Citation
Sex Determination by Discriminant Function Analysis of Native American Crania from Florida and Georgia

Material Information

Title:
Sex Determination by Discriminant Function Analysis of Native American Crania from Florida and Georgia
Creator:
Mcginnes, Michael B
Place of Publication:
[Gainesville, Fla.]
Florida
Publisher:
University of Florida
Publication Date:
Language:
english
Physical Description:
1 online resource (105 p.)

Thesis/Dissertation Information

Degree:
Master's ( M.A.)
Degree Grantor:
University of Florida
Degree Disciplines:
Anthropology
Committee Chair:
Falsetti, Anthony B.
Committee Members:
Hollinger, Thomas G.
Graduation Date:
12/14/2007

Subjects

Subjects / Keywords:
Anthropology ( jstor )
Calipers ( jstor )
Discriminants ( jstor )
Native Americans ( jstor )
Paleoanthropology ( jstor )
Pelvic bones ( jstor )
Sex determination ( jstor )
Skeleton ( jstor )
Skull ( jstor )
Sliding ( jstor )
Anthropology -- Dissertations, Academic -- UF
archaeology, bioarchaeology, cranium, discriminant, florida, georgia, osteology, prehistoric, sex, southeast
City of Gainesville ( local )
Genre:
bibliography ( marcgt )
theses ( marcgt )
government publication (state, provincial, terriorial, dependent) ( marcgt )
born-digital ( sobekcm )
Electronic Thesis or Dissertation
Anthropology thesis, M.A.

Notes

Abstract:
The goal of this research is to determine if the accuracy of discriminant function analysis for sex determination could be improved by using local or regional populations, and by better variable selection. In archaeological contexts, skeletons are usually all that remains of the actual people who once lived there. Information about the sex of those individuals is fundamental to the study of demographics, sex roles, and life ways of past cultures. Determining an individual?s sex is also fundamental to personal identification of skeletal remains from modern contexts, whether from a mass disaster or an unmarked grave. There are several techniques available for determining sex from skeletal remains, and each technique has its place. Discriminant function analysis is valuable in sex determination because it requires relatively little training to use effectively and it serves as an objective check to other methods. The cranium is a reliable indicator of sex and is often the best indicator of sex when other parts of the skeleton have been damaged, destroyed, or separated from the cranium. Previous research assumed that discriminant functions for sex determination developed for one population could easily be used for all populations, with little regard for the skeletal variation between populations. This work tests the hypothesis that sex determination by discriminant function analysis of the crania for Florida and Georgia Native American remains from archaeological contexts is more accurate when the functions are developed using remains from those populations than functions developed from other populations. Cranial measurements and sex identification data are collected from skeletal collections housed at the Florida Museum of Natural History at the University of Florida, and the Smithsonian Institution?s National Museum of Natural History. The sample includes 46 individuals from ten Native American archaeological sites in Florida and Georgia, ranging from the Middle Archaic Period to the Spanish contact era. This research finds that existing discriminant function formulas disproportionately misclassify skeletons from the Florida and Georgia unless the formula is adjusted for that population. Additionally, existing formulas require measurements that are rarely preserved or that do not contribute to identifying sex. New discriminant function formulas based on skeletons from Florida and Georgia are only nominally more accurate than existing formulas, but it is no more difficult to produce new formulas than to adjust existing formulas. Creating new formulas also provides the opportunity to select variables that are more often preserved in archaeological contexts and that also clearly contribute to identifying the sex individuals. This research finds that existing formulas are reliable only if the sectioning point is adjusted for the study population. ( en )
General Note:
In the series University of Florida Digital Collections.
General Note:
Includes vita.
Bibliography:
Includes bibliographical references.
Source of Description:
Description based on online resource; title from PDF title page.
Source of Description:
This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Thesis:
Thesis (M.A.)--University of Florida, 2007.
Local:
Adviser: Falsetti, Anthony B.
Statement of Responsibility:
by Michael B Mcginnes.

Record Information

Source Institution:
UFRGP
Rights Management:
Copyright Mcginnes, Michael B. Permission granted to the University of Florida to digitize, archive and distribute this item for non-profit research and educational purposes. Any reuse of this item in excess of fair use or other copyright exemptions requires permission of the copyright holder.
Classification:
LD1780 2007 ( lcc )

Downloads

This item has the following downloads:


Full Text





measurements, Taylor and DiBennardo (1982) recorded accuracies between 80 and 85% for

white femora.

Dittrick and Suchey (1986) minimize the problem of unknown sex by using only skeletons

with the pubic bone and applying the Phenice method. Suchey had established an accuracy of

99% in sex determination by using a blind test on pubic bone pairs from modern autopsies of

individuals over age 16. In tests on prehistoric central California skeletal remains, they achieved

accuracies of about 90% using linear discriminant function analysis of measurements from the

ends of the long bones. Interestingly, functions based on multiple measurements did not produce

results much better than the best functions using single measurements.

One of the limitations of discriminant function analysis is that the functions should only be

used on skeletons that come from the same population as the one used in development of the

function. Otherwise, results can be unpredictable. Giles and Elliot (1963) tested their own

functions on skeletons from Ireland and on three series of native American skeletons from Indian

Knoll in Ohio, Pecos Pueblo, and Florida. The functions correctly sexed 40 of 42 males (95%)

and 3 of 8 females (37.5%) of the Irish skeletons. Giles and Elliot dismiss the poor female result

and take this as evidence that their formula can be used across populations. A better

interpretation is that the formula disproportionately misclassifies females as male. For the Native

American samples, good results are achieved only after altering the sectioning point. Although it

has been used in the literature, altering sectioning points is not a practical solution to using

discriminant functions across populations (Calcagno 1981; Henke 1977). Kajonoja (1966) found

that the Giles and Elliot functions had an accuracy of only 65% on Finnish crania. It is

interesting to note that the functions developed by Giles and Elliot (1963) for the combined

sample of blacks and whites worked about as well on blacks and whites separately as it did on




























To my grandmother,
Sarah Napps, she loved me for who I was,
no matter what. She supported me in
in every way she could, no matter what.
She saw the best in me, and was proud of me, no matter what.
I strive to live up to her image of me, and her pride in me.









CHAPTER 7
RESULTS

General Results

This section presents the results of the analysis described in Methods and Materials.

Discriminant function scores are calculated for each individual using the Giles and Elliot (1963)

formula for sex determination. Each individual is classified as male or female using the

sectioning point published by Giles and Elliot. Each individual is classified again using a

recalculated sectioning point based on sample data following a method recommended by Giles

and Elliot (1963). Using the published Giles and Elliot (1963) sectioning point, the formula

correctly classifies 13 of 13 males, and 2 of 4 females. The overall accuracy is 88.24%, but the

error is unequally distributed. Using the recalculated sectioning point, the formula correctly

classifies 12 of 13 males, and 3 of 4 females. The overall accuracy is still 88.24%, but the error is

evenly distributed. This error rate is not significantly different than that reported by Giles and

Elliot, but the sample size is unacceptably small due to missing variables.

In order to increase the sample size, missing variables are replaced with the average of the

male and female means for each variable. This allows the calculation of the discriminant

function to be completed without allowing missing variables to influence sex classification.

Discriminant function scores are calculated for each individual using the Giles and Elliot (1963)

formula for sex determination. Each individual is classified as male or female using the

sectioning point published by Giles and Elliot (1963). Each individual is classified again using a

recalculated sectioning point based on sample data following a method recommended by Giles

and Elliot. Using the Giles and Elliot sectioning point, the formula correctly classifies 31 of 32

males, and 2 of 13 females. The overall accuracy is 73.33%, but the error is unequally









measured, the right orbit is measured and the side recorded on the measurement sheet (Martin

1956:477 478; Howells 1973:175).

Orbital Height (OBH): The direct distance between the superior and inferior orbital

margins measured using a sliding caliper. Orbital height is measured perpendicular to orbital

breadth and similarly bisects the orbit. The measuring points are located on the opposing margins

of the orbital borders. Any notches or depressions on either superior or inferior borders are

avoided and the margin is projected when necessary (Bass 1971:69; Martin 1956:478; Montagu

1960:51; Olivier 1969:152).

Biorbital Breadth (ec-ec, EKB): The direct distance from one ectoconchion (ec) to the

other measured using a sliding caliper (Howells 1973:178).

Interorbital Breadth (d-d, DKB): The direct distance between right and left dacryon

measured with a sliding caliper (Martin 1956:477).

Frontal Chord (n-b, FRC): The direct distance from nasion (n) to bregma (b) taken in the

midsagittal plane using a sliding caliper. The skull is rested on its right side to view the left

profile of the frontal region. The tips of the instrument are placed on the bone surface or at the

level of this surface and not in a suture or other depression (Howells 1973:181; Martin

1956:465).

Parietal Chord (b-l, PAC): The direct distance from bregma (b) to lambda (1) taken in the

midsagittal plane measured with a sliding caliper. The skull is left in the same position used to

measure the Frontal Chord (above). The tips of the instrument are placed on the bone surface or

at the level of this surface and not in a suture or other depression (Howells 1973:182; Martin

1956:466 ).









Wilmington Culture

The Wilmington Culture succeeded the Deptford culture along the coast and coastal plain

of Georgia at the end of the Late Woodland period, 500-1150AD. The Wilmington Culture is

defined by Wilmington Plain, Wilmington Cord Marked, and Wilmington Brushed ceramics

(DePratter 1979; Martinez 1975). The typical Wilmington vessel is decorated with large, parallel

individual cord impressions made with a cord wrapped paddle (Caldwell 1952:316). Wilmington

culture is thought to have been influenced by the coeval Weeden Island culture in Florida and by

Mississippian people in the Piedmont (Milanich 1976). Like other Late Woodland cultures,

Wilmington subsistence was based on hunting, fishing and gathering with some horticulture

(Wood et al 1986).

Historic Period

The historic period begins with European contact. This happens at different times in

different areas, and initially has varying degrees of impact. For the Timucua of Northern Florida

and Southern Georgia, contact with Europeans probably begins in 1525 and early 1526 when

scout ships that preceded the Lucas Vasquez de Ayll6n expedition landed on the northern end of

St. Simons Island (Hoffman 1994; Milanich 2004b:225). In 1565 the Spanish began to establish

a missions and colony in Florida with their first permanent New World settlement in St.

Augustine. From this base they established Roman Catholic Missions along the Atlantic Coast to

the Timucua Indians and Guale Indians along the Georgia coast just north of the Timucua. At the

time of the Spanish arrival, many of the Guale and Timucua in northern Florida and Georgia

were already involved in maize agriculture and readily missionized. By 1620 virtually every

Timucuan chiefdom had received Franciscan missions (Milanich 2004b:225). Non-agricultural

groups south of the Timucua, such as the Calusa, were not missionized, despite numerous

Spanish attempts.









the combination of both groups. This suggests that functions developed on a wider population

can achieve good results across that population's constituent groups. (See Henke 1977 for a more

detailed discussion of using discriminant function analysis across populations.)

The methods and criteria discussed above are only applicable to adult remains. A skeleton

may be considered an adult if all long bone epiphyses are fused or if the third molars have

erupted. For sub-adults, if age can be established from dental development, then sex can be

estimated from long bone lengths (see Bass 1995). Other sexing techniques for sub-adults use

sex differences in pelvic measurements (see Krogman and Iscan 1986:200-208).









and MDHA. Discriminant function scores are calculated for 38 individuals, 26 males and 12

females. The new function correctly classifies 11 of 12 females and 24 of 26 males for a

combined accuracy of 91.99%. The discriminant function is then applied to the full sample,

substituting missing values with the average of the male and female means from the test sample.

This function assigned sex correctly to 12 of 13 females and 27 of 32 males, for a total accuracy

of 86.67%. This is not significantly different than the accuracy of the Giles and Elliot function.

Specific Results

Giles and Elliot Discriminant Function

Discriminant function scores are calculated for each individual using Giles and Elliot's

(1963) discriminant function 3 based on a combined sample of black and white individuals (see

Table 2-1, function 3). The purpose of this task is to establish the accuracy of the Giles and Elliot

function on Florida and Georgia Native Americans for comparison to new discriminant

functions. Thirteen males and four females are complete enough to record all eight

measurements needed to use the Giles and Elliot function. Using the sectioning point reported by

Giles and Elliot (1963), the formula classifies individuals with discriminant function scores

above 6237.95 as male, and individuals below that score as female. The formula correctly

classifies all 13 males, but only classifies 2 of 4 females correctly. While the overall accuracy of

88.24% compares well with the Giles and Elliot result of 86%, the error is not equally distributed

between males and females. All males are correctly classified while half of the females are

misclassified. This violates the requirement of DFA that error be spread equally among groups

(Kendal 1957). This type of error is known to occur when a discriminant function is applied to a

group that is not included in the development of the formula.

Giles and Elliot report this type of error where they attempt to use their formula on

different groups. To avoid this problem, Giles and Elliot recommend recalculating the sectioning









Kellog

The Kellog culture was concentrated in northwest Georgia Piedmont between about 800

and 200 BC. Kellog settlements include large, year round camps concentrated in narrow flood

plains adjacent to streams, and small seasonal camps. Kellog base camps covered about an acre

and include abundant artifacts, some sites having middens several feet thick. Seasonal camps are

not as common and contain fewer artifacts. Kellog material culture is dominated by fabric

marked pottery early in the period, with simple stamp and check stamp pottery more popular

later. Stone tools included both stemmed and unstemmed chipped stone points with triangular

blades, slate hoes, and biconvex mortars (Bense 1994:135). Kellog culture subsistence strategies

were not much different than Archaic strategies of hunting, gathering and fishing, but with

perhaps more emphasis on plant foods (Bense 1994:135-136; Hally and Mainfort 2004:266).

Deptford

The Deptford culture was located along the Gulf coast of Florida and the southeast Atlantic

coast between 500 BC and AD 100. The Deptford culture area is located between Mobile Bay

and Cedar Key along the Gulf coast stretching inland approximately 60 miles, and along the

Atlantic coast of South Carolina, Georgia, and northern most Florida extending 30 miles inland.

Modest Deptford shell middens found along the coast are located in hardwood hammocks near

salt marshes and estuaries, while inland sites are usual located in river valleys. Deptford coastal

villages are small and generally contain 5 to 10 houses of either cold weather houses or summer

warm weather pavilions. The cold weather houses are around 20 by 30 foot ovals, and the warm

weather pavilions are approximately 20 by 13 foot ovals. As might be expected from their choice

of site locations, Deptford peoples relied heavily on fish and shellfish gathered from tidal

streams and shallow inshore waters, as well as nearby terrestrial resources. Inland sites are small

as well and may be special use sites, such as hunting camps (Milanich 2004a: 193-194). "In both









variables GOL, BBH, BPL, and MDH. Variables included in the Giles and Elliot formula which

are not significant are: XCB, ZYB, and MAB.

A variable is considered to be poorly preserved if it is not observable in at least 90% of

cases for each sex. For this sample, the variable has to be observable in at least twelve females

and at least 29 males. Five of the significant variables, BBH, BNL, BPL, MDHR, and MDHL,

are excluded because of poor preservation. Five variables, GOL, AUB, WFB, PAC, and MDHA,

meet the criteria for significance and preservation. These five variables are used in discriminant

function analysis for the determination of sex in Florida and Southern Georgia Native

Americans.

Test for Site Effect on Selected Variables

The purpose of testing for site effect is to determine if individuals from different sites in

Florida and Georgia can be treated as a single population. If there is no significant site effect,

then a single discriminant function for sex determination can be used for all sites.

Table 7-3 F values and P values for the hypothesis of no site effect for each variable using type
IV sums of squares and cross products.
Variable DF F P
GOL 8 1.35 0.2581
AUB 8 3.11 0.0116
WFB 8 1.60 0.1688
PAC 8 1.31 0.2756
MDHA 8 0.61 0.7620


An ANOVA is first performed on individual variables to test for the effects of site (Table

7-3). There is no significant site effect for any of the variables at the 0.01 alpha-level. There is a

significant site effect for AUB at the 0.05 alpha-level. This difference is driven by a difference

between the Casey Key site and the Golf Course site, each of which is represented by one

individual. For this variable, the individual from Golf Course is a larger-than-average male, and









pubic bone. This individual is sexed using a combination of cranial and postcranial visual

methods, excluding the pubic bone.

Data are collected from one female cranium recovered from Casey Key. This cranium is

nearly complete, allowing observation of 22 cranial measurements. The postcrania including the

pubic bone is missing. This individual is sexed using visual analysis of the cranium.

Data are collected from one malerecovered from Bay Pines. This individual is nearly

complete, allowing observation of all 25 cranial measurements but not the pubic bone. This

individual is sexed using a combination of cranial and postcranial visual methods, excluding the

pubic bone.

Data are collected from one male recovered from Ballard Estates. This individual is nearly

complete, allowing observation of 23 of 25 cranial measurements but not the pubic bone. This

individual is sexed using a combination of cranial and postcranial visual methods, excluding the

pubic bone.

Cranial Measurement Definitions

Giles and Elliot Measurements

Ten measurements are taken from Giles and Elliot (1962, 1963) following their

descriptions (Table 6-1). Other sources are used to clarify descriptions and measurement

techniques where Giles and Elliot are unclear, particularly Bass (1971, 1995), Howells (1973),

Buikstra and Ubelaker (1994), and FORDISC 2.0 materials (Ousley and Jantz 1996).

The remaining measurements (Table 6-2) are not included in the Giles and Elliot (1962)

study, but are used in more recent research by Bass (1995), Howells (1973), Buikstra and

Ubelaker (1994), and FORDISC 2.0 materials (Ousley and Jantz 1996). The descriptions and

techniques primarily follow Ousley and Jantz (1996).









Live Oak hammock, tidal flats, and estuary habitats. Farther inland, the sandy beaches and dunes

give way to broad estuaries, mudflats, and thickly vegetated shores. These areas serve as place

for marine and inland species to interact (Milanich and Fairbanks 1978; Williams 2004:13).

Inland areas of northern Florida and Georgia with an abundance of fresh water, such as along

streams and lakes, are dominated by hardwood hammocks. Hardwood hammocks are dense

forests characterized by a broad spectrum of plant communities that provide shelter and food for

a vast array of animals (Milanich 1998; Milanich and Fairbanks 1978; Wallace 1978).

Cultural History

Prehistoric occupation of the southeastern United States, or Southeast, is divided into

periods and sub-periods that reflect changes in technology, environment, and subsistence. Breaks

between periods are often subtle, and changes occur at different times across the region.

Archaeological cultures describe specific regional or local units from specific time periods.

Periods include Paleoindian, Archaic, Woodland, and Mississippian. The Archaic, Woodland,

and Mississippian periods are divided into Early, Middle, and Late sub-periods. The Woodland

and Mississippian cultures are comprised of a number of regional cultures.

Paleoindian Period

The initial period of human occupation in North American is termed the Paleoindian

period, and is characterized by the occurrence of fluted stone projectile points or knives, such as

Clovis, Suwannee, and Simpson points. The best available evidence suggests that during the late

Pleistocene bands of highly moble hunters crossed the Bering land bridge from Siberia into

Alaska. The Bering land bridge was exposed during the last glaciation when large amounts of

water were locked up in the polar ice caps and ice sheets, resulting in the lowering of global sea

levels by about 100 meters. The presence of mid-continent ice sheets appear to have prevented

movement of these populations eastward from Alaska until about 13,000 B.C. based on the









LIST OF TABLES


Table pae

2-1 Coefficients for discriminant functions 1,2, and 3 from Giles and Elliot 1963 used to
assign sex based on a white sample, a black sample, and a combined black and white
sam p le .............. ............................. .. .......................................... . ..... ...... 2 3

6-1 Measurements used by Giles and Elliot. ...................... .. .......................... 59

6-2 M measurements not used by Giles and Elliot ................ .................... ................. 60

7-1 Accuracy of the Giles and Elliot Function 3 for sex determination on the study
sample of Florida and Georgia Native Americans ........... ................................ 68

7-2 T-scores and p-values for a difference in mean values of each cranial variable
between males and females. ............ ...... ............................. 70

7-3 F-values and p-values for the hypothesis of no site effect for each variable .................. 71

7-4 MANOVA test criteria and F approximations for the hypothesis of no overall site
effect...... ......................... ........ ................ .. 72

7-5 Coefficients, group means, and sectioning points for Function 1 and Function 2............ 73

7-6 A accuracy of Function 1 .............................................................................. ....... 74

7-7 A accuracy of F unction 2 ................................................................. .. ........................ 75









more adult males than females. A few glass beads are present suggesting that the mound dates

from the 17th century, the same as Fuller Mound A (Stirling 1935:387).

Casey Key (8So17)

The Casey Key site lies on Casey Key, a Gulf coast barrier island located about three miles

south of Osprey, in Sarasota County, Florida. The site was known to residents of the area long

before it was recorded by archaeologists and has never been systematically excavated. The site is

near the Palmer site, discussed below, and is thought to have been roughly contemporaneous

with it, although it is not tightly dated due to a dearth of diagnostic artifacts. The limited pottery

assemblage indicates it is from the Manasota-Weedon Island culture dating from ca. A.D. 250-

750. Casey Key included a village and a burial mound that is thought to have contained over 200

burial but most of these are collected by local residents or sold by high school students. A few

skeletons were donated to the Florida Museum of Natural History in the 1950s and 1960s by

Hilton Leech, who attempted to salvage some data from the mound before it was completely

destroyed (Bullen and Bullen 1976:47-48).

Palmer Burial Mound (8So2a)

The Palmer Site is a complex of sites located near Osprey, in Sarasota County, on Little

Sarasota Bay on the Florida Gulf Coast. The site was the subject of a number of scientific

excavations. The first formal excavations by Ripley Bullen between 1959 and 1962 are the most

extensive and only ones to include the burial mound (Bullen and Bullen 1976). Those

excavations recorded five sites: Hill Cottage Midden, dating to the Archaic period; Shell Ridge,

dating to the Middle Woodland period; Shell Midden, dating from Middle Woodland through

Mississippian periods; the North Creek Area middens; and Palmer Burial Mound. A survey of

the Palmer tract performed in 1974 uncovered four additional sites, including another burial

mound (Miller 1974). Limited investigations were performed in 1979 and 1980 prior to the









CHAPTER 8
CONCLUSIONS AND DISCUSSION

Conclusions

The most significant finding of this research is that the function derived by Giles and Elliot

does not accurately sex skeletons from Florida and Georgia unless the sectioning point is

adjusted. Using the sectioning point published by Giles and Elliot (1963), the function classifies

only 15.38% of females correctly. When the sectioning point is adjusted using male and female

averages for each variable, the function classifies about 84% of males and females correctly. For

Florida and Georgia Native Americans, the sectioning point should be changed to 6513.4 when

using Giles and Elliot's (1963) function 3.

The overall goal of this research is to determine if the accuracy of discriminant function

analysis for sex determination could be improved by using local or regional populations, and by

better variable selection. Previous research has tested the Giles and Elliot (1963) discriminant

functions on Finnish crania and developed functions for that population (Kajanoja 1966). Other

research has developed discriminant functions for fragmentary archaeological post-cranial

remains from other areas (Bass 1995; Black 1978; Krogman and Iscan 1986). None of the

previous studies focuses on comparing the established Giles and Elliot methods to new formula

developed from fragmentary crania from Southeastern archaeological sites, thus making this

research novel. This research is also novel in treating all prehistoric Native Americans in Florida

and Georgia as a single population, regardless of time period. This study addresses the following

hypotheses related to sex determination by discriminant function analysis (DFA):

For archaeological populations, higher accuracy in sex determination by DFA can be
achieved drawing a sample from the archaeological population than by using a dissecting
room sample.

Sex determination by DFA can be accomplished with a smaller number of more robust
measurements without reducing accuracy.









APPENDIX A
CRANIAL MEASUREMENT DEFINITIONS

Giles and Elliot Measurements

The following are the ten measurements used by Giles in Elliot in their 1963 study. The

name of the measurement is followed by the measurement's abbreviation, Giles and Elliot's

description of the measurement, the name of the equivalent Howells/FORDISC measurement,

and the method used to record the measurement. Detailed descriptions of cranial points and

landmarks can be found in Bass (1995), White (1991), and others.

Glabello-occipital length (g-op, GOL): Maximum length of the skull, from the most

anterior point of the frontal in the midline to the most distant point on the occiput in the midline.

This measurement is equivalent to Maximum Cranial Length; the distance of Glabella (g) from

Opisthocranion (op) in the mid sagittal plane measured in a straight line using spreading calipers.

The skull is placed on its side for this measurement. The endpoint of the left branch of the caliper

is placed on Glabella and held with fingers while the endpoint of the right branch of the caliper is

applied similarly to the posterior portion of the skull in the mid sagittal plane until the maximum

length is obtained (Bass 1971:62; Howells 1973:170; Martin 1956:453; Olivier 1969:128).

Maximum width (eu-eu, XCB): The greatest breadth of the cranium perpendicular to the

median sagittal plane, avoiding the supra-mastoid crest. This measurement is equivalent to

Maximum Cranial Breadth; the maximum width of the skull perpendicular to the mid-sagittal

plane wherever it is located with the exception of the inferior temporal line and the immediate

area (i.e. the posterior roots of the zygomatic arches) measured with spreading calipers. The

Maximum Cranial Breadth is measured with the skull resting either on its base or on the occiput.

The two measuring points lie in the same horizontal and frontal planes. The arms of the caliper

are placed at the same level while maintaining the hinge joint of the caliper in the mid sagittal









an excellent example of the kind of where the present research will be of the greatest benefit,

namely in gathering information about isolated remains with uncertain affiliation.

Bay Pines (8Pi64)

The Bay Pines site is located in Pinellas County, Florida on Boca Ciega Bay, just west of

St. Petersburg. The site consists of a shell ridge oriented on a north-south axis, parallel to the

coast of Boca Ciega Bay. The north end of the ridge is near a freshwater lagoon, and two smaller

ridges are present perpendicular to the shore, one in the middle, and one at the south end termed

"shell ridge A." Shell ridge A was excavated by members of the Suncoast Archaeological

Society in 1971 in a salvage operation prior to the construction of a nursing home. Ten burials

are identified in a cemetery and fragmentary remains of at least 14 other individuals are found

scattered throughout what was probably a burial mound. The remains of all 24 individuals are

currently housed at the Florida Museum of Natural History. The site is thought to be multi-

component with occupations from the Deptford period through the early Weeden Island period.

A reanalysis of faunal remains associated with the site and stable isotope analysis of the

skeletons suggest a diet heavy in fish from the nearby Gulf of Mexico and included turtle,

mammal, bird, crab and possibly maize (Gallagher and Warren, 1975; Kelly, Tykot and Milanich

2006).

Canaveral (8Br85)

The Canaveral site is located on Cape Canaveral in Brevard County, in the Indian River

area of Florida's Atlantic coast. Dr. George Woodbury excavated burial mounds on Cape

Canaveral from 1933-1934 as part of the Civil Works Administration relief archaeological

program affiliated with the Bureau of American Ethnology, Smithsonian Institution during the

Great Depression (Milanich 1994:9-10). The site consists of several burial mounds, including the

Burns Mound and Fuller Mounds A, B, and D. The Burns Mound (8Br85) is a burial mound built









LIST OF REFERENCES


Almy, Marion M., and George M. Luer
1993 Guide to the Prehistory of Historic Spanish Point. Historic Spanish Point, Osprey,
Florida.

Austin, Robert J., and Michael Russo
1989 Limited Excavations at the Catfish Creek Site (8So608), Sarasota County,
Florida. Piper Archaeological Research, Inc., St. Petersburg, Florida.

Anderson, David G., and Kenneth E. Sassaman
2004 Early and Middle Holocene Periods, 9500 to 3750 B.C. In .,tiheati, edited by
Raymond D. Fogelson, pp. 87-100. Handbook of North American Indians, Vol.
14, William C. Sturtevant, general editor, Smithsonian Institution, Washington,
D.C.

Bass, William M.
1971 Human Osteology: A Laboratory and FieldManual of the Human .\keleli 2nd
ed. Missouri Archaeological Society, University of Missouri, Columbus.

1987 Human Osteology: A Laboratory and FieldManual of the Human .\keleslu 3rd
ed. Missouri Archaeological Society, University of Missouri, Columbus.

1995 Human Osteology: A Laboratory and FieldManual of the Human .\keleli 4th
ed. Missouri Archaeological Society, University of Missouri, Columbus.

Bense, Judith A.
1994 Archaeology of the .Sniet /hie.i n United States: Paleoindian to World War I.
Academic Press, San Diego.

Birkby, Walter H.
1966 An Evaluation of Race and Sex Identification from Cranial Measurements.
American Journal of Physical Anthropology 24:21-28.

Black, Thomas K.
1978 A New Method for Assessing the Sex of Fragmentary Skeletal Remains: Femoral
Shaft Circumference. American Journal of Physical Anthropology 48:227-232.

Boas, Franz
1910 Report Presented to the 61st Congress on Changes in Bodily Form of
Descendants of immigrants. Government Printing Office, Washington, D.C.

1912 Changes in the Bodily Form of Descendants of Immigrants. American
Anthropologist 14:530-562.

Brothwell, Donald R.
1981 Digging Up Bones: The Excavation, Treatment and Study of Human .,\keletil
Remains. 3rd ed. Cornell University Press, Ithaca, New York.









Wallace, Ronald L.
1975 An Archaeological, Ethnohistorical and Biochemical Investigation of the Guale
Aborigines of the Georgia Coastal Strand. Unpublished Ph.D. dissertation,
Department of Anthropology, University of Florida. Gainesville.

White, Tim D.
1991 Human Osteology. Academic Press, San Diego.

Widmer, Randolph J.
1988 The Evolution of the Calusa: A Nonagricultural Chiefdom on the S.i,,nl e'\t
Florida Coast. University of Alabama Press, Tuscaloosa.

Willey, Gordon R.
1949 Archeology of the Florida Gulf Coast. Smithsonian Miscellaneous Collections
113. Smithsonian Institution, Washington, D.C.

1954 Burial Patterns in the Burns and Fuller Mounds, Cape Canaveral, Florida. Florida
Anthropologist 7:79-90.

Williams, Shanna E.
2004 Dental Asymmetry Through Time in Coastal Florida and Georgia. Unpublished
Master's thesis, Department of Anthropology, University of Florida, Gainesville.

Williams, Wilma B.
1983 Bridge to the Past: Excavations at the Margate-Blount Site. Florida
Anthropologist 36:142-153.

Wilmsen, Edwin N.
1965 An Outline of Early Man Studies in the United States. American Antiquity
31:172-192.

Wood, W. Dean, and William R. Bowen
1995 Woodland Period Archaeology in Northern Georgia. University of Georgia
Laboratory of Archaeology Series Report Number 33. Athens.

Wood, W. Dean, Dan T. Elliot, Teresa P. Rudolph, and Dennis B. Blanton
1986 Prehistory in the Richard B. Russell Reservoir: The Archaic and Woodland
Periods of the Upper Savannah River. The Final Report of Data Recovery at the
Anderson and Elbert County Groups: 38AN8, 38AN126, 9EB17, and 9EB21.
Russell Papers. Interagency Archaeological Services, National Park Service,
Atlanta.

Woodbury, George
n.d. Preliminary Report on Excavation of Mortuary Mounds in Brevard County,
Florida East Coast. Manuscript on file, Department of Anthropology (former
Bureau American Ethnology), Smithsonian Institution, Washington, D.C.
(Archaeological Report of CWA Project 5-F-70, December 18, 1933-February 15,
1934.)









CHAPTER 5
SITES

Introduction

Of the ten archaeological sites used in this study, six (Golf Course, Bay Pines, Canaveral,

Casey Key, Palmer, and Perico Island) are located along the coast of Florida, and three

(Cannon's Point, Taylor Mound, and Couper Field/Indian Field) are located on Saint Simon's

Island, Georgia. The tenth site (Garfield) is located in the Piedmont of Georgia. These sites

represent to Archaic, Deptford, Weeden Island, and contact periods. These sites date from

possibly as early as the Paleoindian through the Contact period. Site numbers and brief

descriptions of the location, the archaeological investigation, and the interpretation of each of

these sites follow below.

Golf Course (8Br44)

The Golf Course site is located on the north edge of the Melbourne Municipal Golf Course

just east of a canal that cuts through the property in Melbourne, Brevard County, Florida. The

site was discovered in 1952 by F.B. Loomis of Amherst College, J.W. Gidley of the U.S.

National Museum (Smithsonian), and C.P. Singleton, a resident of Melbourne, during a survey of

spoil from the nearby canal (Rouse 1951: 153). Loomis and Gidley prevaricated on whether the

remains came from the Pleistocene Melbourne bone bed, or from the lower levels of the

overlying Holocene Van Valkenburg bed. Ales Hrdlicka argued that the "Melbourne Man" was

similar to recent Indians and could not be of great antiquity based on his analysis of the skulls

morphology and his own conviction that humans had not entered the New World more than a

few thousand years ago (Miller 1950; Wilmsen 1965). After reconstructing and reexamining the

skull, Stewart (1946) suggested that it might in fact belong to the Paleoindian period. (Milanich

1994:8; Miller 1950) While the lack of a definitive cultural affiliation is frustrating, this case is









5 S IT E S ...................................................................................... ............. 4 7

In tro d u ctio n ........................................................... ...................................... 4 7
Golf Course (8Br44) ...................................................................... ......... 47
B ay P in e s (8 P i6 4 ) ................................................................. ..................................... 4 8
Canaveral (8Br85) ........................................................ .................. 48
Casey Key (8So17) ....................................................................... ......... 50
Palmer Burial M found (8So2a) ......................................... ................... .. ......... 50
P erico Islan d (8M a6 ) .............................................................. ................................... 5 1
St. Sim ons Island, G eorgia................................... .............. 51
G airfield Site (9B R 57) .......................................................................................... ........ 53

6 M ETHODS AND M ATERIALS ........................................................... ....... 55

Intro du action ..................................................... 5 5
S a m p lin g .......................................................................................................... 5 5
D term nation of Sex ...................................................................... 55
M materials ................................. ......... ..... 56
Cranial Measurement Definitions............................. .............. 58
Giles and Elliot Measurements ........... ...... ........................ 58
Statistical Procedures ................ ......... .................. 62

7 R E S U L T S ............................................................................................................ 6 5

G general R results .............. .... ..... ......................................................... ...... 65
Specific R results .............. .. ................. .................. ...... .............. ........ 67
G iles and Elliot D iscrim inant Function ................................................................. 67
Variable Selection........................................ ............ 69
Test for Site Effect on Selected Variables ........................................ 71
Discriminant Function Analysis ................................. ...................... .. ........... 72
F u n action 1 A ccu racy ......................................................... .................... 73
Function 2 A accuracy ....................................................................................... ........ 75

8 CONCLUSIONS AND DISCUSSION ..................................................................... 77

C o n c lu sio n s ........................................................................................................ 7 7
Summery of Statistical Results............. .... .......... ................... 78
Specific Statistical Results ................................. ........................... .. ....... 79
D iscu ssio n ..................................................... 8 0
Future Research ................................................................. ..... ..... ......... 82

APPENDIX

A CRANIAL MEASUREMENT DEFINITIONS ......................................................... 84

G iles and E lliot M easurem ents ...................... ................................................................. 84
O their M easurem ents .............. ............................................................... ... ........ 87



7









CHAPTER 2
SEX DETERMINATION

Introduction

This chapter describes the various visual, univariate, and multivariate metric methods used

in sex determination using cranial and post-cranial skeletal remains. Estimating sex from a

skeleton relies on sexual dimorphism, the morphological differences between men and women.

Men tend posses a larger body size when compared to females. Female morphology must allow

for both bipedal locomotion and giving birth to relatively large-headed babies when compared to

other primates. With a complete adult skeleton, and particularly a complete pelvis, a physical

anthropologist should be able to correctly assign sex with nearly perfect accuracy. Additionally,

the anthropologist should be able to recognize ambiguous cases where sex identification is less

certain. If the complete skeleton is not available, accuracy depends largely on what bony

elements are available and if the skeleton can be linked to a specific population. If the skeleton is

fragmented or from a sub-adult, then determining the sex is more difficult and less reliable than

with a complete adult skeleton. Methods of sex determination are either visual or metric, and

apply to the crania and post-cranial skeleton. For the best results, the forensic anthropologist

should use all available data.

Visual Methods

Visual methods for estimating sex from the skeleton make use of size differences between

men and women or morphological differences related to childbirth in women. Sexing methods

that rely on morphological differences in the pelvis related to childbirth are the most accurate.

The method for sexing the skeleton by the pubic bone developed by Phenice (1969) is the most

accurate method known for determining the sex of an individual from the skeleton (White 1991).









the sectioning-point, the individual is categorized as male. If it is less than the sectioning point,

the individual is categorized as female.

