A COMPARATIVE EVALUATION OF THE CHEN ET AL. AND SUCHEY-BROOKS
PUBIC BONE AGING METHODS ON A NORTH AMERICAN SAMPLE
By
Julie Michele Fleischman

A THESIS
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of
MASTERS OF SCIENCE
Forensic Science
2011

i

ABSTRACT
A COMPARATIVE EVALUATION OF THE CHEN ET AL. AND SUCHEY-BROOKS
PUBIC BONE AGING METHODS ON A NORTH AMERICAN SAMPLE
By
Julie Michele Fleischman
Accurately assessing age-at-death of adult human skeletons is fundamental in
physical anthropology. The most generally accepted methods for estimating age involve
analysis of the pubic bones. Two such methods—Chen et al. (2008) and Suchey-Brooks
(1990)—were the focus of this study.
The objective of this research was to evaluate the accuracy of the Chen et al. and
Suchey-Brooks methods. The Chen et al. method was developed on a sample of Chinese
Han males. This research utilized a known collection of modern pubic bones curated at
the Forensic Science Center in Phoenix, Arizona. A sample of 296 left male pubic bones
of European ancestry, between the ages of 18 and 70, was evaluated.
Results indicated that there are no statistically significant differences between the
two methods. On average the revised Chen et al. method slightly over-ages the
specimens while the Suchey-Brooks method slightly under-ages. Both methods have an
average error of approximately 9 years from individual’s actual age.
This research demonstrates that the Suchey-Brooks method is most accurate for
aging young adults, while the revised Chen et al. method is most accurate for aging
middle adults. Thus, the Chen et al. method is an important contribution to the field of
physical anthropology for aging older adult skeletal remains. There are some limitations,
such as subjectivity and the intricate scoring system, so the Chen et al. method should be
applied cautiously until further research has been done in the United States.
ii

ACKNOWLEDGMENTS

I would like to thank my thesis committee, Dr. Todd Fenton, Dr. Norman Sauer
and Dr. Jennifer Cobbina, for their support and guidance during every stage of this thesis
project. This study would never have come to fruition without their help. I would like to
thank Dr. Laura Fulginiti for permitting me to study the FSC skeletal sample and for
participating in data collection. Thank you for being one of my biggest supporters, and
as always, I would not be where I am today without you. I would also like to thank
Cheryl Anderson and Jesus Villa for their assistance and Dr. Kristen Hartnett for
providing me with information on the FSC sample. I need to thank Paul Curran at
CSTAT for his statistical assistance; I would still be floundering in a sea of numbers and
data were not for his help.
A very special thank you goes to my family for all of your tireless encouragement
and support throughout my academic career. You have helped to make me the person
that I am, and I would not have made it through without you. I would also like to thank
Andrew Missel for his constant words of encouragement, loving support, and statistical
guidance. You helped me push myself farther than I thought possible. And finally, thank
you to all of my friends and fellow graduate students for pointing me in the right
direction and putting a smile on my face.
Finally, this research could not have been done without financial help from the
National Institute of Justice and the Forensic Science Foundation, as well as Michigan
State University’s Graduate School and Forensic Science Masters Program.

iii

DISCLAIMER

This project was supported by Award No. 2008-DN-BX-K216 awarded by the National
Institute of Justice, Office of Justice Programs, U.S. Department of Justice. The opinions,
findings, and conclusions or recommendations expressed in this thesis are those of the
author and do not necessarily reflect those of the Department of Justice.

iv

TABLE OF CONTENTS

LIST OF TABLES ………………………………………………………………………vii

LIST OF FIGURES …………………………………………………………………….. ix

INTRODUCTION ……………………………………………………………………….. 1

BACKGROUND ………………………………………………………………………… 4
ASSESSMENT OF AGE AT DEATH ………………………………………………….. 4
ACCURACY FOR ADULT AGE AT DEATH ESTIMATION METHODS …………... 5
THE PUBIC BONE AND THE PUBIS SYMPHYSIS JOINT …………………………. 7
HISTORICAL RESEARCH ON THE PUBIC BONE ………………………………….. 8
THE FIRST METHOD TO EVALUATE THE RELATIONSHIP BETWEEN
AGE AND THE PUBIC BONE …………………………………………………. 9
REVISIONS OF THE TODD METHOD ……………………………………………… 10
THE SUCHEY-BROOKS PUBIC AGING METHOD ...……………………………… 18
TESTS OF THE SUCHEY-BROOKS PUBIC BONE AGING METHOD ..………….. 20
THE CHEN ET AL. PUBIC BONE AGING METHOD ..……………………………... 24

STUDY OBJECTIVES ………………………………………………………………….28
RESEARCH QUESTIONS AND HYPOTHESES ……………………………………..29
Research Questions ……………………………………………………………...29
Research Hypotheses ……………………………………………………………29

