AMERICAN JOURNAL OF PHYSICAL ANTHROPOLOGY 000:000–000 (2012)

A New Method to Estimate Adult Age-at-Death

Using the Acetabulum
Stephanie E. Calce*

Department of Anthropology, University of Victoria, Victoria, BC, Canada V8N 1M5

KEY WORDS bioarchaeology; forensic anthropology; age estimation; stepwise multiple regression

ABSTRACT Rissech et al. (J Forensic Sci 51 (2006) descriptions of age categories instead of photos. Three stat-
213–229) described a method to estimate age-at-death of istically significant characteristics highly correlated with
adult males using seven traits of the fused acetabulum. age (P < 0.05) are capable of estimating age-at-death with
This study simplifies Rissech et al.’s technique and extends 81% accuracy, both sexes combined. For misidentified indi-
its application to adult females. Rissech et al.’s original viduals the tendency was to underestimate age. Results of
scoring method was applied to a sample of 100 known-aged both intraobserver error testing and inter-rater reliability
adults, three variables were selected based on stepwise demonstrated a moderate to substantial agreement in scor-
multiple regression, and ages were collapsed into three ing between observers. When estimating the degree of de-
broad ranges: young adult (17–39 years), middle adult (40– velopment of features osteophyte development of the ace-
64 years), and old adult (651 years). The revised method tabular rim was the most inconsistent between observers.
was applied to 249 new known-aged individuals from two The revised acetabular method shows promise in estimat-
other samples. To minimize observer bias, highlight the ing age for adults, particularly for those over the age of 65
most critical traits, and encompass more age-related varia- years. Am J Phys Anthropol 000:000–000, 2012. V 2012
C

tion, unique digital renderings accompany morphological Wiley Periodicals, Inc.

Estimating age-at-death with accuracy is important in England. Using a Bayesian calculation procedure for de-
bioarchaeological and forensic investigations because it velopmental stages of seven age-related traits of the
is a component of both (1) the biological profile, which fused acetabulum, Rissech’s work produced encouraging
establishes the basic physical identity of a person, results in distinguishing stages within the senior age
including sex, stature, and ancestral background and (2) category. Because estimating the age of those over 60
the demographic profile, which helps to form the sample years is notoriously difficult for many morphological age
age distribution. It is well known that activity, genetics, estimation techniques, degeneration of the acetabulum
health, and overall lifestyle affect the rates at which has a valuable role to play in the identification of the
age-related changes are expressed in the skeleton (Is _ can, elderly. Stull and James (2010) recently evaluated
1989; Boldsen et al., 2002). Because individual growth Rougé-Maillart et al.’s (2004) original aging criteria and
rates vary, age-at-death is always reported as a range. found that although acetabular features demonstrate the
To account for variation in the aging process, continual capability to estimate age, an improved coding system
testing of skeletal age-at-death techniques is essential. and percent correct classification approach is necessary
One area of the skeleton that has frequently been used to determine accuracy and precision.
for aging adults is the os coxae. Methods using the auric- Further to the work by Stull and James (2010), a pre-
ular surface and pubic symphysis are both well docu- liminary study was conducted (Calce and Rogers, 2011)
mented (Lovejoy et al., 1985; Suchey and Katz, 1998; to test Rissech et al.’s (2006) adult age estimation
Buckberry and Chamberlain, 2002) and employed often method using the acetabulum on a sample (n 5 100) of
in skeletal aging. Recently, another area of the os coxae, males from the Grant Collection (GRO) housed at the
the acetabulum, has shown promise as an effective age University of Toronto. The objective of such preliminary
indicator (Rissech et al., 2006; Rougé-Maillart et al., research by the current author was to determine the ac-
2007; Stull and James, 2010). Due in part to its postde- curacy and repeatability of several aging criteria pro-
positional preservation in comparison to other osseous posed by Rissech et al. (2006) on a North American pop-
regions, acetabular patterns of observable degeneration ulation. Three main problems were identified. First, that
were first reported in 2004 by Rougé-Maillart et al. who the statistical software program associated with Ris-
identified four potentially age-related skeletal features:
the acetabular rim, the acetabular fossa, the lunate sur-
face, and apical activity. Rougé-Maillart et al. (2004) pro-
posed that these acetabular aging traits complemented
*Correspondence to: Stephanie Calce, Department of Anthropol-
Lovejoy et al.’s (1985) auricular surface technique to esti- ogy, University of Victoria, P.O. Box 3050, STN CSC, Victoria, BC,
mate age, both of which were tested by Rougé-Maillart Canada V8W 3P5. E-mail: scalce@uvic.ca
on a small modern sample of known age European males
(n 5 30) from Spain and France. The high correlation Received 17 February 2011; accepted 28 December 2011
between osseous changes and age-at-death led to further
research (Rissech et al., 2006, 2007) on a larger Euro- DOI 10.1002/ajpa.22026
pean sample (n 5 394) of adult males from 18th to 20th Published online in Wiley Online Library
century cemetery collections in Portugal, Spain, and (wileyonlinelibrary.com).