This analysis presents two discriminant functions. Function 1 uses the same eight variables

used by Giles and Elliot in their discriminant function analysis, but the function is calculated

using a sample of thirteen male and four female Native Americans from Florida and Southern

Georgia (Table 7-5). Function 2 uses the five variables selected based on t-tests and preservation

rates as described above, and it is calculated from 27 males and 12 females. The coefficients,

constants, and sectioning points for this analysis are presented in table 7-5.

Table 7-5 Coefficients, group means, and sectioning points for Function 1 and Function 2.
Cranial variable Function 1 Function 2
coefficients coefficients
AUB 0.0696568389
WFB 0.0138332675
PAC 0.1046833664
GOL -.0717749505 -0.0325698854
MDHA 0.1075559971 0.2188344979
XCB 0.1456008386
BBH -0.0620766571
ZYB 0.2196351470
BPL 0.4499153016
UFHT -0.2704566313
MAB -0.1675593848
Constant -49.19563 -21.50963
Male Mean 1.007282992 0.744143627
Female Mean -3.273669725 -1.674323160
Sectioning Point -1.133193367 -0.470560251
Constant=0 48.062436633 21.039069749
Sectioning Point


Function 1 Accuracy

Only 17 individuals included all of the measurements required for Function 1. Because of

the small sample size, the function is tested on the same individuals used to develop the function.

When applied to the sample of thirteen males and four females from which it is developed,









Function 1 correctly assigned sex to all seventeen individuals (Table 7-6). This result is not

significantly better than the reported Giles and Elliot result (p = 0.0509). It is also not

significantly different than the Giles and Elliot formula applied to the current sample when

observations with missing variables are omitted (p= 0.4848). The accuracy of this function is

inflated in this test due to the small sample size. A better indication of the accuracy of the

function is given by cross validation.

Table 7-6. Accuracy of Function 1, which uses the variables originally used by Giles and Elliot
(1963).
Number
Sex correct Count (n) Percent
Overall 14 17 82.35%
Cross validation Male 11 13 84.62%
Female 3 4 75.00%
Individuals with Overall 17 17 100%
missing variables Male 13 13 100%
omitted Female 4 4 100%
Missing variables Overall 35 45 77.78%
replaced by the Male 24 32 75%
average of the male
and female means Female 11 13 84.62%


Cross validation classifies each observation based on all of the other observations. A

discriminant function is calculated for the dataset minus one observation, and the omitted

observation is classified using the resulting function. This procedure provides a more realistic

estimate of the accuracy of the discriminant function. Using cross validation, a discriminant

function analysis using the Giles and Elliot variables correctly classified eleven of thirteen males

and three of four females, for a combined accuracy of 82.35%. This is not significantly better

than the Giles and Elliot result (p = 0.4130). It is also not significantly different than the Giles

and Elliot formula applied to the current sample when observations with missing variables are

omitted (p= 1.00).

































2007 Michael Bryan McGinnes









110). While this method is useful for analyzing group differences, it is computationally difficult

for the end user to assign individuals to groups.



Z?
Third Group







Y' * *
0 0










.Second Group


First Group X'

Function 1: First Group vs. Second Group


Figure 3-1 Discriminant Function Analysis with three groups using the graphic device proposed
by Rao. Function 1 is plotted on the X-axis, Function 2 is plotted on the Y-axis. Thin
rods X', Y', and Z' are placed to minimize errors (Adapted from Rao 1952; Giles and
Elliot 1962).


A different method is used by Giles and Elliot(1962) to estimate race. Giles and Elliot


(1962) use a pair of canonical discriminant function formulas for the placement of a skull into

white, black, or Native American categories: One formula for black vs. white, and the other for

white vs. Native American. The two functions are then plotted, with the black-white function on

one axis, and the white and Native American function on the other. To place individuals into one

of the three groups, Giles and Elliot use a geometrical device described by Rao (1952:327),









Canonical Discriminant Functions

Canonical discriminant functions determine a combination of variables that separate the

groups as well as possible. Fisher (1936) introduced a simple way to choose the coefficients for a

linear function that maximizes the F-ratio of a one-way analysis of variance for two groups,

which is the ratio of between group variance to within group variance. Specifically, his paper

deals with the discrimination of Iris setosa and Iris versicolor', two species found growing

together in the same colony. His variables are measurements of sepal length, sepal width, petal

length, and petal width. Most of the literature cites Fisher as the originator of discriminant

function analysis, but he cites "Mr. E.S. Martin" and "Miss Mildred Barnard" for applying the

principle to sex differences in the mandible and a secular trend in cranial measurements,

respectively.

The approach involves finding coefficients for a linear combination of n variables:

Z = a + bx, + bx,...b,x,

where Z is the discriminant function score, a is a constant, b1 through b, are discriminant

function coefficients, and xI through x, are independent variables, which maximizes the F-ratio

of Z in a one-way analysis of variance for the two groups. Finding the coefficients of the

canonical discriminant functions is an eigenvalue problem. Details on the computation of DFA

coefficients can be found in Manly (1994).

Discriminant function analysis has uses other than classifying individuals. Discriminant

function coefficients can also be used to evaluate how groups differ (Manly 1994:114). Howells

(1989) used discriminant functions as a form of data reduction, similar to the way others have

used principle components analysis, but that use is not common in the literature.

1 The article is primarily concerned with the discrimination of I setosa and I versicolor, but Fisher extends his
analysis to test the hypothesis that I virginica is a hybrid of I setosa and I versicolor.









on top of a shell midden. The burial mound was composed of a lower sandy layer and an upper

layer composed of "a thick laminated deposit which contained ... charcoal, pot sherds, shells,

etc." (Woodbury n.d.). Woodbury recovered 31 burials from the lower zone and 21 from the

upper zone; all are primary burials with their heads oriented toward the center of the mound in a

spoke pattern. All of the upper zone burials are extended while some from the lower zone are

flexed or semi-flexed (Willey 1954:81). All but one of the burials are adults with approximately

equal numbers of males and females. The Burns Mound is dated to the Malabar II period based

on ceramic types, although a pendant of European silver indicates the mound was used after

European contact.

The excavation at Fuller Mound A recovered in a sample of 96 complete skeletons. All but

twelve are adults, with slightly more females than males. Almost all of the burials are oriented in

the spoke pattern with their heads toward the center of the mound. Most are primary extended

burials lying on the back, although a few are semi-flexed. A few may have been secondary

burials. Iron and metal tools and glass beads of European manufacture are present (Stirling

1935:386). Based on the ceramics and the quantity of European goods, Rouse (1951:197)

suggested that the mound dates from the 17th century.

Fuller Mound B contained the remains of about 20 individuals disarticulated and mixed in

a single secondary burial at the center of the mound. Two primary burials are also found away

from the center with their feet pointing toward the center of the mound (Stirling 1935:387).

Rouse (1951:197) dates the mound to the Malabar I' period based on the ceramics and the lack of

European artifacts.

Fuller Mound D consisted of 16 primary extended burials in the spoke pattern oriented

with heads toward the center of the mound. Five of the individuals are infants, and there are









females and seven males (Wallace 1975:141-144). Indian Field was the location of a large

ceremonial pavilion in which were interred six burials that contained the disturbed, fragmented

,and mixed remains of 22 individuals. Remains of at least 13 of these are recovered from a single

interment (Zahler 1976: 8).

Garfield Site (9BR57)

The Garfield site is the remains of a village located on the confluence of the Etowah River

and Macedonia Slough near Kingston, Bartow County, in northwestern Georgia. Portions of the

site were first excavated by two amateur archaeologists from Decatur, Georgia, James Chapman

and Richard Criscoe during the 1960s. Eighteen burials are recovered from what appeared to be

abandoned storage pits. Those remains were transferred to the Florida Museum of Natural

History in 1974 with other collections they had excavated. Jerald T. Milanich tested the site in

1972 while he was a post-doctoral fellow at the Smithsonian Institution. He recovered an

additional four burials, including two adults, one infant, and one cremation. An additional

fragment of human bone was identified among animal bone excavated from midden deposits.

Artifacts recovered from the site indicate it belongs to the Kellog culture dating to 600 B.C. to

A.D. 100, a date supported by two radiocarbon assays (185570 B.P. and 235060 B.P.) from

charcoal obtained by Milanich (Jerald T. Milanich personal communication 2007; Milanich

1975).

Milanich also recovered a large quantity of floral and faunal remains, ceramics, and lithic

artifacts. As a group they suggest an occupation from early spring into late summer or early fall.

While it seems likely that some maize gardening was carried out toward the end of the site's

period of occupation, wild foods, especially nuts and other plant products, fish and a variety of

mammals, provided most of the diet (Milanich 1975).









Hunt, David R., and John Albanese
2005 History and Demographic Composition of the Robert J. Terry Anatomical
Collection. American Journal of Physical Anthropology 127:406-417.

Hutchinson, Dale L.
2004 Bioarchaeology of the Florida Gulf Coast: Adaptation, Conflict, and Change.
University Press of Florida, Gainesville.

Iscan, Mehmet Y., and Patricia Miller-Shaivitz
1984 Determination of Sex from the Tibia. American Journal of Physical Anthropology
64:53-57.

Johnson, Francis E., and Charles E. Snow
1961 The reassessment of the age and sex of the Indian Knoll skeletal population:
Demographic and methodological aspects. American Journal of Physical
Anthropology 19: 237-244.

Kajanoja, Pauli
1966 Sex determination of Finnish Crania by Discriminant Function Analysis.
American Journal of Physical Anthropology 24:29-34.

Kelly, Jennifer A., Robert H. Tykot and Jerald T. Milanich
2006 Evidence for Early Use of Maize in Peninsular Florida. In Histories ofMaize:
Multidisciplinary Approaches to the Prehistory, Linguistics, Biogeography,
Domestication, and Evolution ofMaize, edited by John Staller, Robert Tykot and
Bruce Benz, pp. 249-261. Elsevier Academic Press, Boston.

Kendall, Maurice G.
1957 A Course in Multivariate Analysis. Hafner Publishing Co., New York.

Key, Patrick J.
1983 Craniometric Relationships Among Plains Indians. Report of Investigations 34.
University of Tennessee, Department of Anthropology, Knoxville.

Konigsberg, Lyle W., and Samantha M. Hens
1998 Use of Ordinal Categorical Variables in Skeletal Assessment of Sex from the
Cranium. American Journal of Physical Anthropology 107:97-112.

Kozuch, Laura
1998 Faunal Remains from the Palmer Site (8So2), with a focus on shark remains. The
Florida Anthropologist 52:208-222.

Krogman, Wilton M.
1962 The Human \ke/leti, in Forensic Medicine. Charles C. Thomas, Springfield,
Illinois.










Table 6-1. Continued
Giles and Modern Symbols Description References
Elliot equivalent
measurement
Palate-external Maxillo- ecm-ecm, The maximum Bass 1971:70;
breadth alveolar MAB breadth across the Howells 1973:176;
breadth, alveolar borders of Martin 1956:480;
external palate the maxilla Montagu 1960:51.
breadth measured at its
widest point,
between each
ectomolare (ecm).
Mastoid length Mastoid length MDH The projection of Howells 1966:6;
the mastoid process 1973:176.
below, and
perpendicular to, the
eye ear (Frankfort
Horizontal) plane in
the vertical plane.
Prosthion- Upper facial n-pr The direct distance Howells 1966:6;
Nasion Height height from nasion (n) to Hrdlicka 1952:143;
prosthion (pr). Martin 1956:476.


Table 6-2. Measurements not used by Giles and Elliot, common symbols, measurement
descriptions, and sources of the measurements. Complete measurement descriptions
and techniques are listed in appendix A along with a sample data collection sheet.
Measurement Symbol Description References
Biorbital ec-ec, EKB The direct distance from one Howells 1973:178
breadth ectoconchion (ec) to the
other.
Interorbital d-d, DKB The direct distance between Martin 1956:477.
breadth right and left dacryon (d).
Maxillo- pr-alv, The direct distance from Bass 1971:70; Hrdlicka
alveolar length, MAL prosthion (pr) to alveolon 1952:146 147; Martin
external palate (alv). 1956:480.
length
Biauricular au-au, ALB The least exterior breadth Howells 1973:173
breadth across the roots of the
zygomatic processes.
Foramen ba-o, FOL The direct distance of basion Martin 1956:455
magnum length (ba) from opisthion (o).









LIST OF FIGURES


Figure pae

3-1 Discriminant Function Analysis with three groups using the graphic device proposed
by Rao ......................................... ........ .......... 29

5-1 Locator map of sites used in this study and major rivers. .............................................. 54









the Atlantic and Gulf areas sand burial mounds appear during the Deptford period. By AD 1

some mounds on the Gulf coast contained items thought to be linked to Hopewellian-related

beliefs and trade" (Milanich 2004a: 194). As the Southeast entered the Middle woodland period,

Deptford was succeeded by the Swift Creek culture in most areas.

St. Johns/Malabar

The Saint Johns culture persisted along the Atlantic coast of Florida for around 2,000

years, from the end of the Archaic in 500 BC into the seventeenth century and contact with the

Spanish. The Saint Johns region includes two sub-regions, the St. Mary's zone in the North, and

the Indian River Zone in the south. The Indian River Zone is located around Brevard, Indian

River, and St. Lucie counties, and with sites found near wetlands of the Saint Johns River Basin,

the Indian River, and along barrier islands (Milanich 1994: 249). The culture of the Indian River

Zone during the Saint Johns period is identified as the Malabar culture, and is divided into two

periods (Rouse 1951). The Malabar I period is approximately contemporaneous with the Saint

Johns I period, present from 500BC to 750 AD, while Malabar II is of the same age as the Saint

Johns II period, present from 750AD to 1565 AD (Milanich 1994: 247,249-250). Malabar sites

can be classified as villages, special use sites, or single use sites. Villages are large,

multicomponent sites that exhibit a wide range of artifacts and large middens. Villages are

always located near wetlands, and are surrounded by special use sites. Special use sites are

smaller multicomponent sites used intermittently for short periods of time. Single use sites were

probably used to gather some specific resource, and all that remains are small artifact scatters or

a few animal remains (Milanich 1994:251-252). Malabar peoples were foragers, and subsistence

patterns were remarkably consistent through both phases, with diets composed of roughly 15%

terrestrial resources, such as deer, raccoons, and rabbits, and 80% fish and shellfish. As time

passed and water levels changed, the Malabar peoples tended to collect larger fish and a wider









Harris, Richard J.
1985 A Primer on Multivariate Statistics. 2nd ed. Academic Press, Orlando.

Hemmings, E. Thomas
1975 An Archaeological Survey of the South Prong of the Alafia River, Florida.
Florida Anthropologist 28:41-51.

Henke, Winfried
1977 On the Method of Discriminant Function Analysis for Sex Determination of the
Skull. Journal of Human Evolution 6:95-100.

Hoffman, Paul E.
1994 Narvaez and Cabeza de Vaca in Florida. In The Forgotten Centuries: Indians and
Europeans in the American .Si/th, 1521-1704, edited by Charles Hudson and
Carmen Chaves Tesser, pp. 50-73. University of Georgia Press, Athens.

Holman, Darryl J., and Kenneth A. Bennett
1991 Determination of sex from arm bone measurements. American Journal of
Physical Anthropology 84:421-426.

Hooton, Earnest A.
1930 The Indians ofPecos Pueblo: a study of their skeletal remains. Department of
Archaeology, Phillips Academy, Andover, Massachusetts. Yale University Press,
New Haven.

Howells, William W.
1966 Craniometry and Multivariate Analysis: The Jomon Population of Japan: A Study
by Discriminant Analysis of Japanese and Ainu Crania. Papers of the Peabody
Museum of Archaeology and Ethnology 57:1-43. Harvard University Press,
Cambridge.

1973 Cranial Variation in Man. A Study by Multivariate Analysis of Patterns of
Difference Among Recent Human Populations. Papers of the Peabody Museum of
Archaeology and Ethnology 67. Harvard University Press, Cambridge.

1989 Skull shapes and the map: Craniometric analysis in the dispersion of modern
homo. Papers of the Peabody Museum ofArchaeology and Ethnology 79.
Harvard University Press, Cambridge.

Hrdlicka, Ales
1940 Catalogue of human crania in the United States National Museum collections:
Indians of the Gulf States. Proceedings of the U.S. National Museum 87, no
3076:315-364.

1952 Practical Anthropometry. 4th ed., edited by T.D. Stewart. Wistar Institute of
Anatomy and Biology, Philadelphia.










ACKNOWLEDGMENTS

This thesis would not have been possible without the support and encouragement of my

supervisory committee, Anthony Falsetti and Thomas Hollinger. I also thank Jerry Milanich,

who was a de facto committee member, but was unable to attend the defense. Special thanks go

to Elaine Oran for her invaluable insight on the structure and writing of a scientific paper, even if

it isn't about rocket science. Finally, I am forever grateful to David Hunt, Collections Manager of

the Smithsonian Institution's Physical Anthropology collections, for his immeasurable

contribution to this research. Collecting the skeletal data would have been impossible without his

permission and assistance. He not only tolerated my haphazard schedule and unannounced visits

with infectious good humor, he was there when I needed a sounding board, provided stimulating

conversation, and answered my many questions. He was always ready to share a grouse or a

laugh, whichever was needed. Thanks in part to a shared interest in food, wine, and the finest

Australian Tupperware; I am honored to call him my friend.

Special thanks are due to the staffs of the Anthropology Departments at the Florida

Museum of Natural History (FLMNH) and the National Museum of Natural History,

Smithsonian Institution, where the collections examined for this research are housed. Ann

Cordell, Elise LeCompte, Diane Kloetzer, and Donna Ruhl provided friendship, assistance, and

knowledge of the collections at the FLMNH. Maggie Dittemore and the rest of the staff at the

John Wesley Powell Library of Anthropology at the Smithsonian Institution provided access to

printed resources that are unobtainable elsewhere.

My immediate family -- Dorothy, Rodney, and Judie McGinnes -- was integral to

completing this project. They contributed in innumerable ways I can't thank them loudly or

strongly enough. I am fortunate to be part of a very large and diverse extended family who









For race, Native Americans are misclassified at a significantly higher rate that non-Native

Americans (83% correct for black and white combined vs. 14% correct for Native Americans).

Snow et al. conclude that the Giles and Elliot functions "provide a useful tool for the

determination of sex and race of unidentified crania submitted for forensic science examination."

The functions, however, did not perform well among Native American subjects. "It thus appears

that the 5000 year old Indian Knoll crania used by Giles and Elliot in developing their functions

do not adequately represent the entire U.S. category of Indian" (Snow et al 1979:459).

The advantages of discriminant function analysis in physical anthropology are that it is

relatively easy to apply, allowing sex and race estimations by those with little training in

osteology, and that it is an objective indicator or race and sex, especially for isolated remains.

The weaknesses include needing to develop functions on the populations from which the subject

comes, and the need to be able to make all of the measurements called for in the discriminant

function, which is not always possible in fragmentary remains typically found in archaeological

sites. Functions developed with fragmentary remains in mind help avoid the problem of missing

measurements.









second maxillary molars. (Bass 1971:70; Howells 1973:176; Martin 1956:480; Montagu

1960:51).

Opisthion-forehead length: The maximum distance from opisthion (the midpoint on the

posterior border of the foramen magnum) to the forehead in the midline. This measurement is not

used in this study because it is not used in the Giles and Elliot formulas, nor has it survived in the

current literature.

Mastoid length (MDH): The length of the mastoid measured perpendicular to the plane

determined by the lower borders of the orbits and the upper borders of the auditory meatuses

(Frankfort horizontal plane). Equivalent to Mastoid Length; The projection of the mastoid

process below, and perpendicular to, the Frankfort horizontal plane in the vertical plane. Both

right and left sides are measured using a sliding caliper. The skull is rested on its right side, and

the calibrated bar of the caliper is applied just behind the mastoid process, with the fixed flat arm

tangent to the upper border of the auditory meatus and pointing (by visual sighting) to the lower

border of the orbit. The calibrated bar is perpendicular to the eye ear plane of the skull (i.e.,

approximately level in the position given). The measuring arm is adjusted until it is level with

the tip of the mastoid process, using the base of the skull generally, and the opposite mastoid

process to control the plane of sighting. (Howells 1966:6, 1973:176; Keen 1950).

Other Measurements

The remaining measurements are not included in the Giles and Elliot (1962) study, but are

used in more recent research, including Bass (1971, 1995), Howells (1973), Buikstra and

Ubelaker (1994), and FORDISC 2.0 materials (Ousley and Jantz 1996). The descriptions and

techniques primarily follow Ousley and Jantz (1996).

Biorbital Breadth (ec-ec, EKB): The direct distance from one ectoconchion (ec) to the

other. This measurement is taken using the sliding caliper (Howells 1973:178).









Discriminant function analysis was developed to solve the problem of predicting group

membership based one or more interval variables. The goal of discriminant function analysis is

to minimize misclassification by maximizing between group differences, or minimizing group

overlap. Discriminant function analysis is also used to determine which variables best

discriminate between groups, and in what ways groups differ. How well a function performs is

usually reported in terms of how many cases would be correctly assigned to their groups using

the discriminant functions (Manly 1994).

One widely used discriminant function for sex determination is calculated by Giles and

Elliot (1963) based on data collected from the Terry and Hamann-Todd skeletal collections. The

Terry and Hamann-Todd collections are comprised of skeletal remains collected from cadavers

used by medical school anatomy classes during the late nineteenth and mid-twentieth centuries

(Hunt and Albanese 2005). The problems presented by a medical school cadaver sample include

possible effects of the inherent socio-economic bias on skeletal morphology. The Terry and

Hamann-Todd skeletal collections are biased, with older individuals and males overrepresented

compared the population as a whole, and they do not include Native Americans. For studying the

sex differences in skeletons, however, such collections are essential because the sex of each

individual is positively known from written records (Giles and Elliot 1963:56). The sample Giles

and Elliot used to calculate their discriminant functions did not include any native Americans,

but the formulas are tested on three series of American Indian crania from Indian Knoll (N>=

344), Pecos Pueblo (N>=110), and Florida (N>=217). These materials are analyzed by Snow

(1948; Johnson and Snow 1961), Hooton (1930), and Hrdlicka (1940) respectively. "On the

whole the discriminant functions described [by Giles and Elliot 1963], assign the correct sex,

assuming that the original estimations are correct, with the same order of magnitude as they do









PROC FREQ procedure with the EXACT BINOMIAL option. The null hypothesis is that the

accuracy of the formulas is equal against the alternative that the new accuracy is higher.

Fisher's exact test is used to compare the accuracy between sets of new results. Fisher's

exact test is a non-parametric statistical hypothesis test used for categorical data where samples

are too small for the Chi-squared test. Fisher's exact test is calculated using the PROC FREQ

procedure with the CHISQ option. The null hypothesis is that the accuracies are equal, which is

tested against the alternative that the accuracies are different.









In order to apply Function 1 to all observations, missing variables are replaced with the

average of the male and female means for each variable. With missing variables replaced,

Function 1 correctly classifies 24 of 32 males and 11 of 13 females, for an overall accuracy of

77.78% (Table 7-6). This is not significantly different from the Giles and Elliot function using

the recalculated sectioning point and missing variables replaced with the average of the male and

female means (p= 0.5912).

Function 2 Accuracy

Function 2 is based on measurements from 39 individuals, including 27 males and 12

females. When applied to the sample from which it is developed, Function 2 correctly assigns

sex to 26 of 27 males and 10 of 12 females, for an overall accuracy of 92.31% (Table 7-7). This

is not significantly different from Giles and Elliot's reported accuracy of 86.4% (p= 0.4087).

Using cross validation, the formula correctly classifies 10 of 12 females and 23 of 27 males for

an overall accuracy of 84.62%. This is not significantly different from the accuracy of the Giles

and Elliot function (p= 0.8819).

Table 7-7. Accuracy of Function 2, which uses variables selected using t-tests and preservation
rates.
Number
Sex correct Count (n) Accuracy
Overall 33 39 84.26%
Cross validation Male 23 27 85.19%
Female 10 12 83.33%
Overall 36 39 92.31%
Individuals with missing Male 26 27 96.30%
variables omittedFemale 26 27 96.30%
variables omitted
Female 10 12 83.33%
Missing variables replaced Overall 39 45 86.87%
by the average of the male Male 28 32 87.50%
and female means Female 11 13 84.62%


To apply Function 2 to all individuals in the sample, missing variables are replaced by the

average of the male and female means for each variable. With the missing variables replaced,









establishment of the Spanish Oaks historical site. The Shell Ridge area was excavated in 1991

under the direction of Corbett Torrence, George Luer, and Marion Almy, and detailed

zooarchaeological analyses were conducted (Hutchinson 2004; see Almy and Luer 1993; Kozuch

1998; Quitmyer 1998).

Bullen and Bullen (1976:35) describe Palmer Mound as "a very unassuming, dome-

shaped, sand mound rising 4 feet above the surrounding land." Despite its modest appearance, it

is actually one of the largest systematically excavated burial mounds in the southeast, with over

400 individuals recovered. The mound was used primarily between A.D. 500 and 800, during the

Manasota period (Bullen and Bullen 1976; Hutchinson 2004:43-59; Williams 2004). Faunal

analysis indicates that fish and shellfish dominated the diet, with little consumption of terrestrial

animals, but some use of terrestrial plants.

Perico Island (8Ma6)

The Perico Island site is located on the western edge of Perico Island in Manatee County,

Florida, west of Bradenton, and between Sarasota and Tampa Bays. The site is composed of

large and small shell middens, a burial mound, and a cemetery area. The site was excavated by

Dr. M.T. Newman in 1933-34 as a relief project. Newman recovered 185 flexed burials from the

burial mound, and 43 primary flexed burials from the cemetery (Willey 1949:176,180). Willey

(1949) initially categorized the site as a local variant of the Glades culture (Willy

1949; 1998:192), but it is now considered part of the Manasota culture (see Milanich 1994; Luer

and Almy 1982).

St. Simons Island, Georgia

Remains are used from three sites located on the northern end of St. Simon's Island,

Martinez B-C, Taylor Mound, and Couper Field/Indian Field. All were excavated as part of the

St Simons Island Archaeological Project operated by the University of Florida between 1972 and









between the means of the sexes (Giles and Elliot 1963), finding the sectioning point graphically

from the bimodal distribution of the scores (Giles and Elliot 1963), or placing the sectioning

point at the grand mean of the function scores (Henke 1977). Henke (1977) tested each of these

methods, and also used the unmodified functions. Henke concluded that only the first two

methods are practical. Calcagno (1981) found that moving the sectioning point by any method

was not a practical solution to the problem. Placing the sectioning point mid-way between the

means of the sexes requires that one first know the sex of the skeletons. The graphic method is

not practice because the distribution of discriminant scores is multimodal, and not bimodal.

Finally, placing the sectioning point at the grand mean of the discriminant scores assumes that

the sexes are equally represented in the sample, which is particularly unlikely to be true in

archaeological series (Henke 1977). Henke's conclusion seems to be that discriminant functions

developed for one group can be used on another, but the sectioning point has to be adjusted.

Birkby (1966) tested the Discriminant Functions developed by Giles and Elliot for race and sex

(1962; 1963). His goal was to determine (1) if discriminant function analysis is applicable in the

assessment of race and sex in human identification and (2) the reliability of such techniques. He

found Indian crania are often misclassified for race and sex. Birkby concludes that the Indian

Knoll sample used by Giles and Elliot (1962, 1963) is not representative of Native Americans as

a whole. Therefore, the functions based on those data are not applicable in the identification of

race and sex in human identification, either forensic or archaeological.

Snow et al. (1979) performed another test of the Giles and Elliot discriminant functions

using forensic cases. The discriminant function for sex determination attained an accuracy that

was not significantly different from Giles and Elliot. With no significant difference in accuracy

between the sexes.









Zahler, James
1976


W., Jr.
A Morphological Analysis of a Protohistoric-Historic Skeletal Population from St.
Simons Island, Georgia. Unpublished Master's thesis, Department of
Anthropology, University of Florida, Gainesville.









For the purpose of sex determination by DFA, Florida and Georgia Native Americans can
be viewed as a single population, regardless of time period.

There is evidence that better discriminant functions can be developed using local

populations, but the results are not statistical significant possibly because of the small sample

size. Sample size is small because of the requirement that individuals include both a relatively

intact cranium and pubic bone. Both of these skeletal elements are relatively delicate and are

seldom preserved intact. Individuals who have both elements intact are rare.

There is evidence that sex determination by DFA can be accomplished with a smaller

number of more robust variables than those used by Giles and Elliot without reducing accuracy.

Five variables are identified that are preserved in at least 90% of males and females selected for

this study. It is possible to apply the discriminant function using these variables to 86.66% of the

sample, compared to just 37.77% for the Giles and Elliot variables. The new function also has

higher accuracy than the function developed using the Giles and Elliot variables, although the

difference is not significant.

Implicit in this research is the assumption that all of the sites used in this study can be

considered a single population. The results of the Multiple Analysis of Variance (MANOVA)

suggest that there is no significant difference between individuals from the different sites used in

this study. It is, therefore, reasonable to treat all of the individuals from these sites as belonging

to a single population.

Summery of Statistical Results

The results presented in the previous chapter can be summarized as follows. When applied

to Florida and Georgia Native Americans:

The Giles and Elliot function 3 and original sectioning point classifies males correctly,
but misclassifies females at a significantly higher rate than other methods.










TABLE OF CONTENTS

page

A C K N O W L E D G M E N T S ......... .................. ........................................................................ 4

LIST OF TA BLE S...................................................... 9

L IST O F F IG U R E S ......... .................. ........................... ........................................ 10

CHAPTER

1 IN T R O D U C T IO N ................................................................. .................................... 15

2 SE X D E TE R M IN A TIO N .................................................................... ........................ 18

In tro d u ctio n ...................................................................... 18
Visual Methods........................................................ 18
M etric M eth o d s ................................................................... ....................................... 2 1

3 DISCRIMINANT FUNCTION ANALYSIS................................ 26

Canonical Discrim inant Functions ............................................................... .. ............ 27
Multiple Discriminant Functions......................................................... 28
Tests of Significance ............................................................................. 30
Prior Probabilities ................................................................. ................. 30
Stepw ise D iscrim inant Function A analysis ................................................................... 30
M ahalanobis D instance .............................................................................................. ........ 31
L ogistic R egression............................. .............. ..... 31
C o n c lu sio n ..................................................................................................... .............. ..... 3 2

4 ARCHAEOLOGICAL CONTEXT......................... ......... ................... 35

Intro du action ..................................................... 3 5
Environm mental Setting .................................................................................................. 35
C u ltu ra l H isto ry ................................................................... ...................................... 3 6
P aleoindian P period ............................................................................................ ........ 36
T he A rchaic P period ........................................................................................ ........ 38
E arly A rchaic............................................ 38
M middle Archaic .................................................................. .... ........ 39
L ate A rc h a ic ................................................................... ............................. 4 0
W oodland and Regional Cultures .......................................................................... ...... 41
K ellog .................................................................................................... .............. 42
D e p tfo rd ................................................................. ..................................... 4 2
St. Johns/M alabar ............................................................... ................ .............. 43
M a n a so ta ................................................................................................ 4 4
W ilm ington Culture .................................. ..................................... 45
H historic P eriod............................. .............. ..... 45


6









distribution of fluted points in the archaeological record. Most of North America appears to have

been occupied by 10,000 B.C. (Bense 1994: 38-39; Milanich 1994: 37-40).

The general consensus is that the Paleoindian life ways were based on the hunting of large

game animals and the gathering of a variety of plant foods. Paleoindians were highly mobile, and

were probably organized into bands with widely ranging patters of movement and inter-group

interaction (Doran 2002:49,50; Bense 1994;Griffin 1979:51). Because of their mobility, the

traditional expectation has been that they should exhibit a basic biological similarity over a wide

geographical area (Key 1983:8; Meiklejohn 1972). Preliminary results from Ross, Ubelaker and

Falsetti (2002), however, indicate that Native Americans are much more biologically

heterogeneous than previously thought.

During the Paleoindian period, the climate was cooler and dryer than today, but with

reduced seasonal variation. Sea levels were as much as 100 meters below current levels (Clausen

et al 1979; Widmer 1988). In Florida, this resulted in more land being exposed, lower water

tables and few sources of surface water. This scarcity of surface water is thought to be a

determining factor in Paleoindians settlement patters in Florida. It is also thought that many sites

that were along the coast during the Paleoindian period have been inundated by rising sea levels

(Bullen 1958; Ruppe 1980). Most Paleoindian sites found to date consist of little more than

limited scatters of lithic debitage (Edwards 1954; Waller 1969).

The human skeletal remains of fewer than 100 individuals from the Paleoindian period

have been recovered from all of North America. In Florida, Paleoindian sites with human

remains include Little Salt Spring, Warm Mineral Springs, and Cutler Ridge, although the

material at Little Salt Spring cannot be incontrovertibly placed within the Paleoindian occupation

(Doran 2002).









Table 6-2. Continued
Measurement Symbol Description References
Minimum ft-ft, WFB The direct distance between Bass 1971:67; Hrdlicka
frontal breadth the two frontotemporale (ft). 1952:142; Martin
1956:457; Olivier
1969:151.
Foramen FOB The distance between the Martin 1956:459
magnum lateral margins of the
breadth Foramen magnum at the point
of greatest lateral curvature.
Upper facial fmt-fmt The direct distance between Martin 1956:475.
breadth each frontomalare temporale
(fmt). This measurement
differs from Howells' FMB
in that the lateral most points
on the suture are used rather
than the most anterior points.
Nasal height n-ns, NLH The direct distance from Bass 1971:68; Howells
nasion (n) to nasospinale (ns). 1966:6; Martin
1956:479; Olivier
1969:153
Orbital breadth d-ec, OBB The laterally sloping distance Martin 1956:477 478;
from dacryon (d) to Howells 1973:175
ectoconchion (ec).
Orbital height OBH The direct distance between Bass 1971:69; Martin
the superior and inferior 1956:478; Montagu
orbital margins. 1960:51; Olivier
1969:152.
Biorbital ec-ec, EKB The direct distance from one Howells 1973:178
breadth ectoconchion (ec) to the
other.
Interorbital d-d, DKB The direct distance between Martin 1956:477.
breadth right and left dacryon (d).
Frontal chord n-b, FRC The direct distance from Howells 1973:181;
nasion (n) to bregma (b) Martin 1956:465.
taken in the midsagittal plane.
Parietal chord b-l, PAC The direct distance from Howells 1973:182;
bregma (b) to lambda (1) Martin 1956:466.
taken in the midsagittal plane.
Occipital chord l-o, OCC The direct distance from Howells 1973:182;
lambda (1) to opisthion (o) Martin 1956:466.
taken in the midsagittal plane.









upper edges of the eye orbits are blunt, and frontal sinuses are larger in males. On the mandible,

the male chin is squarer, and the gonial angle is more acute.

Compared to the male, the female cranium is smaller, smoother and more gracile. In the

female mandible, the chin is more rounded and pointed, and the gonial angle is more oblique.

The smaller size of the female cranium is evident in a smaller palate, and smaller teeth. Females

also display frontal and parietal bossing into adulthood; the upper edges of the eye orbits are

sharp.

Evaluation of these characteristics depends not only on the experience of the osteologist,

but also on matching the specimen to a genetically and temporally close comparative population

(Bass 1995; White 1991). Using the crania, an experienced osteologist should be able to make a

sex determination that is 80-90% accurate. Buikstra and Ubelaker (1994:16-20) provide a

scoring system for several of these visual traits.

For the remainder of the skeleton, males tend to be larger than females, with long bones

that are longer, heavier, and have larger attachment areas for muscles, including the linea aspera,

crests, tuberosities and impressions (Brothwell 1981, Stewart 1948). These criteria are useful if a

related skeletal series is available, but for isolated or fragmentary remains this is only useful if

the bone is at the extreme end of the range, either 'very male' or 'very female.'

Metric Methods

While visual methods can estimate sex quickly and accurately, their evaluation is

subjective and requires experience with sexing techniques and the relevant population. Metric

procedures are based on quantifying the same criteria used in visual sexing. A metric procedure

could be better if the observer is not familiar with visual techniques or the relevant population.

Additionally, metric procedures serve as an objective check to visual methods and can strengthen

the position of the osteologist as expert witness in a courtroom (Stewart 1979). The simplest









The accuracy of the Giles and Elliot function 3 with the sectioning point recalculated is
not significantly different from the Giles and Elliot function applied to black and white
individuals.

The Giles and Elliot function includes some variables that do not contribute to sex
discrimination, and others that preserve poorly.

The accuracy of a new discriminant function developed using the Giles and Elliot
variables (GOL, XCB, BBH, ZYB, BPL, UFHT, MAB, and MDHA) is no more accurate
than the original Giles and Elliot function.

Variables which do contribute to sex discrimination and preserve well include: GOL,
AUB, WFB, PAC, and MDHA.

GOL, AUB, WFB, PAC, and MDHA are not significantly different across the samples
sites in Florida and Georgia individually, or taken together.

The accuracy of a new discriminant function developed using GOL, AUB, WFB, PAC,
and MDHA is more accurate than the Giles and Elliot function, but the difference is not
significant.