MATERIAL AND METHODS ………………………………………………………... 31
MATERIALS …………………………………………………………………………... 31
METHODS …………………………………………………………………………….. 33
The Chen et al. Method ………………………………………………………… 33
The Suchey-Brooks Method ……………………………………………………. 34
Intra- and Inter-Observer Error …………………………………………………. 35
STATISTICAL ANALYSIS PART 1: EVALUATION AND REVISION OF
THE CHEN ET AL. (2008) METHOD ………………………………………… 36
STATISTICAL ANALYSIS PART 2: COMPARISON OF THE CHEN ET
AL. (2008) AND SUCHEY-BROOKS (1990) MODELS ……………………... 40

RESULTS ………………………………………………………………………………. 42
TEST OF THE ORIGINAL CHEN ET AL. MODELS ………………………………... 42
REVISION OF ORIGINAL CHEN ET AL. MODELS ………………………………... 44
v

ACCURACY OF THE MODELS ……………………………………………………… 44
Frequencies of Absolute Mean Error …………………………………………… 44
Absolute Mean Error for All Models …………………………………………… 51
Bias for All Models ……………………………………………………………...52
Variance and Correlation ……………………………………………………….. 53
THE MORPHOLOGICAL FEATURES MOST PREDICTIVE OF AGE …………….. 56
RELATIONSHIPS BETWEEN THE MORPHOLOGICAL FEATURES
AND AGE ……………………………………………………………………….57
RELATIONSHIP BETWEEN ALL MORPHOLOGICAL FEATURES ……………… 58
ACCURACY FOR AGE GROUPS ……………………………………………………. 60
Bias and Absolute Mean Error …………………………………………………. 60
Analysis of Variance (ANOVA) ……………………………………………….. 63
INTRA- AND INTER-OBSERVER ERROR …………………………………………. 69
Intra-Observer Error ……………………………………………………………. 69
Inter-Observer Error ……………………………………………………………. 70

DISCUSSION ………………………………………………………………………….. 72
THE REVISED CHEN ET AL. MODEL COMPARED TO THE SUCHEYBROOKS MODEL …………………………………………………………….. 72
ACCURACY FOR OLDER AGES ……………………………………………………. 74
VARIANCE AND CORRELATION ………………………………………………….. 75
INTRA- AND INTER-OBSERVER ERROR ………………………………………….. 77
LIMITATIONS …………………………………………………………………………. 78

CONCLUSION …………………………………………………………………………. 80

APPENDICES ………………………………………………………………………….. 84
Appendix 1: Nine Morphological Features of the Pubic Bone and Associated
Scores (Chen et al., 2008) ………………………………………………. 85
Appendix 2: Revised Descriptions for the Nine Morphological Features of the
Pubic Bone ……………………………………………………………… 87
Appendix 3: Revised Nine Morphological Features of the Pubic Bone and
Associated Scores ………………………………………………………. 89
Appendix 4: Revised Images of the Nine Chen et al. Morphological Features of
the Pubic Bone ………………………………………………………….. 91

REFERENCES …………………………………………………………………………. 94

vi

LIST OF TABLES

Table 1: Summary Statistics for the FSC Pubic Bone Sample from Hartnett (2007) …...31

Table 2: Age Groups for FSC Sample …………………………………………………. 34

Table 3: Evaluation Criteria of Male Pubic Symphyses and Regression Coefficients for
Original and Revised Chen et al. Multiple Regression Analysis (MRA), Gradual
Regression Analysis (GRA), and Quantification Theory Model-I (QMI) ……………... 45

Table 4: Frequencies of Absolute Mean Error (AME) …………………………………. 48

Table 5: Absolute Mean Error for Each Model ………………………………………… 51

Table 6: Bias for Each Model …………………………………………………………... 52

Table 7: Pearson’s Correlations (r) between Models and Actual Age …………………. 56

Table 8: Forward Stepwise Regression Analysis of the Chen et al. Morphological
Features and Actual Age ………………………………………………………………... 57

Table 9: Pearson’s Correlations (r) between Morphological Features and Actual Age ... 58

Table 10: Pearson’s Correlation Coefficients (r) for All Nine Chen et al. Morphological
Features …………………………………………………………………………………. 59

Table 11: Summary Statistics for FSC Sample by Age Group ………………………… 60

Table 12a: Absolute Mean Error (AME) and Bias for each Model by Age Group ……. 61

Table 12b: Absolute Mean Error (AME) and Bias for each Model by Age Group ……. 61

vii

Table 13: Analysis of Variance (ANOVA) of Absolute Mean Error and Bias for Each
Model by Age Group …………………………………………………………………… 64

Table 14: Analysis of Variance (ANOVA) Post Hoc Comparisons for Absolute Mean
Error for Each Age Group ……………………………………………………………… 65