C 2012
V WILEY PERIODICALS, INC.
2 S.E. CALCE

sech’s (2006) technique required considerable study and

familiarity before operating; misuse of the program
resulted in erroneous age estimates. The second issue
was intra- and interobserver error. Macroscopic morpho-
logical changes for variables two (acetabular rim shape)
and seven (fossa porosity) were difficult to distinguish
and score precisely because descriptions of lesions or
specific regional changes are unclear. This resulted in
subjective coding and high dissimilarity in scoring
between tests where states of rim shape and fossa poros-
ity were scored differently in over half of the cases
(Calce and Rogers, 2011). Lastly, of the seven age-related
traits proposed by Rissech et al. (2006), four did not cor-
relate with chronological age in the GRO collection sam-
ple. The purpose of this current research is to create a
more user-friendly and comprehensible scoring method
using morphological traits of the acetabulum that will
reduce data collection time and maintain the accuracy
and precision of estimating age for adults. At this junc-
ture, emphasis is on the need to clarify some of the more
ambiguous trait descriptions and to improve reproduci-
bility of scoring age-associated features. This study sim-
plifies Rissech et al.’s (2006) technique and extends its
application to adult females. Broad age divisions are
used to ensure that the revisions are capable of captur-
ing age-related changes of the acetabulum between geo-
graphical populations. The revised method will be partic- Fig. 1. Revised definitions of Variables 1–3: acetabular
groove, denoted by curved line; osteophyte development,
ularly useful in bioarchaeological analyses of in situ denoted by circle; and apex growth, denoted by arrow. (1) Ace-
remains. tabular groove: found along the rim, the groove becomes more
pronounced with increased age and covers a larger surface area
MATERIALS AND METHODS between the lunate surface and the acetabular rim. (2) Osteo-
phyte development of the acetabular rim: osteophytes develop
Several steps were employed in the development of the below the anterior inferior iliac spine and travel along the ace-
present method. Rissech et al.’s (2006) seven acetabular tabular rim to invade the superior area of the lunate surface as
age increases. (3) Apex growth: with age, a sharp spicule forms
criteria were originally scored and tested on 100 male
at the posterior horn of the lunate surface, and osteophytic de-
individuals from the GRO (Calce and Rogers, 2011). velopment enters the acetabular notch.
Results of this previous research formed the basis of the
first step in this study to simplify the technique (con-
densing variables from seven to three) by stepwise mul-
tiple regression (to be discussed later herein; see also (P-value [0.05) were eliminated to find the best fitting
Results). To reduce intra- and interobserver error, the model to predict age for the given variables, i.e., could
author refined three age-related variables and associated not reject the null hypothesis (Ho) for such variables
descriptions by seriating male and female ossa coxae (where Ho5 acetabular trait correlation with age). A
and inspecting broad age-related changes across three mixed stepwise selection procedure was used because it
adult age phases. Young, middle, and older aged morpho- results in the most parsimonious model and is essential
logical traits were examined for sex-specific characteris- to identify the minimum number of variables needed to
tics. Through this assessment, adult age ranges were predict age, which in this case were three.
defined and form the revised acetabulum method in the
identification of unknown individuals, which was later
Defining variables and categories of age
tested on two new known-age and -sex samples of male
and female ossa coxae (n 5 249) from the University of Three traits (as determined by stepwise multiple
Tennessee (UTen) and University of New Mexico (UNM). regression), accounted for most of the variation associ-
ated with age: acetabular groove, rim porosity, and apex
Reducing variables activity (see Results). Due to ambiguities with Rissech et
al.’s (2006) original descriptions, these traits were clari-
NCSS 2004 (Hintze, 2006) was used to calculate step- fied and redefined for this research (Fig. 1). Variable 3
wise multiple regression on the GRO data to determine from Rissech et al. (2006; acetabular rim porosity) was
which of Rissech et al.’s (2006) original seven variables changed to ‘‘osteophyte development of the acetabular
were contributing (or affecting) age estimation. Multiple rim’’ to encompass a greater range of variation observed
regression statistics are appropriate because they use in this area, and is designated as Variable 2 in the re-
more than one variable (x1, x2, . . ., xi) to estimate the vised method. To evaluate the usefulness of these varia-
value of ‘‘y,’’ that is, determine which variables (traits of bles to accurately assign unknown individuals as belong-
the acetabulum) are contributing to estimate age (Manly, ing to broad age categories (young, middle, and old),
2005). All possible values of ‘‘x’’ (seven variables) were ossa coxae of 90 males from the GRO and 39 females
listed to estimate ‘‘y’’ (age). In this study, a stepwise from the William M. Bass Donated Skeletal Collection
selection was performed and all ‘‘x’’ variables (traits of (UTen) were seriated based on known ages and exam-
the acetabulum) not strongly correlated with age ined macroscopically to identify age-related trends.

American Journal of Physical Anthropology

 AGE ESTIMATION OF ADULTS AND TRAITS OF THE ACETABULUM 3
TABLE 1. Sample sizes (n) for testing on the Tennessee and New Mexico collections
Intraobserver error
Sample Size (n) sample size (n)
Collection Males Females Males Females
a
William M. Bass Donated Skeletal Collection (UTen; Knoxville, Tennessee) 85 39 25 N/Ab
Maxwell Museum of Anthropology (UNM; Albuquerque, New Mexico) 104c 60c N/Ab 20
a
Females were seriated based on known chronological age to test for the trend of three contributing variables to expressions of
age-related traits as identified by the author. The three variables form the new revised nonsex specific technique, tested blind on
the skeletal collection in New Mexico (UNM).
b
Intraobserver error testing is not applicable (N/A) for this group.
c
Sample sizes were reduced due to poor preservation of remains.

Fig. 2. Orientation image for Variable 1, acetabular groove. To evaluate, hold the os coxae in anterio-lateral aspect and flip hori-
zontally. The acetabular rim should be viewed close-up, under a light source, and at eye level. The area of inspection includes the
lunate surface where it meets the rim, as denoted by the dotted line.