Specific Statistical Results

Other researchers have found that the Giles and Elliot formula cannot be applied to other

populations without modification (Giles and Elliot 1963, Kajanoja 1966). While Giles and Elliot

suggested several methods for recalculating a sectioning point, Henke (1977) and Calcagno

(1981) have determined that those methods are not practical.

Using the Giles and Elliot function with the original sectioning point did not produce

acceptable results. The function is able to classify 13 of 13 males correctly, but misclassified 2 of

4 females. When missing values are replaced with neutral values, the Giles and Elliot function

classifies all 32 males correctly, but misclassifies 11 of 13 females. The Giles and Elliot function

is not a reliable indicator of sex for Florida and Georgia Native Americans using the original

sectioning point. The Giles and Elliot function does produce acceptable results when the

sectioning point is recalculated.









The Timucua and the Guale paid a terrible price for there service to the Spanish crown.

The Spanish used the Indians as a labor force to support the Spanish colony. They did supply the

Indians with technology to increase maize production, but the demand for maize and labor

shifted the native populations from a semi-nomadic hunting, gathering, and farming subsistence

base to sedentary intensive maize agriculture. The maize diet led to nutritional deficiencies,

especially in lysine, tryptophan and iron. These deficiencies can be seen in the remains of

mission Indians in the form of poratic hyperostosis, cribra orbitalia, and enamel hypoplasia. The

increased physical stress imposed by the Spanish need for labor can be seen in the form of

osteoarthritis. They also experienced increased levels of carious lesions and periosteal reactions

due to the increase stresses of mission life. Indian populations under the mission system dropped

sharply as a result of working conditions and a series of Old World disease epidemics in 1595,

1612-1617, 1649-1650, and 1655-1656. Population levels were further impacted by slaving raids

by the English and there allies from 1660 to 1684 (Milanich 1998). As the Timucua populations

declined, the Spanish consolidated the remaining tribes along the Camino Real, giving them

access to the Apalachicola and their labor. By the time the Spanish ceded Florida to the English

in 1763, the Timucua had dwindled from a pre-contact population of approximately 20,000 in the

early sixteenth century to a single adult. Other Indian populations in Florida and Southern

Georgia were similarly decimated, and the few remaining survivors were evacuated to Cuba

when the Spanish left Florida (Milanich 1994, 2004b; Williams 2004).









Krogman, Wilton M., and Mehmet Yasar Iscan
1986 The Human .\le/etl,n in Forensic Medicine. 2nd ed. Charles C. Thomas,
Springfield, Illinois.

Loucks, L. Jill
1976 Early Alachua Tradition Burial Ceremonialism: The Henderson Mound, Alachua
County, Florida. Unpublished Master's thesis, Department of Anthropology,
University of Florida. Gainesville.

Luer, George M., and Marion M. Almy
1979 Three Aboriginal Shell Middens on Longboat Key, Florida: Manasota Period
Sites of Barrier Island Exploitation. Florida Anthropologist 32:33-45.

1982 A Definition of the Manasota Culture. Florida Anthropologist 35:34-58.

Luer, George M., Marion M. Almy, Dana Ste. Claire, and Robert Austin
1987 The Myakkahatchee Site (8So397), A Large Multi-Period Inland from the Shore
Site in Sarasota County, Florida. Florida Anthropologist 40:137-153.

Manly, Bryan F.J.
1994 Multivariate Statistical Methods: A Primer. Chapman & Hall, New York.

Martin, Rudolf
1956 Lehrbuch der Anthropologie in Systematischer Darstellung. 3rd ed. Gustav
Fischer, Stuttgart.

Martinez, Carlos A.
1975 Culture Sequence on the Central Georgia Coast, 1000 B.C.-1650 A.D.
Unpublished Master's thesis, Department of Anthropology, University of Florida,
Gainesville.

Meiklejohn, Christopher
1972 Biological concomitants of a model of band society. In International Conference
on the Prehistory and Paleoecology of Western North America Arctic and
Subarctic, edited by S. Raymond and P. Schledermann, pp. 133-141.
Archaeological Association, Department of Archaeology, University of Calgary,
Calgary.

Milanich, Jerald T.
1975 Fabric Impressed Pottery Site, Bartow County, Georgia. Early Georgia
Newsletter, Society for Georgia Archaeology 11:4-5.

1976 Georgia Origins for the Alachua Tradition. Bulletin No. 5. Florida Bureau of
Historic Sites and Properties, Tallahassee.

1977 A Chronology for the Aboriginal Cultures of St. Simons Island, Georgia. Florida
Anthropologist 30:134-144.









APPENDIX B
TABLE OF SITES


Site name Sex Catalog
overall #
Golf Course site (8Br44), Brevard county, Florida. "Melbourne Man" male 331422
Bay Pines (8Pi64) Pinellas county, Florida. (Gallagher and Warren
1975) 24 burials male s0272
Canaveral (8Br85), Brevard county, Florida. female 377440
The Canaveral site includes the Burns and Fuller mounds. Rouse 377475
(1951) places the mounds in the Malabar I, Malabar I', and Malabar II 377503
cultures based on ceramic types. Malabar is an Indian River variation male 377427
of St. Johns (500 B.C.-AD1763). 377428
377441
377465
377466
377467
377471
377478
377481
377484
377489
377502
377507
377513
377602
377605
Cannon's Point, St. Simons Island, Glynn County, Georgia.
Late Wilmington culture ceramic assemblage, radiocarbon dated to
990+75 (A.D. 960) (Martinez 1975; Milanich 1977) male s0418
Casey Key (8So17) Sarasota County, Florida. A Manasota Weeden
Island culture mound dating from ca. A.D. 250-750. (Bullen and Bullen
1976). female 92733
Couper Field/Indian Field, Glynn County, Georgia. St. Simons Island. female s0300
Late Pre-Columbian/early Spanish colonial/mission period village. s0301
s0302
s0311
male s0299
s0303
s0304
s0305
s0309
Garfield Site (9Br57) Bartow County, Georgia. Kellog culture (Early
Woodland, ca. 600BC-AD 100) village on Etowah River. 18 Burials.
(Milanich 1975; Wood and Bowan 1995:8). male s0325


female


97533









CHAPTER 1
INTRODUCTION

Sex identification has long been an important part of skeletal analysis in the archaeological

setting. Accurate sex estimations are basic to studies of past adaptations of humans to new

environments and demographic histories. Archaeologists may use sex information to establish

demographic patterns, study the affect of sex on status within a particular archaeological culture,

or to examine migration patterns (Buikstra and Ubelaker 1994:15). Additionally, forensic

anthropologists may use this information to help identify an individual in a medico-legal context.

Methods for sex identification have evolved from visual methods to include metric

methods based on univariate and multivariate statistical analysis, particularly discriminant

function analysis. Using visual methods to determine sex, an osteologist may examine the overall

size of the subject, the size of the mastoid process, the shape of the frontal bone, or the gonial

angle of the mandible. The determination of sex using visual method relies primarily on the

experience and judgment of the osteologist. Using these methods, osteologists typically achieve

an accuracy of 80% to 90% (Giles and Elliot 1962). Metric methods use various measurements

of the skeleton, many of which attempt to capture aspects used in visual methods (David Hunt,

personal communication 2006). In univariate analysis, a single measurement of an individual is

compared to the distribution of measurements from a sample of known sex specimens to arrive at

the likely sex of the individual. Using only one measurement, however, does not account for

differences in shape of a skeletal element between males and females, and for any single

measurement there is considerable overlap between the range of variation for males and females.

Multivariate methods use multiple measurements which can capture the shape of a skeletal

element and minimize the amount of overlap between males and females. Discriminant function

analysis is a multivariate method designed for this problem.









offered all sorts of support through the years, including kind words, empathy, sympathy, a roof, a

meal, companionship, and even nagging when needed. Barbara and Max Skidmore have been

positive role models since childhood. More recently, Elizabeth and Jay Boris, and Dan and

Elaine Oran have provided unflagging support I thank them for their extreme generosity, good

cheer, and for allowing me to deep fry the Thanksgiving turkey. Last, but not least, thank you to

Dr. Ruth Trocolli, my wife, my best friend, an exemplary scholar and outstanding human being. I

could not have done it without her love, support, and encouragement. We shared the long road

that culminated in this thesis.









good results, but found the method "cumbersome." They preferred a probit model that used

categorical independent variables.

Conclusion

On the whole, the accuracy of DFA in identifying sex is no more accurate than visual

methods when used by a trained osteologist. The advantage is that someone who is not an

osteologist, with brief instructions in measuring, can perform discriminant function sexing

quickly and objectively. This group includes medical examiners and archeologists who have had

some training in osteology, but who may be inexperienced or 'rusty' (Krogman 1962;

Birkbyl966; Giles 1970). For the trained osteologist and forensic anthropologist, discriminant

function analysis can be used as an objective check to visual methods and adds weight to expert

testimony (Snow 1979).

Discriminant function analysis has two critical limitations when used for sex estimation of

skeletal remains. The first is the need for all measurements used in the function to be observable.

The second is that the functions can only be used on individuals who come from the population

from which the function was developed.

The first problem is fairly straightforward, as is its solution. A discriminant function score

cannot be calculated if an observation is missing. One solution is to generate multiple functions

using different combinations of measurements. Another solution is to generate functions using

skeletal elements that are robust, and therefore likely to be preserved. This approach was taken

by Black [femur] (1978); DiBennardo and Taylor [femur] (1982); Taylor and DiBennardo

[stepwise-femur] (1982); Iscan and Miller-Shaivitz [tibia] (1984); Dittrick and Suchey [femur

and humerus] (1986).

The problem of using functions across populations is more difficult. Some authors have

suggested moving the sectioning point by various methods, including using the mid-point









SEX DETERMINATION BY DISCRIMINANT FUNCTION ANALYSIS OF NATIVE
AMERICAN CRANIA FROM FLORIDA AND GEORGIA





















By

MICHAEL BRYAN MCGINNES


A THESIS PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
MASTER OF ARTS

UNIVERSITY OF FLORIDA


2007









metric methods use a single measurement, and compare it to a distribution of that measurement

from a collection of known sex individuals. A sectioning point is placed such that males and

females are equally likely to be classified correctly, and misclassifications are minimized. The

sectioning points are usually arrived at by discriminant function analysis. Sectioning points for

several different long bones are given by Bass (1995), and an exhaustive list of discriminant

function studies to determine sex and their accuracies can be found in Rathbun and Buikstra

(1984:212-216). There are a host of discriminant function tests based on cranial and post-cranial

measurements.

Giles and Elliot (1963) provide one of the best established discriminant functions for sex

determination using the skull. From combinations of nine cranial measurements a total of 21

discriminant functions are described to indicate sex in whites, blacks and whites and blacks taken

together. The measurements are:

Glabello-occipital length: The maximum length of the skull, from the most anterior point of the
frontal in the midline to the most distant point on the occiput in the midline.
Maximum width: The greatest breadth of the cranium perpendicular to the median sagittal plane,
avoiding the supra-mastoid crest.
Basion-bregma height: Cranial height measured from basion to bregma.
Maximum diameter bi-zygomatic: The maximum width between the lateral surfaces of the
zygomatic arches measured perpendicular to the median sagittal plane.
Basion-nasion: The direct distance from basion to nasion.
Basion-prosthion: The direct distance from basion to the most anterior point on the maxilla in the
median sagittal plane.
Nasion breadth: The maximum breadth of the nasal aperture perpendicular to nasal height.
Palate-external breadth: The maximum breadth of the palate taken on the outside of the alveolar
borders.
Opisthion-forehead length: The maximum distance from opisthion (the midpoint on the posterior
border of the foramen magnum) to the forehead in the midline.
Mastoid length: The length of the mastoid measured perpendicular to the plane determined by
the lower borders of the orbits and the upper borders of the auditory meat uses (Frankfort
plane).

Functions 1, 2, and 3 use 8 of the 9 measurements, are the most accurate for each group,

and use the same measurements from each group. The sectioning point is halfway between the









the present case, it is used to test the hypothesis that all of the individuals from the different sites

in this research can be considered to be a single group for the purposes of this study. The

MANOVA is preformed using SAS PROC GLM.

PROC GLM Data=thesisdata ORDER=freq;
CLASS site;
MODEL GOL XCB ZYB BBH BNL BPL AUB WFB UFBR EKB FRC PAC MDHA= site;
MANOVA H=site / PRINTED;


The next task is to calculate the results of the Giles and Elliot (1962) function 3 for sex

discrimination on the study sample in order to evaluate its accuracy. Sex is first determined using

the sectioning point given by Giles and Elliot. The sectioning point is then recalculated using the

method described by Giles and Elliot. The average of the function is calculated using the average

value of each function for each sex separately,

f(f,,ff)

and the sectioning point is placed midway between the two results. The function is calculated

again, replacing missing values with the average of the male and female means for that variable.

Accuracy is then evaluated again using the original sectioning point, and the recalculated

sectioning point described above.

Once the accuracy of the Giles and Elliot formula is determined, new discriminant

functions are calculated based on the data collected. The first function calculated uses the same

variables Giles and Elliot used. The second function calculated uses the variables selected using

t-tests and that showed good preservation as indicated by the proportion of crania from which the

measurement could be observed.

Exact binomial probabilities are used to compare the accuracy of new results to Giles and

Elliot's (1963) reported accuracy of 86.4%. Exact probabilities are calculated using the SAS









The Middle Archaic was a time of dramatic cultural change in the Southeast. Middle

Archaic peoples occupied new types of locations for the first time and created new types of sites,

including freshwater and marine shell middens (Milanich 1994). Ceremonial shell and earthen

mound construction were initiated in several areas, long-distance trade networks appeared, and

new tool forms were adopted (Anderson and Sassaman 2004:95). Both Early and Middle Archaic

peoples in peninsular Florida began using aquatic environments, such as Windover pond, for

burial (Milanich 1994:81).

Late Archaic

The beginning of the Late Archaic coincides with the beginning of the late Holocene

Period and essentially modern environmental conditions by 3000 B.C. (Milanich 1998;

Sassaman and Anderson 2004; Watts and Hansen 1988:310). This period is marked by greater

regionalization and cultural diversity as human populations adapted to specific environmental

zones. Cultures were no longer faced with the challenge of long-term environmental and climatic

fluctuations (Milanich 1994).

During the Late Archaic period, the firing of clay pottery, along with other technological

innovations, appeared in Florida. Ceramic vessel technology gradually spread across the

southeast, and was adopted by virtually all regional populations by about 650 B.C. "Local

variations in pottery technology and style reflect growing diversity of cultural expression.

Regional exchange and intergroup ritual at locations of ceremonial earthworks were among the

means by which members of different populations interacted" (Anderson and Sassaman

2004:101). More Late Archaic sites are know than sites from any earlier period.

Despite changes in technology, there are few apparent differences between Late Archaic

subsistence strategies and earlier periods. Populations in the Southeast continued to expand on

the hunting and gathering economies of ancestral populations, with shellfish, fish and other food









In terms of accuracy, Function 1 correctly categorizes all individuals in the initial sample,

but did not do as well overall in cross validation or when missing values are substituted with

neutral values. The accuracy of Function 1 is not significantly different from the Giles and Elliot

function. Since it is based on the same variables, Function 1 suffers from the same issues of

applicability as the Giles and Elliot function. Function 1 could only be applied to 2 of 13

females, and 15 of 32 males.

Function 2 is more accurate across the board than the Giles and Elliot function, but the

difference is not statistically significant. Accuracies are almost identical in cross validation and

when missing values are substituted with neutral values. Function 2 offers better applicability

than the Giles and Elliot function. Where the Giles and Elliot function could be applied to 2 of

13 females and 15 of 32 males, Function 2 could be applied to 12 of 13 females and 27 of 32

males.

Discussion

The goal of the research is to determine if new discriminant functions for sex

determination could be developed for Florida and Georgia Native American populations that are

better than the functions presented by Giles and Elliot (1963). While neither Function 1, which

uses the Giles and Elliot variables, nor Function 2, which uses more robust variables, did

significantly better than the Giles and Elliot function, Function 2 could be used on more than

twice as many individuals without having to substitute missing variables.

While neither of the two new functions are significantly better than the Giles and Elliot

function with a recalculated sectioning point, there is reason be believe that creating new

functions might be beneficial. The first has to do with the problems involved with calculating the

new sectioning point. The second problem has to do with sample size and the power of the

statistics used here.









the Casey Key individual is an unusually small female. There is not enough evidence to justify

rejecting the null hypothesis, eliminating the variable AUB, or eliminating either site.

The MANOVA test is the multivariate equivalent to an ANOVA. It is used in cases where

there are multiple metric dependent variables, and one or more categorical independent variables.

It is used to test the null hypothesis that there is no overall difference between sites for the array

of variables included in this study. The sex of the individual is also used as an independent

variable so that all individuals can be included, and site difference is controlled for sex

differences.

Table 7-4 MANOVA test criteria and F approximations for the hypothesis of no overall site
effect, using type IV sums of squares and cross products.
Statistic Value F Value DF Pr > F
Wilks' Lambda .1636 1.44 40 0.0709
Pillai's Trace 1.387 1.39 40 0.0817
Hotelling Lawley Trace 2.483 1.48 40 0.0830


In the MANOVA test, all variables are considered as a single array to test the null

hypothesis that site has an effect on the array of cranial measurements against the alternative that

there is a site effect. SAS provides three approximations of the F for a MANOVA, Wilk's

Lambda, Pillai's Trace and the Hotelling- Lawley Trace. None of the statistics are significant at

the 0.05 alpha-level, so there is not enough evidence to reject the null hypothesis (Table 7-4).

This indicates that it is reasonable to treat all sites as a single group for this analysis.

Discriminant Function Analysis

The goal of the discriminant function analysis is to produce a linear formula that can be

used with cranial data to accurately assign sex to Native American skeletal remains recovered

from Florida archaeological sites. To use the function, each variable is multiplied by the

corresponding coefficient and the results and constant are summed. If the result is greater than









Mahalanobis Distance

The Mahalanobis distance (D2) is a measure of distance that takes into account

correlations between variables. For classification purposes, Mahalanobis distance can be used to

measure the distance of individuals to group centers and each individual can be allocated to the

group to which it is closest. If x,,x,,...,x, are the values of variables X,,X,,...,Xp for the

individual, with corresponding population mean values of 1, 12,..., then:

P P
D 1= (X, Y,)vs (Xs Ys)
r=l s=l

where v'"is the element in the rth row and sth column of the inverse of the covariance matrix for

thep variables (Manly 1994:63).

Assumptions

Both canonical and Malahanobis distance methods are based on two assumptions. The first

is that the within-group covariance matrix is the same for all groups. The second is that the data

is normally distributed within groups. The second assumption is important for the validity of

tests of significance.

Logistic Regression

A different approach to discrimination between two groups uses logistic regression.

Logistic regression is a variation of multiple linear regression where the dependent variable is

assigned as either 1 or 0, usually used as 'success' or 'failure.' The regression formula then returns

a probability of success or failure. Rather than representing success or failure, 1 and 0 can be

used to represent groups. When applied to unknown individuals, the regression will return the

posterior probability of group membership, with a sectioning point of 0.5. Konigsberg and Hens

(1998) use logistic regression in sex estimation on measurements of the crania. They reported









The Archaic Period

The Archaic Period describes is differentiated from the earlier Paleoindian period on the

basis of stylistic differences in point types, the appearance of new artifacts types, and apparent

changes in economic orientation (Anderson and Sassaman 2004). The change in material culture

from Paleoindian to Archaic coincided with a change to a warmer, less arid climate. As the

temperature increased, the glaciers retreated and sea levels rose. Because of these environmental

changes many species that had previously thrived in the Southeast went extinct or disappeared

from the region. Changes in flora and fauna led to changes in subsistence patterns and material

culture.

The regional chronology for the Archaic period in the Southeast was established by

correlating changes in dagnotic artifacts from excavations at deeply stratified sites such as Ice

House Bottoms in Tennessee, Russell Cave in Alabama, Indian Knoll in Kentucky, and the

Hardaway and Doerschuk sites in North Carolina (Stoltman 2004). The Archaic is traditionally

divided into Early, Middle, and Late Archaic phases.

Early Archaic

The Early Archaic can be seen as a transition from the Paleoindian period to the Middle

Archaic. Population increased throughout the Archaic, and people were shifting from nomadic

hunting to somewhat more sedentary lifestyles near coastal and riverine settings (Milanich

1994). The shift in subsistence strategies coincided with a period of transition to warmer, less

arid conditions. Projectile points transition from lanceolate forms present during the Paleoindian

period to stemmed, side- and corner-notched, and hafted forms with bifurcated bases (Anderson

and Sassaman 2004; Milanich 1994:63). The Early Archaic appears to reflect a continuation of

the Paleoindian hunting and gathering lifestyle with increased regional specialization. Early









OCC, l-o

p


PAC, b-1

UFBR, fmt-fmt

UFHT, n-pr

WFB, ft-ft

XCB, eu-eu

ZYB, zy-zy


Occipital chord

P-value, the probability of getting a value at least as extreme as the
observed value by chance alone.

Parietal chord

Upper facial breadth

Upper facial height

Minimum frontal breadth

Maximum cranial breadth

Bizygomatic breadth









Phenice (1969) identifies three indicators of sex in the pubic bone: the ventral arch, the subpubic

concavity, and the medial aspect of the ischiopubic ramus.

The ventral arch is a slightly elevated ridge of bone that sweeps inferiorly and laterally

across the ventral surface of the pubis, merging with medial border of the ischiopubic ramus. The

ventral arch is evaluated by orienting the pubis so that its rough ventral surface faces the

observer, who looks down along the plane of the pubic symphysis surface. The ventral arch,

when present, sets off the inferior, medial corner of the pubic bone in ventral view. The ventral

arch is present only in females. Male pubic bones may have elevated ridges in this area, but these

do not take the wide, evenly arching path of the female's ventral arch or set off the lower medial

quadrant of the pubis.

The subpubic concavity is a concave curve on the medial edge of the ischiopubic ramus

displayed in female os coxae. The female ischiopubic ramus is concave, while male edges are

straight or very slightly concave. The subpubic concavity is evaluated by turning the pubis over,

orienting it so that its smooth, convex dorsal surface faces the observer, who is once again

sighting along the midline. From this position it is possible to observe the medial edge of the

ischiopubic ramus. For females, the edge of the ramus is concave in this view. Males do not

show the dramatic concavity here. Male edges are straight or very slightly concave. (If the bone

is in good shape, and there is no danger of damage, another method for evaluating the subpubic

concavity is to lay the ischiopubic ramus on a flat surface. If it can be rocked, it indicates a male,

if it cannot be rocked, it indicates a female).

To evaluate the medial aspect of the ischiopubic ramus, the observer turns the pubis 90,

orienting the symphysis surface so that the observer is looking directly perpendicular to it. From

this position it is possible to observe the ischiopubic ramus in the region immediately inferior to









CHAPTER 4
ARCHAEOLOGICAL CONTEXT

Introduction

This chapter provides a general cultural history of the southeast, introduces the

archaeological sites that provide the skeletal materials for this study. It places those sites

geographically, temporally and within the framework of the area's cultural history.

Environmental Setting

Human occupation in what is now Florida and Georgia began about 13,000 B.P.2 during

the end of the Pleistocene epoch. From the start of human occupation until about 7000 B.P.,

Florida and Georgia were undergoing tremendous environmental change. The environment

gradually went from cold and dry at the end of the Pleistocene, to the warm and humid modern

climate. That change in temperature was accompanied by higher sea levels, which have reduced

Florida to half the land area of what it was when people first entered the state. Along with a

warming climate and rising sea levels, many of the plants and animals in Florida and Georgia

were supplanted by species better adapted to the changing environment. By 7000 B.P. sea levels

reached about the levels where they are today, and the flora and fauna present were essentially

the same species present in the region today (Milanich 1994).

The coastal zone of the region, including the Gulf of Mexico and Atlantic coast of Florida

and southern Georgia, is characterized by a generally inhospitable beach and foreshore. This area

is subject to seasonal exposure to storms and scarce resources. Chains of barrier islands are also

found along the eastern coastline of Florida and Georgia containing beach and dune landscapes,



2 B.P. stands for years before present based on radiocarbon dating, with present defined as the year 1950. It is an
alternative to traditional dates of A.D. and B.C.. B.P. dates can be roughly converted to traditional dates by
subtracting 1950, but radiocarbon years are not equivalent to calendar years. For this reason, the dating framework
used in the primary reports for each site and region are followed without attempting to impose imprecise
conversions.









Archaic sites have been found at a number of locations in Georgia (Wauchope 1966) and Florida

(Milanich 1994), many near permanent water sources.

Because of the rapidity of post-glacial sea level changes, the complex estuary systems that

would eventually become dominant resource procurement areas for many coastal populations

had not stabilized. Stable, productive estuary systems were not present until after 3,000 B.C. Sea

level rise caused ground water levels to rise increasing the amount of surface water available

(Widmer 1988; Watts 1975; Doran 2002:50). The increase in surface water made new locations

suitable for occupation, allowing Archaic peoples to move into new ecotones (Milanich

1994:63). Additionally, "because water sources were large and more numerous, the Early

Archaic peoples could sustain larger populations, occupy sites for longer periods, and perform

activities that required longer occupation at a specific locale" (Milanich 1994:69). A period of

greater aridity returned near the end of the Early Archaic, about 6000 B.C., though less arid than

at the end of the Pleistocene.

Middle Archaic

Average annual temperatures during the Middle Archaic were not much different than

modern temperatures, but temperature variance was more extreme. Summers were hotter and

winters were colder. While lake water levels were lower throughout much of the North American

continent, sea and ground water levels were higher in Florida and Georgia (Anderson and

Sassaman 2004). During the Middle Archaic period more and larger surface water sources were

available in Florida, and increasingly moist conditions appeared after about 4000 B.C. (Milanich

1994:84; Watts 1969, 1971; Watts and Hansen 1988). A gradual change in forest cover occurred

with pines and mixed forests replacing oaks. By about 3000 B.C., vegetation and climate become

essentially modern, and sea level rise tapered off (Milanich 1994:75, 84).









plane. The ends of the caliper are held in each hand and applied to the lateral portions of the

skull, making circular motions until the maximum breadth is obtained. Areas below the

squamosal suture are included, where the maximum is sometimes found (Bass 1971:62; Howells

1973:172; Hrdlicka 1952:140; Martin 1956:455; Montagu 1960:44).

Basion-bregma height (ba-b, BBH): Cranial height measured from basion to bregma. This

measurement is equivalent to Basion Bregma Height; the direct distance from the lowest point on

the anterior margin of the foramen magnum, basion (ba), to bregma (b) is measured with the

spreading caliper. The skull is placed on its side and the endpoint of one of the arms of the

caliper is placed at the most inferior point of the margin of the foramen magnum in the mid

sagittal plane and supported with fingers. Then the endpoint of the second arm of the caliper is

applied to bregma (Bass 1971:62; Howells 1966:6; Martin 1956:459; Olivier 1969:129).

Maximum diameter bi-zygomatic (zy-zy, ZYB): Maximum width between the lateral

surfaces of the zygomatic arches measured perpendicular to the median sagittal plane. This

measurement is equivalent to Bizygomatic Breadth; The direct distance between each zygion

(zy), located at the most lateral points of the zygomatic arches measured with a sliding caliper.

The skull is placed on its base, and he blunt points of the caliper are applied to the zygomatic

arches and the maximum breadth is recorded (Bass 1971:67; Martin 1956:476).

Basion-nasion(ba-n, BNL): Distance from basion to nasion. Equivalent to Cranial Base

Length: The direct distance from nasion (n) to basion (ba) measured using the spreading caliper.

The skull is placed with the cranial vault down on a cork or sandbag skull-ring. The endpoint of

the one arm of caliper is applied to nasion (n) while the other is applied to the anterior border of

the foramen magnum in the mid sagital plane. This measurement is not taken where anomalous









BIOGRAPHICAL SKETCH

Michael McGinnes, who grew up in Gainesville, Florida, earned an Associate of Science

degree in photography from the Southeast Center for Photo/Graphic Studies at Daytona Beach

Community College, an Associate of Arts degree from Santa Fe Community College, and a

bachelor's degree from the University of Florida. He has been employed as a professional

archaeologist since 1995 and specializes in mortuary archaeology. He has conducted field

projects in Panama, Florida, Georgia, South Carolina, Delaware, Oklahoma, Texas, Alabama,

Maryland, Virginia, and the District of Columbia. At the National Museum of Natural History,

Smithsonian Institution, he worked as the bibliographic research assistant for the Southeast

volume of the Handbook of North American Indians, and was an intern for the collections

manager of the physical anthropology collections. He is married to Dr. Ruth Trocolli, the

archaeologist of the District of Columbia.









B TABLE OF SITES............................................................... 92

L IST O F R E F E R E N C E S ......... ..... ............ ................. ......................................................94

B IO G R A PH IC A L SK E T C H ........... .... ............. ................. ....................... ....................... 105









for the black and white sample. For the Florida Indians, however, this is true only when the

sectioning point is based on mean values of these same Indians." (Giles and Elliot 1963:66-67).

Can the accuracy of sex determination by discriminant function analysis be improved over

Giles and Elliot's results by deriving new formulas based on regional populations? This

hypothesis is tested using skeletal remains recovered from archaeological sites in Florida and

Georgia.









1975. All three locations are on Cannon's Point at the northern end of the island. Wallace (1975)

used ceramic and mortuary analysis to examine the relationship between these sites. He found that

the sites are from a single, contemporaneous group of Guale Indians, but represent different

hierarchical groups within the culture. That conclusion will not be challenged here.

Martinez B-C is located at the tip of Cannon's Point near the Hampton River. The test pit

was excavated there in 1974 and initially two burials were recovered. A third burial discovered

adjacent to the test pit was recovered in 1975. The individuals include one infant and two adult

males, and all are primary extended burials. This location is thought to be part of a primary

living area. Based on associated ceramics, the burials are thought to be from the Wilmington

Period (Martinez 1975:56-58).

Taylor Mound is a Historic period (ca. A.D. 1600-1650) ceremonial mound with associated

burials. While some historic artifacts are associated with the mound, it does not appear to have

been heavily impacted by European contact (Wallace 1975:39-78; Zahler 1976:2). Eleven burials

were recovered from this location, The sex of one skeleton could not be identified because it was

too fragmented and too young. A second was excluded from analysis because its stratagraphic

affiliation was uncertain (Wallace 1975:44). The remaining nine individuals included seven

females and two males. Prior to formal excavation, thirteen burials were excavated by local

residents, but information for these individuals was not recorded.

Couper Field and Indian Field are the northern and southern parts, respectively, of the

same village, and is part of the same sociocultural population as Taylor Mound, but represents

different levels of the social hierarchy. Couper Field lies immediately south of the remains of the

antebellum Couper Mansion. Although the area had been heavily plowed the majority of burials

are undisturbed. There are 16 interments containing 18 individuals, including one infant, ten












AUB, au-au

B.P.

BBH, ba-b

BNL, ba-n

BPL, ba pr

DFA

DKB, d-d

DKB, d-d

EKB, ec-ec

EKB, ec-ec

FOB

FOL, ba-o

FRC, n-b

GOL, g-op

MAB, ecm-ecm

MAL, pr-alv

MDHA

MDHL

MDHR

MNI

NLB, al-al

NLH, n-ns

OBB, d-ec

OBH


LIST OF ABBREVIATIONS

Biauricular breadth

Radiocarbon years before present, with present defined as the year 1950.

Basion bregma height

Cranial base length

Basion prosthion length

Discriminant function analysis.

Interorbital breadth

Interorbital breadth

Biorbital breadth

Biorbital breadth

Foramen magnum breadth

Foramen magnum length

Frontal chord

Maximum cranial length

Maxillo-alveolar breadth external palate breadth

Maxillo-alveolar length, external palate length

Average mastoid height

Left mastoid height

Right mastoid height

Minimum Number of Individuals

Nasal breadth

Nasal height

Orbital breadth

Orbital height









Stewart, Thomas Dale
1946 A Re-examination of the Fossil Human Skeletal Remains from Melbourne,
Florida, with Further Data on the Vero Skull. .Suithi,iuial Miscellaneous
Collections 106(10): 1-28.

1948 Medico-legal aspects of the skeleton: Age, sex, race and stature. American
Journal of Physical Anthropology 6:3 15-321.

1979 Essentials ofForensic Anthropology. Charles C. Thomas, Springfield, Illinois.

Stirling, Matthew W,
1935 Smithsonian Archaeological Projects Conducted under the Federal Emergency
Relief Administration, 1933-1934. Annual Report of the S.itli,,i, tnin Institution
1934, pp. 371-400. Washington, D.C.

Stoltman, James B.
2004 History of Archaeological Research. In So.,inheati, edited by Raymond D.
Fogelson, pp. 14-30. Handbook of North American Indians, Vol. 14, William C.
Sturtevant, general editor, Smithsonian Institution, Washington, D.C.

Taylor, James V., and Robert DiBennardo
1982 Determination of Sex of White Femora by Discriminant Function Analysis:
Forensic Science Applications. Journal of Forensic Sciences 27:417-423.

Wauchope, Robert W.
1966 Archaeological Survey of Northern Georgia with a Test of Some Cultural
Hypotheses. American Antiquity 35; Memoirs of the Society for American
Archaeology 21. Washington, D.C.

Waller, Ben I.
1969 Paleoindian and other artifacts from a Florida stream bed. Florida Anthropologist
22: 37-39.

Watts, William A.
1969 A pollen diagram from Mud Lake, Marion County, north-central Florida.
Geological Society ofAmerica Bulletin 80: 631-642.

1971 Post-Glacial and Interglacial Vegetation History of Southern Georgia and Central
Florida. Ecology 52:676-689.

1975 A late Quaternary record of vegetation from Lake Annie, south-central Florida.
Geology 3: 344-346.

Watts, William A., and Barbara C.S. Hansen
1988 Environments of Florida in the Late Wisconsin and Holocene. In Wet Site
Archaeology, edited by Barbara A. Purdy, pp. 307-323. Telford Press, West
Caldwell, New Jersey.









resources becoming increasingly important (Milanich 1994:85). With the exception of areas that

eventually adopted intensive agriculture, the general subsistence patterns of the late Archaic

period continued largely unchanged into the colonial period (Milanich 1994; Sassaman and

Anderson 2004). Diminished rates of sea-level rise promoted the establishment of increasingly

productive estuarine environments, and maturing floodplain habitat. During this time, evidence

of coastal populations in Florida is much more abundant, and Late Archaic shell middens are

preserved in many locales (Milanich 1994; Sassaman and Anderson 2004:101). "The panregional

spread of mortuary ceremonial institutions and greater use of native cultigens in certain sub-

regions mark the end of the period at about 650 B.C." (Sassaman and Anderson 2004:101).

Woodland and Regional Cultures

The Woodland period follows the Archaic. It is characterized by increasing population and

social complexity through time, and limited adoption of horticulture. With more people on the

landscape, mobility decreased and local manifestations of the culture emerged. Trade and

exchange with regions outside the Southeast occurred to some extent. As groups settled into their

local environment and became more sedentary, regional cultures began to emerge. Regional

cultures are distinguished by variations in potter styles, projectile point styles, house types, and

settlement patterns. Due to the abundance of regional cultures, only the woodland cultures

represented in the skeletal sample are detailed here.

The advent of the Woodland period occurs at different times throughout Eastern North

America, but began earlier in Florida and Georgia and lasted until the European contact in some

areas. After 500 B.C., there is archaeological evidence for occupation of every environment

within Florida, including the forested interior uplands of northern Florida (Milanich 1994: 106).

After about A.D. 750, there is evidence for more intensive cultivation of plants, including the

possible introduction of maize (Milanich 1994: 108).









which is not common in the literature (see fig. 3-1; Giles and Elliot 1962). The Giles and Elliot

functions correctly identified race for 82.6% of males 88.1% of females (Giles and Elliot 1962).

Tests of Significance

Various tests of significance are available for evaluating the difference between the mean

values for any pair of groups, overall differences between the means of several groups, and if the

mean of a discriminant function differs from group to group. See Harris (1985) for a discussion

of the difficulties surrounding these tests.

Prior Probabilities

Some computer programs can allow for prior probabilities of group membership. This

could be useful in sex determination if reliable data can be generated for the proportion of male

and female skeletons recovered from a particular environment, whether due to taphonomy,

demographics, or other causes. Care should be taken that assignment of prior probabilities

reflects actual or known proportions (e.g. number of males and females in a population) and not

any form of prior bias.

Stepwise Discriminant Function Analysis

Stepwise discriminant function analysis simply applies a stepwise selection to the variables

included in a discriminant function analysis. Variables are added to the discriminant functions,

one at a time, until adding additional variables does not give significantly better discrimination.

Several authors have used stepwise discriminant function analysis in estimating sex from skeletal

measurements. Holman and Bennett (1991) use the procedure built into SAS (STEPDISC) on the

bones of the arm and wrist with good results. Taylor and DiBennardo (1982), DiBennardo and

Taylor (1983), and Iscan and Miller-Shaivitz (1984) each use the procedure built into SPSS for

the femur, femur and pelvis, and tibia, respectively, with good results.