Table 15: Analysis of Variance (ANOVA) Post Hoc Comparisons for Bias for Each Age
Group …………………………………………………………………………………… 67

Table 16: Intra-Observer Cohen’s Kappa Measure of Agreement (κ) for the Chen et al.
Morphological Features and the Suchey-Brooks Method ……………………………… 69

Table 17: Intra-Observer Pearson’s Correlations (r) for the Chen et al. and SucheyBrooks Methods ………………………………………………………………………… 70

Table 18: Inter-Observer Pearson’s Correlations (r) …………………………………… 71

Table 19: Nine Morphological Features of the Pubic Bone and Associated Scores (Chen
et al., 2008) ……………………………………………………………………………... 85

Table 20: Revised Nine Morphological Features of the Pubic Bone and Associated
Scores …………………………………………………………………………………… 89

viii

LIST OF FIGURES

Figure 1: The os coxa. The shaded region is the pubic bone (Modified from Scheuer and
Black, 2004) ………………………………………………….………………………….. 7

Figure 2: Pubic symphyseal surfaces (Todd, 1920) ……………………………………… 8

Figure 3: Todd’s ten pubic symphysis phases (McKern and Stewart, 1957) …………... 10

Figure 4: McKern and Stewart pubic symphyseal components (1957) ………………… 12

Figure 5: Line drawings of the six male Suchey-Brooks pubic bone age phases (Buikstra
and Ubelaker, 1994) …………………………………………………………………….. 19

Figure 6: Ridges and furrows on the symphyseal surface: (0-4) (Chen et al., 2008) …... 26

Figure 7: Age Distribution of Sample ………………………………….………………. 32

Figure 8: Original Chen et al. Absolute Mean Error Histograms ………………….…… 43

Figure 9: Original Chen et al. Bias Histograms ………………….……….……….……. 43

Figure 10: Frequency of Absolute Mean Error: Original Chen MRA+GRA …………... 49

Figure 11: Frequency of Absolute Mean Error: Original Chen QMI+GRA …………… 49

Figure 12: Frequency of Absolute Mean Error: Revised Chen MRA+GRA …………... 50

Figure 13: Frequency of Absolute Mean Error: Revised Chen QMI+GRA ……………. 50

Figure 14: Frequency of Absolute Mean Error: Suchey-Brooks Model ………………...51

ix

Figure 15: Absolute Mean Error ………………………………………………………... 52

Figure 16: Bias ………………………………………………….………………………. 53

Figure 17: Revised Chen et al. MRA+GRA Scatterplot ……………………………….. 54

Figure 18: Revised Chen et al. QMI+GRA Scatterplot ………………………………… 54

Figure 19: Variance Explained by the Regression Models …………………………….. 55

Figure 20: Absolute Mean Error for Age Groups ……………………………………… 62

Figure 21: Bias for Age Groups ………………………………………………………… 63

Figure 22: Comparison of Actual Age and Estimated Age using the Suchey-Brooks
Method with a Superimposition of the Suchey-Brooks Age Phase Ranges ……………. 73

Figure 23: Feature 1: Ridges and Furrows of the Symphyseal Surface ………………… 91

Figure 24: Feature 2: Ridges of the Pubic Tubercle ……………………………………. 91

Figure 25: Feature 3: Lower Extremity of the Symphysial Surface ……………………. 91

Figure 26: Feature 4: Ventral Rampart …………………………………………………. 92

Figure 27: Feature 5: Ossific Nodule …………………………………………………… 92

Figure 28: Feature 6: Dorsal Margin …………………………………………………… 92

Figure 29: Feature 7: Ventral Beveling ………………………………………………… 93

x

Figure 30: Feature 8: General Macroscopic Changes of the Symphyseal Surface ……... 93

Figure 31: Feature 9: Bone Density of the Symphyseal Surface ……………………….. 93

xi

INTRODUCTION
Accurately assessing the age-at-death of adult human skeletal remains is one of
the most fundamental aspects of creating a biological profile. A biological profile
comprises the demographic characteristics of an individual including sex, age-at-death,
ancestry, and living stature (Stewart, 1979; Byers, 2008). Establishing age-at-death,
presented in the form of an age range, can assist forensic or medicolegal authorities in
properly identifying an unknown decedent (Komar and Buikstra, 2008). It is equally
valuable for bioarchaeologists when deducing past conditions of health, disease, and
demography (Lovejoy et al., 1985a). However, estimating age-at-death is often
challenging. “The estimation of age-at-death of adult skeletal material is,” according to
Buckberry and Chamberlain (2002:231), “one of the more difficult tasks undertaken by
physical anthropologists.” The challenge arises due to the variability of genetic and
environmental factors which influence skeletal remodeling and degeneration throughout
an individual’s life (Mulhern and Jones, 2005; Buckberry and Chamberlain, 2002).
Despite these challenges, there are many techniques and methods available to
forensic anthropologists and bioarchaeologists to facilitate estimation of age-at-death.
The variety can complicate the decision of which technique should be used and which is
likely to be the most accurate (Baccino et al., 1999). The most accepted and widely used
methods for age assessment rely on changes of osseous and dental elements (Byers, 2008;
Hens et al. 2008). Methods for estimating age from the skeleton can be divided into two
phases: modeling changes and remodeling changes. Modeling changes are based on the
growth and development of the bones (and dentition) and are applicable to subadults.