Specimens were selected from each of the GRO and tions of age were established: young adult (17–39 years),
UTen collections based on the number of usable skele- middle adult (40–64 years), and old adult (651 years).
tons (Table 1). The GRO and UTen were appropriate col- Variable descriptions for each broad category of age form
lections to evaluate the usefulness of acetabular criteria the new revised method by the author to estimate age-
to estimate age-at-death because they are made up of an at-death from the acetabulum (Figs. 2–6). Figures 2 and
adequate number of male (n 5 147) and female (n 5 87) 3 demonstrate: [1] how the os coxa was oriented for
specimens (Bedford et al., 1993; UTen Forensic Anthro- inspection of variables and [2] point to anatomical loca-
pology Centre, 2005). Seriation of the male GRO and tions where age-related characteristics were observed.
female UTen samples revealed similar age-related Figures 4–6 were used to estimate age for an unknown
changes at the acetabular rim, the anterior inferior iliac individual. Left os coxae were examined, with the right
spine, and at the apex of the posterior horn of the lunate being substituted if poor preservation precluded analysis
surface. Based on trait expression observed within the of the left. The acetabulum was morphologically
GRO and UTen collections of individuals, broad defini- inspected, focusing on the total pattern of observable

American Journal of Physical Anthropology

4 S.E. CALCE

Fig. 3. Orientation image for Variables 2 and 3, osteophyte development of the acetabular rim, and apex growth. To examine
age-related characteristics, hold os coxae in anterio-lateral position. An adequate light source is favorable to distinguish features of
age-related morphological variation. Areas of examination include the superior portion of the lunate surface below the anterior infe-
rior iliac spine, and the posterior horn of the lunate surface, also known as the apex.

traits to determine whether the individual expressed Testing the revised method
similarities associated with either the young, middle, or
older age phase. Based on analogous expressions of The revised method was tested blind on two contempo-
young, middle, or older age qualities (Figs. 4–6), age-at- rary North American skeletal populations (n 5 249). The
death was estimated by reporting a range, 17–39 years, UTen collection (N 5 650), housed at the University of
40–64 years, or 651 years. Tennessee Forensic Anthropology Research Center, con-

American Journal of Physical Anthropology

 AGE ESTIMATION OF ADULTS AND TRAITS OF THE ACETABULUM 5

Fig. 4. Line drawings and associated descriptions of Variables 1, 2, and 3 for the young adult age category, 17–39 years. To
assess age of an unknown individual, morphologically inspect the acetabulum and focus on the total pattern of observable traits to
determine whether the individual expresses similarities associated with the young adult age phase. Based on analogous expressions
of young adult age qualities, estimate age-at-death by reporting a range, 17–39 years.

American Journal of Physical Anthropology

6 S.E. CALCE

Fig. 5. Line drawings and associated descriptions of Variables 1, 2, and 3 for the middle adult age category, 40–64 years. Mor-
phologically inspect the acetabulum and focus on the total pattern of observable traits to determine whether the individual
expresses similarities associated with the middle adult age phase. Based on analogous expressions of middle adult age qualities,
estimate age-at-death by reporting a range, 40–64 years.

tains donated complete skeletons predominantly made Males were tested using both the UTen and UNM
up of European-derived (White) or African-derived documented collections. Because the test for females was
(Black) persons (UTen Forensic Anthropology Centre, developed on the UTen sample, age estimation of female
2005). The UNM Documented Collection, housed at the specimens was tested on individuals belonging only to
Maxwell Museum of Anthropology at the UNM (N 5 the UNM collection. The samples were chosen on the ba-
257), contains complete skeletons from donated persons sis of high-quality preservation of skeletal elements;
and positively identified forensic cases with documented however, of 100 female and 200 male ossa coxae only 60
demographic information, representing diverse socioeco- and 189 respectively were in a condition suitable for
nomic classes and ethnic affinities (Komar and Lathrop, evaluation. In total, 51 specimens were not evaluated
2006). Age-at-death is known for 77% of individuals (n 5 because they were either: poorly preserved; altered by
698), and 79% of persons (n 5 716) are over 40 years of surgical implant at the site of the acetabular–joint; miss-
age; both robust samples, the UTen and UNM collections ing from storage box; or damaged at the rim /apex due
are appropriate models to test age estimation for older to mishandling. Eighty-five individuals were tested from
adults (UNM Maxwell Museum of Anthropology, 2003; the UTen Collection, 164 from the UNM Collection. They
UTen Forensic Anthropology Centre, 2005; Komar and range in age from 19 to 101 years, and died between
Lathrop, 2006; Grivas and Komar, 2008). 1984 and 2006 (UNM Maxwell Museum of Anthropology,

American Journal of Physical Anthropology

 AGE ESTIMATION OF ADULTS AND TRAITS OF THE ACETABULUM 7

Fig. 6. Line drawings and associated descriptions of variables 1, 2, and 3 for the old adult age category, 651 years. Morphologi-
cally inspect the acetabulum and focus on the total pattern of observable traits to determine whether the individual expresses simi-
larities associated with the older adult age phase. Based on analogous expressions of older adult age qualities, estimate age-at-
death by reporting a range, 651 years.