N
N


. Simons Island


Atlantic
Ocean


Gulf of Mexico

0 50 100
S 5 0 0Miles
6 50 160Kilometers


Course


Figure 5-1 Locator map of sites used in this study and major rivers.









Buikstra, Jane E., and Douglas H. Ubelaker (editors)
1994 Standards for Data Collection from Human .\kel/ei Remains: Proceedings of a
Seminar at the Field Museum ofNatural History. Arkansas Archaeological
Survey Research Series 44. Fayetteville, Arkansas.

Bullen, Ripley P.
1958 Six Sites Near the Chattahoochee River in the Jim Woodruff Reservoir Area,
Florida. In River Basin Survey Papers, edited by Frank H. H. Roberts, Jr., pp.
315-357. Bureau of American Ethnology, Bulletin 169. Smithsonian Institution,
Washington, D.C.

Bullen, Ripley P., and Adelaide K. Bullen
1976 The Palmer Site. Florida Anthropological Society, Gainesville. Issued as Florida
Anthropologist 29, no.2, part 2.

Calcagno, James M.
1981 On the Applicability of Sexing Human Skeletal Material by Discriminant
Function Analysis. Journal ofHuman Evolution 10:189-198.

Caldwell, Joseph R.
1952 The Archeology of Eastern Georgia and South Carolina. In Archeology ofEastern
United States, edited by James B. Griffin, pp. 312-321. University of Chicago
Press, Chicago.

Clausen, Carl J., A. D. Cohen, Cesare Emiliani, J. A. Holman, and J. J. Stipp
1979 Little Salt Spring, Florida: A Unique Underwater Site. Science 203:609-614.

DePratter, Chester B.
1979 Ceramics. In The Anthropology of St. Citwhei ine, Island, The Refuge-Deptford
Mortuary Complex, edited by D. H. Thomas and C. S. Larsen, pp. 109-132.
Anthropological Papers 56 (1). American Museum of Natural History, New York.

DiBennardo, Robert, and James V. Taylor
1979 Sex Assessment of the Femur: A Test of a New Method. American Journal of
Physical Anthropology 50:635-637.

1982 Classification and Misclassification in Sexing the Black Femur by Discriminant
Function Analysis. American Journal of Physical Anthropology 58:145-151.

1983 Multiple Discriminant Function Analysis of Sex and Race in the Postcranial
Skeleton. American Journal of Physical Anthropology 61:305-314.

Dittrick, Jean and Judy M. Suchey
1986 Sex Determination of Prehistoric Central California Skeletal Remains Using
Discriminant Analysis of the Femur and Humerus. American Journal of Physical
Anthropology 70:3-9.









This study demonstrates that when using discriminant function analysis missing variables

can be replaced with the average of the male and female means with good results, although the

limits of this method are not explored. One potential line of future research is to determine how

robust discriminant functions are to this method of replacing missing variables.

One of the purported advantages of discriminant function analysis is that the functions can

be used by individuals with relatively little training. This assumption needs to be tested. A study

where students with little or no osteology training are asked to measure crania and use

discriminant functions to determine sex should be performed. Their results could be compared to

a group of students given an identical amount of training in visual methods.









mean score for males and the mean score for females. A score above the sectioning point is

designated male; one below is designated female.

Table 2-1. Coefficients for discriminant functions 1,2, and 3 from Giles and Elliot 1963 used to
assign sex based on a white sample, a black sample, and a combined black and white
sample.
Measurements Whites Blacks Combined
(Function 1) (Function 2) (Function 3)
Glabello-Occipital Length (GOL) 3.107 9.222 6.083
Maximum Width (XCB) -4.643 7.000 -1.000
Basion-Bregma height (BBH) 5.786 1.000 9.500
Max Diameter Bi-zygomatic (ZYB) 14.821 31.111 28.250
Basion-Prosthion (BPL) 1.000 5.889 2.250
Prosthion-Nasion Height (UFHT) 2.714 20.222 9.917
Palate--External breadth (MAB) -5.179 -30.556 -19.167
Mastoid Length (MDH) 6.071 47.111 25.417
Sectioning Points 2676.39 8171.53 6237.95
Male mean 2779.66 8487.56 6466.17
Female Mean 2573.12 7855.50 6009.72
Sample Accuracy (Percent correct) 86.1% 84.6% 86.0%
Expected Accuracy 86.6% 87.6% 86.4%

A large variety of metric methods using the post-cranial skeleton are available in Krogman

and Iscan (1986) and Bass (1995). These are commonly used in forensic anthropology. Several

methods were developed for fragmentary remains that focus on bioarchaeology collections.

Using Bioarchaeology collections has the advantage of using the population of interest to

develop criteria, but in most cases the true sex of the individuals in the study is unknown.

Therefore, true accuracy cannot be determined, only how consistent the method is with other,

established methods.

Several studies have been focused on identifying sex from fragmentary remains. Using a

discriminant function based on only the midshaft femoral circumference of prehistoric skeletons

from Ohio, Black (1978) recorded an accuracy of 85%. Using the same measurement,

DiBennardo and Taylor (1979) developed and tested discriminant functions on black and white

femura of known sex and achieved an accuracy of 82%. Using various combinations of three









Among the most widely cited uses of canonical discriminant function analysis in

anthropology come from Giles and Elliot, who used the technique for the estimation of sex

(1963; Giles 1964) and race (Giles and Elliot 1962). For these studies, Giles and Elliot used

measurements of Native American remains from the Indian Knoll, KY site originally published

by Snow (1948), while black and white subjects came from the Terry and Todd Collections.

Their use of discriminant functions for estimation of sex from the crania is discussed in detail in

the chapter on sex estimation. In their calculations, Giles and Elliot used formula from Kendall

(1957), which are equivalent to those presented by Fisher (1936). Using Black and White

individuals from the same sample, Giles and Elliot (1963) developed discriminant functions for

sex using nine cranial measurements in different combinations to form 21 discriminant functions

for sex determination. An accuracy of 82-89% is attained with the Black and White material.

This compares favorable with the 77-87% accuracy expected from visual sex estimation using

the cranium alone.

Multiple Discriminant Functions

When canonical discriminant analysis is used, it may be possible to determine several

linear combinations of variables for separating groups where there are multiple groups and

variables. The number of functions available is either the number of variables or one less than

the number of groups, which ever is smaller. All functions maximize the F-ratio subject to the

condition that they are uncorrelated with previous functions within groups. The canonical

discriminant functions are, therefore, linear combinations of the original variables chosen such

that the first function reflects group differences as much as possible, and subsequent functions

capture as much as possible of group differences not displayed by previous functions. Group

assignment is then accomplished by calculating the distances to group means (Manly 1994:108-









growths occurred on the anterior border of the foramen magnum (Howells 1966:6; Martin

1956:455).

Basion-prosthion (ba pr, BPL): Distance from basion to the most anterior point on the

maxilla in the median sagittal plane. Equivalent to Basion Prosthion Length; The direct distance

from basion (ba) to prosthion (pr) measured using a spreading caliper, or sliding caliper where

the foramen magnum obstructs the use of spreading calipers or in crania in which the central

incisors have been lost. The fixed point of the sliding caliper or one tip of the spreading caliper is

applied to the most anterior point on the alveolar process in the mid sagittal plane. The movable

point of the sliding caliper or the other tip of the spreading caliper is then brought to the margin

of the anterior border of the foramen magnum in the mid sagittal plane (Martin 1956:474).

Nasion breadth (al-al, NLB): Maximum breadth of the nasal aperture perpendicular to

nasal height. Equivalent to Nasal Breadth; the maximum breadth of the nasal aperture measured

with a sliding caliper. The points of the instrument are placed on the sharp lateral margins of the

nasal aperture at its most lateral curvature. The measurement is taken perpendicular to the mid

sagittal plane and recorded to the nearest millimeter (Bass 1971:68; Howells 1973:176; Martin

1956:479; Montagu 1960:50; Olivier 1969:153).

Palate-external breadth (ecm-ecm, MAB): The maximum breadth of the palate taken on the

outside of the alveolar borders. Equivalent to Maxillo-Alveolar Breadth and External Palate

Breadth; the maximum breadth across the alveolar borders of the maxilla measured at its widest

point between each ectomolare (ecm). The maximum breadth is usually found at the level of the

second molars. Using a spreading caliper, both arms of the caliper are applied to the alveolar

borders above the tooth row from an anterior position. The points of measurement (ecm) are

usually not found on the alveolar processes, but are located on the bony segment above the









Table 7-2. T-scores and p-values for a difference in mean values of each cranial variable
between males and females. Significant values are in italics. Variables with
acceptable )reservation that are used in further analysis are in bold face.
Cranial Measurement Male Count Female Count Total Count T-statistic P-value
GOL 30 12 42 -3.25 0.0024
XCB 29 12 41 -2.25 0.0310
ZYB 15 4 19 -2.84 0.0113
BBH 27 11 38 -3.12 0.0036
BNL 27 10 37 -3.28 0.0023
BPL 26 9 35 -2.96 0.0057
MAB 25 8 33 -0.15 0.8835
MAL 24 10 34 -1.34 0.1889
AUB 30 13 43 -3.63 0.0008
UFHT 28 10 38 -0.79 0.4356
WFB 31 12 43 -3.16 0.0030
UFBR 29 11 40 -2.73 0.0096
NLH 28 10 38 -1.13 0.2679
NLB 27 10 37 -0.53 0.6000
OBB 26 10 36 -1.44 0.1577
OBH 27 10 37 -0.25 0.8047
EKB 25 10 35 -2.73 0.0100
DKB 25 10 35 -2.36 0.0242
FRC 32 12 44 -1.40 0.1696
PAC 32 13 45 -3.63 0.0008
OCC 30 11 41 -.11 0.9135
FOL 27 10 37 -1.53 0.1360
FOB 25 11 36 -1.82 0.0770
MDHR 28 13 41 -4.01 0.0003
MDHL 27 9 36 -2.70 0.0106
MDHA 31 13 44 -3.92 0.0003


Thirteen of the 26 measurements are significant at the alpha=0.01 significance level,

including left, right and average mastoid length (Table 7-2). The significant variables are:

Glabello-occipital length (GOL); Basion-Bregma height (BBH); Basion-nasion (BNL); Basion-

prosthion (BPL); Left, right and average Mastoid height (MDHL, MDHR, MDHA); Parietal

chord (PAC); Biauricular breadth (AUB); Minimum frontal breadth (WFB); Upper facial breadth

(UFBR); and Biorbital breadth (EKB). Giles and Elliot's function 3 includes the significant









variety of species (Milanich 1994:251,253). The Malabar I pottery assemblages include Saint

Johns sponge-spiculate tempered pottery, but are dominated by undecorated pottery tempered

with quartz sand. Malabar II pottery is characterized by the appearance of Saint Johns Check

Stamped pottery. Analysis of pottery from Malabar sites shows continuity of manufacturing

methods through both periods (Cordell in Sigler-Eisenberg et al 1985:118-134; Milanich

1994:250) and a link to Saint Johns pottery (Espenshade 1983; Milanich 1994:250).

Manasota

The Manasota culture found along the central peninsular Gulf coast region coincided with

the Deptford and early Weeden Island cultures, lasting from about 500 B.C. to A.D. 700. This

region, which surrounds Tampa Bay, extends along the Gulf from Pasco county south to Sarasota

County, and stretches inland nearly to the Peace River drainage. Most Manasota village sites are

multicomponent shell middens of various sizes found on or near the shore. Some of the coastal

shell middens include shell ramps constructed to provide access to the tops. Intensively occupied

interior villages with dirt middens have been found in wetland locals (Hemmings 1975; Padgett

1976; Luer et al. 1987; Milanich 1994). Other types of Manasota sites are found away from the

coast, in interior pine flatwoods on higher ground near water sources and wetland habitats. These

are presumed to be short-term villages and special use camps (Austin and Russo 1989). The

evidence from these sites suggests that the Manasota economy was based on fishing, hunting,

and shellfish gathering. Most of the Manasota meat diet was derived from aquatic species,

including fish, shark, rays, and shellfish. The Manasota peoples also consumed terrestrial species

such as deer, canines, rodents, birds, reptiles and amphibians (Milanich 1994). Manasota

material culture is dominated by the use of shell tools with some bone tools, but little use of

stone tools. Ceramics were limited to plain sand-tempered pottery (Luer and Almy 1979:40-41 in

Milanich 1994:222-223).









Statistical Procedures

Calculations are performed using Microsoft Excel 2004 for Macintosh, Version 11.3.3 and

SAS 8.3 running on Unix at grove.ufl.edu. A t-test is a statistical hypothesis test that follows a

Student's t-distribution if the null hypothesis is true and the variable is normally distributed. The

t-test is used to test the null hypothesis that the mean of a variable is the same for males and

females against the alternative that the means are different. The value of the t-statistic is the

difference between the means of the two groups divided by the standard error of the difference.

The result is used to identify which variables are likely to be useful in discriminating between

male and female crania. The t-statistic is calculated using the formula:


sx1 -X2
Xi-X2


where Xi X2 is the difference between the sample means, and s- -l is the standard error of

the differences between the two means. For groups of unequal size, s 2is computed by the

formula:

(N, 1)s + (N2 1)s2 1 1
X 2N, + N, -2 N, N,


where s2 (variance) is calculated by the formula:

2 (x- x)2
(n -1)

and s2 is calculated using the VAR function in Excel. The t-statistic is then compared to a

Student's t-distribution with n-2 degrees of freedom. This is accomplished using the 'TDIST'

function in Excel.

Multiple Analysis of Variance, or MANOVA, is the multivariate equivalent of a t-test. It is

used to test the null hypothesis that more than two groups do not differ for multiple variables. In









distributed. Using the recalculated sectioning point, the formula correctly classifies 27 of 32

males, and 11 of 13 females. The overall accuracy is 84.44%, and the error is evenly distributed.

Variables for further analysis are selected using t-tests and variable preservation rates. T-

tests are performed to identify variables that are likely to be useful in discriminating between

males and females. Ten variables are significant, including: GOL, BBH, BNL, BPL, AUB,

WFB, UFBR, EKB, PAC, and MDH. Five of these variables are excluded from further analysis

because their low preservation would have decreased the number of usable individuals to

unacceptable levels. The excluded significant variables are: BBH, BNL, BPL, UFBR, and EKB.

The variables retained for further analysis are GOL, AUB, WFB, PAC, and MDHA.

A MANOVA is performed using the retained variables to test the null hypothesis that there

is no difference between groups. There is not enough evidence to reject the null hypothesis that

there is no difference between groups. Therefore, it is concluded that all individuals could be

treated as a single group.

Two new discriminant functions are created. The first is created from the sample data

using the same variables used by Giles and Elliot (1963). The second is created using the

variables identified in the variable selection step. The first step is to create a new formula using

the same variables as Giles and Elliot (1963). This is done using PROC DISCRIM in SAS and

variables GOL XCB BBH ZYB BPL UFHT MAB MDHA. Again, discriminant function scores

could only be calculated for 17 individuals (13 Male and 4 female) due to missing variables. The

new function correctly classifies 13 of 13 males and 4 of 4 females, for a combined accuracy of

100%. This is not significantly different from the accuracy of the Giles and Elliot function.

Next, a new function is calculated using the variables identified during variable selection.

This is done using PROC DISCRIM in SAS and variables GOL, XCB, AUB, WFB, FRC, PAC,









CHAPTER 3
DISCRIMINANT FUNCTION ANALYSIS

Discriminant function analysis (DFA) has been used extensively to determine sex by

archaeologists and forensic anthropologists (e.g. Giles and Elliot 1963; Black 1978; DiBennardo

and Taylor 1979, 1982, 1983; Dittrick and Suchey 1986). The results are comparable to those of

traditional methods, but requires far less training and experience. This section describes the goals

and capabilities of discriminant function analysis, some of the methods for calculating

discriminant functions, and introduces research using discriminant function analysis in sex

determination.

Discriminant function analysis addresses the problem of how well it is possible to separate

two or more groups of individuals using multiple combinations of weighted variables. DFA

requires classes that are predetermined, such as male or female. The object is not to create

classes or populations that divide heterogeneous material. With two groups, there are two

specific errors one can make: mistaking a member of one group for being from the other. For

example, misclassifying (1) a male as a female, or (2) a female as a male. Both types of mistakes

should occur at an equal rate, and there should be as few mistakes as possible. Finally, each

subject must be assigned to one population or the other so that "Unknown" is not an option

(Kendall 1957).

There are several approaches, including Canonical discriminant functions, Mahalanobis

distances and logistic regression. These methods are described below, followed by the

advantages and disadvantages of DFA and by a literature review of DFA used in sex

determination.









1994 Archaeology ofPrecolumbian Florida. University Press of Florida, Gainesville.

1998 Florida's Indians From Ancient Time to the Present. University Press of Florida,
Gainesville.

2004a Prehistory of Florida After 500 B.C. In .Si,,nluhei, edited by Raymond D.
Fogelson, pp. 191-203. Handbook of North American Indians, Vol. 14, William
C. Sturtevant, general editor, Smithsonian Institution, Washington, D.C.

2004b Timucua. In .SNi,nlheti, edited by Raymond D. Fogelson, pp. 219-228. Handbook
of North American Indians, Vol. 14, William C. Sturtevant, general editor,
Smithsonian Institution, Washington, D.C.

Milanich, Jerald T., and Charles H. Fairbanks
1978 Florida Archaeology. Academic Press, New York.

Miller, Carl F.
1950 Early Cultural Horizons in the Southeastern United States. American Antiquity
15(4):273-288.

Miller, James J.
1974 An Archaeological Survey of the Palmer Oaks Tract in Sarasota County.
Miscellaneous Project Report Series, No. 20. Florida Bureau of Historic Sites and
Properties, Tallahassee.

Montagu, M.F. Ashley
1960 A Handbook ofAnthropometry. Charles C. Thomas, Springfield, Illinois.

Padgett, Thomas J.
1976 Hinterland Exploitation in the Central Gulf Coast-Manatee Region during the
Safety Harbor Period. Florida Anthropologist 29:39-48.

Olivier, Georges
1969 PracticalAnthropology. Charles C. Thomas, Springfield, Illinois.

Ousley, Stephen D., and Richard L. Jantz
1996 FORDISC 2.0: Personal Computer Forensic Discriminant Functions. Forensic
Anthropology Center, Department of Anthropology, University of Tennessee,
Knoxville.

Phenice, T.W.
1969 A Newly Developed Visual method of Sexing the Os Pubis. American Journal of
Physical Anthropology 30:297-302.

Quitmyer, Irv
1998 Zooarchaeological Indicators of Habitat Exploitation and Seasonality from the
Shell Ridge Midden, Palmer Site (8So2), Osprey, Florida. Florida Anthropologist
51:193-203.









1 is able to correctly classify 36 of 45 individuals (80%), while the Giles and Elliot function

correctly classified 15 of 45 individuals (33.33%).

This research draws on sites and individuals ranging from the Archaic to the Spanish

mission periods, and should be applicable to any remains from that time range found in Florida.

This research does not use any Paleoindian remains because they are so rare. Therefore, the

functions developed here are not recommended for use in determining the sex of Paleoindians.

One might argue that all Native Americans from the Archaic to the Spanish Mission period

from Florida and Georgia is too diverse a group to be considered a single population. The

MANOVA test results show that there is no significant between-site difference for these groups.

The Giles and Elliot function uses blacks and whites from dissecting room collections, and that

function works well for both of those groups. It is hard to argue that blacks and whites are a

single population, but that Native Americans from a limited geographic area are not, even if the

Native Americans are from a broad time span. The Giles and Elliot discriminant function works

well for both blacks and whites, and there is no reason to believe that the functions developed in

this research would not work as well for any Native Americans recovered from Florida or

Georgia.

Future Research

Because of the small sample size used in this research and the lack of significant

improvement over the established Giles and Elliot function, using Function 1 is not

recommended without further testing. It should be possible to generate larger samples using

Function 1 because the variables it uses are more robust than the ones used by Giles and Elliot.

Because of Function 1's potential for application to a larger number of less well-preserved crania

and potentially greater accuracy, it should be tested on a larger sample. Future tests should

include as many sites from Florida and Georgia as possible.









indicators. In sexing skeletons, the heaviest weight is given to the Phenice method of sexing the

pubic bone because of its high accuracy, followed by other features of the pelvis and postcrania.

Cranial indicators are given the least weight, and sexing by crania alone is avoided.

Materials

The samples used in this study come from a total of ten sites. Samples from Canaveral,

Perico Island, and Ballard Estates are housed at the Smithsonian Institution's National Museum

of Natural History. Samples from Garfield Site, Couper Field, Taylor Mound, Cannon's Point,

Palmer, Casey Key, and Bay Pines are from the Collections of the Anthropology Division of the

Florida Museum of Natural History. Measurements are taken from a total of 46 individuals.

Because some individuals are incomplete, not all measurements could be taken for every

individual, so not all individuals are included in each stage of the analysis.

Data are collected from one complete adult male recovered from the Garfield site. The

complete set of 25 cranial measurements is recorded. The postcrania, including the pubic bone is

missing, so sexing is accomplished using visual evaluation the crania.

Data are collected from nine individuals recovered from Couper Field, five males and four

females. Two of the four females are excluded from further analysis because they lacked a

measurable pubic bone and have fragmentary crania. The remaining two females are sexed with

using visual analysis of the crania and postcrania. Two males have an observable pubic bone.

The remaining three males are sexed using visual analysis of the crania.

Data are collected from seven individuals recovered from Taylor mound, including four

males, two females and one individual of unidentified sex. One male and one female are

excluded from further analysis because they lacked postcranial remains including the pubic bone

and have fragmentary crania. Two males are sexed using the pubic bone, although their crania

are fragmentary, and with only seven and 14 observable measurements. The remaining male and









In order to increase sample size, missing variables are replaced with the average of the

male mean and the female mean for each variable. This allows the discriminant function to be

calculated without the missing observation influencing the final classification.

With the missing variables replaced and using the Giles and Elliot sectioning point, 31 out

of 32 males (p=0.1123) are correctly classified, but only 2 out of 13 females (p<.0001) are

correctly classified. Both of these results are significantly different from both Giles and Elliot's

result and from each other (p<0.0001). With the sectioning point recalculated, 28 out of 32 males

are correctly classified, and 10 out of 13 females are correctly classified. The results using the

repositioned sectioning point are not significantly different from the Giles and Elliot result for

males (p=1.0000), females (p=0.5112), or male and females combined (p=0.8283).

Variable Selection

The goal of variable selection is to select for further analysis those variables that are likely

to aid in the discrimination of males and females, and that are likely to be preserved for

measurement. Variables are selected for further analysis using t-tests and variable preservation

rates. T-tests are performed for all cranial variables for a difference in mean between males and

females (Table 7-2). This is done in order to identify variables likely to useful in discriminant

function analysis. The t-test tests the null hypothesis that there is no difference between the male

and female means of a variable against the alternative that there is a difference. A variable whose

mean is not significantly different between males and females is unlikely to contribute to sex

discrimination. A variable that is poorly preserved in the present sample would reduce the

sample size to unacceptable levels, and is likely to limit the applicability of this research in other

cases.









CHAPTER 6
METHODS AND MATERIALS

Introduction

This section describes sampling methods and the criteria used for selecting skeletal

materials for measurement. It also describes the skeletal materials used for this study and the

methods used in determining the sex of each skeleton. Finally, it describes the procedures used in

data collection, including measurements taken and their description, and the software and

methods used in calculating the discriminant functions.

Sampling

In terms of research design, the ideal approach is to take a simple random sample or

random block sample from all pre-contact Native American skeletons excavated from Florida or

Georgia archaeological sites, and sex them using the pubic bone. Unfortunately, this ideal is not

practical. The remains of Florida and Georgia Native Americans are housed in several

institutions around the country, and not all are readily accessible. Of the accessible remains, not

all have measurable crania. Of those, few have an intact pubic bone, and none are of known sex.

Therefore, it is necessary to draw the sample from those remains that can provide the needed

data. The best practical method is to use all of the available remains that have a measurable

cranium, and which can be reliably sexed. This 'total sample' approach is likely to introduce bias

into the sample. For example, individuals who are robust, buried in shell middens, or from more

recent populations are likely to be over represented.

Determination of Sex

The sex of the specimens is determined by using visual assessment of the crania and

postcrania described in chapter 2. The Phenice (1969) method is used in conjunction with other

postcranial indicators, such as the sciatic notch and pre-auricular sulcus, and with cranial









Table 6-1. Measurements used by Giles and Elliot, the measurement's modern equivalent,
common symbols, Giles and Elliot descriptions, and references used. Complete
measurement descriptions and techniques are listed in appendix A.
Giles and Modern Symbols Description References
Elliot equivalent
measurement
Glabello- Maximum g-op, The distance of Bass 1971:62;
occipital cranial length GOL glabella (g) from Howells 1973:170;
length opisthocranion (op) Martin 1956:453 ;
in the mid sagittal Olivier 1969:128.
plane measured in a
straight line.
Maximum Maximum eu-eu, The maximum Bass 1971:62;
width cranial breadth XCB width of the skull Howells 1973:172;
perpendicular to the Hrdlicka 1952:140;
mid-sagittal plane. Martin 1956:455 ;
Montagu 1960:44.
Basion-bregma Basion bregma ba-b, The direct distance Bass 1971:62;
height height BBH from the lowest Howells 1966:6;
point on the anterior Martin 1956:459;
margin of the Olivier 1969:129.
foramen magnum,
basion (ba), to
bregma (b).
Maximum Bizygomatic zy-zy, The direct distance Bass 1971:67;
diameter bi- breadth ZYB between each Martin 1956:476.
zygomatic zygion (zy), located
at the most lateral
points of the
zygomatic arches.
Basion-nasion Cranial base ba-n, The direct distance Howells 1966:6;
length BNL from nasion (n) to Martin 1956:455
basion (ba).
Basion- Basion ba pr, The direct distance Martin 1956:474.
prosthion prosthion BPL from basion (ba) to
length prosthion (pr).
Nasion breadth Nasal breadth al-al, The maximum Bass 1971:68;
NLB breadth of the nasal Howells 1973:176;
aperture. Martin 1956:479;
Montagu 1960:50;
Olivier 1969:153.









Several authors point out the problem of using discriminant functions developed for one

population on another and conclude the problem of recalculating the sectioning point has no

practical solution (Henke 1977, Calcagnol981). Using the midpoint of the male and female

mean score for a function produces a workable sectioning point, but requires a sufficient number

of male and female individuals whose sex has already been identified. This is a problem akin to

opening a crate with the enclosed crowbar. If there are enough individuals whose sex has been

identified it is almost as easy to create a new discriminant function as it is to recalculate the

sectioning point. The Giles and Elliot function can be made to work across populations by

recalculating the sectioning point, but does not take full advantage of the power of DFA to

separate groups. Regional differences in shape are going to add within-group variance, which is

going to work to reduce the effectiveness of the function. A function developed on a group

includes regional differences as part of the calculation. A discriminant function created from the

population under study should have a higher ratio of between-group to within-group variation

and be more accurate if the sample size in sufficiently large. If the sample size is too small, the

function may be skewed by idiosyncratic variation within the sample.

Another problem with small sample sizes is a lack of statistical power. Statistical power is

the ability of a hypothesis test to detect a difference. The larger the sample size, the higher the

statistical power, and the smaller the difference a hypothesis test can detect. In the present study,

Function 1 is more accurate than the Giles and Elliot function, but there is not enough statistical

power for the difference to be significant because of the small sample size.

If one considers individuals who could not be sexed due to missing variables as incorrect,

then Function 1 does significantly better than the Giles and Elliot function (p< 0.0001). Function









Foramen Magnum Length (ba-o, FOL): The direct distance of basion (ba) from opisthion

(o) measured using a sliding caliper. The tips of the instrument are applied on the opposing edges

of the border of the foramen magnum along the sagittal plane (Martin 1956:455).

Minimum Frontal Breadth (ft-ft, WFB): The direct distance between the two

frontotemporale measured with a sliding caliper. With the skill placed on its base, the two

endpoints of the caliper are placed on the temporal ridges at the two frontotemporale, and the

least distance between both temporal lines on the frontal bone is recorded (Bass 1971:67;

Hrdlicka 1952:142; Martin 1956:457; Olivier 1969:151).

Foramen Magnum Breadth (FOB): The distance between the lateral margins of the

Foramen magnum at the point of greatest lateral curvature measured with a sliding caliper

(Martin, 1956:459).

Upper Facial Breadth (UFBR, fmt-fmt): The direct distance between each frontomalare

temporale measured using a sliding caliper. The measurement is taken between the two external

points on the frontomalar suture (Martin 1956:475). UFBR differs from Howells' FMB in that

the lateral most points on the suture are used rather than the most anterior points.

Nasal Height (n-ns, NLH): The direct distance from nasion (n) to nasospinale (ns)

measured using a sliding caliper. The direct distance from nasion to the midpoint of a line

connecting the lowest points of the inferior margin of the nasal notches is measured (Bass

1971:68; Howells 1966:6; Martin 1956:479; Olivier 1969:153).

Orbital Breadth (d-ec, OBB): The laterally sloping distance from dacryon (d) to

ectoconchion (ec) measured using a sliding caliper. The left orbit is measured for standardization

and practical reasons where available. If the left orbit is damaged or otherwise could not be









Function 2 correctly classified 28 of 32 males and 11 of 13 females for an overall accuracy of

86.87%. The accuracy of Function 2 is not significantly different than the Giles and Elliot (1963)

formula applied to the current sample using the recalculated sectioning point and missing

variables replaced with the average of the male and female means (p= 1.00).









point based on the sample data. Their method is to create a centroid discriminant function score

for each sex by using the mean value of each variable to calculate discriminant function scores.

The average of the male and female centroids is used as the sectioning point.

"We can determine the mean value of the discriminant function scores for males by taking
the mean male value for each measurement and entering them into the discriminant
function. If this is likewise done for the females, the arithmetic mean of the two scores
provides a sectioning point to use when we have no a priori reason to believe that a
specimen is more likely to be male than female (Kendall 1957). Following this procedure,
we will say that any specimen falling on the side of this line toward the male mean will be
called male, and any specimen falling on the other side will be called female. So doing
should minimize the probability of misclassification (Giles and Elliot 1963).


Table 7-1


Accuracy of the Giles and Elliot function 3 for sex determination on the study sample
f o Florida and Georgia Nat ve Americans


Number
Sectioning Point Sex Correct Total (n) Accuracy
Giles and Elliot overall 15 17 88.24%
Individuals Sectioning Point male 13 13 100.00%
with missing (6237.95) female 2 4 50.00%
variables Recalculated overall 15 17 88.24%
omitted Sectioning Point male 12 13 92.31%
(6513.392571) female 3 4 75.00%
Missing Giles and Elliot overall 33 45 75.56%
variables Sectioning Point male 32 32 100%
replaced with (6237.95) female 2 13 15.38%
average of the Recalculated overall 38 45 84.44%
male and Sectioning Point male 27 32 84.375%
female means (6513.392571) female 11 13 84.615%

The male centroid is 6731.518743, the female centroid is 6295.266398, and the

recalculated sectioning point is 6513.392571. Using the recalculated sectioning point, the overall

accuracy remains the same, but errors are evenly distributed. The function correctly sexed 12 of

13 males and 3 out of 4 females. The accuracy of the formula using the recalculated sectioning

point is not significantly different from the result reported by Giles and Elliot for males

(p=0.9109), females (p=0.8855) or males and females combined (p=1.0000).









Abstract of Thesis Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Master of Arts

SEX DETERMINATION BY DISCRIMINANT FUNCTION ANALYSIS OF NATIVE
AMERICAN CRANIA FROM FLORIDA AND GEORGIA

By

Michael Bryan McGinnes

December 2007

Chair: Anthony Falsetti
Major: Anthropology

The goal of this research is to determine if the accuracy of discriminant function analysis

for sex determination could be improved by using local or regional populations, and by better

variable selection. In archaeological contexts, skeletons are usually all that remains of the actual

people who once lived there. Information about the sex of those individuals is fundamental to the

study of demographics, sex roles, and life ways of past cultures. Determining an individual's sex

is also fundamental to personal identification of skeletal remains from modern contexts, whether

from a mass disaster or an unmarked grave.

There are several techniques available for determining sex from skeletal remains, and each

technique has its place. Discriminant function analysis is valuable in sex determination because it

requires relatively little training to use effectively and it serves as an objective check to other

methods. The cranium is a reliable indicator of sex and is often the best indicator of sex when

other parts of the skeleton have been damaged, destroyed, or separated from the cranium.

Previous research assumed that discriminant functions for sex determination developed for

one population could easily be used for all populations, with little regard for the skeletal

variation between populations. This work tests the hypothesis that sex determination by

discriminant function analysis of the crania for Florida and Georgia Native American remains









Doran, Glen H.
2002 Windover: Multidisciplinary Investigations of an Early Archaic Florida
Cemetery. University Press of Florida, Gainesville.

Edwards, William Edward
1954 The Helen Blazes Site of Central-eastern Florida: A study in Method Utilizing the
Disciplines of Archaeology, Geology, and Pedology. Unpublished Ph.D.
dissertation, Department of Anthropology, Columbia University, New York.

Espenshade, Christopher T.
1983 Ceramic Ecology and Aboriginal Household Pottery Production at the Gauthier
Site, Florida. Unpublished Master's thesis, Department of Anthropology,
University of Florida, Gainesville.

Fisher, Ronald A.
1936 The Utilization of Multiple Measurements in Taxonomic Problems. Annals of
Eugenics 7:179-188.

Gallagher, John C., and Lyman O. Warren
1975 The Bay Pines Site, Pinellas County, Florida. Florida Anthropologist 28:96-116.

Giles, Eugene and Orville Elliot
1962 Race Identification from Cranial Measurements. Journal of Forensic Science
7:149-184.

1963 Sex Determination by Discriminant Function Analysis of Crania. American
Journal ofPhysical Anthropology 21:53-68.

Giles, Eugene
1964 Sex Determination by Discriminant Function Analysis of the Mandible. American
Journal of Physical Anthropology 22:129-136.

1970 Discriminant Function Sexing of the Human Skeleton. In Personal Identification
in Mass Disasters, edited by Thomas Dale Stewart, pp. 99-107. National Museum
of Natural History, Smithsonian Institution, Washington, D.C.

Griffin, John W.
1949 Notes on the Archaeology of Useppa Island. Florida Anthropologist 2:92-93.

1979 The Origin and Dispersion of American Indians in North America. In The First
Americans: Origins, Affinities, and Adaptations, edited by W.S. Laughlin and
A.B. Harper, pp. 43-56. Gustav Fischer, New York.

Hally, David J., and Robert C. Mainfort, Jr.
2004 Prehistory of the Eastern Interior after 500 B.C. In S.,iniheav\i, edited by Raymond
D. Fogelson, pp. 265-285. Handbook of North American Indians, Vol. 14,
William C. Sturtevant, general editor, Smithsonian Institution, Washington, D.C.









from archaeological contexts is more accurate when the functions are developed using remains

from those populations than functions developed from other populations.

Cranial measurements and sex identification data are collected from skeletal collections

housed at the Florida Museum of Natural History at the University of Florida, and the

Smithsonian Institution's National Museum of Natural History. The sample includes 46

individuals from ten Native American archaeological sites in Florida and Georgia, ranging from

the Middle Archaic Period to the Spanish contact era.

This research finds that existing discriminant function formulas disproportionately

misclassify skeletons from the Florida and Georgia unless the formula is adjusted for that

population. Additionally, existing formulas require measurements that are rarely preserved or

that do not contribute to identifying sex. New discriminant function formulas based on skeletons

from Florida and Georgia are only nominally more accurate than existing formulas, but it is no

more difficult to produce new formulas than to adjust existing formulas. Creating new formulas

also provides the opportunity to select variables that are more often preserved in archaeological

contexts and that also clearly contribute to identifying the sex individuals. This research finds

that existing formulas are reliable only if the sectioning point is adjusted for the study

population.




Full Text

PAGE 1

SEX DETERMINATION BY DISCRIMINANT FUNCTION ANALYSIS OF NATIVE AMERICAN CRANIA FROM FLORIDA AND GEORGIA By MICHAEL BRYAN MCGINNES A THESIS PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF ARTS UNIVERSITY OF FLORIDA 2007

PAGE 2

2 2007 Michael Bryan McGinnes

PAGE 3

3 To my grandmother, Sarah Napps, she loved me for who I was, no matter what. She supported me in in every way she could, no matter w hat. She saw the best in me, and was proud of me, no matter what. I strive to live up to her image of me, and her pride in me.

