzyme loci in Salmonid fishes. Am. Nat. 124:540-541. Arenshurg B, Kohyliansky E, and Hershkovitz I (1987) The effect of isolation and endogamy on mor hological Livshits G and Kohyliansky E (1984) Comparative analand metrical traits amon the Bedouin orsouthern ysis of morphological traits in biochemically homozySinai. In G Gvirtzman, A Shmueli, Y Gradus, I Beitgous and heterozygous individuals from a single popArieh, and M Har-El (eds.): Sinai. Human Geograulation. J . Hum. Evol. 13,161-171. SEPARATING H L W POPULATIONS Livshits G and Kobyliansky E (1987) Dermatoglyphic traits as possible markers of developmental processes in humans. Am. J . Med. Genet. 26~111-122. Livshits G and Kobyliansky E (1989) Study of genetic variance in the fluctuating asymmetry of anthropometrical traits. Ann. Hum. Biol. 16:121-129. Marx E (1974) Bedouin of the Negev (in Hebrew). Tel Aviv, Israel: Reshafim. Martin R and Saller K (1957) Lehrbuch der Anthropologie. Stuttgart: Gustav Fisher Verlag. Milanesi Q and Petrucci S (1980)Aspetto fisico e misure antro ometriche di un gruppo di Beduini nell'ambiente lesertico della Transgiordania Meridionale. Archivio per YAntropologia 110~267-276. Moss ML and Young RW (1960)A functional approach to craniology. Am. J . Phys. Anthropol. 18~281-292. Muhsam HV (1966)Bedouin or the Negev: Eight Demographic Studies. Jerusalem: Academic Press. Nakata M, Yu P, Davis B, and Nance WE (1974) Genetic determinants of craniofacial morphology: A twin study. Ann. Hum. Genet. 37~431-443. Nie NH, Hull CH, Jenkins JG, Steinbrenner K and Bent DH (1975) SPSS-Statistical Package for the Social Sciences. New York: McGraw-Hill. Palmer AR and Strobeck C (1986) Fluctuating asymmetry: Measurements analysis patterns. Annu. Rev. Ecol. Syst. 17:391-421. Relethford JH (1988) Effects of English admixture and geographic distance on anthropometric variation and genetic structure in 18th-century Ireland. Am. J. Phys. Anthropol. 76~111-124. Relethford JH, Lee FC, and Crawford MH (1980) Population structure and anthropometric variation in rural western Ireland: mi ation and biological differentiation. Ann. Hum. Bior7:411-428. Rudan P, Roberts DF, Janicijevic B, SmolejN, Szirovicza L, and Kastelan A (1986) Anthropometry and the biological structure of the Hvar population. Am. J. Phys. Anthropol. 70~231-240. 319 Sharma K (1987) Familial resemblance for craniofacial traits in a Punjabi population. Z. Morphol. Anthropol. 77~157-165. Sharon M (1977) The Bedouin and Eretz-Israel Under Islamic government. In E Sohar (ed.): The Desert, Tel Aviv, Israel: Reshafim, pp. 199-211. Siege1 MI and Smookler HH (1973) Fluctuating dental asymmetry and audiogenic stress. Growth 37:35-39. Soule ME (1979) Heterozygosity and developmental stability: Another look. Evolution 33~396-401, Stinson S (19851Sex differences in environmental sensitivity during growth and development. Yrbk. Phys. Anthropol. 28:123-147. Stouch TR and Jurs PC (1985a)Monte Carlo studies of the classifications made by nonparametric linear discriminant functions. J. Chem. Inf. Comput. Sci. 25:45en c)v. Stouch TR and Jurs PC (1985b) Monte Carlo studies of the classifications made by nonparametric linear discriminant functions. 2. Effects of nonideal data. J . Chem. Inf. Comput. Sci. 25~92-98. Stuper AJ and Jurs PC (1976) A computer system for automated data analysis using pattern recognition techniques. J. Chem. Inf. Comput. Sci. 4:99-105. Susanne C (1977)Heritability of anthropological characters. Hum. Biol. 49:573-580. Vallois HV (1937) Les Bedouins Taamre du Desert de Judee. Etude anthropologique. L'Anthropologie 63:62-92. Varmuza K (1980) Pattern Recognition in Chemistry. New York Springer-Verlag. Vig P and Hewitt AB (19741 Is craniofacial asymmetry and adaptation for masticatory function an evolutionary process? Nature 248~165. Woo TL (1931) On the asymmetry of the human skull. Biometrika 22:324-351. Young R (1957) Postnatal growth of the frontal and parietal bones in white males. Am. J . Phys. Anthropol. 15:367-386.