American Journal of Physical Anthropology

8 S.E. CALCE

2003; UTen Forensic Anthropology Centre, 2005). Sam- or true value) was calculated by dividing the number of
ple sizes are displayed in Table 1, and the age distribu- correct calculations by the total number of specimens. To
tion of the sample is presented in Figure 7. test for intraobserver error, 45 individuals were selected
Specimens were chosen randomly from each age cate- from the UTen and UNM samples (UTen males, n 5 25;
gory but resulted in a test population made up of largely UNM females, n 5 20) and subjected to re-examination
‘‘American Whites’’ (n 5 207, males and females com- by the author (Table 1). Intraobserver error was calcu-
bined). Asian descent is not represented in this sample lated by the difference in categorical assignment
and ‘‘American Blacks’’ only make up 7% of the total between the first and second age estimates. To quantify
sample population, whereas ‘‘Hispanic’’ persons are rep- interobserver constancy and evaluate the utility of ace-
resented by 6%. Of the 249 individuals studied, 11 per- tabular descriptions with associated line drawings, 55
sons were not identified as belonging to any ancestral ossa coxae were observed under identical conditions by
group; reasons for this may include failure to collect rele- three additional independent observers with varied
vant information on intake of body donation. Males and osteological experience.
females were examined separately to identify sex-specific An inter-rater reliability analysis using the weighted
differences or problems with the technique. Individuals Kappa statistic was performed to determine consistency
with noninflammatory osteoarthritis were not excluded among raters (Fleiss and Cohen, 1973; Landis and Koch,
because such manifestations are related to age and/or ac- 1977). Kappa assesses the proportion of agreement
tivity. Known ages for each individual were not docu- between observers corrected for chance and the standard
mented until each specimen was examined, to eliminate measure of interobserver reliability with nominal data
observer bias. If the individual’s known age fell within (e.g., young, middle, and old). Scaled on a range from 21
the estimated age class (young, middle, or old), the age- to 11, a negative kappa value indicates a poorer than
at-death calculation was recorded as correct. A measure chance agreement, zero indicates agreement totally
of accuracy (how close a measured value is to the actual by chance alone, a positive value indicates a better than
chance agreement, and 11 indicates perfect agreement
(Fleiss and Cohen, 1973; Walrath et al., 2004). Kappa
does not take into account the degree of disagreement
between observers and all disagreement is treated
equally as total disagreement. Therefore when the cate-
gories are ordered, it is preferable to use weighted
Kappa, and assign different weights wi to subjects for
whom the raters differ by i categories, so that different
levels of agreement can contribute to the value of Kappa.
Based on the assessment criteria of Landis and Koch
(1977) for the adequacy of Kappa, a weighted Kappa
score of 0.41–0.60 indicated a ‘‘moderate agreement,’’
and 0.61–0.79 was used as an indicator of ‘‘substantial
agreement’’ between observers for all traits to estimate
age (Landis and Koch, 1977). The Intraclass Correlation
Coefficient (ICC) was used to measure the reliability of
ratings and absolute agreement between three observers
Fig. 7. The sample age distribution for both UTen and UNM (Shrout and Fleiss, 1979). ICC is useful when two or
collections of individuals (males and females combined) based
more raters are used to rate the same study subjects; in
on known chronological ages. Across all age classes, age was
estimated for 249 persons. Fewer numbers of individuals are this study, the measurement of absolute agreement is
represented by the young adult age category, likely because valid because systematic differences are relevant to
older aged persons have an increased susceptibility to mortality. show consistencies (or irregularities) in scoring (Bartko,
A bell shaped curve symmetrical about the mean (60.5 years) 1976). ICC ranges from 21 to 11 where 11 indicates
represents the frequency distribution of the test population dis- identical ratings for a subject, or perfect agreement.
played in 10-year age classes; the sample population is normally Four separate scores for weighted Kappa and ICC were
distributed. Above each bar, the number of individuals in each generated to assess interobserver reliability: (1) the over-
10-year age class (n) is also reported. all ability to estimate age-at-death based on the combi-

TABLE 2. Results of stepwise-multiple-regression for seven variables as defined by Rissech et al. (2006)
Standard R2 R2 other Probability Pct change
In Variable coefficient increment X’s T-Value level Sqrt (MSE)
Yes v1 (groove) 0.2523 0.041516 0.348036 2.6006 0.010776 2.9279
Yes v3 (rim porosity) 0.2790 0.036181 0.535084 2.4278 0.017056 2.4916
Yes v4 (apex activity) 0.2444 0.037373 0.374115 2.4674 0.015379 2.5893
No v2 (rim shape) 0.004172 0.659325 0.8231 0.412540 0.1684
No v5 (outer edge of fossa) 0.002102 0.309018 0.5831 0.561199 0.3455
No v6 (fossa activity) 0.004254 0.256713 0.8311 0.408011 0.1615
No v7 (fossa porosity) 0.005560 0.340952 0.9513 0.343878 0.0496

R2 5 0.410699 Sqrt (MSE) 5 12.82071.

List of variables selected v1, v3, and v4.
Note that v3 (rim porosity) as described here as one of the original variables by Rissech et al. (2006) is referred to as Variable 2,
Osteophyte development of the acetabular rim in the revised method.

American Journal of Physical Anthropology

 AGE ESTIMATION OF ADULTS AND TRAITS OF THE ACETABULUM 9
TABLE 3. Interobserver reliability using the weighted Kappa statistic (P < 0.001) for (1) the overall age-at-death estimate and (2)
for each individually assessed trait
Observer Weighted Standard 95% confidence
comparison Kappaa error interval
Age-at-death estimate 1 3 2 0.631 0.080 0.474–0.788
2 3 3 0.585 0.081 0.426–0.743
1 3 3 0.492 0.088 0.319–0.665
Trait 1 (acetabular groove) 1 3 2 0.663 0.078 0.511–0.815
2 3 3 0.506 0.080 0.348–0.664
1 3 3 0.653 0.078 0.501–0.805
Trait 2 (osteophyte development of the acetabular rim) 1 3 2 0.449 0.093 0.268–0.631
2 3 3 0.486 0.091 0.307–0.665
1 3 3 0.337 0.096 0.148–0.525
Trait 3 (apex growth) 1 3 2 0.659 0.075 0.511–0.806
2 3 3 0.698 0.074 0.554–0.843
1 3 3 0.624 0.080 0.467–0.782
a
Kappa: 0.41–0.60 5 moderate inter-rater agreement; 0.61–0.80 5 substantial inter-rater agreement; 0.81–1.00 5 almost perfect
agreement (Landis and Koch, 1977).