PAGE 4

4 ACKNOWLEDGMENTS This thesis would not have been possible without the support and encouragement of my supervisory committee, Anthony Falsetti and Thomas Hollinger. I also thank Jerry Milanich, who was a de facto committee member, but was unable to attend the defense. Special thanks go to Elaine Oran for her invaluable insight on the structure and writing of a scientific paper, e ven if it isn't about rocket science. Finally, I am forever grateful to David Hunt, Collections Manager of the Smithsonian Institutions Physical Anthropology collections, for his immeasurable contribution to this research. Collecting the skeletal data wou ld have been impossible without his permission and assistance. He not only tolerated my haphazard schedule and unannounced visits with infectious good humor, he was there when I needed a sounding board, provided stimulating conversation, and answered my ma ny questions. He was always ready to share a grouse or a laugh, whichever was needed. Thanks in part to a shared interest in food, wine, and the finest Australian Tupperware; I am honored to call him my friend. Special thanks are due to the staffs of the Anthropology Departments at the Florida Museum of Natural History (FLMNH) and the National Museum of Natural History, Smithsonian Institution, where the collections examined for this research are housed. Ann Cordell, Elise LeCompte, Diane Kloetzer, and Do nna Ruhl provided friendship, assistance, and knowledge of the collections at the FLMNH. Maggie Dittemore and the rest of the staff at the John Wesley Powell Library of Anthropology at the Smithsonian Institution provided access to printed resources that a re unobtainable elsewhere. My immediate family -Dorothy, Rodney, and Judie McGinnes -was integral to completing this project. They contributed in innumerable ways I cant thank them loudly or strongly enough. I am fortunate to be part of a very larg e and diverse extended family who

PAGE 5

5 offered all sorts of support through the years, including kind words, empathy, sympathy, a roof, a meal, companionship, and even nagging when needed. Barbara and Max Skidmore have been positive role models since childhood. More recently, Elizabeth and Jay Boris, and Dan and Elaine Oran have provided unflagging support I thank them for their extreme generosity, good cheer, and for allowing me to deep fry the Thanksgiving turkey. Last, but not least, thank you to Dr. Ruth T rocolli, my wife, my best friend, an exemplary scholar and outstanding human being. I could not have done it without her love, support, and encouragement. We shared the long road that culminated in this thesis.

PAGE 6

6 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ .......... 4 LIST OF TABLES ................................ ................................ ................................ ...................... 9 LIST OF FIGURES ................................ ................................ ................................ .................. 10 CHAPTER 1 INTRODUCTION ................................ ................................ ................................ ............. 15 2 SEX DETERMINATION ................................ ................................ ................................ .. 18 Introduction ................................ ................................ ................................ ....................... 18 Visual Methods ................................ ................................ ................................ .................. 18 Metric Methods ................................ ................................ ................................ ................. 21 3 DISCRIMINANT FUNCTION ANALYSIS ................................ ................................ ...... 26 Canonical Discriminant Functions ................................ ................................ ..................... 27 Multiple Discri minant Functions ................................ ................................ ........................ 28 Tests of Significance ................................ ................................ ................................ ... 30 Prior Probabilities ................................ ................................ ................................ ....... 30 Stepwise Discriminant Function Analysis ................................ ................................ .......... 30 Mahalanobis Distance ................................ ................................ ................................ ........ 31 Logistic Regression ................................ ................................ ................................ ............ 31 Conclusion ................................ ................................ ................................ ......................... 32 4 ARCHAEOLOGICAL CONTEXT ................................ ................................ .................... 35 Introduction ................................ ................................ ................................ ....................... 35 Environmental Setting ................................ ................................ ................................ ....... 35 Cultural History ................................ ................................ ................................ ................. 36 Paleoindian Period ................................ ................................ ................................ ...... 36 The Archaic Period ................................ ................................ ................................ ..... 38 Early Archaic ................................ ................................ ................................ ....... 38 Middle Archaic ................................ ................................ ................................ .... 39 Late Archaic ................................ ................................ ................................ ........ 40 Woodland and Regional Cultures ................................ ................................ ................ 41 Kellog ................................ ................................ ................................ .................. 42 Deptford ................................ ................................ ................................ .............. 42 St. Johns/Malabar ................................ ................................ ................................ 43 Manasota ................................ ................................ ................................ ............. 44 Wilmington Culture ................................ ................................ ............................. 45 Historic Period ................................ ................................ ................................ ............ 45

PAGE 7

7 5 SITES ................................ ................................ ................................ ................................ 47 Introduction ................................ ................................ ................................ ....................... 47 Golf Course (8Br44) ................................ ................................ ................................ .......... 47 Bay Pines (8Pi64) ................................ ................................ ................................ .............. 48 Canaveral (8Br85) ................................ ................................ ................................ ............. 48 Casey Key (8So17) ................................ ................................ ................................ ............ 50 Palmer Burial Mound (8So2a) ................................ ................................ ........................... 50 Perico Island (8Ma6) ................................ ................................ ................................ ......... 51 St. Simons Island, Georgia ................................ ................................ ................................ 51 Garfield Site (9BR57) ................................ ................................ ................................ ........ 53 6 METHODS AND MATERIALS ................................ ................................ ....................... 55 Introduction.......................................................................................................................55 Sampling....................................................................................................................55 Determination of Sex..................................................................................................55 Materials.....................................................................................................................56 Cranial Measurement Definitions.......................................................................................58 Giles and Elliot Measurements....................................................................................58 Statistical Procedures..................................................................................................62 7 RESULTS ................................ ................................ ................................ ......................... 65 General Results ................................ ................................ ................................ .................. 65 Specific Results ................................ ................................ ................................ ................. 67 Giles and Elliot Discriminant Function ................................ ................................ ....... 67 Variable S election ................................ ................................ ................................ ....... 69 Test for Site Effect on Selected Variables ................................ ................................ ... 71 Discriminant Function Analysis ................................ ................................ .................. 72 Function 1 Accuracy ................................ ................................ ................................ ... 73 Function 2 Accuracy ................................ ................................ ................................ ... 75 8 CONCLUSIONS AND DISCUSSION ................................ ................................ .............. 77 Conclusions ................................ ................................ ................................ ....................... 77 Summery of Statistical Results ................................ ................................ .................... 78 Specific Statistical Results ................................ ................................ .......................... 79 Discussion ................................ ................................ ................................ ......................... 80 Future Research ................................ ................................ ................................ ................. 82 APPENDIX A CRANIAL MEASUREMENT DEFINITIONS ................................ ................................ .. 84 Giles and Elliot Measurements ................................ ................................ ........................... 84 Other Measurements ................................ ................................ ................................ .......... 87

PAGE 8

8 B TABLE OF SITES ................................ ................................ ................................ ............. 92 LIST OF REFERENCES ................................ ................................ ................................ .......... 94 BIOGRAPHICAL SKETCH ................................ ................................ ................................ ... 105

PAGE 9

9 LIST OF TABLES Table page 2 1 Coeff icients for discriminant functions 1,2, and 3 from Giles and Elliot 1963 used to assign sex based on a white sample, a black sample, and a combined black and white sample. ................................ ................................ ................................ .......................... 23 6 1 Measurements used by Giles and Elliot. ................................ ................................ ........ 59 6 2 Measurements not used by Giles and Elliot ................................ ................................ ... 60 7 1 Accuracy of the Giles and Elliot Function 3 for sex determination on the study sample of Florida and Georgia Native Americans. ................................ ......................... 68 7 2 T scores and p values for a difference in mean values of each cranial variable between males and females. ................................ ................................ .......................... 70 7 3 F values and p values for the hypothesis of no site effect for each variable ................... 71 7 4 MANOVA test criteria and F approximations for the hypothesis of no overall site effect ................................ ................................ ................................ ............................. 72 7 5 Coefficients, group means, and sectioning points for Function 1 and Function 2. ........... 73 7 6 Accura cy of Function 1 ................................ ................................ ................................ 74 7 7 Accuracy of Function 2 ................................ ................................ ................................ 75

PAGE 10

10 LIST OF FIGURES Figure page 3 1 Discriminant Function Analysis with three groups using the graphic de vice proposed by Rao ................................ ................................ ................................ ........................... 29 5 1 Locator map of sites used in this study and major rivers. ................................ ............... 54

PAGE 11

11 LIST OF ABBREVIATIONS AUB, au au Biauricular breadth B.P. Radiocarbon years before present, with present defined as the year 1950. BBH, ba b Basion bregma height BNL, ba n Cranial base length BPL, ba pr Basion prosthion length DFA Discriminant function analysis. DKB, d d Interorbital breadth DKB, d d Interorbital breadth EKB, ec ec Biorbital breadth EKB, ec ec Biorbital breadth FOB Foramen magnum breadth FOL, ba o Foramen magnum length FRC, n b Frontal chord GOL, g op Maximum cranial length MAB, ecm ecm Maxillo alveolar breadth external palate breadth MAL, pr alv Maxillo alveolar length, external palate length MDHA Average mastoid height MDHL Left mastoid height MDHR Right mastoid height MNI Minimum Number of Individuals NLB, al al Nasal breadth NLH, n ns Nasal height OBB, d ec Orbital breadth OBH Orbital height

PAGE 12

12 OCC, l o Occipital chord p P value, th e probability of getting a value at least as extreme as the observed value by chance alone. PAC, b l Parietal chord UFBR, fmt fmt Upper facial breadth UFHT, n pr Upper facial height WFB, ft ft Minimum frontal breadth XCB, eu eu Maximum cranial breadth Z YB, zy zy Bizygomatic breadth

PAGE 13

13 Abstract of Thesis Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Master of Arts SEX DETERMINATION BY DISCRIMINANT FUNCTION ANALYS IS OF NATIVE AMERICAN CRANIA FROM FLORIDA AND GEORGIA By Michael Bryan McGinnes December 2007 Chair: Anthony Falsetti Major: Anthropology The goal of this research is to determine if the accuracy of discriminant function analysis for sex determination co uld be improved by using local or regional populations, and by better variable selection. In archaeological contexts, skeletons are usually all that remains of the actual people who once lived there. Information about the sex of those individuals is fundam ental to the study of demographics, sex roles, and life ways of past cultures. Determining an individuals sex is also fundamental to personal identification of skeletal remains from modern contexts, whether from a mass disaster or an unmarked grave. There are several techniques available for determining sex from skeletal remains, and each technique has its place. Discriminant function analysis is valuable in sex determination because it requires relatively little training to use effectively and it serves a s an objective check to other methods. The cranium is a reliable indicator of sex and is often the best indicator of sex when other parts of the skeleton have been damaged, destroyed, or separated from the cranium. Previous research assumed that discrimin ant functions for sex determination developed for one population could easily be used for all populations, with little regard for the skeletal variation between populations. This work tests the hypothesis that sex determination by discriminant function ana lysis of the crania for Florida and Georgia Native American remains

PAGE 14

14 from archaeological contexts is more accurate when the functions are developed using remains from those populations than functions developed from other populations. Cranial measurements an d sex identification data are collected from skeletal collections housed at the Florida Museum of Natural History at the University of Florida, and the Smithsonian Institutions National Museum of Natural History. The sample includes 46 individuals from te n Native American archaeological sites in Florida and Georgia, ranging from the Middle Archaic Period to the Spanish contact era. This research finds that existing discriminant function formulas disproportionately misclassify skeletons from the Florida and Georgia unless the formula is adjusted for that population. Additionally, existing formulas require measurements that are rarely preserved or that do not contribute to identifying sex. New discriminant function formulas based on skeletons from Florida and Georgia are only nominally more accurate than existing formulas, but it is no more difficult to produce new formulas than to adjust existing formulas. Creating new formulas also provides the opportunity to select variables that are more often preserved in archaeological contexts and that also clearly contribute to identifying the sex individuals. This research finds that existing formulas are reliable only if the sectioning point is adjusted for the study population.

PAGE 15

15 CHAPTER 1 INTRODUCTION Sex identificati on has long been an important part of skeletal analysis in the archaeological setting. Accurate sex estimations are basic to studies of past adaptations of humans to new environments and demographic histories. Archaeologists may use sex information to esta blish demographic patterns, study the affect of sex on status within a particular archaeological culture, or to examine migration patterns ( Buikstra and Ubelaker 1994:15). Additionally, forensic anthropologists may use this information to help identify an individual in a medico legal context. Methods for sex identification have evolved from visual methods to include metric methods based on univariate and multivariate statistical analysis, particularly discriminant function analysis. Using visual methods to determine sex, an osteologist may examine the overall size of the subject, the size of the mastoid process, the shape of the frontal bone, or the gonial angle of the mandible. The determination of sex using visual method relies primarily on the experience and judgment of the osteologist. Using these methods, osteologists typically achieve an accuracy of 80% to 90% (Giles and Elliot 1962). Metric methods use various measurements of the skeleton, many of which attempt to capture aspects used in visual methods (David Hunt, personal communication 2006). In univariate analysis, a single measurement of an individual is compared to the distribution of measurements from a sample of known sex specimens to arrive at the likely sex of the individual. Using only one mea surement, however, does not account for differences in shape of a skeletal element between males and females, and for any single measurement there is considerable overlap between the range of variation for males and females. Multivariate methods use multip le measurements which can capture the shape of a skeletal element and minimize the amount of overlap between males and females. Discriminant function analysis is a multivariate method designed for this problem.

PAGE 16

16 Discriminant function analysis was developed to solve the problem of predicting group membership based one or more interval variables. The goal of discriminant function analysis is to minimize misclassification by maximizing between group differences, or minimizing group overlap. Discriminant functio n analysis is also used to determine which variables best discriminate between groups, and in what ways groups differ. How well a function performs is usually reported in terms of how many cases would be correctly assigned to their groups using the discrim inant functions (Manly 1994). One widely used discriminant function for sex determination is calculated by Giles and Elliot (1963) based on data collected from the Terry and Hamann Todd skeletal collections. The Terry and Hamann Todd collections are compri sed of skeletal remains collected from cadavers used by medical school anatomy classes during the late nineteenth and mid twentieth centuries (Hunt and Albanese 2005). The problems presented by a medical school cadaver sample include possible effects of th e inherent socio economic bias on skeletal morphology. The Terry and Hamann Todd skeletal collections are biased, with older individuals and males overrepresented compared the population as a whole, and they do not include Native Americans. For studying th e sex differences in skeletons, however, such collections are essential because the sex of each individual is positively known from written records (Giles and Elliot 1963:56). The sample Giles and Elliot used to calculate their discriminant functions did n ot include any native Americans, but the formulas are tested on three series of American Indian crania from Indian Knoll (N>= 344), Pecos Pueblo (N>=110), and Florida (N>=217). These materials are analyzed by Snow (1948; Johnson and Snow 1961), Hooton (193 0), and Hrdli ka (1940) respectively. On the whole the discriminant functions described [by Giles and Elliot 1963], assign the correct sex, assuming that the original estimations are correct, with the same order of magnitude as they do

PAGE 17

17 for the black and w hite sample. For the Florida Indians, however, this is true only when the sectioning point is based on mean values of these same Indians. (Giles and Elliot 1963:66 67). Can the accuracy of sex determination by discriminant function analysis be improved ov er Giles and Elliot's results by deriving new formulas based on regional populations? This hypothesis is tested using skeletal remains recovered from archaeological sites in Florida and Georgia.

PAGE 18

18 CHAPTER 2 SEX DETERMINATION Introduction This chapter desc ribes the various visual, univariate, and multivariate metric methods used in sex determination using cranial and post cranial skeletal remains. Estimating sex from a skeleton relies on sexual dimorphism, the morphological differences between men and women Men tend posses a larger body size when compared to females. Female morphology must allow for both bipedal locomotion and giving birth to relatively large headed babies when compared to other primates. With a complete adult skeleton, and particularly a c omplete pelvis, a physical anthropologist should be able to correctly assign sex with nearly perfect accuracy. Additionally, the anthropologist should be able to recognize ambiguous cases where sex identification is less certain. If the complete skeleton i s not available, accuracy depends largely on what bony elements are available and if the skeleton can be linked to a specific population. If the skeleton is fragmented or from a sub adult, then determining the sex is more difficult and less reliable than w ith a complete adult skeleton. Methods of sex determination are either visual or metric, and apply to the crania and post cranial skeleton. For the best results, the forensic anthropologist should use all available data. Visual Methods Visual methods for e stimating sex from the skeleton make use of size differences between men and women or morphological differences related to childbirth in women. Sexing methods that rely on morphological differences in the pelvis related to childbirth are the most accurate. The method for sexing the skeleton by the pubic bone developed by Phenice (1969) is the most accurate method known for determining the sex of an individual from the skeleton (White 1991).

PAGE 19

19 Phenice (1969) identifies three indicators of sex in the pubic bone : the ventral arch, the subpubic concavity, and the medial aspect of the ischiopubic ramus. The ventral arch is a slightly elevated ridge of bone that sweeps inferiorly and laterally across the ventral surface of the pubis, merging with medial border of t he ischiopubic ramus. The ventral arch is evaluated by orienting the pubis so that its rough ventral surface faces the observer, who looks down along the plane of the pubic symphysis surface. The ventral arch, when present, sets off the inferior, medial co rner of the pubic bone in ventral view. The ventral arch is present only in females. Male pubic bones may have elevated ridges in this area, but these do not take the wide, evenly arching path of the female's ventral arch or set off the lower medial quadra nt of the pubis. The subpubic concavity is a concave curve on the medial edge of the ischiopubic ramus displayed in female os coxae The female ischiopubic ramus is concave, while male edges are straight or very slightly concave. The subpubic concavity is evaluated by turning the pubis over, orienting it so that its smooth, convex dorsal surface faces the observer, who is once again sighting along the midline. From this position it is possible to observe the medial edge of the ischiopubic ramus. For females the edge of the ramus is concave in this view. Males do not show the dramatic concavity here. Male edges are straight or very slightly concave. (If the bone is in good shape, and there is no danger of damage, another method for evaluating the subpubic co ncavity is to lay the ischiopubic ramus on a flat surface. If it can be rocked, it indicates a male, if it cannot be rocked, it indicates a female). To evaluate the medial aspect of the ischiopubic ramus, the observer turns the pubis 90, orienting the sym physis surface so that the observer is looking directly perpendicular to it. From this position it is possible to observe the ischiopubic ramus in the region immediately inferior to

PAGE 20

20 the symphysis. This medial aspect of the ischiopubic ramus displays a shar p edge in females. In males the surface is fairly flat, broad, and blunt. In the Phenice method, some criteria may not obviously sex the specimen, so those criteria should be discarded. If there is some ambiguity concerning one or two of the criteria, the re is usually one of the remaining criteria that clearly indicate the subjects sex. Accuracy of sexing based on this method ranges from 96 to 100% (White 1991:325). Because of its position in childbirth, the pelvis includes many other characteristics that can be used to visually estimate sex. Compared to the male, the female pelvis is broader, has a wider sciatic notch, and normally includes a pre auricular sulcus (a groove between the auricular area and the sciatic notch). It has a smaller acetabulum (the socket that holds the head of the femur), a longer pubic bone, and a wider subpubic angle. The sacrum is shorter and broader, and the obturator foramen smaller and triangular in females. Compared to females, the male pelvis may be heavier and more robust, and the auricular area tends to be flatter. The pre auricular sulcus seldom occurs in males, but if present in males, it is shallower than in females. The obturator foramen is larger and ovoid in males. Evaluation of these criteria individually yields acc uracies from 83 to 94%. In combination, accuracies range from 95 to 98% (Rogers and Saunders 1994:1050 1051). After the pelvis, the next best indicator of sex is the cranium. Estimation of sex is based on the generalization that the male is more robust and has muscle attachment points that are larger and rougher. Male muscle attachment points are especially pronounced on the occipital bone, where they may form a nuchal crest. Males also have larger mastoid processes, more prominent supraorbital ridges, and the posterior end of the zygomatic process extends farther as a crest. The

PAGE 21

21 upper edges of the eye orbits are blunt, and frontal sinuses are larger in males. On the mandible, the male chin is squarer, and the gonial angle is more acute. Compared to the male the female cranium is smaller, smoother and more gracile. In the female mandible, the chin is more rounded and pointed, and the gonial angle is more oblique. The smaller size of the female cranium is evident in a smaller palate, and smaller teeth. Female s also display frontal and parietal bossing into adulthood; the upper edges of the eye orbits are sharp. Evaluation of these characteristics depends not only on the experience of the osteologist, but also on matching the specimen to a genetically and tempo rally close comparative population (Bass 1995; White 1991). Using the crania, an experienced osteologist should be able to make a sex determination that is 80 90% accurate. Buikstra and Ubelaker (1994:16 20) provide a scoring system for several of these vi sual traits. For the remainder of the skeleton, males tend to be larger than females, with long bones that are longer, heavier, and have larger attachment areas for muscles, including the linea aspera, crests, tuberosities and impressions (Brothwell 1981, Stewart 1948). These criteria are useful if a related skeletal series is available, but for isolated or fragmentary remains this is only useful if the bone is at the extreme end of the range, either 'very male' or 'very female.' Metric Methods While visua l methods can estimate sex quickly and accurately, their evaluation is subjective and requires experience with sexing techniques and the relevant population. Metric procedures are based on quantifying the same criteria used in visual sexing. A metric proce dure could be better if the observer is not familiar with visual techniques or the relevant population. Additionally, metric procedures serve as an objective check to visual methods and can strengthen the position of the osteologist as expert witness in a courtroom (Stewart 1979). The simplest

PAGE 22

22 metric methods use a single measurement, and compare it to a distribution of that measurement from a collection of known sex individuals. A sectioning point is placed such that males and females are equally likely to be classified correctly, and misclassifications are minimized. The sectioning points are usually arrived at by discriminant function analysis. Sectioning points for several different long bones are given by Bass (1995), and an exhaustive list of discrimina nt function studies to determine sex and their accuracies can be found in Rathbun and Buikstra (1984:212 216). There are a host of discriminant function tests based on cranial and post cranial measurements. Giles and Elliot (1963) provide one of the best e stablished discriminant functions for sex determination using the skull. From combinations of nine cranial measurements a total of 21 discriminant functions are described to indicate sex in whites, blacks and whites and blacks taken together. The measureme nts are: Glabello occipital length : The maximum length of the skull, from the most anterior point of the frontal in the midline to the most distant point on the occiput in the midline. Maximum width : The greatest breadth of the cranium perpendicular to the median sagittal plane, avoiding the supra mastoid crest. Basion bregma height : Cranial height measured from basion to bregma. Maximum diameter bi zygomatic : The maximum width between the lateral surfaces of the zygomatic arches measured perpendicular to t he median sagittal plane. Basion nasion : The direct distance from basion to nasion. Basion prosthion : The direct distance from basion to the most anterior point on the maxilla in the median sagittal plane. Nasion breadth : The maximum breadth of the nasal a perture perpendicular to nasal height. Palate external breadth : The maximum breadth of the palate taken on the outside of the alveolar borders. Opisthion forehead length : The maximum distance from opisthion (the midpoint on the posterior border of the fora men magnum) to the forehead in the midline. Mastoid length : The length of the mastoid measured perpendicular to the plane determined by the lower borders of the orbits and the upper borders of the auditory meat uses (Frankfort plane). Functions 1, 2, and 3 use 8 of the 9 measurements, are the most accurate for each group, and use the same measurements from each group. The sectioning point is halfway between the

PAGE 23

23 mean score for males and the mean score for females. A score above the sectioning point is desig nated male; one below is designated female. Table 2 1. Coefficients for discriminant functions 1,2, and 3 from Giles and Elliot 1963 used to assign sex based on a white sample, a black sample, and a combined black and white sample. Measurements Whites (F unction 1) Blacks (Function 2) Combined (Function 3) Glabello Occipital Length (GOL) 3.107 9.222 6.083 Maximum Width (XCB) 4.643 7.000 1.000 Basion Bregma height (BBH) 5.786 1.000 9.500 Max Diameter Bi zygomatic (ZYB) 14.821 31.111 28.250 Basion P rosthion (BPL) 1.000 5.889 2.250 Prosthion Nasion Height (UFHT) 2.714 20.222 9.917 Palate -External breadth (MAB) 5.179 30.556 19.167 Mastoid Length (MDH) 6.071 47.111 25.417 Sectioning Points 2676.39 8171.53 6237.95 Male mean 2779.66 8487.56 6466. 17 Female Mean 2573.12 7855.50 6009.72 Sample Accuracy (Percent correct) 86.1% 84.6% 86.0% Expected Accuracy 86.6% 87.6% 86.4% A large variety of metric methods using the post cranial skeleton are available in Krogman and Iscan (1986) and Bass (1995). These are commonly used in forensic anthropology. Several methods were developed for fragmentary remains that focus on bioarchaeology collections. Using Bioarchaeology collections has the advantage of using the population of interest to develop criteria, but in most cases the true sex of the individuals in the study is unknown. Therefore, true accuracy cannot be determined, only how consistent the method is with other, established methods. Several studies have been focused on identifying sex from fragmenta ry remains. Using a discriminant function based on only the midshaft femoral circumference of prehistoric skeletons from Ohio, Black (1978) recorded an accuracy of 85%. Using the same measurement, DiBennardo and Taylor (1979) developed and tested discrimin ant functions on black and white femura of known sex and achieved an accuracy of 82%. Using various combinations of three

PAGE 24

24 measurements, Taylor and DiBennardo (1982) recorded accuracies between 80 and 85% for white femora. Dittrick and Suchey (1986) minimiz e the problem of unknown sex by using only skeletons with the pubic bone and applying the Phenice method. Suchey had established an accuracy of 99% in sex determination by using a blind test on pubic bone pairs from modern autopsies of individuals over age 16. In tests on prehistoric central California skeletal remains, they achieved accuracies of about 90% using linear discriminant function analysis of measurements from the ends of the long bones. Interestingly, functions based on multiple measurements did not produce results much better than the best functions using single measurements. One of the limitations of discriminant function analysis is that the functions should only be used on skeletons that come from the same population as the one used in develo pment of the function. Otherwise, results can be unpredictable. Giles and Elliot (1963) tested their own functions on skeletons from Ireland and on three series of native American skeletons from Indian Knoll in Ohio, Pecos Pueblo, and Florida. The function s correctly sexed 40 of 42 males (95%) and 3 of 8 females (37.5%) of the Irish skeletons. Giles and Elliot dismiss the poor female result and take this as evidence that their formula can be used across populations. A better interpretation is that the formu la disproportionately misclassifies females as male. For the Native American samples, good results are achieved only after altering the sectioning point. Although it has been used in the literature, altering sectioning points is not a practical solution to using discriminant functions across populations (Calcagno 1981; Henke 1977). Kajonoja (1966) found that the Giles and Elliot functions had an accuracy of only 65% on Finnish crania. It is interesting to note that the functions developed by Giles and Ellio t (1963) for the combined sample of blacks and whites worked about as well on blacks and whites separately as it did on

PAGE 25

25 the combination of both groups. This suggests that functions developed on a wider population can achieve good results across that popula tion's constituent groups. (See Henke 1977 for a more detailed discussion of using discriminant function analysis across populations.) The methods and criteria discussed above are only applicable to adult remains. A skeleton may be considered an adult if a ll long bone epiphyses are fused or if the third molars have erupted. For sub adults, if age can be established from dental development, then sex can be estimated from long bone lengths (see Bass 1995). Other sexing techniques for sub adults use sex differ ences in pelvic measurements (see Krogman and Iscan 1986:200 208).

PAGE 26

26 CHAPTER 3 DISCRIMINANT FUNCTIO N ANALYSIS Discriminant function analysis (DFA) has been used extensively to determine sex by archaeologists and forensic anthropologists (e.g. Giles and E lliot 1963; Black 1978; DiBennardo and Taylor 1979, 1982, 1983; Dittrick and Suchey 1986). The results are comparable to those of traditional methods, but requires far less training and experience. This section describes the goals and capabilities of discr iminant function analysis, some of the methods for calculating discriminant functions, and introduces research using discriminant function analysis in sex determination. Discriminant function analysis addresses the problem of how well it is possible to se parate two or more groups of individuals using multiple combinations of weighted variables. DFA requires classes that are predetermined, such as male or female. The object is not to create classes or populations that divide heterogeneous material. With two groups, there are two specific errors one can make: mistaking a member of one group for being from the other. For example, misclassifying (1) a male as a female, or (2) a female as a male. Both types of mistakes should occur at an equal rate, and there sh ould be as few mistakes as possible. Finally, each subject must be assigned to one population or the other so that Unknown is not an option (Kendall 1957). There are several approaches, including Canonical discriminant functions, Mahalanobis distances an d logistic regression. These methods are described below, followed by the advantages and disadvantages of DFA and by a literature review of DFA used in sex determination.

PAGE 27

27 Canonical Discriminant Functions Canonical discriminant functions determine a combin ation of variables that separate the groups as well as possible. Fisher (1936) introduced a simple way to choose the coefficients for a linear function that maximizes the F ratio of a one way analysis of variance for two groups, which is the ratio of betwe en group variance to within group variance. Specifically, his paper deals with the discrimination of Iris setosa and Iris versicolor 1 two species found growing together in the same colony. His variables are measurements of sepal length, sepal width, petal length, and petal width. Most of the literature cites Fisher as the originator of discriminant function analysis, but he cites "Mr. E.S. Martin" and "Miss Mildred Barnard" for applying the principle to sex differences in the mandible and a secular trend i n cranial measurements, respectively. The approach involves finding coefficients for a linear combination of n variables: where Z is the discriminant function score, a is a constant, through are discriminant function coefficients, and through are independent variables, which maximizes the F ratio of Z in a one way analysis of variance for the two groups. Finding the coefficients of the can onical discriminant functions is an eigenvalue problem. Details on the computation of DFA coefficients can be found in Manly (1994). Discriminant function analysis has uses other than classifying individuals. Discriminant function coefficients can also be used to evaluate how groups differ (Manly 1994:114). Howells (1989) used discriminant functions as a form of data reduction, similar to the way others have used principle components analysis, but that use is not common in the literature. 1 The article is primarily concerned with the discrimination of I setosa and I versicolor but Fisher extends his analysis to test the hypothesis that I virginica is a hybrid of I setosa and I versic olor.

PAGE 28

28 Among the most wid ely cited uses of canonical discriminant function analysis in anthropology come from Giles and Elliot, who used the technique for the estimation of sex (1963; Giles 1964) and race (Giles and Elliot 1962). For these studies, Giles and Elliot used measuremen ts of Native American remains from the Indian Knoll, KY site originally published by Snow (1948), while black and white subjects came from the Terry and Todd Collections. Their use of discriminant functions for estimation of sex from the crania is discusse d in detail in the chapter on sex estimation. In their calculations, Giles and Elliot used formula from Kendall (1957), which are equivalent to those presented by Fisher (1936). Using Black and White individuals from the same sample, Giles and Elliot (1963 ) developed discriminant functions for sex using nine cranial measurements in different combinations to form 21 discriminant functions for sex determination. An accuracy of 82 89% is attained with the Black and White material. This compares favorable with the 77 87% accuracy expected from visual sex estimation using the cranium alone. Multiple Discriminant Functions When canonical discriminant analysis is used, it may be possible to determine several linear combinations of variables for separating groups wh ere there are multiple groups and variables. The number of functions available is either the number of variables or one less than the number of groups, which ever is smaller. All functions maximize the F ratio subject to the condition that they are uncorre lated with previous functions within groups. The canonical discriminant functions are, therefore, linear combinations of the original variables chosen such that the first function reflects group differences as much as possible, and subsequent functions cap ture as much as possible of group differences not displayed by previous functions. Group assignment is then accomplished by calculating the distances to group means (Manly 1994:108

PAGE 29

29 110). While this method is useful for analyzing group differences, it is co mputationally difficult for the end user to assign individuals to groups. Figure 3 1 Discriminant Function Analysis with three groups using the graphic device proposed by Rao. Function 1 is plotted on the X axis, Function 2 is plotted on the Y axis. Thi n rods and are placed to minimize errors (Adapted from Rao 1952; Giles and Elliot 1962). A different method is used by Giles and Elliot (1962) to estimate race. Giles and Elliot (196 2) use a pair of canonical discriminant function formulas for the placement of a skull into white, black, or Native American categories: One formula for black vs. white, and the other for white vs. Native American. The two functions are then plotted, with the black white function on one axis, and the white and Native American function on the other. To place individuals into one of the three groups, Giles and Elliot use a geometrical device described by Rao (1952:327),

PAGE 30

30 which is not common in the literature ( see fig. 3 1; Giles and Elliot 1962). The Giles and Elliot functions correctly identified race for 82.6% of males 88.1% of females (Giles and Elliot 1962). Tests of Significance Various tests of significance are available for evaluating the difference bet ween the mean values for any pair of groups, overall differences between the means of several groups, and if the mean of a discriminant function differs from group to group. See Harris (1985) for a discussion of the difficulties surrounding these tests. Pr ior Probabilities Some computer programs can allow for prior probabilities of group membership. This could be useful in sex determination if reliable data can be generated for the proportion of male and female skeletons recovered from a particular environm ent, whether due to taphonomy, demographics, or other causes. Care should be taken that assignment of prior probabilities reflects actual or known proportions (e.g. number of males and females in a population) and not any form of prior bias. Stepwise Disc riminant Function Analysis Stepwise discriminant function analysis simply applies a stepwise selection to the variables included in a discriminant function analysis. Variables are added to the discriminant functions, one at a time, until adding additional variables does not give significantly better discrimination. Several authors have used stepwise discriminant function analysis in estimating sex from skeletal measurements. Holman and Bennett (1991) use the procedure built into SAS (STEPDISC) on the bones of the arm and wrist with good results. Taylor and DiBennardo (1982), DiBennardo and Taylor (1983), and Iscan and Miller Shaivitz (1984) each use the procedure built into SPSS for the femur, femur and pelvis, and tibia, respectively, with good results.

PAGE 31

31 Mah alanobis Distance The Mahalanobis distance ( ) is a measure of distance that takes into account correlations between variables. For classification purposes, Mahalanobis distance can be used to measure the distance of individuals to gr oup centers and each individual can be allocated to the group to which it is closest. If are the values of variables for the individual, with corresponding population mean values of t hen: where is the element in the r th row and s th column of the inverse of the covariance matrix for the p variables (Manly 1994:63). Assumptions Both canonical and Malahanobis distance methods are based on two assumptions. The first is that the within group covariance matrix is the same for all groups. The second is that the data is normally distributed within groups. The second assumption is important for the validity of tests of significance. Logistic Regress ion A different approach to discrimination between two groups uses logistic regression. Logistic regression is a variation of multiple linear regression where the dependent variable is assigned as either 1 or 0, usually used as 'success' or 'failure.' The regression formula then returns a probability of success or failure. Rather than representing success or failure, 1 and 0 can be used to represent groups. When applied to unknown individuals, the regression will return the posterior probability of group me mbership, with a sectioning point of 0.5. Konigsberg and Hens (1998) use logistic regression in sex estimation on measurements of the crania. They reported

PAGE 32

32 good results, but found the method cumbersome. They preferred a probit model that used categorical independent variables. Conclusion On the whole, the accuracy of DFA in identifying sex is no more accurate than visual methods when used by a trained osteologist. The advantage is that someone who is not an osteologist, with brief instructions in measurin g, can perform discriminant function sexing quickly and objectively. This group includes medical examiners and archeologists who have had some training in osteology, but who may be inexperienced or rusty (Krogman 1962; Birkby1966; Giles 1970). For the tr ained osteologist and forensic anthropologist, discriminant function analysis can be used as an objective check to visual methods and adds weight to expert testimony (Snow 1979). Discriminant function analysis has two critical limitations when used for sex estimation of skeletal remains. The first is the need for all measurements used in the function to be observable. The second is that the functions can only be used on individuals who come from the population from which the function was developed. The firs t problem is fairly straightforward, as is its solution. A discriminant function score cannot be calculated if an observation is missing. One solution is to generate multiple functions using different combinations of measurements. Another solution is to ge nerate functions using skeletal elements that are robust, and therefore likely to be preserved. This approach was taken by Black [femur] (1978); DiBennardo and Taylor [femur] (1982) ; Taylor and DiBennardo [stepwise femur] (1982); Iscan and Miller Shaivitz [tibia] (1984); Dittrick and Suchey [femur and humerus] (1986). The problem of using functions across populations is more difficult. Some authors have suggested moving the sectioning point by various methods, including using the mid point

PAGE 33

33 between the mean s of the sexes (Giles and Elliot 1963), finding the sectioning point graphically from the bimodal distribution of the scores (Giles and Elliot 1963), or placing the sectioning point at the grand mean of the function scores (Henke 1977). Henke (1977) tested each of these methods, and also used the unmodified functions. Henke concluded that only the first two methods are practical. Calcagno (1981) found that moving the sectioning point by any method was not a practical solution to the problem. Placing the sec tioning point mid way between the means of the sexes requires that one first know the sex of the skeletons. The graphic method is not practicle because the distribution of discriminant scores is multimodal, and not bimodal. Finally, placing the sectioning point at the grand mean of the discriminant scores assumes that the sexes are equally represented in the sample, which is particularly unlikely to be true in archaeological series (Henke 1977). Henke's conclusion seems to be that discriminant functions dev eloped for one group can be used on another, but the sectioning point has to be adjusted. Birkby (1966) tested the Discriminant Functions developed by Giles and Elliot for race and sex (1962; 1963). His goal was to determine (1) if discriminant function an alysis is applicable in the assessment of race and sex in human identification and (2) the reliability of such techniques. He found Indian crania are often misclassified for race and sex. Birkby concludes that the Indian Knoll sample used by Giles and Elli ot (1962, 1963) is not representative of Native Americans as a whole. Therefore, the functions based on those data are not applicable in the identification of race and sex in human identification, either forensic or archaeological. Snow et al. (1979) perf ormed another test of the Giles and Elliot discriminant functions using forensic cases. The discriminant function for sex determination attained an accuracy that was not significantly different from Giles and Elliot. With no significant difference in accur acy between the sexes.