1

After the skeleton has finished maturing, bone continues to remodel throughout
adulthood (Byers, 2008).
The most frequently used and well accepted techniques for assessment of age in
adults involve an analysis of the pubic bones (Meindl et al., 1985; Meindl and Lovejoy,
1989; Djurić et al., 2007). The reliability of the pubic bones for age estimation is
primarily based on two factors. First, numerous studies have demonstrated that the pubic
bones advance through observable changes into adulthood after other skeletal elements
have ceased their developmental changes (McKern and Stewart, 1957; Meindl and
Lovejoy, 1989). The second factor is that the age-related changes of the pubic bone, and
the pubic symphyseal face, are distinctive and clearly perceptible (Meindl et al., 1985).
Some of the most distinct age-related changes of the pubic bones are visible on
the pubic symphyseal face. The symphyseal face in young individuals is billowy and is
described as having marked ridges and furrows. With age the ridges and furrows begin to
diminish and eventually the symphyseal face will appear flat or concave (Todd, 1920).
The ventral (anterior) and dorsal (posterior) margins of the symphyseal face also proceed
through distinct stages of bone deposition and degradation. Features such as the dorsal
plateau and the ventral bevel appear later in life, and when both margins are fully
developed, a rim will be visible around the circumference of the symphyseal face. The
texture and consistency of the bone are also features that transform with age; in young
individuals the bone is dense and heavy, but with age it begins to degrade resulting in
visible porosity (Todd, 1920; McKern and Stewart, 1957; Meindl et al., 1985).
These changes have been distinctly correlated with age ranges and can be utilized
to estimate the age of an unknown individual. Numerous methods for age assessment via

2

the pubic bones are available, including the Chen, Zhang, and Tao (2008) (subsequently
referred to as Chen et al.) and Suchey-Brooks (1990) methods. Developed in 2008, the
Chen et al. method focuses on nine morphological features of the male pubic bone for the
Chinese Han population. Chen and colleagues analyzed 262 pairs of pubic bones from
which they deduced four statistical equations for male age estimation. In addition to
developing a method for age-at-death estimation using morphological changes, the aim of
the Chen et al. study was to improve upon current pubic bone aging methods such as
Suchey-Brooks (1990).
Chen and colleagues (2008) state that with the use of their statistical formulae,
evaluating males only, and subdividing each feature, age-at-death can be quantitatively
estimated with a high degree of accuracy and reliability. In fact, their study notes that in
China this method can accurately age males within one year, and that an average of 98%
of the variance in age can be explained by their 9 features. Based upon these impressive
claims I wanted to evaluate the Chen et al. method to determine if its utility is as great in
the United States as it is in China.

3

BACKGROUND
Assessment of Age at Death
In both forensic and paleodemographic contexts, age is one of the essential
biological parameters which facilitates the identification of human skeletal remains.
Although numerous methods exist for estimating age at death, adult age estimation can be
one of the more difficult elements of the biological profile (Đşcan, 1989; Buckberry and
Chamberlain, 2002; Osborne et al., 2004; Mulhern and Jones, 2005). More than any
other aspect of the biological profile, as Dr. William Maples notes, “the accuracy of ageestimation techniques are very much at the mercy of the individual applying
them…[T]here is always a certain amount of deviation between results obtained in the
reference laboratory and the application of aging techniques by individuals in the field”
(Đşcan, 1989:13).
The techniques and processes for assessing age for juvenile and adult skeletal
remains are distinct, and the latter is often more challenging. An estimation of age for
young individuals is based primarily upon the predictable sequence of bone growth and
development which is fairly constant across populations (Đşcan, 1989). For adults, the
estimation of age is more difficult since aging patterns are less obvious and an
accumulation of years allows for greater internal differences to develop among the
various skeletal age indicators (Đşcan, 1989; Ubelaker, 2008). For example, “a skeleton
may present a relatively youthful-appearing pubic symphysis and sternal end of the fourth
rib and yet show premature arthritic development and extensive tooth loss” (Ubelaker,
2008:40). Despite these challenges, the pubic bone demonstrates significant age-related