TABLE 4. Interobserver reliability using the Intraclass Correlation Coefficient (ICC) (P < 0.001) for (1) the overall age-at-death esti-
mate and (2) for each individually assessed trait
# of subjects # of raters Single measure of ICCa 95% confidence interval
Age-at-death estimate 55 3 0.701 0.569–0.796
Trait 1 (acetabular groove) 55 3 0.725 0.583–0.826
Trait 2 (osteophyte development at rim) 55 3 0.543 0.388–0.681
Trait 3 (apex growth) 55 3 0.760 0.654–0.843
a
Intraclass correlation coefficient, ICC: 0.5–0.6 5 moderate agreement; 0.7–0.8 5 strong agreement; and [0.8 5 almost perfect
agreement (Portney and Watkins, 2000).

nation of three observable traits, i.e., which individuals bles (Figs. 2–6). Forty-three out of 45 individuals were
were classified as young, middle, or old adults and (2) aged the same when subjected to re-examination by the
how each of the three individual traits (acetabular author (error in scoring 5 4.4%). When estimating age,
groove, osteophyte development of the acetabular rim, the average inter-rater reliability was Kappa 5 0.569
and apical growth) were scored by independent observers (95% CI 0.407–0.732, SE 0.08) and ICC 5 0.701 (95% CI
as being young, middle or old. 0.569–0.796), indicating a moderate to substantial agree-
Box plots, Spearman’s Rank correlation, and the Krus- ment between observers (Tables 3 and 4). Despite their
kal–Wallis test were used to compare and analyze popu- varied osteological experience, observers were able to
lation distributions between young, middle, and older identify acetabular variables consistently and use them
age categories to detect differences and associations to accurately estimate age. Among all traits, apical activ-
between trait expression and known chronological age. ity was the easiest to differentiate (Kappa 5 0.660; ICC
Statistical analyses were performed using SPSS for Win- 5 0.905); osteophyte development of the acetabular rim
dows release 19.0.0 (1 August 2010) and MedCalc for proved the most difficult (K 5 0.424; ICC 5 0.780;
Windows, version 11.6.1.0 (MedCalc Software, Maria- Tables 3 and 4). Age dependence of individual traits is
kerke, Belgium). presented in Table 5 where the correlation between esti-
mated trait scoring and actual age is given by three
RESULTS independent observers. Experience of the observer and
familiarity with defined character traits can improve
Stepwise multiple regression analysis of data using interobserver reliability. Results of both intra- and inter-
Rissech et al.’s (2006) original variables on 100 male observer error testing demonstrate a reasonable degree
skeletons from the GRO (Calce and Rogers, 2011) indi- of reproducibility indicating that only a modest amount
cates that there are three statistically significant varia- of error exists when estimating the degree of develop-
bles highly correlated with age: Variable 1, acetabular ment of features. Investigators should be aware that del-
groove (P-value 0.01); Variable 3, acetabular rim porosity icate features of the acetabulum are more difficult to dis-
(P-value 0.017); and Variable 4, apex activity (P-value tinguish on greasy bone, and specimens may appear
0.015). Note that Variable 3, as defined by Rissech et al. younger in these cases.
(2006) becomes Variable 2 (osteophyte development Males (n 5 189) and females (n 5 60) were examined
of the acetabular rim) in the revised method. R2 and separately; no significant sex-specific differences were
P-value coefficients for data collected on the GRO for found. In this early stage, it appears the method is not
each of Rissech et al.’s (2006) seven variables are pro- dependent on sex, although further testing on a larger
vided in Table 2. female sample is favorable to support these preliminary
results. The three traits identified in this analysis are
Testing the revised method highly correlated with age (P \ 0.05), and together are
capable of estimating age-at-death with 81% accuracy,
Each observer was able to score traits effectively using both sexes combined. Of the percentage wrong (19%),
only the information given in the descriptions of varia- the tendency was to underestimate age by one age cate-

American Journal of Physical Anthropology

10 S.E. CALCE
TABLE 5. Correspondence between trait scores and actual age for three independent observers
Young (n 5 15) Middle (n 5 20) Old (n 5 20)
Obs 1 Obs 2 Obs 3 Obs 1 Obs 2 Obs 3 Obs 1 Obs 2 Obs 3
Trait 1 10 9 10 10 14 9 14 10 16
Trait 2 11 15 15 13 14 14 6 6 4
Trait 3 9 15 12 13 8 12 9 11 15

TABLE 6. Cross tabulation between actual and

estimated age category
Estimated age category
Actual age category (n) Young Middle Old
Young 34 29 5 0
Middle 105 4 87 14
Old 110 2 22 86

gory. Overall, errors in assigning an incorrect age group-

ing were low accounting for 15%, 17%, and 22% of age-
at-death estimates for young, middle, and older aged
persons respectively (Table 6).