PAGE 34

34 For race, Native Americans are misclassified at a significantly higher rate that non Native Americans (83% correct for black and white combined vs. 14% correct for Native Americans). Snow et al. conclude that the Giles and Elliot fu nctions provide a useful tool for the determination of sex and race of unidentified crania submitted for forensic science examination. The functions, however, did not perform well among Native American subjects. It thus appears that the 5000 year old In dian Knoll crania used by Giles and Elliot in developing their functions do not adequately represent the entire U.S. category of Indian (Snow et al 1979:459). The advantages of discriminant function analysis in physical anthropology are that it is relativ ely easy to apply, allowing sex and race estimations by those with little training in osteology, and that it is an objective indicator or race and sex, especially for isolated remains. The weaknesses include needing to develop functions on the populations from which the subject comes, and the need to be able to make all of the measurements called for in the discriminant function, which is not always possible in fragmentary remains typically found in archaeological sites. Functions developed with fragmentary remains in mind help avoid the problem of missing measurements.

PAGE 35

35 CHAPTER 4 ARCHAEOLOGICAL CONTE XT Introduction This chapter provides a general cultural history of the southeast, introduces the archaeological sites that provide the skeletal materials for this study. It places those sites geographically, temporally and within the framework of the area's cultural history. Environmental Setting Human occupation in what is now Florida and Georgia began about 13,000 B.P. 2 during the end of the Pleistocene epoc h. From the start of human occupation until about 7000 B.P., Florida and Georgia were undergoing tremendous environmental change. The environment gradually went from cold and dry at the end of the Pleistocene, to the warm and humid modern climate. That cha nge in temperature was accompanied by higher sea levels, which have reduced Florida to half the land area of what it was when people first entered the state. Along with a warming climate and rising sea levels, many of the plants and animals in Florida and Georgia were supplanted by species better adapted to the changing environment. By 7000 B.P. sea levels reached about the levels where they are today, and the flora and fauna present were essentially the same species present in the region today (Milanich 19 94). The coastal zone of the region, including the Gulf of Mexico and Atlantic coast of Florida and southern Georgia, is characterized by a generally inhospitable beach and foreshore. This area is subject to seasonal exposure to storms and scarce resources Chains of barrier islands are also found along the eastern coastline of Florida and Georgia containing beach and dune landscapes, 2 B.P. stands for years before present based on radiocarbon dating, with present defined as the year 1950. It is an alternative to traditional dates of A.D. and B.C.. B.P. dates can be roughly converted to traditional dates by subtracting 1950, but r adiocarbon years are not equivalent to calendar years. For this reason, the dating framework used in the primary reports for each site and region are followed without attempting to impose imprecise conversions.

PAGE 36

36 Live Oak hammock, tidal flats, and estuary habitats. Farther inland, the sandy beaches and dunes give way to broad estuaries mudflats, and thickly vegetated shores. These areas serve as place for marine and inland species to interact (Milanich and Fairbanks 1978; Williams 2004:13). Inland areas of northern Florida and Georgia with an abundance of fresh water, such as along str eams and lakes, are dominated by hardwood hammocks. Hardwood hammocks are dense forests characterized by a broad spectrum of plant communities that provide shelter and food for a vast array of animals (Milanich 1998; Milanich and Fairbanks 1978; Wallace 19 78). Cultural History Prehistoric occupation of the southeastern United States, or Southeast, is divided into periods and sub periods that reflect changes in technology, environment, and subsistence. Breaks between periods are often subtle, and changes occ ur at different times across the region. Archaeological cultures describe specific regional or local units from specific time periods. Periods include Paleoindian, Archaic, Woodland, and Mississippian. The Archaic, Woodland, and Mississippian periods are d ivided into Early, Middle, and Late sub periods. The Woodland and Mississippian cultures are comprised of a number of regional cultures. Paleoindian Period The initial period of human occupation in North American is termed the Paleoindian period, and is ch aracterized by the occurrence of fluted stone projectile points or knives, such as Clovis, Suwannee, and Simpson points. The best available evidence suggests that during the late Pleistocene bands of highly moble hunters crossed the Bering land bridge from Siberia into Alaska. The Bering land bridge was exposed during the last glaciation when large amounts of water were locked up in the polar ice caps and ice sheets, resulting in the lowering of global sea levels by about 100 meters. The presence of mid con tinent ice sheets appear to have prevented movement of these populations eastward from Alaska until about 13,000 B.C. based on the

PAGE 37

37 distribution of fluted points in the archaeological record. Most of North America appears to have been occupied by 10,000 B.C (Bense 1994: 38 39; Milanich 1994: 37 40). The general consensus is that the Paleoindian life ways were based on the hunting of large game animals and the gathering of a variety of plant foods. Paleoindians were highly mobile, and were probably organize d into bands with widely ranging patters of movement and inter group interaction (Doran 2002:49,50; Bense 1994;Griffin 1979:51). Because of their mobility, the traditional expectation has been that they should exhibit a basic biological similarity over a w ide geographical area (Key 1983:8; Meiklejohn 1972). Preliminary results from Ross, Ubelaker and Falsetti (2002), however, indicate that Native Americans are much more biologically heterogeneous than previously thought. During the Paleoindian period, the c limate was cooler and dryer than today, but with reduced seasonal variation. Sea levels were as much as 100 meters below current levels (Clausen et al 1979; Widmer 1988). In Florida, this resulted in more land being exposed, lower water tables and few sour ces of surface water. This scarcity of surface water is thought to be a determining factor in Paleoindians settlement patters in Florida. It is also thought that many sites that were along the coast during the Paleoindian period have been inundated by risi ng sea levels (Bullen 1958; Rupp 1980). Most Paleoindian sites found to date consist of little more than limited scatters of lithic debitage ( Edwards 1954; Waller 1969). The human skeletal remains of fewer than 100 individuals from the Paleoindian period have been recovered from all of North America. In Florida, Paleoindian sites with human remains include Little Salt Spring, Warm Mineral Springs, and Cutler Ridge, although the material at Little Salt Spring cannot be incontrovertibly placed within the Pal eoindian occupation (Doran 2002).

PAGE 38

38 The Archaic Period The Archaic Period describes is differentiated from the earlier Paleoindian period on the basis of stylistic differences in point types, the appearance of new artifacts types, and apparent changes in ec onomic orientation (Anderson and Sassaman 2004). T he change in material culture from Paleoindian to Archaic coincided with a change to a warmer, less arid climate. As the temperature increased, the glaciers retreated and sea levels rose. Because of these e nvironmental changes many species that had previously thrived in the Southeast went extinct or disappeared from the region. Changes in flora and fauna led to changes in subsistence patterns and material culture. The regional chronology for the Archaic peri od in the Southeast was established by correlating changes in dagnotic artifacts from excavations at deeply stratified sites such as Ice House Bottoms in Tennessee, Russell Cave in Alabama, Indian Knoll in Kentucky, and the Hardaway and Doerschuk sites in North Carolina (Stoltman 2004). The Archaic is traditionally divided into Early, Middle, and Late Archaic phases. Early Archaic The Early Archaic can be seen as a transition from the Paleoindian period to the Middle Archaic. Population increased throughout the Archaic, and people were shifting from nomadic hunting to somewhat more sedentary lifestyles near coastal and riverine settings (Milanich 1994). The shift in subsistence strategies coincided with a period of transition to warmer, less arid conditions. Projectile points transition from lanceolate forms present during the Paleoindian period to stemmed, side and corner notched, and hafted forms with bifurcated bases (Anderson and Sassaman 2004; Milanich 1994:63). The Early Archaic appears to reflect a co ntinuation of the Paleoindian hunting and gathering lifestyle with increased regional specialization. Early

PAGE 39

39 Archaic sites have been found at a number of locations in Georgia (Wauchope 1966) and Florida (Milanich 1994), many near permanent water sources. Be cause of the rapidity of post glacial sea level changes, the complex estuary systems that would eventually become dominant resource procurement areas for many coastal populations had not stabilized. Stable, productive estuary systems were not present until after 3,000 B.C. Sea level rise caused ground water levels to rise increasing the amount of surface water available (Widmer 1988; Watts 1975; Doran 2002:50). The increase in surface water made new locations suitable for occupation, allowing Archaic people s to move into new ecotones (Milanich 1994:63). Additionally, because water sources were large and more numerous, the Early Archaic peoples could sustain larger populations, occupy sites for longer periods, and perform activities that required longer occu pation at a specific locale (Milanich 1994:69). A period of greater aridity returned near the end of the Early Archaic, about 6000 B.C., though less arid than at the end of the Pleistocene. Middle Archaic Average annual temperatures during the Middle Arc haic were not much different than modern temperatures, but temperature variance was more extreme. Summers were hotter and winters were colder. While lake water levels were lower throughout much of the North American continent, sea and ground water levels w ere higher in Florida and Georgia (Anderson and Sassaman 2004). During the Middle Archaic period more and larger surface water sources were available in Florida, and increasingly moist conditions appeared after about 4000 B.C. (Milanich 1994:84; Watts 1969 1971; Watts and Hansen 1988). A gradual change in forest cover occurred with pines and mixed forests replacing oaks. By about 3000 B.C., vegetation and climate become essentially modern, and sea level rise tapered off (Milanich 1994:75, 84).

PAGE 40

40 The Middle Archaic was a time of dramatic cultural change in the Southeast. Middle Archaic peoples occupied new types of locations for the first time and created new types of sites, including freshwater and marine shell middens (Milanich 1994). Ceremonial shell and e arthen mound construction were initiated in several areas, long distance trade networks appeared, and new tool forms were adopted (Anderson and Sassaman 2004:95). Both Early and Middle Archaic peoples in peninsular Florida began using aquatic environments, such as Windover pond, for burial (Milanich 1994:81). Late Archaic The beginning of the Late Archaic coincides with the beginning of the late Holocene Period and essentially modern environmental conditions by 3000 B.C. (Milanich 1998; Sassaman and Anders on 2004; Watts and Hansen 1988:310). This period is marked by greater regionalization and cultural diversity as human populations adapted to specific environmental zones. Cultures were no longer faced with the challenge of long term environmental and clima tic fluctuations (Milanich 1994). During the Late Archaic period, the firing of clay pottery, along with other technological innovations, appeared in Florida. Ceramic vessel technology gradually spread across the southeast, and was adopted by virtually al l regional populations by about 650 B.C. Local variations in pottery technology and style reflect growing diversity of cultural expression. Regional exchange and intergroup ritual at locations of ceremonial earthworks were among the means by which members of different populations interacted (Anderson and Sassaman 2004:101). More Late Archaic sites are know than sites from any earlier period. Despite changes in technology, there are few apparent differences between Late Archaic subsistence strategies and earlier periods. Populations in the Southeast continued to expand on the hunting and gathering economies of ancestral populations, with shellfish, fish and other food

PAGE 41

41 resources becoming increasingly important (Milanich 1994:85). With the exception of areas that eventually adopted intensive agriculture, the general subsistence patterns of the late Archaic period continued largely unchanged into the colonial period (Milanich 1994; Sassaman and Anderson 2004). Diminished rates of sea level rise promoted the es tablishment of increasingly productive estuarine environments, and maturing floodplain habitat. During this time, evidence of coastal populations in Florida is much more abundant, and Late Archaic shell middens are preserved in many locales (Milanich 1994; Sassaman and Anderson 2004:101). The panregional spread of mortuary ceremonial institutions and greater use of native cultigens in certain sub regions mark the end of the period at about 650 B.C. (Sassaman and Anderson 2004:101). Woodland and Regional C ultures The Woodland period follows the Archaic. It is characterized by increasing population and social complexity through time, and limited adoption of horticulture. With more people on the landscape, mobility decreased and local manifestations of the cu lture emerged. Trade and exchange with regions outside the Southeast occurred to some extent. As groups settled into their local environment and became more sedentary, regional cultures began to emerge. Regional cultures are distinguished by variations in potter styles, projectile point styles, house types, and settlement patterns. Due to the abundance of regional cultures, only the woodland cultures represented in the skeletal sample are detailed here. The advent of the Woodland period occurs at different times throughout Eastern North America, but began earlier in Florida and Georgia and lasted until the European contact in some areas. After 500 B.C., there is archaeological evidence for occupation of every environment within Florida, including the foreste d interior uplands of northern Florida (Milanich 1994: 106). After about A.D. 750, there is evidence for more intensive cultivation of plants, including the possible introduction of maize (Milanich 1994: 108).

PAGE 42

42 Kellog The Kellog culture was concentrated in northwest Georgia Piedmont between about 800 and 200 BC. Kellog settlements include large, year round camps concentrated in narrow flood plains adjacent to streams, and small seasonal camps. Kellog base camps covered about an acre and include abundant art ifacts, some sites having middens several feet thick. Seasonal camps are not as common and contain fewer artifacts. Kellog material culture is dominated by fabric marked pottery early in the period, with simple stamp and check stamp pottery more popular la ter. Stone tools included both stemmed and unstemmed chipped stone points with triangular blades, slate hoes, and biconvex mortars (Bense 1994:135). Kellog culture subsistence strategies were not much different than Archaic strategies of hunting, gathering and fishing, but with perhaps more emphasis on plant foods (Bense 1994:135 136; Hally and Mainfort 2004:266). Deptford The Deptford culture was located along the Gulf coast of Florida and the southeast Atlantic coast between 500 BC and AD 100. The Deptfor d culture area is located between Mobile Bay and Cedar Key along the Gulf coast stretching inland approximately 60 miles, and along the Atlantic coast of South Carolina, Georgia, and northern most Florida extending 30 miles inland. Modest Deptford shell mi ddens found along the coast are located in hardwood hammocks near salt marshes and estuaries, while inland sites are usual located in river valleys. Deptford coastal villages are small and generally contain 5 to 10 houses of either cold weather houses or s ummer warm weather pavilions. The cold weather houses are around 20 by 30 foot ovals, and the warm weather pavilions are approximately 20 by 13 foot ovals. As might be expected from their choice of site locations, Deptford peoples relied heavily on fish an d shellfish gathered from tidal streams and shallow inshore waters, as well as nearby terrestrial resources. Inland sites are small as well and may be special use sites, such as hunting camps (Milanich 2004a:193 194). In both

PAGE 43

43 the Atlantic and Gulf areas s and burial mounds appear during the Deptford period. By AD 1 some mounds on the Gulf coast contained items thought to be linked to Hopewellian related beliefs and trade (Milanich 2004a:194). As the Southeast entered the Middle woodland period, Deptford wa s succeeded by the Swift Creek culture in most areas. St. Johns/Malabar The Saint Johns culture persisted along the Atlantic coast of Florida for around 2,000 years, from the end of the Archaic in 500 BC into the seventeenth century and contact with the Sp anish. The Saint Johns region includes two sub regions, the St. Marys zone in the North, and the Indian River Zone in the south. The Indian River Zone is located around Brevard, Indian River, and St. Lucie counties, and with sites found near wetlands of t he Saint Johns River Basin, the Indian River, and along barrier islands (Milanich 1994: 249). The culture of the Indian River Zone during the Saint Johns period is identified as the Malabar culture, and is divided into two periods (Rouse 1951). The Malabar I period is approximately contemporaneous with the Saint Johns I period, present from 500BC to 750 AD, while Malabar II is of the same age as the Saint Johns II period, present from 750AD to 1565 AD (Milanich 1994: 247,249 250). Malabar sites can be class ified as villages, special use sites, or single use sites. Villages are large, multicomponent sites that exhibit a wide range of artifacts and large middens. Villages are always located near wetlands, and are surrounded by special use sites. Special use si tes are smaller multicomponent sites used intermittently for short periods of time. Single use sites were probably used to gather some specific resource, and all that remains are small artifact scatters or a few animal remains (Milanich 1994:251 252). Mala bar peoples were foragers, and subsistence patterns were remarkably consistent through both phases, with diets composed of roughly 15% terrestrial resources, such as deer, raccoons, and rabbits, and 80% fish and shellfish. As time passed and water levels c hanged, the Malabar peoples tended to collect larger fish and a wider

PAGE 44

44 variety of species (Milanich 1994:251,253). The Malabar I pottery assemblages include Saint Johns sponge spiculate tempered pottery, but are dominated by undecorated pottery tempered wit h quartz sand. Malabar II pottery is characterized by the appearance of Saint Johns Check Stamped pottery. Analysis of pottery from Malabar sites shows continuity of manufacturing methods through both periods (Cordell in Sigler Eisenberg et al 1985:118 134 ; Milanich 1994:250) and a link to Saint Johns pottery (Espenshade 1983; Milanich 1994:250). Manasota The Manasota culture found along the central peninsular Gulf coast region coincided with the Deptford and early Weeden Island cultures, lasting from about 500 B.C. to A.D. 700. This region, which surrounds Tampa Bay, extends along the Gulf from Pasco county south to Sarasota County, and stretches inland nearly to the Peace River drainage. Most Manasota village sites are multicomponent shell middens of vario us sizes found on or near the shore. Some of the coastal shell middens include shell ramps constructed to provide access to the tops. Intensively occupied interior villages with dirt middens have been found in wetland locals (Hemmings 1975; Padgett 1976; L uer et al. 1987; Milanich 1994). Other types of Manasota sites are found away from the coast, in interior pine flatwoods on higher ground near water sources and wetland habitats. These are presumed to be short term villages and special use camps (Austin an d Russo 1989). The evidence from these sites suggests that the Manasota economy was based on fishing, hunting, and shellfish gathering. Most of the Manasota meat diet was derived from aquatic species, including fish, shark, rays, and shellfish. The Manasot a peoples also consumed terrestrial species such as deer, canines, rodents, birds, reptiles and amphibians (Milanich 1994). Manasota material culture is dominated by the use of shell tools with some bone tools, but little use of stone tools. Ceramics were limited to plain sand tempered pottery (Luer and Almy 1979:40 41 in Milanich 1994:222 223).

PAGE 45

45 Wilmington Culture The Wilmington Culture succeeded the Deptford culture along the coast and coastal plain of Georgia at the end of the Late Woodland period, 500 11 50AD. The Wilmington Culture is defined by Wilmington Plain, Wilmington Cord Marked, and Wilmington Brushed ceramics (DePratter 1979; Martinez 1975). The typical Wilmington vessel is decorated with large, parallel individual cord impressions made with a co rd wrapped paddle (Caldwell 1952:316). Wilmington culture is thought to have been influenced by the coeval Weeden Island culture in Florida and by Mississippian people in the Piedmont (Milanich 1976). Like other Late Woodland cultures, Wilmington subsisten ce was based on hunting, fishing and gathering with some horticulture (Wood et al 1986). Historic Period The historic period begins with European contact. This happens at different times in different areas, and initially has varying degrees of impact. For the Timucua of Northern Florida and Southern Georgia, contact with Europeans probably begins in 1525 and early 1526 when scout ships that preceded the Lucas Vsquez de Aylln expedition landed on the northern end of St. Simons Island (Hoffman 1994; Milanic h 2004b:225). In 1565 the Spanish began to establish a missions and colony in Florida with their first permanent New World settlement in St. Augustine. From this base they established Roman Catholic Missions along the Atlantic Coast to the Timucua Indians and Guale Indians along the Georgia coast just north of the Timucua. At the time of the Spanish arrival, many of the Guale and Timucua in northern Florida and Georgia were already involved in maize agriculture and readily missionized. By 1620 virtually eve ry Timucuan chiefdom had received Franciscan missions (Milanich 2004b:225). Non agricultural groups south of the Timucua, such as the Calusa, were not missionized, despite numerous Spanish attempts.

PAGE 46

46 The Timucua and the Guale paid a terrible price for there service to the Spanish crown. The Spanish used the Indians as a labor force to support the Spanish colony. They did supply the Indians with technology to increase maize production, but the demand for maize and labor shifted the native populations from a s emi nomadic hunting, gathering, and farming subsistence base to sedentary intensive maize agriculture. The maize diet led to nutritional deficiencies, especially in lysine, tryptophan and iron. These deficiencies can be seen in the remains of mission India ns in the form of poratic hyperostosis, cribra orbitalia, and enamel hypoplasia. The increased physical stress imposed by the Spanish need for labor can be seen in the form of osteoarthritis. They also experienced increased levels of carious lesions and pe riosteal reactions due to the increase stresses of mission life. Indian populations under the mission system dropped sharply as a result of working conditions and a series of Old World disease epidemics in 1595, 1612 1617, 1649 1650, and 1655 1656. Populat ion levels were further impacted by slaving raids by the English and there allies from 1660 to 1684 (Milanich 1998). As the Timucua populations declined, the Spanish consolidated the remaining tribes along the Camino Real, giving them access to the Apalach icola and their labor. By the time the Spanish ceded Florida to the English in 1763, the Timucua had dwindled from a pre contact population of approximately 20,000 in the early sixteenth century to a single adult. Other Indian populations in Florida and So uthern Georgia were similarly decimated, and the few remaining survivors were evacuated to Cuba when the Spanish left Florida (Milanich 1994, 2004b; Williams 2004).

PAGE 47

47 CHAPTER 5 SITES Introduction Of the ten archaeological sites used in this study, six (Gol f Course, Bay Pines, Canaveral, Casey Key, Palmer, and Perico Island) are located along the coast of Florida, and three (Cannons Point, Taylor Mound, and Couper Field/Indian Field) are located on Saint Simons Island, Georgia. The tenth site (Garfield) is located in the Piedmont of Georgia. These sites represent to Archaic, Deptford, Weeden Island, and contact periods. These sites date from possibly as early as the Paleoindian through the Contact period. Site numbers and brief descriptions of the location, the archaeological investigation, and the interpretation of each of these sites follow below. Golf Course (8Br44) The Golf Course site is located on the north edge of the Melbourne Municipal Golf Course just east of a canal that cuts through the property in Melbourne, Brevard County, Florida. The site was discovered in 1952 by F.B. Loomis of Amherst College, J.W. Gidley of the U.S. National Museum (Smithsonian), and C.P. Singleton, a resident of Melbourne, during a survey of spoil from the nearby canal (Ro use 1951: 153). Loomis and Gidley prevaricated on whether the remains came from the Pleistocene Melbourne bone bed, or from the lower levels of the overlying Holocene Van Valkenburg bed. Ale Hrdli ka argued that the Melbourne Man was similar to recent I ndians and could not be of great antiquity based on his analysis of the skulls morphology and his own conviction that humans had not entered the New World more than a few thousand years ago (Miller 1950; Wilmsen 1965). After reconstructing and reexamining the skull, Stewart (1946) suggested that it might in fact belong to the Paleoindian period. (Milanich 1994:8; Miller 1950) While the lack of a definitive cultural affiliation is frustrating, this case is

PAGE 48

4 8 an excellent example of the kind of where the presen t research will be of the greatest benefit, namely in gathering information about isolated remains with uncertain affiliation. Bay Pines (8Pi64) The Bay Pines site is located in Pinellas County, Florida on Boca Ciega Bay, just west of St. Petersburg. The s ite consists of a shell ridge oriented on a north south axis, parallel to the coast of Boca Ciega Bay. The north end of the ridge is near a freshwater lagoon, and two smaller ridges are present perpendicular to the shore, one in the middle, and one at the south end termed "shell ridge A." Shell ridge A was excavated by members of the Suncoast Archaeological Society in 1971 in a salvage operation prior to the construction of a nursing home. Ten burials are identified in a cemetery and fragmentary remains of at least 14 other individuals are found scattered throughout what was probably a burial mound. The remains of all 24 individuals are currently housed at the Florida Museum of Natural History. The site is thought to be multi component with occupations from the Deptford period through the early Weeden Island period. A reanalysis of faunal remains associated with the site and stable isotope analysis of the skeletons suggest a diet heavy in fish from the nearby Gulf of Mexico and included turtle, mammal, bird, crab and possibly maize (Gallagher and Warren, 1975; Kelly, Tykot and Milanich 2006). Canaveral (8Br85) The Canaveral site is located on Cape Canaveral in Brevard County, in the Indian River area of Floridas Atlantic coast. Dr. George Woodbury excavated burial mounds on Cape Canaveral from 1933 1934 as part of the Civil Works Administration relief archaeological program affiliated with the Bureau of American Ethnology, Smithsonian Institution during the Great Depression (Milanich 1994:9 10). The site con sists of several burial mounds, including the Burns Mound and Fuller Mounds A, B, and D. The Burns Mound (8Br85) is a burial mound built

PAGE 49

49 on top of a shell midden. The burial mound was composed of a lower sandy layer and an upper layer composed of a thick laminated deposit which contained charcoal, pot sherds, shells, etc. (Woodbury n.d.). Woodbury recovered 31 burials from the lower zone and 21 from the upper zone; all are primary burials with their heads oriented toward the center of the mound in a spo ke pattern. All of the upper zone burials are extended while some from the lower zone are flexed or semi flexed (Willey 1954:81). All but one of the burials are adults with approximately equal numbers of males and females. The Burns Mound is dated to the M alabar II period based on ceramic types, although a pendant of European silver indicates the mound was used after European contact. The excavation at Fuller Mound A recovered in a sample of 96 complete skeletons. All but twelve are adults, with slightly mo re females than males. Almost all of the burials are oriented in the spoke pattern with their heads toward the center of the mound. Most are primary extended burials lying on the back, although a few are semi flexed. A few may have been secondary burials. Iron and metal tools and glass beads of European manufacture are present (Stirling 1935:386). Based on the ceramics and the quantity of European goods, Rouse (1951:197) suggested that the mound dates from the 17 th century. Fuller Mound B contained the rema ins of about 20 individuals disarticulated and mixed in a single secondary burial at the center of the mound. Two primary burials are also found away from the center with their feet pointing toward the center of the mound (Stirling 1935:387). Rouse (1951:1 97) dates the mound to the Malabar I' period based on the ceramics and the lack of European artifacts. Fuller Mound D consisted of 16 primary extended burials in the spoke pattern oriented with heads toward the center of the mound. Five of the individuals are infants, and there are

PAGE 50

50 more adult males than females. A few glass beads are present suggesting that the mound dates from the 17 th century, the same as Fuller Mound A (Stirling 1935:387). Casey Key (8So17) The Casey Key site lies on Casey Key, a Gulf co ast barrier island located about three miles south of Osprey, in Sarasota County, Florida. The site was known to residents of the area long before it was recorded by archaeologists and has never been systematically excavated. The site is near the Palmer si te, discussed below, and is thought to have been roughly contemporaneous with it, although it is not tightly dated due to a dearth of diagnostic artifacts. The limited pottery assemblage indicates it is from the Manasota Weedon Island culture dating from c a. A.D. 250 750. Casey Key included a village and a burial mound that is thought to have contained over 200 burial but most of these are collected by local residents or sold by high school students. A few skeletons were donated to the Florida Museum of Nat ural History in the 1950s and 1960s by Hilton Leech, who attempted to salvage some data from the mound before it was completely destroyed (Bullen and Bullen 1976:47 48). Palmer Burial Mound (8So2a) The Palmer Site is a complex of sites located near Osprey in Sarasota County, on Little Sarasota Bay on the Florida Gulf Coast. The site was the subject of a number of scientific excavations. The first formal excavations by Ripley Bullen between 1959 and 1962 are the most extensive and only ones to include the burial mound (Bullen and Bullen 1976). Those excavations recorded five sites: Hill Cottage Midden, dating to the Archaic period; Shell Ridge, dating to the Middle Woodland period; Shell Midden, dating from Middle Woodland through Mississippian periods; the North Creek Area middens; and Palmer Burial Mound. A survey of the Palmer tract performed in 1974 uncovered four additional sites, including another burial mound (Miller 1974). Limited investigations were performed in 1979 and 1980 prior to the

PAGE 51

51 establishm ent of the Spanish Oaks historical site. The Shell Ridge area was excavated in 1991 under the direction of Corbett Torrence, George Luer, and Marion Almy, and detailed zooarchaeological analyses were conducted (Hutchinson 2004; see Almy and Luer 1993; Kozu ch 1998; Quitmyer 1998). Bullen and Bullen (1976:35) describe Palmer Mound as a very unassuming, dome shaped, sand mound rising 4 feet above the surrounding land. Despite its modest appearance, it is actually one of the largest systematically excavated b urial mounds in the southeast, with over 400 individuals recovered. The mound was used primarily between A.D. 500 and 800, during the Manasota period (Bullen and Bullen 1976; Hutchinson 2004:43 59; Williams 2004). Faunal analysis indicates that fish and sh ellfish dominated the diet, with little consumption of terrestrial animals, but some use of terrestrial plants. Perico Island (8Ma6) The Perico Island site is located on the western edge of Perico Island in Manatee County, Florida, west of Bradenton, and between Sarasota and Tampa Bays. The site is composed of large and small shell middens, a burial mound, and a cemetery area. The site was excavated by Dr. M.T. Newman in 1933 34 as a relief project. Newman recovered 185 flexed burials from the burial mound and 43 primary flexed burials from the cemetery (Willey 1949:176,180). Willey (1949) initially categorized the site as a local variant of the Glades culture (Willy 1949;1998:192), but it is now considered part of the Manasota culture (see Milanich 1994; Luer and Almy 1982). St. Simons Island, Georgia Remains are used from three sites located on the northern end of St. Simons Island, Martinez B C, Taylor Mound, and Couper Field/Indian Field. All were excavated as part of the St Simons Island Archaeologic al Project operated by the University of Florida between 1972 and

PAGE 52

52 1975. All three locations are on Cannons Point at the northern end of the island. Wallace (1975) used ceramic and mortuary analysis to examin the relationship between these sites. He found that the sites are from a single, contemporaneous group of Guale Indians, but represent different hierarchical groups within the culture. That conclusion will not be challenged here. Martinez B C is located at the tip of Cannons Point near the Hampton Riv er. The test pit was excavated there in 1974 and initially two burials were recovered. A third burial discovered adjacent to the test pit was recovered in 1975. The individuals include one infant and two adult males, and all are primary extended burials. T his location is thought to be part of a primary living area. Based on associated ceramics, the burials are thought to be from the Wilmington Period (Martinez 1975:56 58). Taylor Mound is a Historic period (ca. A.D.1600 1650) ceremonial mound with associate d burials. While some historic artifacts are associated with the mound, it does not appear to have been heavily impacted by European contact (Wallace 1975:39 78; Zahler 1976:2). Eleven burials were recovered from this location, The sex of one skeleton coul d not be identified because it was too fragmented and too young. A second was excluded from analysis because its stratagraphic affiliation was uncertain (Wallace 1975:44). The remaining nine individuals included seven females and two males. Prior to formal excavation, thirteen burials were excavated by local residents, but information for these individuals was not recorded. Couper Field and Indian Field are the northern and southern parts, respectively, of the same village, and is part of the same sociocult ural population as Taylor Mound, but represents different levels of the social hierarchy. Couper Field lies immediately south of the remains of the antebellum Couper Mansion. Although the area had been heavily plowed the majority of burials are undisturbed There are 16 interments containing 18 individuals, including one infant, ten

PAGE 53

53 females and seven males (Wallace 1975:141 144). Indian Field was the location of a large ceremonial pavilion in which were interred six burials that contained the disturbed, fra gmented ,and mixed remains of 22 individuals. Remains of at least 13 of these are recovered from a single interment (Zahler 1976: 8). Garfield Site (9BR57) The Garfield site is the remains of a village located on the confluence of the Etowah River and Mace donia Slough near Kingston, Bartow County, in northwestern Georgia. Portions of the site were first excavated by two amateur archaeologists from Decatur, Georgia, James Chapman and Richard Criscoe during the 1960s. Eighteen burials are recovered from what appeared to be abandoned storage pits. Those remains were transferred to the Florida Museum of Natural History in 1974 with other collections they had excavated. Jerald T. Milanich tested the site in 1972 while he was a post doctoral fellow at the Smithson ian Institution. He recovered an additional four burials, including two adults, one infant, and one cremation. An additional fragment of human bone was identified among animal bone excavated from midden deposits. Artifacts recovered from the site indicate it belongs to the Kellog culture dating to 600 B.C. to A.D. 100, a date supported by two radiocarbon assays (185570 B.P. and 235060 B.P.) from charcoal obtained by Milanich (Jerald T. Milanich personal communication 2007; Milanich 1975). Milanich also re covered a large quantity of floral and faunal remains, ceramics, and lithic artifacts. As a group they suggest an occupation from early spring into late summer or early fall. While it seems likely that some maize gardening was carried out toward the end of the sites period of occupation, wild foods, especially nuts and other plant products, fish and a variety of mammals, provided most of the diet (Milanich 1975).

PAGE 54

54 Figure 5 1 Locator map of sites used in this study and major rivers.

PAGE 55

55 CHAPTER 6 METHODS AND MATERIALS Introduction This section describes sampling methods and the criteria used for selecting skeletal materials for measurement. It also describes the skeletal materials used for this study and the methods used in determining the sex of each skeleto n. Finally, it describes the procedures used in data collection, including measurements taken and their description, and the software and methods used in calculating the discriminant functions. Sampling In terms of research design, the ideal approach is to take a simple random sample or random block sample from all pre contact Native American skeletons excavated from Florida or Georgia archaeological sites, and sex them using the pubic bone. Unfortunately, this ideal is not practical. The remains of Florida and Georgia Native Americans are housed in several institutions around the country, and not all are readily accessible. Of the accessible remains, not all have measurable crania. Of those, few have an intact pubic bone, and none are of known sex. Therefor e, it is necessary to draw the sample from those remains that can provide the needed data. The best practical method is to use all of the available remains that have a measurable cranium, and which can be reliably sexed. This total sample approach is lik ely to introduce bias into the sample. For example, individuals who are robust, buried in shell middens, or from more recent populations are likely to be over represented. Determination of Sex The sex of the specimens is determined by using visual assessme nt of the crania and postcrania described in chapter 2. The Phenice (1969) method is used in conjunction with other postcranial indicators, such as the sciatic notch and pre auricular sulcus, and with cranial

PAGE 56

56 indicators. In sexing skeletons, the heaviest w eight is given to the Phenice method of sexing the pubic bone because of its high accuracy, followed by other features of the pelvis and postcrania. Cranial indicators are given the least weight, and sexing by crania alone is avoided. Materials The samples used in this study come from a total of ten sites. Samples from Canaveral, Perico Island, and Ballard Estates are housed at the Smithsonian Institutions National Museum of Natural History. Samples from Garfield Site, Couper Field, Taylor Mound, Cannons Point, Palmer, Casey Key, and Bay Pines are from the Collections of the Anthropology Division of the Florida Museum of Natural History. Measu rements are taken from a total of 46 individuals. Because some individuals are incomplete, not all measurements cou ld be taken for every individual, so not all individuals are included in each stage of the analysis. Data are collected from one complete adult male recovered from the Garfield site. The complete set of 25 cranial measurements is recorded. The postcrania, including the pubic bone is missing, so sexing is accomplished using visual evaluation the crania. Data are collected from nine individuals recovered from Couper Field, five males and four females. Two of the four females are excluded from further analysis because they lacked a measurable pubic bone and have fragmentary crania. The remaining two females are sexed with using visual analysis of the crania and postcrania. Two males have an observable pubic bone. The remaining three males are sexed using visual analysis of the crania. Data are collected from seven individuals recovered from Taylor mound, including four males, two females and one individual of unidentified sex. One male and one female are excluded from further analysis because they lacked postcra nial remains including the pubic bone and have fragmentary crania. Two males are sexed using the pubic bone, although their crania are fragmentary, and with only seven and 14 observable measurements. The remaining male and

PAGE 57

57 female are sexed using a combinat ion of cranial and visual analysis of the postcrania, but did not include the pubic bone. The individual of unidentified sex is missing postcranial sex indicators, and the cranium is fragmentary. Data are collected from one male recovered from Cannons Po int. This individual has a fragmentary cranium, allowing observation of 9 of 25 cranial measurements; but possesses a relatively complete postcrania that included the pubic bone. This individual is sexed using cranial and postcranial visual methods, includ ing observation of the pubic bone. Data are collected from nineteen individuals recovered from Canaveral, including three females and 16 males. All three females and 14 males have crania that are complete or nearly complete, and have relatively complete po stcrania including the pubic bone. All 17 are sexed using a combination of cranial and postcranial visual methods, including observation of the pubic bone. The remaining two male crania and postcrania are nearly complete, but lack an observable pubic bone. These two individuals are sexed using a combination of cranial and postcranial visual methods, excluding observation of the pubic bone. Data are collected from four individuals recovered from from Perico Island, including two males and two females. All fo ur individuals are complete, allowing observation of all 25 cranial measurements and the pubic bone. All are sexed using a combination of cranial and postcranial visual methods, including observation of the pubic bone. Data are collected from two individua ls recovered from the Palmer site, including one male and one female. The female is nearly complete, allowing observation 24 of 25 cranial measurements and the pubic bone. This individual is sexed using a combination of cranial and postcranial visual metho ds, including observations of the pubic bone. The second individual is a fragmentary male, allowing observation of only 7 of 25 cranial measurements, and is missing the

PAGE 58

58 pubic bone. This individual is sexed using a combination of cranial and postcranial vis ual methods, excluding the pubic bone. Data are collected from one female cranium recovered from Casey Key. This cranium is nearly complete, allowing observation of 22 cranial measurements. The postcrania including the pubic bone is missing. This individua l is sexed using visual analysis of the cranium. Data are collected from one malerecovered from Bay Pines. This individual is nearly complete, allowing observation of all 25 cranial measurements but not the pubic bone. This individual is sexed using a comb ination of cranial and postcranial visual methods, excluding the pubic bone. Data are collected from one male recovered from Ballard Estates. This individual is nearly complete, allowing observation of 23 of 25 cranial measurements but not the pubic bone. This individual is sexed using a combination of cranial and postcranial visual methods, excluding the pubic bone. Cranial Measurement Definitions Giles and Elliot Measurements Ten measurements are taken from Giles and Elliot (1962, 1963) following their de scriptions (Table 6 1). Other sources are used to clarify descriptions and measurement techniques where Giles and Elliot are unclear, particularly Bass (1971, 1995), Howells (1973), Buikstra and Ubelaker (1994), and FORDISC 2.0 materials (Ousley and Jantz 1996). The remaining measurements (Table 6 2) are not included in the Giles and Elliot (1962) study, but are used in more recent research by Bass (1995), Howells (1973), Buikstra and Ubelaker (1994), and FORDISC 2.0 materials (Ousley and Jantz 1996). The d escriptions and techniques primarily follow Ousley and Jantz (1996).