4

changes, which is why it is considered a reliable skeletal region for assessing age at death
in both archaeological and forensic investigations (Scheuer and Black, 2000).
There are numerous explanations for the irregularity of adult aging, the most
significant being human variation. Human aging is progressive and universal but agerelated processes are highly variable (Schmitt et al., 2002). Rates of skeletal remodeling
are directly affected by both genetic and non-genetic processes and are inconsistent
between and among individuals and populations. Skeletal remodeling can be influenced
by multiple factors including diet, health, cultural practices, living environment, and
trauma, and since the majority of adult skeletal aging methods rely on evaluation of
remodeling changes, adult age estimation can be complicated for physical anthropologists
(Đşcan, 1989; Buckberry and Chamberlain, 2002; Schmitt et al., 2002; Mulhern and Jones,
2005; Falys et al., 2006; Hens et al., 2008). As a result, methods designed to estimate age
at death for human skeletal remains are continually being developed and revised (Schmitt
et al., 2002).
Accuracy for Adult Age at Death Estimation Methods
Within the field of forensic anthropology, accuracy is a relative term since each
morphoscopic age-at-death estimation technique has unique measures of accuracy and
precision (Buikstra and Komar, 2008). For example, each phase of the Suchey-Brooks
(1990) pubic symphyseal surface aging method (discussed in the following chapter) has a
mean, standard deviation, and 95% age range, which is not the same as two standard
deviations due to the non-normal distribution of the sample (Klepinger, 2006; Buikstra
and Komar, 2008). Unfortunately, as Schmitt (2004) and Klepinger (2006) note, even
this large 95% range underestimates the extent of variability. One study, however, states

5

that the Suchey-Brooks (1990) method accurately assigns 72.0% of females and 89.74%
of males into the correct phase in a Balkan sample (Djurić et al. 2007).
For the medial surface of the right fourth rib, Đşcan and Loth (1984; 1985) provide
a mean, standard deviation, standard error, 95% confidence interval and an age range to
assess accuracy. However, as Klepinger (2006) emphasizes, the small sample sizes for
these studies result in misleading confidence intervals. When the Đşcan and Loth (1984;
1985) studies were tested for inter-observer error, Krogman and Đşcan (1986) note that
accuracy is within one age phase (which has mean accuracies between 4 and 20 years
from actual age at death).
For the auricular surface method, originally developed by Lovejoy et al (1985a),
the eight modal age stages are distributed in non-overlapping groups of 5 to 10 years.
Thus, Lovejoy et al. (1985a) were confident that age could be accurately estimated within
5 to 10 years. These age ranges were deemed to be too narrow by Buckberry and
Chamberlain (2002) who revised the Lovejoy et al (1985a) method. Buckberry and
Chamberlain (2002) present a mean, standard deviation, median age, and age range for
each stage as measures of accuracy. A test of Buckberry and Chamberlain’s (2002)
method, using inaccuracy and bias as the measures of accuracy, find that the original
Lovejoy et al. (1985a) method is on average 8 years more accurate than the revised
method for individuals aged 20 to 49; but for individuals 50 to 69 years, the revised
method is almost 7.5 years more accurate than Lovejoy et al. (1985a) (Mulhern and Jones,
2005).
As these examples demonstrate, there is not a standard age, age range, or
confidence interval which identifies an age-at-death estimation as “accurate.” Each

6

method employs different means for measuring accuracy, as well as different values for
what is considered “accurate.” While Lovejoy et al. (1985a) consider a range of 5 to 10
years to be an accurate estimation of age at death, Suchey-Brooks (1990) consider 58
years to be an accurate range (Suchey-Brooks Phase V for Females can range between 25
to 83 years). Therefore, accuracy, in forensic anthropology, is contextually related to
individual methods.
The Pubic Bones and the Pubic Symphysis Joint
The pubic bone (or pubis) is one of three
elements that, when mature, form the large
irregular-shaped bone of the hip—the os coxa.
Before maturity, each of the paired os coxae is
composed of three bones (ilium, ischium, and
pubis) and several epiphyses (Figure 1) (McKern
and Stewart, 1957). Around the time of puberty
these three elements begin to unite and fuse at the
acetabulum (Scheuer and Black, 2000).

Figure 1: The os coxa. The shaded
region is the pubic bone (Modified from
Scheuer and Black, 2004).