Population statistics and nonparametric tests

Using Spearman’s Rank correlation (for three varia- Fig. 8. Data is summarized using a box plot to display and
bles combined and across all age classes) a positive asso- compare sample population distributions between three age cat-
ciation between trait expression and known chronologi- egories. Medians for young (32.5), middle (54), and older (76)
cal age is occurring within the data set (rs 5 0.7551; P \ aged adults differ. Distributions do not overlap indicating that
it is possible to discern changes in trait expression between
0.001). Results demonstrate that variables are correlated
these three age groups.
with age in such a way that as trait expression changes
from young to middle to older adult phases, chronologi-
cal age increases. Because a component system of scor- Low error rate, consistency in scoring and reduction in
ing was not used by the author on the UTen and UNM data collection time are considerable advantages to using
test samples, data for each variable was not collected in- the author’s technique, but it is important to investigate
dependently, so there is no way to determine whether bone remodeling and biomechanics to explain behavioral
correlation between age and trait expression is weaker changes in bone structure relative to the effects of genet-
when Variables 1, 2, and 3 are analyzed separately. ics and the environment. Because skeletal morphology is
A Kruskal–Wallis (K–W) test was conducted to com- influenced by a combination of mechanical forces (or
pare three groupings of ordinal data: young (n 5 34), loadings), activity is an important variable to consider,
middle (n 5 105), and older (n 5 110) aged categories. the effects of which may potentially be a major limita-
Robust differences were detected among them (P-value tion of the method in general. Teasing apart whether the
\0.001), indicating that samples are from dissimilar pop- observed changes reflect age or activity at the weight-
ulations and all medians are distinct (Fig. 8). The K–W bearing site of the acetabulum requires further study.
test demonstrates that it is possible to detect differences
between broad categories of age through morphological A holistic approach to scoring age-related traits
inspection of three variables: acetabular groove, osteo-
phyte development of the acetabular rim, and apex Hoppa and Vaupel (2002), Rissech et al. (2006), and
growth. Standard deviation representing the sample more recently Konigsberg et al. (2008) describe the
spread of individuals around the means for categories of promise of a more rigorous and intuitive statistical ex-
actual age in young, middle, and old age classes are planation of estimating age-at-death using the Bayesian
found in Table 7. Based on the revised morphological approach. Rissech et al.’s (2006) method relies on the use
descriptions of three acetabular variables that form the of a known comparative collection to generate age ranges
basis of this analysis, the probability that young, middle, each time an investigator scores an unknown skeleton,
and older aged adults display similar characteristics is therefore there are no fixed age categories in the original
nil (0.000). method. Regrettably, age-at-death estimates using Ris-
sech’s method are biased in the direction of the known-
DISCUSSION AND CONCLUSIONS age reference sample, used as a standard of calibration
to estimate age for unknown persons (Boldsen et al.,
Methods to accurately estimate age-at-death for the 2002; Ross and Kimmerle, 2009). To date, many age esti-
adult skeleton are a challenging, but worthwhile investi- mation studies report on the problem of test populations
gation. Precise definitions of fewer, more encompassing mimicking age distributions of the original sample (Boc-
variables reduce subjectivity in scoring and increase the quet-Appel and Masset, 1982; Lucy et al., 1996; Boldsen
utility of the method, particularly for older aged individ- et al., 2002; Hoppa and Vaupel, 2002; Ross and Kim-
uals where joint surfaces are complicated by extrinsic merle, 2009), and although the benefit of computing a
factors such as weight, infection, and physical activity. likelihood function for the combination of individually

American Journal of Physical Anthropology

 AGE ESTIMATION OF ADULTS AND TRAITS OF THE ACETABULUM 11
TABLE 7. Descriptive statistics for each age phase young, middle, and old adults representing males and females of the test sample
(n 5 249) from the UTen and UNM collections of individuals
Standard Standard
Count Mean deviation error Minimum Median Maximum Range IQR
Young adult 17–39 years 34 31.79 5.772 0.991 19 32.5 39 20 9.5
Middle adult 40–64 years 105 52.97 7.324 0.718 40 54 64 24 13
Old adult 651 years 110 76.74 8.388 0.803 65 76 101 36 14

age related traits is propitious, so far, methods using the TABLE 8. Sample age distribution for males and females
acetabulum to estimate age have not employed Bayes’ in the UTen and UNM collections
theorem to correct this problem.
Age class Males (n) Females (n)
A component system was not used because numerical
scoring of qualitative variables requires more rigid 16–20 2 0
descriptions of progressive biological phenomena (Meindl 21–30 8 2
et al., 1985). Component scoring systems fail to capture 31–40 23 2
age-related variability and can be problematic if the in- 41–50 31 3
51–60 42 8
vestigator does not understand descriptions distinguish- 61–70 34 13
ing key features. For example, differentiating between 71–80 28 15
what classifies a trait as ‘‘1’’ from ‘‘2’’ or ‘‘2’’ from ‘‘3,’’ etc. 81–90 18 12
For these reasons, biological assessment is based on a 91–99 3 4
basic sequence of modal changes that allows for full inte- 100–109 0 1
gration of all age-related variables and reduces intra- TOTAL 189 60
and interobserver error (Lovejoy et al., 1985; Brooks and Young adult (19–39) 30 4
Suchey, 1990; Meindl et al., 1990). To employ the Middle adult (40–64) 91 14
author’s new method effectively, all three traits of the Old adult (651) 68 42
TOTAL 189 60
acetabulum for each individual must be in good condi-
tion for evaluation: that is, not altered by surgical
implant at the site of the acetabular–joint or damaged at
the rim /apex. Additional research is necessary to deter- the smallest number of individuals belong to this age
mine which trait (1, 2, or 3) is most accurate to estimate category (n 5 34; Table 6), but may also reflect less vari-
age in adult samples. For now, investigators should ation and absolute change to pristine morphology com-
examine the total pattern of observable traits and select monly observed in early adulthood. Larger error was
the age category that most closely resembles the acetab- observed for middle and older aged adults (17% and 22%
ular region of the unknown specimen (Lovejoy et al., respectively). The author considers two explanations for
1985; Meindl et al., 1990; Suchey and Katz, 1998; Rog- this result. First, large inaccuracies are merely a statis-
ers, 2009). For use with ordinal categorical variables, tical artifact of the sample, because the largest propor-
modal phase analysis is favorable over the more compli- tion of UTen and UNM individuals for whom age was
cated component scoring system because it incorporates estimated (86% of total sample) belong to these age
both early and late phases of development into one age classes (Tables 6 and 8). Second, age-related morphologi-
class (Rogers, 2009). A more robust age class, as is used cal changes are highly variable for individuals 401
in this study, is necessary to provide a more accurate years, where further research into specific degenerative
representation of the variation associated with acetabu- modifications targeting middle-aged and elderly persons
lar morphology (Franklin, 2010). Doing so in a forensic is required (Calce and Rogers, 2011).
context avoids excluding decedents based on chronologi- Issues of donor-based and forensic skeletal collections
cal age and instead, preserves attributes of physiology are a current topic of debate in physical anthropology.
(Suchey and Katz, 1998; Cunha et al., 2009). In compari- Multiple biases of sample distributions must be consid-
son to the pubic symphyseal and auricular surface aging ered to determine whether these populations are repre-
methods, the acetabulum technique performs better at sentative of normal variation both in living and decedent
reducing broad categories of age for estimates of groups (Usher, 2002; Grivas and Komar, 2008). The
unknown individuals, particularly for those over 50 research design of this project focused on maximizing
years (Suchey and Katz, 1998; Buckberry and Chamber- the sample size to include as many usable specimens as
lain, 2002). Unlike other areas of the skeleton, continued was possible. With respect to sex ratio, males are overre-
age-related changes are observed in the acetabulum presented in the test sample accounting for 76% of all
beyond the sixth decade; ‘‘the acetabulum in the future, age estimates. Due to the limited number of females
may in the end be the most reliable for 60-year-old or comprised in the entire UNM collection (38.2%), this
older individuals’’ (Cunha et al., 2009: 5). The next (and result is not considered unusual (Grivas and Komar,
ongoing) step of this research is to develop more narrow 2008). Although no significant sex-specific differences
age categories from morphological descriptions particu- were found, the sample of female specimens used in this
larly for individuals over 65 years. investigation was small and should be increased. The
author suggests expanding the research to focus more
Directions for future research specifically on the female expression of age-related ace-
tabular traits to determine whether differences in carry-
Assignment to an incorrect age group was low (19% ing capacity at the femoroacetabular joint produce vari-
across all age classes; Table 6). Percentage wrong esti- able morphology between sexes.
mates for each age category varied with the smallest Although the number of individuals in the middle and
error occurring for younger persons (15%), likely because older age categories are consistent (Table 8), younger