PAGE 59

59 Table 6 1. Measurements used by Giles and Elliot, the measurements modern equivalent, common symbols, Giles and Elliot descriptions, and references used. Complete measurement descriptio ns and techniques are listed in appendix A. Giles and Elliot measurement Modern equivalent Symbols Description References Glabello occipital length Maximum cranial length g op, GOL The distance of glabella (g) from opisthocranion (op) in the mid sagittal plane measured in a straight line. Bass 1971:62; Howells 1973:170; Martin 1956:453 ; Olivier 1969:128. Maximum width Maximum cranial breadth eu eu, XCB The maximum width of the skull perpendicular to the mid sagittal plane. Bass 1971:62; Howells 1973:172 ; Hrdli ka 1952:140; Martin 1956:455 ; Montagu 1960:44. Basion bregma height Basion bregma height ba b, BBH The direct distance from the lowest point on the anterior margin of the foramen magnum, basion (ba), to bregma (b). Bass 1971:62; Howells 1966:6; M artin 1956:459 ; Olivier 1969:129. Maximum diameter bi zygomatic Bizygomatic breadth zy zy, ZYB The direct distance between each zygion (zy), located at the most lateral points of the zygomatic arches. Bass 1971:67; Martin 1956:476 Basion nasion Crania l base length ba n, BNL The direct distance from nasion (n) to basion (ba). Howells 1966:6; Martin 1956:455 Basion prosthion Basion prosthion length ba pr, BPL The direct distance from basion (ba) to prosthion (pr). Martin 1956:474. Nasion breadth Nasal breadth al al, NLB The maximum breadth of the nasal aperture. Bass 1971:68; Howells 1973:176; Martin 1956:479; Montagu 1960:50; Olivier 1969:153.

PAGE 60

60 Table 6 1. Continued Giles and Elliot measurement Modern equivalent Symbols Description References Pala te external breadth Maxillo alveolar breadth, external palate breadth ecm ecm, MAB The maximum breadth across the alveolar borders of the maxilla measured at its widest point, between each ectomolare (ecm). Bass 1971:70; Howells 1973:176; Martin 1956:480; Montagu 1960:51. Mastoid length Mastoid length MDH The projection of the mastoid process below, and perpendicular to, the eye ear (Frankfort Horizontal) plane in the vertical plane. Howells 1966:6; 1973:176. Prosthion Nasion Height Upper facial height n pr The direct distance from nasion (n) to prosthion (pr). Howells 1966:6; Hrdli ka 1952:143; Martin 1956:476. Table 6 2. Measurements not used by Giles and Elliot, common symbols, measurement descriptions, and sources of the measurements. Complete measurement descriptions and techniques are listed in appendix A along with a sample data collection sheet. Measurement Symbol Description References Biorbital breadth ec ec, EKB The direct distance from one ectoconchion (ec) to the other. Howells 1973:178 Interorbital breadth d d, DKB The direct distance between r ight and left dacryon (d). Martin 1956:477. Maxillo alveolar length, external palate length pr alv, MAL The direct distance from prosthion (pr) to alveolon (alv). Bass 1971:70; Hrdli ka 1952:146 147; Martin 1956:480. Biauricular breadth au au, ALB The least exterior breadth across the roots of the zygomatic processes. Howells 1973:173 Foramen magnum length ba o, FOL The direct distance of basion (ba) from opist hion (o). Martin 1956:455

PAGE 61

61 Table 6 2. Continued Measurement Symbol Description References Minimum frontal breadth ft ft, WFB The direct distance between the two frontotemporale (ft). Bass 1971:67; Hrdli ka 1952:142; Martin 1956:457; Olivier 1969:151. Foramen magnum breadth FOB The distance between the lateral margins of the Foramen magnum at the point of greatest lateral curvature. Martin 1956:459 Upper facial breadth fmt fmt The direct distance between each frontomalare temporale (fmt). This mea surement differs from Howells FMB in that the lateral most points on the suture are used rather than the most anterior points. Martin 1956:475. Nasal height n ns, NLH The direct distance from nasion (n) to nasospinale (ns). Bass 1971:68; Howells 1966: 6; Martin 1956:479; Olivier 1969:153 Orbital breadth d ec, OBB The laterally sloping distance from dacryon (d) to ectoconchion (ec). Martin 1956:477 478; Howells 1973:175 Orbital height OBH The direct distance between the superior and inferior orbit al margins. Bass 1971:69; Martin 1956:478; Montagu 1960:51; Olivier 1969:152. Biorbital breadth ec ec, EKB The direct distance from one ectoconchion (ec) to the other. Howells 1973:178 Interorbital breadth d d, DKB The direct distance between right and left dacryon (d). Martin 1956:477. Frontal chord n b, FRC The direct distance from nasion (n) to bregma (b) taken in the midsagittal plane. Howells 1973:181; Martin 1956:465. Parietal chord b l, PAC The direct distance from bregma (b) to lambda (l) taken in the midsagittal plane. Howells 1973:182; Martin 1956:466. Occipital chord l o, OCC The direct distance from lambda (l) to opisthion (o) taken in the midsagittal plane. Howells 1973:182; Martin 1956:466.

PAGE 62

62 Statistical Procedures Calculation s are performed using Microsoft Excel 2004 for Macintosh, Version 11.3.3 and SAS 8.3 running on Unix at grove.ufl.edu. A t test is a statistical hypothesis test that follows a Students t distribution if the null hypothesis is true and the variable is norm ally distributed. The t test is used to test the null hypothesis that the mean of a variable is the same for males and females against the alternative that the means are different. The value of the t statistic is the difference between the means of the two groups divided by the standard error of the difference. The result is used to identify which variables are likely to be useful in discriminating between male and female crania. The t statistic is calculated using the formula: wher e is the difference between the sample means, and is the standard error of the differences between the two means. For groups of unequal size, is computed by the formula: where (variance) is calculated by the formula: and is calculated using the VAR function in Excel. The t statistic is then compared to a Students t distribution with n 2 degree s of freedom. This is accomplished using the TDIST function in Excel. Multiple Analysis of Variance, or MANOVA, is the multivariate equivalent of a t test. It is used to test the null hypothesis that more than two groups do not differ for multiple variab les. In

PAGE 63

63 the present case, it is used to test the hypothesis that all of the individuals from the different sites in this research can be considered to be a single group for the purposes of this study. The MANOVA is preformed using SAS PROC GLM. PROC GLM Da ta=thesisdata ORDER=freq; CLASS site; MODEL GOL XCB ZYB BBH BNL BPL AUB WFB UFBR EKB FRC PAC MDHA= site; MANOVA H=site / PRINTE; The next task is to calculate the results of the Giles and Elliot (1962) function 3 for sex discrimination on the study sam ple in order to evaluate its accuracy. Sex is first determined using the sectioning point given by Giles and Elliot. The sectioning point is then recalculated using the method described by Giles and Elliot. The average of the function is calculated using t he average value of each function for each sex separately, and the sectioning point is placed midway between the two results. The function is calculated again, replacing missing values with the average of the male and female means for that variable. Accuracy is then evaluated again using the original sectioning point, and the recalculated sectioning point described above. Once the accuracy of the Giles and Elliot formula is determined, new discriminant functions are calculated based on the data collected. The first function calculated uses the same variables Giles and Elliot used. The second function calculated uses the variables selected using t tests and that showed good preservation as indicated by the proportion of crania from wh ich the measurement could be observed. Exact binomial probabilities are used to compare the accuracy of new results to Giles and Elliots (1963) reported accuracy of 86.4%. Exact probabilities are calculated using the SAS

PAGE 64

64 PROC FREQ procedure with the EXAC T BINOMIAL option. The null hypothesis is that the accuracy of the formulas is equal against the alternative that the new accuracy is higher. Fishers exact test is used to compare the accuracy between sets of new results. Fishers exact test is a non para metric statistical hypothesis test used for categorical data where samples are too small for the Chi squared test. Fishers exact test is calculated using the PROC FREQ procedure with the CHISQ option. The null hypothesis is that the accuracies are equal, which is tested against the alternative that the accuracies are different.

PAGE 65

65 CHAPTER 7 RESULTS General Results This section presents the results of the analysis described in Methods and Materials. Discri minant function scores are calculated for each individual using the Giles and Elliot (1963) formula for sex determination. Each individual is classified as male or female using the sectioning point published by Giles and Elliot. Each individual is classifi ed again using a recalculated sectioning point based on sample data following a method recommended by Giles and Elliot (1963). Using the published Giles and Elliot (1963) sectioning point, the formula correctly classifies 13 of 13 males, and 2 of 4 females The overall accuracy is 88.24%, but the error is unequally distributed. Using the recalculated sectioning point, the formula correctly classifies 12 of 13 males, and 3 of 4 females. The overall accuracy is still 88.24%, but the error is evenly distribute d. This error rate is not significantly different than that reported by Giles and Elliot, but the sample size is unacceptably small due to missing variables. In order to increase the sample size, missing variables are replaced with the average of the male and female means for each variable. This allows the calculation of the discriminant function to be completed without allowing missing variables to influence sex classification. Discriminant function scores are calculated for each individual using the Gile s and Elliot (1963) formula for sex determination. Each individual is classified as male or female using the sectioning point published by Giles and Elliot (1963). Each individual is classified again using a recalculated sectioning point based on sample da ta following a method recommended by Giles and Elliot. Using the Giles and Elliot sectioning point, the formula correctly classifies 31 of 32 males, and 2 of 13 females. The overall accuracy is 73.33%, but the error is unequally

PAGE 66

66 distributed. Using the reca lculated sectioning point, the formula correctly classifies 27 of 32 males, and 11 of 13 females. The overall accuracy is 84.44%, and the error is evenly distributed. Variables for further analysis are selected using t tests and variable preservation rates T tests are performed to identify variables that are likely to be useful in discriminating between males and females. Ten variables are significant, including: GOL, BBH, BNL, BPL, AUB, WFB, UFBR, EKB, PAC, and MDH. Five of these variables are excluded fr om further analysis because their low preservation would have decreased the number of usable individuals to unacceptable levels. The excluded significant variables are: BBH, BNL, BPL, UFBR, and EKB. The variables retained for further analysis are GOL, AUB, WFB, PAC, and MDHA. A MANOVA is performed using the retained variables to test the null hypothesis that there is no difference between groups. There is not enough evidence to reject the null hypothesis that there is no difference between groups. Therefore it is concluded that all individuals could be treated as a single group. Two new discriminant functions are created. The first is created from the sample data using the same variables used by Giles and Elliot (1963). The second is created using the varia bles identified in the variable selection step. The first step is to create a new formula using the same variables as Giles and Elliot (1963). This is done using PROC DISCRIM in SAS and variables GOL XCB BBH ZYB BPL UFHT MAB MDHA. Again, discriminant funct ion scores could only be calculated for 17 individuals (13 Male and 4 female) due to missing variables. The new function correctly classifies 13 of 13 males and 4 of 4 females, for a combined accuracy of 100%. This is not significantly different from the a ccuracy of the Giles and Elliot function. Next, a new function is calculated using the variables identified during variable selection. This is done using PROC DISCRIM in SAS and variables GOL, XCB, AUB, WFB, FRC, PAC,

PAGE 67

67 and MDHA. Discriminant function scores are calculated for 38 individuals, 26 males and 12 females. The new function correctly classifies 11 of 12 females and 24 of 26 males for a combined accuracy of 91.99%. The discriminant function is then applied to the full sample, substituting missing val ues with the average of the male and female means from the test sample. This function assigned sex correctly to 12 of 13 females and 27 of 32 males, for a total accuracy of 86.67%. This is not significantly different than the accuracy of the Giles and Elli ot function. Specific Results Giles and Elliot Discriminant Function Discriminant function scores are calculated for each individual using Giles and Elliots (1963) discriminant function 3 based on a combined sample of black and white individuals (see Tabl e 2 1, function 3). The purpose of this task is to establish the accuracy of the Giles and Elliot function on Florida and Georgia Native Americans for comparison to new discriminant functions. Thirteen males and four females are complete enough to record a ll eight measurements needed to use the Giles and Elliot function. Using the sectioning point reported by Giles and Elliot (1963), the formula classifies individuals with discriminant function scores above 6237.95 as male, and individuals below that score as female. The formula correctly classifies all 13 males, but only classifies 2 of 4 females correctly. While the overall accuracy of 88.24% compares well with the Giles and Elliot result of 86%, the error is not equally distributed between males and femal es. All males are correctly classified while half of the females are misclassified. This violates the requirement of DFA that error be spread equally among groups (Kendal 1957). This type of error is known to occur when a discriminant function is applied t o a group that is not included in the development of the formula. Giles and Elliot report this type of error where they attempt to use their formula on different groups. To avoid this problem, Giles and Elliot recommend recalculating the sectioning

PAGE 68

68 point b ased on the sample data. Their method is to create a centroid discriminant function score for each sex by using the mean value of each variable to calculate discriminant function scores. The average of the male and female centroids is used as the sectionin g point. "We can determine the mean value of the discriminant function scores for males by taking the mean male value for each measurement and entering them into the discriminant function. If this is likewise done for the females, the arithmetic mean of t he two scores provides a sectioning point to use when we have no a priori reason to believe that a specimen is more likely to be male than female (Kendall 1957). Following this procedure, we will say that any specimen falling on the side of this line towar d the male mean will be called male, and any specimen falling on the other side will be called female. So doing should minimize the probability of misclassification (Giles and Elliot 1963). Table 7 1 Accuracy of the Giles and Elliot function 3 for sex dete rmination on the study sample of Florida and Georgia Native Americans. Sectioning Point Sex Number Correct Total (n) Accuracy overall 15 17 88.24% male 13 13 100.00 % Giles and Elliot Sectioning Point (6237.95) female 2 4 50.00% overall 15 17 88.24% male 12 13 92.31% Individuals with missing variables omitted Recalculated Sectioning Point (6513.392571) female 3 4 75.00% overall 33 45 75.56% male 32 32 100% Giles and Elliot Sectioning Point (6237.95) female 2 13 15.38% overall 38 45 84.44% male 27 32 84.375% Missing variables replaced with average of the male and female means Recalculated Sectioning Point (6513.392571) female 11 13 84.615% The male centroid is 6731.518743, the female centroid is 6295.266398, and the recalculate d sectioning point is 6513.392571. Using the recalculated sectioning point, the overall accuracy remains the same, but errors are evenly distributed. The function correctly sexed 12 of 13 males and 3 out of 4 females. The accuracy of the formula using the recalculated sectioning point is not significantly different from the result reported by Giles and Elliot for males (p=0.9109), females (p=0.8855) or males and females combined (p=1.0000).

PAGE 69

69 In order to increase sample size, missing variables are replaced w ith the average of the male mean and the female mean for each variable. This allows the discriminant function to be calculated without the missing observation influencing the final classification. With the missing variables replaced and using the Giles an d Elliot sectioning point, 31 out of 32 males (p=0.1123) are correctly classified, but only 2 out of 13 females (p<.0001) are correctly classified. Both of these results are significantly different from both Giles and Elliots result and from each other (p <0.0001). With the sectioning point recalculated, 28 out of 32 males are correctly classified, and 10 out of 13 females are correctly classified. The results using the repositioned sectioning point are not significantly different from the Giles and Elliot result for males (p=1.0000), females (p=0.5112), or male and females combined (p=0.8283). Variable Selection The goal of variable selection is to select for further analysis those variables that are likely to aid in the discrimination of males and females, and that are likely to be preserved for measurement. Variables are selected for further analysis using t tests and variable preservation rates. T tests are performed for all cranial variables for a difference in mean between males and females (Table 7 2). This is done in order to identify variables likely to useful in discriminant function analysis. The t test tests the null hypothesis that there is no difference between the male and female means of a variable against the alternative that there is a differ ence. A variable whose mean is not significantly different between males and females is unlikely to contribute to sex discrimination. A variable that is poorly preserved in the present sample would reduce the sample size to unacceptable levels, and is like ly to limit the applicability of this research in other cases.

PAGE 70

70 Table 7 2. T scores and p values for a difference in mean values of each cranial variable between males and females. Significant values are in italics. Variables with acceptable preservation t hat are used in further analysis are in bold face. Cranial Measurement Male Count Female Count Total Count T statistic P value GOL 30 12 42 3.25 0.0024 XCB 29 12 41 2.25 0.0310 ZYB 15 4 19 2.84 0.0113 BBH 27 11 38 3.12 0.0036 BNL 27 10 37 3.28 0. 0023 BPL 26 9 35 2.96 0.0057 MAB 25 8 33 0.15 0.8835 MAL 24 10 34 1.34 0.1889 AUB 30 13 43 3.63 0.0008 UFHT 28 10 38 0.79 0.4356 WFB 31 12 43 3.16 0.0030 UFBR 29 11 40 2.73 0.0096 NLH 28 10 38 1.13 0.2679 NLB 27 10 37 0.53 0.6000 OBB 26 10 36 1.44 0.1577 OBH 27 10 37 0.25 0.8047 EKB 25 10 35 2.73 0.0100 DKB 25 10 35 2.36 0.0242 FRC 32 12 44 1.40 0.1696 PAC 32 13 45 3.63 0.0008 OCC 30 11 41 .11 0.9135 FOL 27 10 37 1.53 0.1360 FOB 25 11 36 1.82 0.0770 MDHR 28 13 41 4.01 0 .0003 MDHL 27 9 36 2.70 0.0106 MDHA 31 13 44 3.92 0.0003 Thirteen of the 26 measurements are significant at the alpha=0.01 significance level, including left, right and average mastoid length (Table 7 2). The significant variables are: Glabello occip ital length (GOL); Basion Bregma height (BBH); Basion nasion (BNL); Basion prosthion (BPL); Left, right and average Mastoid height (MDHL, MDHR, MDHA); Parietal chord (PAC); Biauricular breadth (AUB); Minimum frontal breadth (WFB); Upper facial breadth (UFB R); and Biorbital breadth (EKB). Giles and Elliots function 3 includes the significant

PAGE 71

71 variables GOL, BBH, BPL, and MDH. Variables included in the Giles and Elliot formula which are not significant are: XCB, ZYB, and MAB. A variable is considered to be po orly preserved if it is not observable in at least 90% of cases for each sex. For this sample, the variable has to be observable in at least twelve females and at least 29 males. Five of the significant variables, BBH, BNL, BPL, MDHR, and MDHL, are exclude d because of poor preservation. Five variables, GOL, AUB, WFB, PAC, and MDHA, meet the criteria for significance and preservation. These five variables are used in discriminant function analysis for the determination of sex in Florida and Southern Georgia Native Americans. Test for Site Effect on Selected Variables The purpose of testing for site effect is to determine if individuals from different sites in Florida and Georgia can be treated as a single population. If there is no significant site effect, th en a single discriminant function for sex determination can be used for all sites. Table 7 3 F values and P values for the hypothesis of no site effect for each variable using type IV sums of squares and cross products. Variable DF F P GOL 8 1.35 0.2581 AUB 8 3.11 0.0116 WFB 8 1.60 0.1688 PAC 8 1.31 0.2756 MDHA 8 0.61 0.7620 An ANOVA is first performed on individual variables to test for the effects of site (Table 7 3). There is no significant site effect for any of the variables at the 0.01 alpha le vel. There is a significant site effect for AUB at the 0.05 alpha level. This difference is driven by a difference between the Casey Key site and the Golf Course site, each of which is represented by one individual. For this variable, the individual from G olf Course is a larger than average male, and

PAGE 72

72 the Casey Key individual is an unusually small female. There is not enough evidence to justify rejecting the null hypothesis, eliminating the variable AUB, or eliminating either site. The MANOVA test is the mu ltivariate equivalent to an ANOVA. It is used in cases where there are multiple metric dependent variables, and one or more categorical independent variables. It is used to test the null hypothesis that there is no overall difference between sites for the array of variables included in this study. The sex of the individual is also used as an independent variable so that all individuals can be included, and site difference is controlled for sex differences. Table 7 4 MANOVA test criteria and F approximation s for the hypothesis of no overall site effect, using type IV sums of squares and cross products. Statistic Value F Value DF Pr > F Wilks Lambda .1636 1.44 40 0.0709 Pillais Trace 1.387 1.39 40 0.0817 Hotelling Lawley Trace 2.483 1.48 40 0.0830 In t he MANOVA test, all variables are considered as a single array to test the null hypothesis that site has an effect on the array of cranial measurements against the alternative that there is a site effect. SAS provides three approximations of the F for a MA NOVA, Wilks Lambda, Pillais Trace and the Hotelling Lawley Trace. None of the statistics are significant at the 0.05 alpha level, so there is not enough evidence to reject the null hypothesis (Table 7 4). This indicates that it is reasonable to treat al l sites as a single group for this analysis. Discriminant Function Analysis The goal of the discriminant function analysis is to produce a linear formula that can be used with cranial data to accurately assign sex to Native American skeletal remains recov ered from Florida archaeological sites. To use the function, each variable is multiplied by the corresponding coefficient and the results and constant are summed. If the result is greater than

PAGE 73

73 the sectioning point, the individual is categorized as male. If it is less than the sectioning point, the individual is categorized as female. This analysis presents two discriminant functions. Function 1 uses the same eight variables used by Giles and Elliot in their discriminant function analysis, but the function is calculated using a sample of thirteen male and four female Native Americans from Florida and Southern Georgia (Table 7 5). Function 2 uses the five variables selected based on t tests and preservation rates as described above, and it is calculated from 27 males and 12 females. The coefficients, constants, and sectioning points for this analysis are presented in table 7 5. Table 7 5 Coefficients, group means, and sectioning points for Function 1 and Function 2. Cranial variable Function 1 coefficients F unction 2 coefficients AUB 0.0696568389 WFB 0.0138332675 PAC 0.1046833664 GOL .0717749505 0.0325698854 MDHA 0.1075559971 0.2188344979 XCB 0.1456008386 BBH 0.0620766571 ZYB 0.2196351470 BPL 0.4499153016 UFHT 0.2704566313 MAB 0.167559 3848 Constant 49.19563 21.50963 Male Mean 1.007282992 0.744143627 Female Mean 3.273669725 1.674323160 Sectioning Point 1.133193367 0.470560251 Constant=0 Sectioning Point 48.062436633 21.039069749 Function 1 Accuracy Only 17 individuals included all of the measurements required for Function 1. Because of the small sample size, the function is tested on the same individuals used to develop the function. When applied to the sample of thirteen males a nd four females from which it is developed,

PAGE 74

74 Function 1 correctly assigned sex to all seventeen individuals (Table 7 6). This result is not significantly better than the reported Giles and Elliot result (p = 0.0509). It is also not significantly different t han the Giles and Elliot formula applied to the current sample when observations with missing variables are omitted (p= 0.4848). The accuracy of this function is inflated in this test due to the small sample size. A better indication of the accuracy of the function is given by cross validation. Table 7 6. Accuracy of Function 1, which uses the variables originally used by Giles and Elliot (1963). Sex Number correct Count (n) Percent Overall 14 17 82.35% Male 11 13 84.62% Cross validation Female 3 4 75.00% Overall 17 17 100% Male 13 13 100% Individuals with missing variables omitted Female 4 4 100% Overall 35 45 77.78% Male 2 4 32 75% Missing variables replaced by the average of the male and female means Female 11 13 84.62% Cross validation classifies each observation based on all of the other observations. A discriminant function is calculated for the dataset minus one observation, and the omitted observation is classified using the resulting function. This procedure provides a more realistic estimate of the accuracy of the discriminant function. Using cross validation, a discriminant function analysis using the Giles and Elliot variables correctly classified eleven of thirteen males and three of four females, for a combined accuracy of 82.35%. This is not significantly better than the Giles and Elliot result (p = 0.4130). It is also not significantly different than the Giles and Elliot formula applied to the current sample when observations wi th missing variables are omitted (p= 1.00).

PAGE 75

75 In order to apply Function 1 to all observations, missing variables are replaced with the average of the male and female means for each variable. With missing variables replaced, Function 1 correctly classifies 2 4 of 32 males and 11 of 13 females, for an overall accuracy of 77.78% (Table 7 6). This is not significantly different from the Giles and Elliot function using the recalculated sectioning point and missing variables replaced with the average of the male an d female means (p= 0.5912). Function 2 Accuracy Function 2 is based on measurements from 39 individuals, including 27 males and 12 females. When applied to the sample from which it is developed, Function 2 correctly assigns sex to 26 of 27 males and 10 of 12 females, for an overall accuracy of 92.31% (Table 7 7). This is not significantly different from Giles and Elliots reported accuracy of 86.4% (p= 0.4087). Using cross validation, the formula correctly classifies 10 of 12 females and 23 of 27 males for an overall accuracy of 84.62%. This is not significantly different from the accuracy of the Giles and Elliot function (p= 0.8819). Table 7 7. Accuracy of Function 2, which uses variables selected using t tests and preservation rates. Sex Number correct Count (n) Accuracy Overall 33 39 84.26% Male 23 27 85.19% Cross validation Female 10 12 83.33% Overall 36 39 92.31% Male 26 27 96.30% Individuals with missing variables omitted Female 10 12 83.33% Overall 39 45 86.87% Male 28 32 87.50% Missing variables replaced by the average of the male and female means Female 11 13 84.62% To apply Function 2 to all individuals in the sample, missing variables are replaced by the average of the male and female means for each variable. With the missing var iables replaced,

PAGE 76

76 Function 2 correctly classified 28 of 32 males and 11 of 13 females for an overall accuracy of 86.87%. The accuracy of Function 2 is not significantly different than the Giles and Elliot (1963) formula applied to the current sample using t he recalculated sectioning point and missing variables replaced with the average of the male and female means (p= 1.00).

PAGE 77

77 CHAPTER 8 CONCLUSIONS AND DISC USSION Conclusions The most significant finding of this research is that the function derived by Giles an d Elliot does not accurately sex skeletons from Florida and Georgia unless the sectioning point is adjusted. Using the sectioning point published by Giles and Elliot (1963), the function classifies only 15.38% of females correctly. When the sectioning poin t is adjusted using male and female averages for each variable, the function classifies about 84% of males and females correctly. For Florida and Georgia Native Americans, the sectioning point should be changed to 6513.4 when using Giles and Elliots (1963 ) function 3. The overall goal of this research is to determine if the accuracy of discriminant function analysis for sex determination could be improved by using local or regional populations, and by better variable selection. Previous research has tested the Giles and Elliot (1963) discriminant functions on Finnish crania and developed functions for that population (Kajanoja 1966). Other research has developed discriminant functions for fragmentary archaeological post cranial remains from other areas (Bas s 1995; Black 1978; Krogman and Iscan 1986). None of the previous studies focuses on comparing the established Giles and Elliot methods to new formula developed from fragmentary crania from Southeastern archaeological sites, thus making this research novel This research is also novel in treating all prehistoric Native Americans in Florida and Georgia as a single population, regardless of time period. This study addresses the following hypotheses related to sex determination by discriminant function analysi s (DFA): For archaeological populations, higher accuracy in sex determination by DFA can be achieved drawing a sample from the archaeological population than by using a dissecting room sample. Sex determination by DFA can be accomplished with a smaller num ber of more robust measurements without reducing accuracy.

PAGE 78

78 For the purpose of sex determination by DFA, Florida and Georgia Native Americans can be viewed as a single population, regardless of time period. There is evidence that better discriminant functio ns can be developed using local populations, but the results are not statistical significant possibly because of the small sample size. Sample size is small because of the requirement that individuals include both a relatively intact cranium and pubic bone Both of these skeletal elements are relatively delicate and are seldom preserved intact. Individuals who have both elements intact are rare. There is evidence that sex determination by DFA can be accomplished with a smaller number of more robust variable s than those used by Giles and Elliot without reducing accuracy. Five variables are identified that are preserved in at least 90% of males and females selected for this study. It is possible to apply the discriminant function using these variables to 86.66 % of the sample, compared to just 37.77% for the Giles and Elliot variables. The new function also has higher accuracy than the function developed using the Giles and Elliot variables, although the difference is not significant. Implicit in this research i s the assumption that all of the sites used in this study can be considered a single population. The results of the Multiple Analysis of Variance (MANOVA) suggest that there is no significant difference between individuals from the different sites used in this study. It is, therefore, reasonable to treat all of the individuals from these sites as belonging to a single population. Summery of Statistical Results The results presented in the previous chapter can be summarized as follows. When applied to Florid a and Georgia Native Americans: The Giles and Elliot function 3 and original sectioning point classifies males correctly, but misclassifies females at a significantly higher rate than other methods.

PAGE 79

79 The accuracy of the Giles and Elliot function 3 with the sectioning point recalculated is not significantly different from the Giles and Elliot function applied to black and white individuals. The Giles and Elliot function includes some variables that do not contribute to sex discrimination, and others that pres erve poorly. The accuracy of a new discriminant function developed using the Giles and Elliot variables (GOL, XCB, BBH, ZYB, BPL, UFHT, MAB, and MDHA) is no more accurate than the original Giles and Elliot function. Variables which do contribute to sex dis crimination and preserve well include: GOL, AUB, WFB, PAC, and MDHA. GOL, AUB, WFB, PAC, and MDHA are not significantly different across the samples sites in Florida and Georgia individually, or taken together. The accuracy of a new discriminant function d eveloped using GOL, AUB, WFB, PAC, and MDHA is more accurate than the Giles and Elliot function, but the difference is not significant. Specific Statistical Results Other researchers have found that the Giles and Elliot formula cannot be applied to other p opulations without modification (Giles and Elliot 1963, Kajanoja 1966). While Giles and Elliot suggested several methods for recalculating a sectioning point, Henke (1977) and Calcagno (1981) have determined that those methods are not practical. Using the Giles and Elliot function with the original sectioning point did not produce acceptable results. The function is able to classify 13 of 13 males correctly, but misclassified 2 of 4 females. When missing values are replaced with neutral values, the Giles an d Elliot function classifies all 32 males correctly, but misclassifies 11 of 13 females. The Giles and Elliot function is not a reliable indicator of sex for Florida and Georgia Native Americans using the original sectioning point. The Giles and Elliot fun ction does produce acceptable results when the sectioning point is recalculated.

PAGE 80

80 In terms of accuracy, Function 1 correctly categorizes all individuals in the initial sample, but did not do as well overall in cross validation or when missing values are su bstituted with neutral values. The accuracy of Function 1 is not significantly different from the Giles and Elliot function. Since it is based on the same variables, Function 1 suffers from the same issues of applicability as the Giles and Elliot function. Function 1 could only be applied to 2 of 13 females, and 15 of 32 males. Function 2 is more accurate across the board than the Giles and Elliot function, but the difference is not statistically significant. Accuracies are almost identical in cross valida tion and when missing values are substituted with neutral values. Function 2 offers better applicability than the Giles and Elliot function. Where the Giles and Elliot function could be applied to 2 of 13 females and 15 of 32 males, Function 2 could be app lied to 12 of 13 females and 27 of 32 males. Discussion The goal of the research is to determine if new discriminant functions for sex determination could be developed for Florida and Georgia Native American populations that are better than the functions p resented by Giles and Elliot (1963). While neither Function 1, which uses the Giles and Elliot variables, nor Function 2, which uses more robust variables, did significantly better than the Giles and Elliot function, Function 2 could be used on more than t wice as many individuals without having to substitute missing variables. While neither of the two new functions are significantly better than the Giles and Elliot function with a recalculated sectioning point, there is reason be believe that creating new f unctions might be beneficial. The first has to do with the problems involved with calculating the new sectioning point. The second problem has to do with sample size and the power of the statistics used here.

PAGE 81

81 Several authors point out the problem of using discriminant functions developed for one population on another and conclude the problem of recalculating the sectioning point has no practical solution (Henke 1977, Calcagno1981). Using the midpoint of the male and female mean score for a function produces a workable sectioning point, but requires a sufficient number of male and female individuals whose sex has already been identified. This is a problem akin to opening a crate with the enclosed crowbar. If there are enough individuals whose sex has been ide ntified it is almost as easy to create a new discriminant function as it is to recalculate the sectioning point. The Giles and Elliot function can be made to work across populations by recalculating the sectioning point, but does not take full advantage of the power of DFA to separate groups. Regional differences in shape are going to add within group variance, which is going to work to reduce the effectiveness of the function. A function developed on a group includes regional differences as part of the cal culation. A discriminant function created from the population under study should have a higher ratio of between group to within group variation and be more accurate if the sample size in sufficiently large. If the sample size is too small, the function may be skewed by idiosyncratic variation within the sample. Another problem with small sample sizes is a lack of statistical power. Statistical power is the ability of a hypothesis test to detect a difference. The larger the sample size, the higher the statis tical power, and the smaller the difference a hypothesis test can detect. In the present study, Function 1 is more accurate than the Giles and Elliot function, but there is not enough statistical power for the difference to be significant because of the s mall sample size. If one considers individuals who could not be sexed due to missing variables as incorrect, then Function 1 does significantly better than the Giles and Elliot function (p< 0.0001). Function

PAGE 82

82 1 is able to correctly classify 36 of 45 individ uals (80%), while the Giles and Elliot function correctly classified 15 of 45 individuals (33.33%). This research draws on sites and individuals ranging from the Archaic to the Spanish mission periods, and should be applicable to any remains from that time range found in Florida. This research does not use any Paleoindian remains because they are so rare. Therefore, the functions developed here are not recommended for use in determining the sex of Paleoindians. One might argue that all Native Americans from the Archaic to the Spanish Mission period from Florida and Georgia is too diverse a group to be considered a single population. The MANOVA test results show that there is no significant between site difference for these groups. The Giles and Elliot functi on uses blacks and whites from dissecting room collections, and that function works well for both of those groups. It is hard to argue that blacks and whites are a single population, but that Native Americans from a limited geographic area are not, even if the Native Americans are from a broad time span. The Giles and Elliot discriminant function works well for both blacks and whites, and there is no reason to believe that the functions developed in this research would not work as well for any Native Americ ans recovered from Florida or Georgia. Future Research Because of the small sample size used in this research and the lack of significant improvement over the established Giles and Elliot function, using Function 1 is not recommended without further testin g. It should be possible to generate larger samples using Function 1 because the variables it uses are more robust than the ones used by Giles and Elliot. Because of Function 1s potential for application to a larger number of less well preserved crania an d potentially greater accuracy, it should be tested on a larger sample. Future tests should include as many sites from Florida and Georgia as possible.

PAGE 83

83 This study demonstrates that when using discriminant function analysis missing variables can be replaced with the average of the male and female means with good results, although the limits of this method are not explored. One potential line of future research is to determine how robust discriminant functions are to this method of replacing missing variables One of the purported advantages of discriminant function analysis is that the functions can be used by individuals with relatively little training. This assumption needs to be tested. A study where students with little or no osteology training are asked to measure crania and use discriminant functions to determine sex should be performed. Their results could be compared to a group of students given an identical amount of training in visual methods.