The pubic bone, according to Scheuer and Black (2000), forms the anterior lower
portion of the os coxa and is composed of a body, a superior ramus which fuses with the
illium, and an inferior ramus which fuses with the ischium. The left and right pubic
bones converge, although separated by symphyseal cartilage, at the midsagittal plane to
form the pubic symphysis joint (Figure 2) (Krogman and Đşcan, 1986). This is a slightly
moveable (amphiarthoris) joint with a fibrocartilaginous disk between the two bones
(Suchey and Katz, 1998). The joint surface of each pubic bone, the symphyseal surface,

7

undergoes systematic and
observable changes over the
course of a lifetime (Todd,
1920; McKern and Stewart,
1957). The symphyseal
surface in young individuals
has alternating horizontal

Figure 2: Pubic symphyseal surfaces (Todd, 1920)

ridges and depressions, but with age, the ridges and depressions begin to diminish and
eventually the surface appears flat or concave (Todd, 1920).
Historical Research on the Pubic Bone
Morphological changes in the pubic bone throughout life, and their relationship to
skeletal age, have been recognized for centuries (Todd, 1920; McKern and Stewart,
1957). Beginning in 1761, according to Todd (1920), the anatomist Dr. William Hunter
was the first to observe the similarities between the pubic symphysis and the joints
between each vertebra; both being fibrocatilaginous symphyses. Dr. Hunter was
followed by Swiss anatomist Christoph Aeby, who in 1858, described the symphyseal
face of the pubis “as a more or less irregular convex surface bounded by an oval outline”
(Todd, 1920:294). In 1872, German physician and anatomist Friedrich Henle recognized
that with age, the symphyseal face experienced variation in texture and dimensions, and
in 1889, anatomist John Cleland described changes in the pubic bone which he attributed
to differences between males and females. These differences were later attributed to agerelated modifications rather than sex (Todd, 1920). Although these early scientists

8

recognized that the pubic bone underwent gross morphological changes, they did not
associate specific changes with particular age stages (McKern and Stewart, 1957).
The First Method to Evaluate the Relationship between Age and the Pubic Bone
The first systematic study and formal method for evaluating age-related
morphological changes of the pubic bone was conducted by T. Wingate Todd in 1920
(Suchey, Wisely and Katz, 1986; Meindl and Lovejoy, 1989). Todd notes that certain
bones associated with joints, especially amphiarthroses such as the pubic symphysis,
change with age. Based upon these changes, Todd began to study the pubic symphysis as
a region for age estimation (Todd, 1920). Prior to this time researchers felt there was a
distinct lack of data from which an accurate skeletal age estimate could be made. In
particular, there were no reliable criteria to estimate age for individuals between 25 and
55 years old (Todd, 1920).
In light of the need for aging data, Todd and his colleagues began amassing a
skeletal collection with the most accurate antemortem records that could be obtained at
the time. The collection, in 1920, (later to become the Hamann-Todd Osteological
Collection) consisted of 650 male and female skeletons of European and African ancestry.
For this first systematic study, 306 males of European ancestry were used to assess age
changes of the pubic bone. Todd notes that there was a problem with obtaining accurate
ages at death since civil and hospital records were not dutifully maintained; he also
observes that individuals tended to provide their ages in round numbers which produced
distinct increases in the numbers of individuals within certain age categories (Todd,
1920). Thus, it is possible that many individuals in the sample had age estimates “which
are more than a couple of years off” (Suchey, Wiseley, and Katz, 1986:36). Todd also

9

distinguishes between two different types of skeletal aging: normal developmental
progression and morphological changes induced by external factors such as disease
(Todd, 1920).
Based upon his investigations, Todd developed a ten-phase system which
describes changes of the male pubic bone for individuals between the ages of 18 and 50
years (Figure 3) (Todd,
1920). He concludes
that the pubic bone is
best for aging
individuals between
20 and 40 years old
and that the technique
should not be used in
isolation. He suggests
using the pubic bone

Figure 3: Todd’s ten pubic symphysis phases (Suchey, Wisely,
and Katz, 1986)

in conjunction with other aging methods when the whole skeleton is available (Todd,
1920). Todd’s publication “Age changes in the pubic bone. I. The white male pubis,” is
heavily illustrated with photographs which serve as a guide for identification. The
photographs are useful visual references which demonstrate human variation and provide
examples of young skeletons, to which many scientists at that time did not typically have
access (Todd, 1920).
Revisions of the Todd Method

10

Numerous studies have reviewed or improved upon the Todd method for pubic
bone age estimation. The first to evaluate this method was Kazuro Hanihara who
conducted a study in 1952 in which Japanese pubic bones were aged using Todd’s tenphase system (Hanihara and Suzuki, 1978). Hanihara concludes that there is little
difference between Caucasian and Japanese individuals when the Todd method is applied.
However, if Caucasian standards are used, Japanese individuals tend to present as three
years older than their actual age (Hanihara and Suzuki, 1978).
The first validation and systematic study of Todd’s pubic aging method was
conducted by Sheilagh T. Brooks in 1955 (Suchey, Wiseley, and Katz, 1986). Two
samples were used for this study. The first was the University of California Museum of
Anthropology (UCMA) series which consisted of approximately 400 archaeological
individuals. The second was the Western Reserve University (WRU) series which was
the original sample upon which Todd based his aging method (Brooks, 1955). Brooks
notes that the Todd system has three general categories: Phases I-III are post-adolescent,
Phases IV-VI show the build up of the symphyseal outline, and Phases VII-X
demonstrate quiescence and breakdown of the symphyseal face. Based upon her
statistical analyses, Brooks states that Todd’s Phases V-VIII should be shifted three years
younger to accommodate for the method’s prevalence to over-age individuals. With
these slight modifications, she states that the pubic bone could be made into one of the
more reliable adult age indicators (Brooks, 1955).
Focusing particular attention on the pubic symphyseal face, McKern and Stewart
(1957) were the next group to amend the Todd system. The pubic symphyseal face is
ideal because its modifications and epiphyseal-like additions of bone continue into later