American Journal of Physical Anthropology

12 S.E. CALCE

adults (17–39 years) are underrepresented (n 5 34, 14% tons. In: Hoppa RD, Vaupel JW, editors. Paleodemography:
of total sample; Table 6). As a person ages, they become age distributions from skeletal samples. Cambridge: Cam-
more susceptible to mortality which explains the large bridge University Press. p 73–106.
number of individuals in the documented UTen and Brooks SP, Suchey JM. 1990. Skeletal age determination based
on the os pubis: a comparison of the Ascádi–Nemeskéri and
UNM skeletal collections over 40 years of age, resulting
Suchey–Brooks methods. J Hum Evol 5:227–238.
in an inevitably smaller sample size for persons less Buckberry JL, Chamberlain AT. 2002. Age estimation from the
than 39 years. Even though the sample size for young auricular surface of the ilium: a revised method. Am J Phys
adults (males and females combined) is not as sizeable Anthropol 119:231–239.
as the middle and older age categories, young traits of Calce SE, Rogers TL. 2011. Evaluation of age estimation tech-
the acetabulum are more easily discernible due to the nique: Testing traits of the acetabulum. J Forensic Sci 56:
density, general smoothness, robusticity, and minimal 302–311.
porosity. Young adult females (n 5 4) are largely under- Cunha E, Baccino E, Martrille L, Ramsthaler F, Prieto J, Schu-
represented in the test sample (Table 8), which is attrib- liar Y, Lynnerup N, Cattaneo C. 2009. The problem of aging
uted to a general lack in available female specimens for human remains and living individuals: a review. Forensic Sci
Int 193:1–13.
study in modern late 20th century samples (Marks,
Fleiss JL, Cohen J. 1973. The equivalence of weighted kappa
1995; Grivas and Komar, 2008). That said, skeletal col- and the intraclass correlation coefficient as measures of reli-
lections with a large percentage of female specimens, ability. Educ Psychol Meas 33:613–619.
such as Hamann-Todd or Robert J. Terry would also be Franklin D. 2010. Forensic age estimation in human skeletal
appropriate. remains: current concepts and future directions. Legal Med
Applicability of this method to nonwhite populations 12:1–7.
must be explored. The investigator plans to test this Grivas CR, Komar DA. 2008. Kumho, Daubert, and the nature
method on available collections of various geographical of scientific inquiry: implications for forensic anthropology.
origins and encourages other researchers to do the same. J Forensic Sci 53:771–776.
Figures 1–6 may serve as datasheets in the estimation Haglund WD. 1997. Dogs and coyotes: postmortem involvement
with human remains. In: Haglund WD, Sorg MH, editors.
of age for an unknown individual. Morphological fea-
Forensic taphonomy: the postmortem fate of human remains.
tures of the acetabulum can be used as an effective age Boca Raton: CRC Press LLC. p 367–381.
indicator. The acetabulum is a durable part of skeleton, Hintze J. 2006. NCSS, PASS and GESS. NCSS. Kaysville, Utah.
which survives well in postdepositional environments Hoppa RD, Vaupel JW. 2002. The Rostock manifesto for paleode-
(Haglund, 1997), making the os coxae an important mography: the way from stage to age. In: Hoppa RD, Vaupel
region of study for methods of personal identification. In- JW, editors. Paleodemography: age distributions from skeletal
vestigator confidence using the author’s method is high, samples. Cambridge: Cambridge University Press. p 1–8.
but whenever possible should be combined with multiple _ can MY. 1989. Age markers in the human skeleton. Spring-
Is
age indicators for the most precise estimation of age-at- field, IL: Charles C. Thomas.
death. The present technique produces encouraging Komar DA, Lathrop S. 2006. Frequencies of morphological char-
acteristics in two contemporary forensic collections: implica-
results, can be completed on a small budget, and within
tions for identification. J Forensic Sci 51:974–978.
a reasonable amount of time. Consistency in scoring, Konigsberg LW, Herrmann NP, Wescott DJ, Kimmerle EH.
reduction in data collection time, and low error rates are 2008. Estimation and evidence in forensic anthropology: age-
significant advantages to using the technique, which is at-death. J Forensic Sci 53:541–557.
flexible and useful in forensic and bioarchaeological anal- Landis JR, Koch GG. 1977. The measurement of observer agree-
yses of human remains. ment for categorical data. Biometrics 33:159–174.
Lin LI-K. 1989. A concordance correlation coefficient to evaluate
ACKNOWLEDGMENTS reproducibility. Biometrics 45:255–268.
Lovejoy CO, Meindl RS, Pryzbeck TR, Mensforth RP. 1985.
The author thanks the following people: Drs. Helen K. Chronological metamorphosis of the auricular surface of the
Kurki, Tracy L. Rogers, and Michael Schillaci for critical ilium: a new method of determination of adult skeletal age at
review of the study proposal and/or editorial suggestions death. Am J Phys Anthropol 68:15–28.
of the manuscript; Joyce Hui, biomedical communications Lucy D, Aykroyd RD, Pollard AM, Solheim T. 1996. A Bayesian
approach to adult human age estimation from dental obser-
artist, for producing acetabular renderings; Drs. Heather
vations by Johanson’s age changes. J Forensic Sci 41:189–
J. Edgar and Lee Meadows Jantz for granting access to 194.
skeletal collections; graduate students Catherine Merritt, Manly BFJ. 2005. Multivariate statistical methods: a primer,
Lelia Watamaniuk, and Johanna Kelly for collaborating in 3rd ed. Boca Raton: Chapman and Hall.
testing observational consistency of the variables; and Dr. Marks J. 1995. Human biodiversity: genes, race, and history.
Christopher Ruff as well as three anonymous reviewers New York: Aldine de Gruyter.
for helpful comments on the manuscript. Meindl RS, Lovejoy CO, Mensforth RP, Walker RA. 1985. A re-
vised method of age determination using the os pubis, with a
review and tests of accuracy of other current methods of pubic
symphyseal aging. Am J Phys Anthropol 68:29–45.
LITERATURE CITED Meindl RS, Russell KF, Lovejoy CO. 1990. Reliability of age at
death in the Hamann–Todd collection: validity of subselection
Bartko JJ. 1976. On various intraclass correlation reliability procedures used in blind test of the summary age technique.
coefficients. Psychol Bull 83:762–765. Am J Phys Anthropol 83:349–357.
Bocquet-Appel JP, Masset C. 1982. Farewell to paleodemogra- Portney LG, Watkins MP. 2000. Foundations of clinical
phy. J Hum Evol 12:321–333. research: applications to practice, 2nd ed. Upper Saddle River,
Bedford ME, Russell KF, Lovejoy CO, Meindl RS, Simpson SW, NJ: Prentice-Hall.
Stuart-Macadam PL. 1993. Test of the multifactorial aging Rissech C, Estabrook GF, Cunha E, Malgosa A. 2006. Using the
method using skeletons with known ages-at-death from the acetabulum to estimate age-at-death of adult males. J Foren-
Grant Collection. Am J Phys Anthropol 91:287–297. sic Sci 51:213–229.
Boldsen JL, Milner GR, Konigsberg LW, Wood JW. 2002. Transi- Rissech C, Estabrook GF, Cunha E, Malgosa A. 2007. Estima-
tion analysis: a new method for estimating age from skele- tion of age-at-death for adult males using the acetabulum,