PAGE 84

84 APPENDIX A CRANIAL MEASUREMENT DEFINITIONS Giles and E lliot Measurements The following are the ten measurements used by Giles in Elliot in their 1963 study. The name of the measurement is followed by the measurements abbreviation, Giles and Elliots description of the measurement, the name of the equivalent Howells/FORDISC measurement, and the method used to record the measurement. Detailed descriptions of cranial points and landmarks can be found in Bass (1995), White (1991), and others. Glabello occipital length (g op, GOL): Maximum length of the skull, fro m the most anterior point of the frontal in the midline to the most distant point on the occiput in the midline. This measurement is equivalent to Maximum Cranial Length; the distance of Glabella (g) from Opisthocranion (op) in the mid sagittal plane measu red in a straight line using spreading calipers. The skull is placed on its side for this measurement. The endpoint of the left branch of the caliper is placed on Glabella and held with fingers while the endpoint of the right branch of the caliper is appli ed similarly to the posterior portion of the skull in the mid sagittal plane until the maximum length is obtained (Bass 1971:62; Howells 1973:170; Martin 1956:453; Olivier 1969:128). Maximum width (eu eu, XCB) : The greatest breadth of the cranium perpendi cular to the median sagittal plane, avoiding the supra mastoid crest. This measurement is equivalent to Maximum Cranial Breadth; the maximum width of the skull perpendicular to the mid sagittal plane wherever it is located with the exception of the inferio r temporal line and the immediate area (i.e. the posterior roots of the zygomatic arches) measured with spreading calipers. The Maximum Cranial Breadth is measured with the skull resting either on its base or on the occiput. The two measuring points lie in the same horizontal and frontal planes. The arms of the caliper are placed at the same level while maintaining the hinge joint of the caliper in the mid sagittal

PAGE 85

85 plane. The ends of the caliper are held in each hand and applied to the lateral portions of t he skull, making circular motions until the maximum breadth is obtained. Areas below the squamosal suture are included, where the maximum is sometimes found (Bass 1971:62; Howells 1973:172; Hrdli ka 1952:140; Martin 1956:455; Montagu 1960:44). Basion bregm a height (ba b, BBH) : Cranial height measured from basion to bregma. This measurement is equivalent to Basion Bregma Height ; the direct distance from the lowest point on the anterior margin of the foramen magnum, basion (ba), to bregma (b) is measured with the spreading caliper. The skull is placed on its side and the endpoint of one of the arms of the caliper is placed at the most inferior point of the margin of the foramen magnum in the mid sagittal plane and supported with fingers. Then the endpoint of t he second arm of the caliper is applied to bregma (Bass 1971:62; Howells 1966:6; Martin 1956:459; Olivier 1969:129). Maximum diameter bi zygomatic (zy zy, ZYB) : Maximum width between the lateral surfaces of the zygomatic arches measured perpendicular to th e median sagittal plane. This measurement is equivalent to Bizygomatic Breadth; The direct distance between each zygion (zy), located at the most lateral points of the zygomatic arches measured with a sliding caliper. The skull is placed on its base, and h e blunt points of the caliper are applied to the zygomatic arches and the maximum breadth is recorded (Bass 1971:67; Martin 1956:476). Basion nasion (ba n, BNL) : Distance from basion to nasion. Equivalent to Cranial Base Length: The direct distance from nas ion (n) to basion (ba) measured using the spreading caliper. The skull is placed with the cranial vault down on a cork or sandbag skull ring. The endpoint of the one arm of caliper is applied to nasion (n) while the other is applied to the anterior border of the foramen magnum in the mid sagital plane. This measurement is not taken where anomalous

PAGE 86

86 growths occurred on the anterior border of the foramen magnum (Howells 1966:6; Martin 1956:455). Basion prosthion (ba pr, BPL) : Distance from basion to the most a nterior point on the maxilla in the median sagittal plane. Equivalent to Basion Prosthion Length; The direct distance from basion (ba) to prosthion (pr) measured using a spreading caliper, or sliding caliper where the foramen magnum obstructs the use of sp reading calipers or in crania in which the central incisors have been lost. The fixed point of the sliding caliper or one tip of the spreading caliper is applied to the most anterior point on the alveolar process in the mid sagittal plane. The movable poin t of the sliding caliper or the other tip of the spreading caliper is then brought to the margin of the anterior border of the foramen magnum in the mid sagittal plane (Martin 1956:474). Nasion breadth (al al, NLB) : Maximum breadth of the nasal aperture pe rpendicular to nasal height. Equivalent to Nasal Breadth; the maximum breadth of the nasal aperture measured with a sliding caliper. The points of the instrument are placed on the sharp lateral margins of the nasal aperture at its most lateral curvature. T he measurement is taken perpendicular to the mid sagittal plane and recorded to the nearest millimeter (Bass 1971:68; Howells 1973:176; Martin 1956:479; Montagu 1960:50; Olivier 1969:153). Palate external breadth (ecm ecm, MAB) : The maximum breadth of the palate taken on the outside of the alveolar borders. Equivalent to Maxillo Alveolar Breadth and External Palate Breadth; the maximum breadth across the alveolar borders of the maxilla measured at its widest point between each ectomolare (ecm). The maximum breadth is usually found at the level of the second molars. Using a spreading caliper, both arms of the caliper are applied to the alveolar borders above the tooth row from an anterior position. The points of measurement (ecm) are usually not found on the alveolar processes, but are located on the bony segment above the

PAGE 87

87 second maxillary molars. (Bass 1971:70; Howells 1973:176; Martin 1956:480; Montagu 1960:51). Opisthion forehead length: The maximum distance from opisthion (the midpoint on the posterior bor der of the foramen magnum) to the forehead in the midline. This measurement is not used in this study because it is not used in the Giles and Elliot formulas, nor has it survived in the current literature. Mastoid length (MDH) : The length of the mastoid me asured perpendicular to the plane determined by the lower borders of the orbits and the upper borders of the auditory meatuses (Frankfort horizontal plane). Equivalent to Mastoid Length; The projection of the mastoid process below, and perpendicular to, th e Frankfort horizontal plane in the vertical plane. Both right and left sides are measured using a sliding caliper. The skull is rested on its right side, and the calibrated bar of the caliper is applied just behind the mastoid process, with the fixed flat arm tangent to the upper border of the auditory meatus and pointing (by visual sighting) to the lower border of the orbit. The calibrated bar is perpendicular to the eye ear plane of the skull (i.e., approximately level in the position given). The measuri ng arm is adjusted until it is level with the tip of the mastoid process, using the base of the skull generally, and the opposite mastoid process to control the plane of sighting. (Howells 1966:6, 1973:176; Keen 1950 ). Other Measurements The remaining meas urements are not included in the Giles and Elliot (1962) study, but are used in more recent research, including Bass (1971, 1995), Howells (1973), Buikstra and Ubelaker (1994), and FORDISC 2.0 materials (Ousley and Jantz 1996). The descriptions and techniq ues primarily follow Ousley and Jantz (1996). Biorbital Breadth (ec ec, EKB): The direct distance from one ectoconchion (ec) to the other. This measurement is taken using the sliding caliper (Howells 1973:178).

PAGE 88

88 Interorbital Breadth (d d, DKB): The direct distance between right and left dacryon measured using a sliding caliper (Martin 1956:477). Maxillo Alveolar Length, External Palate Length (pr alv, MAL): The direct distance from prosthion (Hrdli ka's prealveolar point) to alveolon (alv) measured using a Spreading or sliding caliper. A sliding caliper is used only in crania in which the central incisor teeth have been lost. The skull is placed with the cranial vault down on a cork or sandbag skull ring so the base is facing up. A rubber band is applied to the posterior borders of the alveolar arch and the distance measured from the anterior prosthion to the middle of the band in the midsagittal plane (Bass 1971:70; Hrdli ka 1952:146 147; Martin 1956:480). Biauricular Breadth (au au, ALB): The least exterior breadth across the roots of the zygomatic processes, wherever found, measured using a sliding caliper. With the skull resting on the occiput and with the base toward the observer, the outside of the roots of the zygomatic process are measured at their dee pest incurvature, generally slightly anterior to the external auditory meatus, with the sharp points of the caliper. This measurement makes no reference to standard landmarks of the ear region. (Howells 1973:173). Upper Facial Height (n pr, UFHT): The dire ct distance from nasion (n) to prosthion (pr) measured using a sliding caliper. The fixed point of the caliper is placed on nasion and the movable point is applied to the tip of the alveolar border between the upper central incisors. If the alveolar proces s exhibited slight resorption or erosion at the point of prosthion, the projection of the process is estimated when the alveolar process of the lateral incisors is still intact. This measurement is not taken when resorption or erosion is more pronounced (H owells 1966:6; Hrdli ka 1952:143; Martin 1956:476). This differs from Howells NPH in using the inferior border of the alveolar process rather than the most anterior point.

PAGE 89

89 Foramen Magnum Length (ba o, FOL): The direct distance of basion (ba) from opisthio n (o) measured using a sliding caliper. The tips of the instrument are applied on the opposing edges of the border of the foramen magnum along the sagittal plane (Martin 1956:455). Minimum Frontal Breadth (ft ft, WFB): The direct distance between the two f rontotemporale measured with a sliding caliper. With the skill placed on its base, the two endpoints of the caliper are placed on the temporal ridges at the two frontotemporale, and the least distance between both temporal lines on the frontal bone is reco rded (Bass 1971:67; Hrdli ka 1952:142; Martin 1956:457; Olivier 1969:151). Foramen Magnum Breadth (FOB): The distance between the lateral margins of the Foramen magnum at the point of greatest lateral curvature measured with a sliding caliper (Martin, 1956 :459). Upper Facial Breadth (UFBR, fmt fmt): The direct distance between each frontomalare temporale measured using a sliding caliper. The measurement is taken between the two external points on the frontomalar suture (Martin 1956:475). UFBR differs from H owells FMB in that the lateral most points on the suture are used rather than the most anterior points. Nasal Height (n ns, NLH): The direct distance from nasion (n) to nasospinale (ns) measured using a sliding caliper. The direct distance from nasion to the midpoint of a line connecting the lowest points of the inferior margin of the nasal notches is measured (Bass 1971:68; Howells 1966:6; Martin 1956:479; Olivier 1969:153). Orbital Breadth (d ec, OBB): The laterally sloping distance from dacryon (d) to e ctoconchion (ec) measured using a sliding caliper. The left orbit is measured for standardization and practical reasons where available. If the left orbit is damaged or otherwise could not be

PAGE 90

90 measured, the right orbit is measured and the side recorded on t he measurement sheet (Martin 1956:477 478; Howells 1973:175). Orbital Height (OBH): The direct distance between the superior and inferior orbital margins measured using a sliding caliper. Orbital height is measured perpendicular to orbital breadth and simi larly bisects the orbit. The measuring points are located on the opposing margins of the orbital borders. Any notches or depressions on either superior or inferior borders are avoided and the margin is projected when necessary (Bass 1971:69; Martin 1956:47 8; Montagu 1960:51; Olivier 1969:152). Biorbital Breadth (ec ec, EKB): The direct distance from one ectoconchion (ec) to the other measured using a sliding caliper (Howells 1973:178). Interorbital Breadth (d d, DKB): The direct distance between right and l eft dacryon measured with a sliding caliper (Martin 1956:477). Frontal Chord (n b, FRC): The direct distance from nasion (n) to bregma (b) taken in the midsagittal plane using a sliding caliper. The skull is rested on its right side to view the left profil e of the frontal region. The tips of the instrument are placed on the bone surface or at the level of this surface and not in a suture or other depression (Howells 1973:181; Martin 1956:465). Parietal Chord (b l, PAC): The direct distance from bregma (b) t o lambda (l) taken in the midsagittal plane measured with a sliding caliper. The skull is left in the same position used to measure the Frontal Chord (above). The tips of the instrument are placed on the bone surface or at the level of this surface and not in a suture or other depression (Howells 1973:182; Martin 1956:466 ).

PAGE 91

91 Occipital Chord (l o, OCC): The direct distance from lambda (l) to opisthion (o) taken in the midsagittal plane measured using a sliding caliper. The skull is left in the same position used to measure the Frontal Chord (above). The tips of the instrument are placed on the bone surface or at the level of this surface and not in a suture or other depression. The point of the movable branch of the caliper is placed against the posterior bor der of the foramen magnum and held in place with the right thumb (Howells 1973:182; Martin 1956:466 ).

PAGE 92

92 APPENDIX B TABLE OF SITES Site name Sex overall Catalog # Golf Course site (8Br44), Brevard county, Florida. Melbourne Man male 331422 Bay Pines (8 Pi64) Pinellas county, Florida. (Gallagher and Warren 1975) 24 burials male s0272 female 377440 377475 377503 male 377427 377428 377441 377465 377466 377467 377471 377478 377481 377484 377489 377502 377507 377513 377602 Canaveral (8Br85), Brevard county, Florida. The Canaveral site includes the Burns and Fuller mounds. Rouse (1951) places the mounds in the Malabar I, Malabar I, and Malabar II cultures based on ceramic types. Malabar is an Indian River variation of St. Johns (500 B.C. AD1763). 377605 Cannon's Point, St. Simons Island, Glynn County, Georgia. Late Wilmington culture ceramic assemblage, radiocarbon dated to 99075 (A.D. 960) (Martinez 1975; Milanich 1977) male s0418 Casey Key (8So17) Sarasota County, Florida. A Manasota Weeden Island culture mound dating from ca. A.D. 250 750. (Bullen and Bullen 1976). female 92733 female s0300 s0301 s0302 s0311 male s0299 s0303 s0304 s0305 Couper Field/Indian Field, Glynn County, Georgia. St. Simons Island. Late Pre Columbian/early Spanish colonia l/mission period village. s0309 Garfield Site (9Br57) Bartow County, Georgia. Kellog culture (Early Woodland, ca. 600BC AD 100) village on Etowah River. 18 Burials. (Milanich 1975; Wood and Bowan 1995:8). male s0325 female 97533

PAGE 93

93 Palmer Site (8So2a) Sarasota County, Florida. Gulf coast. Archaic, Early Woodland, Middle Woodland, and Late Woodland. Final MNI 429. (Bullen and Bullen 1976, Miller 1974, Almy and Luer 1993, Kozuch 1998 Hutchinson 2004:49) female 97533 Palmer Site (8So2a) Sarasota County, Florida. Gulf coast. Archaic, Early Woodland, Middle Woodland, and Late Woodland. Final MNI 429. (Bullen and Bullen 1976, Miller 1974, Almy and Luer 1993, Kozuch 1998 Hutchinson 2004:49) male 97555 female 373499 373552 male 373493 Perico Island (8Ma6) Manatee County, Florida. Glades I pottery with some Biscayne and some Deptford (Willey 1949:172 182). 373530 Unknown s0385 female s0361 s0368 male s0362 s0365 s0367 Taylor Mound, St. Simons Island, Georgia. Historic period (AD1600 1650) ceremonial mound. 11 total burials. Same sociocultural population as Couper Field (Wallace 1975). s0371

PAGE 94

94 LIST OF REFERENCES Almy, Marion M ., and George M. Luer 1993 Guide to the Prehistory of Historic Spanish Point. Historic Spanish Point, Osprey, Florida. Austin, Robert J., and Michael Russo 1989 Limited Excavations at the Catfish Creek Site (8So608), Sarasota County, Florida. Piper Archaeo logical Research, Inc., St. Petersburg, Florida. Anderson, David G., and Kenneth E. Sassaman 2004 Early and Middle Holocene Periods, 9500 to 3750 B.C. In Southeast, edited by Raymond D. Fogelson, pp. 87 100. Handbook of North American Indians, Vol. 14, Wil liam C. Sturtevant, general editor, Smithsonian Institution, Washington, D.C. Bass, William M. 1971 Human Osteology: A Laboratory and Field Manual of the Human Skeleton. 2nd ed. Missouri Archaeological Society, University of Missouri, Columbus. 1987 Human Osteology: A Laboratory and Field Manual of the Human Skeleton. 3rd ed. Missouri Archaeological Society, University of Missouri, Columbus. 1995 Human Osteology: A Laboratory and Field Manual of the Human Skeleton. 4th ed. Missouri Archaeological Society, University of Missouri, Columbus. Bense, Judith A. 1994 Archaeology of the Southeastern United States: Paleoindian to World War I. Academic Press, San Diego. Birkby, Walter H. 1966 An Evaluation of Race and Sex Identification from Cranial Measurements. Ame rican Journal of Physical Anthropology 24:21 28. Black, Thomas K. 1978 A New Method for Assessing the Sex of Fragmentary Skeletal Remains: Femoral Shaft Circumference. American Journal of Physical Anthropology 48:227 232. Boas, Franz 1910 Report Present ed to the 61st Congress on Changes in Bodily Form of Descendants of Immigrants. Government Printing Office, Washington, D.C. 1912 Changes in the Bodily Form of Descendants of Immigrants. American Anthropologist 14:530 562. Brothwell, Donald R. 1981 Digging Up Bones: The Excavation, Treatment and Study of Human Skeletal Remains. 3rd ed. Cornell University Press, Ithaca, New York.

PAGE 95

95 Buikstra, Jane E., and Douglas H. Ubelaker (editors) 1994 Standards for Data Collection from Human Skeletal Remains: Proceedings o f a Seminar at the Field Museum of Natural History. Arkansas Archaeological Survey Research Series 44. Fayetteville, Arkansas. Bullen, Ripley P. 1958 Six Sites Near the Chattahoochee River in the Jim Woodruff Reservoir Area, Florida. In River Basin Survey Papers edited by Frank H. H. Roberts, Jr., pp. 315 357. Bureau of American Ethnology, Bulletin 169. Smithsonian Institution, Washington, D.C. Bullen, Ripley P., and Adelaide K. Bullen 1976 The Palmer Site. Florida Anthropological Society, Gainesville. Iss ued as Florida Anthropologist 29, no.2, part 2. Calcagno, James M. 1981 On the Applicability of Sexing Human Skeletal Material by Discriminant Function Analysis. Journal of Human Evolution 10:189 198. Caldwell, Joseph R. 1952 The Archeology of Eastern Geo rgia and South Carolina. In Archeology of Eastern United States edited by James B. Griffin, pp. 312 321. University of Chicago Press, Chicago. Clausen, Carl J., A. D. Cohen, Cesare Emiliani, J. A. Holman, and J. J. Stipp 1979 Little Salt Spring, Florida: A Unique Underwater Site. Science 203:609 614. DePratter, Chester B. 1979 Ceramics. In The Anthropology of St. Catherines Island, The Refuge Deptford Mortuary Complex edited by D. H. Thomas and C. S. Larsen, pp. 109 132. Anthropological Papers 56 (1). American Museum of Natural History, New York. DiBennardo, Robert, and James V. Taylor 1979 Sex Assessment of the Femur: A Test of a New Method. American Journal of Physical Anthropology 50:635 637. 1982 Classification and Misclassification in Sexing the Black Femur by Discriminant Function Analysis. American Journal of Physical Anthropology 58:145 151. 1983 Multiple Discriminant Function Analysis of Sex and Race in the Postcranial Skeleton. American Journal of Physical Anthropology 61:305 314. Dittrick, J ean and Judy M. Suchey 1986 Sex Determination of Prehistoric Central California Skeletal Remains Using Discriminant Analysis of the Femur and Humerus. American Journal of Physical Anthropology 70:3 9.

PAGE 96

96 Doran, Glen H. 2002 Windover: Multidisciplinary Inves tigations of an Early Archaic Florida Cemetery University Press of Florida, Gainesville. Edwards, William Edward 1954 The Helen Blazes Site of Central eastern Florida: A study in Method Utilizing the Disciplines of Archaeology, Geology, and Pedology. Unpu blished Ph.D. dissertation, Department of Anthropology, Columbia University, New York. Espenshade, Christopher T. 1983 Ceramic Ecology and Aboriginal Household Pottery Production at the Gauthier Site, Florida. Unpublished Masters thesis, Department of Ant hropology, University of Florida, Gainesville. Fisher, Ronald A. 1936 The Utilization of Multiple Measurements in Taxonomic Problems. Annals of Eugenics 7:179 188. Gallagher, John C., and Lyman O. Warren 1975 The Bay Pines Site, Pinellas County, Florida. Florida Anthropologist 28:96 116. Giles, Eugene and Orville Elliot 1962 Race Identification from Cranial Measurements. Journal of Forensic Science 7:149 184. 1963 Sex Determination by Discriminant Function Analysis of Crania. American Journal of Physical A nthropology 21:53 68. Giles, Eugene 1964 Sex Determination by Discriminant Function Analysis of the Mandible. American Journal of Physical Anthropology 22:129 136. 1970 Discriminant Function Sexing of the Human Skeleton. In Personal Identification in Mass Disasters, edited by Thomas Dale Stewart, pp. 99 107. National Museum of Natural History, Smithsonian Institution, Washington, D.C. Griffin, John W. 1949 Notes on the Archaeology of Useppa Island. Florida Anthropologist 2:92 93. 1979 The Origin and Dispers ion of American Indians in North America. In The First Americans : Origins, Affinities, and Adaptations edited by W.S. Laughlin and A.B. Harper, pp. 43 56. Gustav Fischer, New York. Hally, David J., and Robert C. Mainfort, Jr. 2004 Prehistory of the Easter n Interior after 500 B.C. In Southeast, edited by Raymond D. Fogelson, pp. 265 285. Handbook of North American Indians, Vol. 14, William C. Sturtevant, general editor, Smithsonian Institution, Washington, D.C.

PAGE 97

97 Harris, Richard J. 1985 A Primer on Multivaria te Statistics. 2nd ed. Academic Press, Orlando. Hemmings, E. Thomas 1975 An Archaeological Survey of the South Prong of the Alafia River, Florida. Florida Anthropologist 28:41 51. Henke, Winfried 1977 On the Method of Discriminant Function Analysis for Se x Determination of the Skull. Journal of Human Evolution 6:95 100. Hoffman, Paul E. 1994 Narvez and Cabeza de Vaca in Florida. In The Forgotten Centuries: Indians and Europeans in the American South, 1521 1704 edited by Charles Hudson and Carmen Chaves T esser, pp. 50 73. University of Georgia Press, Athens. Holman, Darryl J., and Kenneth A. Bennett 1991 Determination of sex from arm bone measurements. American Journal of Physical Anthropology 84:421 426. Hooton, Earnest A. 1930 The Indians of Pecos Puebl o: a study of their skeletal remains. Department of Archaeology, Phillips Academy, Andover, Massachusetts. Yale University Press, New Haven. Howells, William W. 1966 Craniometry and Multivariate Analysis: The Jomon Population of Japan: A Study by Discrimi nant Analysis of Japanese and Ainu Crania Papers of the Peabody Museum of Archaeology and Ethnology 57:1 43. Harvard University Press, Cambridge. 1973 Cranial Variation in Man. A Study by Multivariate Analysis of Patterns of Difference Among Recent Human Populations. Papers of the Peabody Museum of Archaeology and Ethnology 67. Harvard University Press, Cambridge. 1989 Skull shapes and the map: Craniometric analysis in the dispersion of modern homo. Papers of the Peabody Museum of Archaeology and Ethnology 79. Harvard University Press, Cambridge. Hrdli ka, Ales 1940 Catalogue of human crania in the United States National Museum collections: Indians of the Gulf States. Proceedings of the U.S. National Museum 87, no 3076:315 364. 1952 Practical Anthropometry 4th ed., edited by T.D. Stewart. Wistar I nstitute of Anatomy and Biology, Philadelphia.

PAGE 98

98 Hunt, David R., and John Albanese 2005 History and Demographic Composition of the Robert J. Terry Anatomical Collection. American Journal of Physical Anthropology 127:406 417. Hutchinson, Dale L. 2004 Bioarcha eology of the Florida Gulf Coast: Adaptation, Conflict, and Change. University Press of Florida, Gainesville. Iscan, Mehmet Y., and Patricia Miller Shaivitz 1984 Determination of Sex from the Tibia. American Journal of Physical Anthropology 64:53 57. Joh nson, Francis E., and Charles E. Snow 1961 The reassessment of the age and sex of the Indian Knoll skeletal population: Demographic and methodological aspects. American Journal of Physical Anthropology 19: 237 244. Kajanoja, Pauli 1966 Sex determination of Finnish Crania by Discriminant Function Analysis. American Journal of Physical Anthropology 24:29 34. Kelly, Jennifer A., Robert H. Tykot and Jerald T. Milanich 2006 Evidence for Early Use of Maize in Peninsular Florida. In Histories of Maize: Multidisci plinary Approaches to the Prehistory, Linguistics, Biogeography, Domestication, and Evolution of Maize edited by John Staller, Robert Tykot and Bruce Benz, pp. 249 261. Elsevier Academic Press, Boston. Kendall, Maurice G. 1957 A Course in Multivariate An alysis. Hafner Publishing Co., New York. Key, Patrick J. 1983 Craniometric Relationships Among Plains Indians. Report of Investigations 34. University of Tennessee, Department of Anthropology, Knoxville. Konigsberg, Lyle W., and Samantha M. Hens 1998 Use of Ordinal Categorical Variables in Skeletal Assessment of Sex from the Cranium. American Journal of Physical Anthropology 107:97 112. Kozuch, Laura 1998 Faunal Remains from the Palmer Site (8So2), with a focus on shark remains. The Florida Anthropologist 52:208 222. Krogman, Wilton M. 1962 The Human Skeleton in Forensic Medicine. Charles C. Thomas, Springfield, Illinois.

PAGE 99

99 Krogman, Wilton M., and Mehmet Yasar Iscan 1986 The Human Skeleton in Forensic Medicine 2nd ed Charles C. Thomas, Springfield, Illinois Loucks, L. Jill 1976 Early Alachua Tradition Burial Ceremonialism: The Henderson Mound, Alachua County, Florida. Unpublished Masters thesis, Department of Anthropology, University of Florida. Gainesville. Luer, George M., and Marion M. Almy 1979 Three A boriginal Shell Middens on Longboat Key, Florida: Manasota Period Sites of Barrier Island Exploitation. Florida Anthropologist 32:33 45. 1982 A Definition of the Manasota Culture. Florida Anthropologist 35:34 58. Luer, George M., Marion M. Almy, Dana Ste. Claire, and Robert Austin 1987 The Myakkahatchee Site (8So397), A Large Multi Period Inland from the Shore Site in Sarasota County, Florida. Florida Anthropologist 40:137 153. Manly, Bryan F.J. 1994 Multivariate Statistical Methods: A Primer Chapman & Hal l, New York. Martin, Rudolf 1956 Lehrbuch der Anthropologie in Systematischer Darstellung 3rd ed. Gustav Fischer, Stuttgart. Martinez, Carlos A. 1975 Culture Sequence on the Central Georgia Coast, 1000 B.C. 1650 A.D. Unpublished Master's thesis, Departm ent of Anthropology, University of Florida, Gainesville. Meiklejohn, Christopher 1972 Biological concomitants of a model of band society. In International Conference on the Prehistory and Paleoecology of Western North America Arctic and Subarctic edited b y S. Raymond and P. Schledermann, pp. 133 141. Archaeological Association, Department of Archaeology, University of Calgary, Calgary. Milanich, Jerald T. 1975 Fabric Impressed Pottery Site, Bartow County, Georgia. Early Georgia Newsletter, Society for Geo rgia Archaeology 11:4 5. 1976 Georgia Origins for the Alachua Tradition. Bulletin No 5. Florida Bureau of Historic Sites and Properties, Tallahassee. 1977 A Chronology for the Aboriginal Cultures of St. Simons Island, Georgia. Florida Anthropologist 30:13 4 144.

PAGE 100

100 1994 Archaeology of Precolumbian Florida University Press of Florida, Gainesville. 1998 Florida's Indians From Ancient Time to the Present University Press of Florida, Gainesville. 2004a Prehistory of Florida After 500 B.C. In Southeast, edited by Raymond D. Fogelson, pp. 191 203. Handbook of North American Indians, Vol. 14, William C. Sturtevant, general editor, Smithsonian Institution, Washington, D.C. 2004b Timucua. In Southeast, edited by Raymond D. Fogelson, pp. 219 228. Handbook of North Amer ican Indians, Vol. 14, William C. Sturtevant, general editor, Smithsonian Institution, Washington, D.C. Milanich, Jerald T., and Charles H. Fairbanks 1978 Florida Archaeology. Academic Press, New York. Miller, Carl F. 1950 Early Cultural Horizons in the So utheastern United States. American Antiquity 15(4):273 288. Miller, James J. 1974 An Archaeological Survey of the Palmer Oaks Tract in Sarasota County. Miscellaneous Project Report Series, No. 20. Florida Bureau of Historic Sites and Properties, Tallahasse e. Montagu, M.F. Ashley 1960 A Handbook of Anthropometry. Charles C. Thomas, Springfield, Illinois. Padgett, Thomas J. 1976 Hinterland Exploitation in the Central Gulf Coast Manatee Region during the Safety Harbor Period. Florida Anthropologist 29:39 48. O livier, Georges 1969 Practical Anthropology. Charles C. Thomas, Springfield, Illinois. Ousley, Stephen D., and Richard L. Jantz 1996 FORDISC 2.0: Personal Computer Forensic Discriminant Functions. Forensic Anthropology Center, Department of Anthropology, University of Tennessee, Knoxville. Phenice, T.W. 1969 A Newly Developed Visual method of Sexing the Os Pubis. American Journal of Physical Anthropology 30:297 302. Quitmyer, Irv 1998 Zooarchaeological Indicators of Habitat Exploitation and Seasonality fr om the Shell Ridge Midden, Palmer Site (8So2), Osprey, Florida. Florida Anthropologist 51:193 203.

PAGE 101

101 Rathbun Ted A. and Jane E. Buikstra 1984 Human Identification: Case Studies in Forensic Anthropology. Charles C. Thomas, Springfield, Illinois. Rao, C Radha krishna 1952 Advanced Statistical Methods in Biometric Research John Wiley and Sons, New York. Rogers, Tracy L., and Shelley R. Saunders 1994 Accuracy of sex determination using morphological traits of the human pelvis. Journal of Forensic Science 39:10 47 1056. Ross, Ann H., Douglas H. Ubelaker, and Anthony B. Falsetti 2002 Craniometric Variation in the Americas. Human Biology 74:807 818. Rouse, Irving 1951 A Survey of Indian River Archeology, Florida Yale University Publications in Anthropology 44, New Haven. Rupp, Reynold J. 1980 The Archaeology of Drowned Terrestrial Sites: A Preliminary Report. Bulletin No 6:35 45. Florida Bureau of Historic Sites and Properties, Tallahassee. SAS Institute, Inc. 1979 SAS User's Guide. SAS Institute Inc., Raleigh, N orth Carolina. Sassaman, Kenneth E., and David G. Anderson 2004 Late Holocene Period, 3750 to 650 B.C. In Southeast, edited by Raymond D. Fogelson, pp. 115 127. Handbook of North American Indians, Vol. 14, William C. Sturtevant, general editor, Smithsonia n Institution, Washington, D.C. Sigler Eisenberg, Brenda, Ann Cordell, Richard Estabrook, Elizabeth Horvath, Lee A. Newsom, and Michael Russo 1985 Archaeological Site Types, Distribution, and Preservation within the Upper St. Johns River Basin, Florida. Miscellaneous Project Report 27. Department of Anthropology, Florida Museum of Natural History, Gainesville. Snow, Charles E. 1948 Indian Knoll Skeletons of Site Oh 2, Ohio County, Kentucky. Reports in Anthropology 4, no. 3, part 2. Department of Anthropo logy, University of Kentucky, Lexington. Snow, Charles E., Steve Hartman, Eugene Giles, and Fontaine Young 1979 Sex and Race Determination of Crania by Calipers and Computer: A test of the Giles and Elliot Discriminant Functions in 52 Forensic Science Cas es. Journal of Forensic Sciences 24:448 460.

PAGE 102

102 Stewart, Thomas Dale 1946 A Re examination of the Fossil Human Skeletal Remains from Melbourne, Florida, with Further Data on the Vero Skull. Smithsonian Miscellaneous Collections 106(10): 1 28. 1948 Medico lega l aspects of the skeleton: Age, sex, race and stature. American Journal of Physical Anthropology 6:315 321. 1979 Essentials of Forensic Anthropology Charles C. Thomas, Springfield, Illinois. Stirling, Matthew W, 1935 Smithsonian Archaeological Projects C onducted under the Federal Emergency Relief Administration, 1933 1934. Annual Report of the Smithsonian Institution 1934, pp. 371 400. Washington, D.C. Stoltman, James B. 2004 History of Archaeological Research. In Southeast, edited by Raymond D. Fogelson, pp. 14 30. Handbook of North American Indians, Vol. 14, William C. Sturtevant, general editor, Smithsonian Institution, Washington, D.C. Taylor, James V., and Robert DiBennardo 1982 Determination of Sex of White Femora by Discriminant Function Analysis: Forensic Science Applications. Journal of Forensic Sciences 27:417 423. Wauchope, Robert W. 1966 Archaeological Survey of Northern Georgia with a Test of Some Cultural Hypotheses. American Antiquity 35; Memoirs of the Society for American Archaeology 21. Washington, D.C. Waller, Ben I. 1969 Paleoindian and other artifacts from a Florida stream bed. Florida Anthropologist 22: 37 39. Watts, William A. 1969 A pollen diagram from Mud Lake, Marion County, north central Florida. Geological Society of America Bu lletin 80: 631 642. 1971 Post Glacial and Interglacial Vegetation History of Southern Georgia and Central Florida. Ecology 52:676 689. 1975 A late Quaternary record of vegetation from Lake Annie, south central Florida. Geology 3: 344 346. Watts, William A. and Barbara C.S. Hansen 1988 Environments of Florida in the Late Wisconsin and Holocene. In Wet Site Archaeology edited by Barbara A. Purdy, pp. 307 323. Telford Press, West Caldwell, New Jersey.

PAGE 103

103 Wallace, Ronald L. 1975 An Archaeological, Ethnohistorica l and Biochemical Investigation of the Guale Aborigines of the Georgia Coastal Strand. Unpublished Ph.D. dissertation, Department of Anthropology, University of Florida. Gainesville. White, Tim D. 1991 Human Osteology. Academic Press, San Diego. Widmer, Ra ndolph J. 1988 The Evolution of the Calusa: A Nonagricultural Chiefdom on the Southwest Florida Coast. University of Alabama Press, Tuscaloosa. Willey, Gordon R. 1949 Archeology of the Florida Gulf Coast Smithsonian Miscellaneous Collections 113. Smithso nian Institution, Washington, D.C. 1954 Burial Patterns in the Burns and Fuller Mounds, Cape Canaveral, Florida. Florida Anthropologist 7:79 90. Williams, Shanna E. 2004 Dental Asymmetry Through Time in Coastal Florida and Georgia. Unpublished Masters the sis, Department of Anthropology, University of Florida, Gainesville. Williams, Wilma B. 1983 Bridge to the Past: Excavations at the Margate Blount Site. Florida Anthropologist 36:142 153. Wilmsen, Edwin N. 1965 An Outline of Early Man Studies in the Unite d States. American Antiquity 31:172 192. Wood, W. Dean, and William R. Bowen 1995 Woodland Period Archaeology in Northern Georgia University of Georgia Laboratory of Archaeology Series Report Number 33. Athens. Wood, W. Dean, Dan T. Elliot, Teresa P. Rudo lph, and Dennis B. Blanton 1986 Prehistory in the Richard B. Russell Reservoir: The Archaic and Woodland Periods of the Upper Savannah River. The Final Report of Data Recovery at the Anderson and Elbert County Groups: 38AN8, 38AN126, 9EB17, and 9EB21. Russ ell Papers. Interagency Archaeological Services, National Park Service, Atlanta. Woodbury, George n.d. Preliminary Report on Excavation of Mortuary Mounds in Brevard County, Florida East Coast. Manuscript on file, Department of Anthropology (former Bureau American Ethnology), Smithsonian Institution, Washington, D.C. (Archaeological Report of CWA Project 5 F 70, December 18, 1933 February 15, 1934.)

PAGE 104

104 Zahler, James W., Jr. 1976 A Morphological Analysis of a Protohistoric Historic Skeletal Population from St. Simons Island, Georgia. Unpublished Master's thesis, Department of Anthropology, University of Florida, Gainesville.

PAGE 105

105 BIOGRAPHICAL SKETCH Michael McGinnes, who grew up in Gainesville, Florida, earned an Associate of Science degree in photography from the Southeast Center for Photo/Graphic Studies at Daytona Beach Community College, an Associate of Arts degree from Santa Fe Community College, and a bachelors degree from the University of Florida. He has been employed as a professional archaeologist since 1995 and specializes in mortuary archaeology. He has conducted field projects in Panama, Florida, Georgia, South Carolina, Delaware, Oklahoma, Texas, Alabama, Maryland, Virginia, and the District of Columbia. At the National Museum of Natural History, Smit hsonian Institution, he worked as the bibliographic research assistant for the Southeast volume of the Handbook of North American Indians and was an intern for the collections manager of the physical anthropology collections. He is married to Dr. Ruth Tro colli, the archaeologist of the District of Columbia.