11

adulthood, while most other elements of the skeleton have quiesced (Meindl and Lovejoy,
1989; McKern and Stewart, 1957). In their 1957 study, McKern and Stewart note that
Todd’s method only addresses typical phases, with “typical” indicating regular
progression through each phase rather than considering symphyseal variation. To
ameliorate this problem they developed a new system which accounted for all variation.
They chose a formula which evaluates three components, each with individual
chronological age sequences: the dorsal plateau, ventral rampart and the symphyseal rim.
Each component includes five subdivisions that are scored independently (Figure 4)
(McKern and Stewart, 1957). A score of 0-5 is assigned for each of the three components
and the sum of those three scores is used to derive the estimated age of the individual
(Suchey, Wiseley, and Katz, 1986).

Figure 4: McKern and Stewart pubic symphyseal components (1957)
12

McKern and Stewart (1957) describe the age-related changes of the symphyseal
face as follows: The dorsal plateau begins on the dorsal demi-face and gradually fills in
until it becomes flat. The ventral rampart is more variable and may remain incomplete
leaving a hiatus. The ventral rampart begins as a beveling (flattening) of the ventral
demi-face and then the rampart begins to form as extensions of either the superior or
inferior surfaces. The symphyseal rim is the final stage of maturation. At the same time
the rim is completing, the bone texture begins to change from a granular to a more finegrained and dense bone. With increased age, the rim begins to breakdown and becomes
smooth.
The McKern and Stewart system is flexible, so that with a slight change in
scoring, the age estimation will be within an accurate range (McKern and Stewart, 1957).
This study, however, does not incorporate many individuals in older age groups (only
15% of the sample was older than 31 years) and is developed exclusively on military
servicemen returned from the Korean War (Meindl and Lovejoy, 1989; Suchey, Wiseley,
and Katz, 1986). With increased age the pubic symphysis tends to decelerate its
metamorphosis, and this is not well documented (McKern and Stewart, 1957). This is
problematic because the method does not provide a complete representation of the
changes that occur in older adults. Nonetheless, McKern and Stewart feel that their new
method “expresses the true nature of symphyseal variability and does not confine the
observer to the narrow limits of typical phases” (1957:88).
A study conducted by Suchey, Wiseley and Katz in 1986 evaluated both the Todd
(1920) and McKern-Stewart (1957) methods. Using a modern autopsy sample of 739
males collected from Los Angles County autopsies (first described by Katz and Suchey,

13

1986) the authors reexamined Todd’s phases of pubic development and the McKernStewart method. Suchey, Wiseley and Katz (1986) conclude that the Todd method tends
to overestimate the ages in the sample by an average of 3 to 4 years, and that the
McKern-Stewart method works fairly well for individuals under the age of 25 (standard
deviations are all less than 5.0), but when individuals are older than this, the method
becomes less useful (standard deviations are between 6.52 and 13.89). These results are
to be expected as the McKern-Stewart sample had few individuals over the age of 30
years. Suchey and colleagues suggest that both methods poorly predict ages in older age
groups and that the cutoff age for using these methods should be 40 years. Finally they
state that modifications are needed for both aging methods before they are applicable to
modern, diverse and older populations (Suchey, Wiseley, and Katz, 1986).
In 1978, Hanihara and Suzuki developed yet another method based on the Todd
system. Their study assesses age from the pubic symphysis by employing multiple
regression analysis and quantification theory model I analysis (defined as a multiple
regression analysis with dummy variables) (Igarashi et al., 2005). This study is limited to
individuals between the ages of 18 and 38, because after the age of 40, the variability
becomes too great for the method to accurately age (Hanihara and Suzuki, 1978). Only
70 pairs of Japanese pubic bones were used to conduct this study and both males and
females were combined, even though Hanihara and Suzuki (1978) state that it is desirable
to separate males and. They examine seven features for their model: ridges and furrows,
pubic tubercle, lower end, dorsal margin, superior ossific nodule, ventral beveling and
symphyseal rim. Hanihara and Suzuki describe both multiple regression analysis (MRA)
and quantification theory model I (QMI) and state that QMI is more reliable than MRA,