American Journal of Physical Anthropology

 AGE ESTIMATION OF ADULTS AND TRAITS OF THE ACETABULUM 13
applied to four Western European populations. J Forensic Sci Suchey JM, Katz D. 1998. Applications of pubic age determina-
52:774–778. tion in a forensic setting. In: Reichs KJ, editor. Forensic oste-
Rogers TL. 2009. Skeletal age estimation. In: Blau S, Ubelaker ology: advances in the identification of human remains, 2nd
DH, editors. Handbook of forensic anthropology and archaeol- ed. Springfield, IL: Charles C. Thomas. p 204–236.
ogy. Walnut Creek, CA: Left Coast Press, Inc. p 208–221. University of New Mexico Maxwell Museum of Anthropology,
Ross AH, Kimmerle EH. 2009. Contribution of quantitative meth- 2003. Documented skeletal collection. Available from:http://
ods in forensic anthropology: a new era. In: Blau S, Ubelaker www.unm.edu/osteolab/coll_doc.html. Accessed 2008 Decem-
DH, editors. Handbook of forensic anthropology and archaeol- ber 5.
ogy. Walnut Creek, CA: Left Coast Press, Inc. p 479–489. University of Tennessee. Forensic Anthropology Center, 2005.
Rougé-Maillart C, Jousset N, Vielle B, Gaudin A, Telmon N. Available from:http://web.utk.edu/fac/facilities.html. Accessed
2007. Contribution of the study of acetabulum for the estima- 2010 July 5.
tion of adult subjects. Forensic Sci Int 171:103–110. Usher BM. 2002. Reference samples: The first step in linking
Shrout PE, Fleiss JL. 1979. Intraclass correlations: uses in biology and age in the human skeleton. In: Hoppa RD, Vaupel
assessing rater reliability. Psychol Bull 86:420–428. JW, editors. Paleodemography: age distributions from skele-
SPSS for Windows, Rel. 19.0.0. 2010. Chicago: SPSS Inc. tal samples. Cambridge: Cambridge University Press. p 29–
Stull KA, James DM. 2010. Determination of age-at-death using 47.
the acetabulum. In: Latham KE, Finnegan M, editors. Age Walrath DA, Turner P, Bruzek J. 2004. Reliability test of the
estimation of the human skeleton. Springfield, IL: Charles C. visual assessment of cranial traits for sex determination. Am
Thomas. p 134–146. J Phys Anthropol 125:132–137.

American Journal of Physical Anthropology