14

especially for individuals younger than 30 years old (confidence ranges for QMI are
between -6.51 and 4.89 while for MRA they are between -5.36 and 6.46) . They also
note that ages can be calculated using normalized scores (Hanihara and Suzuki, 1978).
A study conducted in 1985 by Meindl and colleagues use blind tests to assess
current pubic aging methods for accuracy. The McKern-Stewart, Gilbert-McKern (pubic
aging method for females), Hanihara-Suzuki, and Todd systems were examined. Meindl
et al. (1985) address the concern of ancestry and sex on the application of pubic aging,
since a major critique of aging standards is that they are strictly based on European
populations. Statistical analyses show that no race-sex combination produced a
significant bias in age determination. The results of this work also demonstrate that a
subjective aging application is superior to a typological aging approach. “Archetypal
aging,” according to Meindl et al., “is less accurate than a more generalized and
interpretive method” (1985:33). By this they mean that examining each specimen in
accordance with descriptive explanations of the aging pattern is more accurate than
matching each specimen to a typological cast. The Todd system, according to the authors,
was being used as a typological standard when it was intended to be used in a more
subjective manner to describe age changes (Meindl et al., 1985). Meindl and colleagues
revise the Todd method and establish five major biological phases: pre-epiphyseal, active
epiphyseal, immediate postepiphyseal, maturing/pre-degenerative, and degenerative.
Based upon their results, they conclude that their modified Todd system is the most
reliable, however all of the methods underestimate the ages of the sample individuals
(Meindl et al., 1985).
Another study evaluating methods for determining age at death from the pubic

15

symphysis was undertaken by Katz and Suchey in 1986. In particular, the Todd system
and the McKern-Stewart system were examined. This study evaluates a sample of 739
documented male pubic bones from Los Angeles county autopsies. The specimens range
in age from 14 to 92 years old (Katz and Suchey, 1986). The main purpose of this study
is to use a large sample and apply regression analysis to study the performance of aging
methods. The sample for this study, as mentioned above, has a wide age distribution and
all individuals have known birth and death dates. The individuals were born both in the
United States and in foreign countries, and pubic bones outside of normal morphology
were removed from the study sample. The ancestry of each individual was determined
by a typological approach focusing on appearance, which today is inherently problematic.
Katz and Suchey feel that that their sample is representative and is superior to the
samples used for the Todd and McKern-Stewart methods (Katz and Suchey, 1986).
Katz and Suchey heavily critique both the Todd and McKern-Stewart methods.
They note that the collection used by Todd consisted mostly of males over the age of 40,
and that some of the specimens did not have known ages (only three individuals had
recorded documents of birth). They state that the original Todd system usually over-ages
individuals, so they recommend a modified system with only six phases. The McKernStewart sample consisted predominantly of white males in their twenties, all of whom
had died in the Korean War. The McKern-Stewart method focuses on three components
of the symphyseal face, but Katz and Suchey feel that the system is difficult for
inexperienced individuals to use (Katz and Suchey, 1986).
Katz and Suchey conclude that the Todd system tends to over-age individuals by
approximately 5 years, and neither the Todd nor the McKern-Stewart system accounts for

16

variability in older individuals. At the time this article was written, it was generally
accepted that pubic bone morphology could not accurately be used to determine the age
of older individuals. As such, the authors eliminated specimens over the age of 40 for
their statistical analyses. The regression analyses improved when individuals over this
age were removed, with the standard deviations being reduced by an average of 5 units
(Katz and Suchey, 1986). When the statistical data were reviewed, the authors note that
the Todd (especially the modified six-phase Todd) system is more accurate than the
McKern-Stewart method. They also state that these standards can not be used to assess
ages for females because an analysis of females pubic bones demonstrated that females
present more variability than males (Katz and Suchey, 1986).
Yet another study was conducted by Sinha and Gupta in 1995 to determine if
Indian pubic bones could be accurately aged using Western standards. The sample for
this study consisted of 82 pairs of male pubic bones autopsied at Lady Hardinge Medical
College and Associated Hospitals. The ages of each individual were recorded and
verified via police records and birth certificates; the ages ranged from 12 to 75 years old.
Samples were excluded from the study if they were pathological, if they showed signs of
fracture, of if there was doubt as to the age of the individual (Sinha and Gupta, 1995).
Multiple features of the pubic symphysis were examined for this study: ridges and
furrows, dorsal margin, ventral beveling, lower extremity, ossific nodule, upper extremity,
ventral rampart, dorsal plateau, and symphyseal rim. Sinha and Gupta examined and
documented the formation and completion/disappearance of each trait. The pubic bones
were then aged according to the Todd, McKern-Stewart, and Hanihara-Suzuki aging
methods (Sinha and Gupta, 1995).

17

Most studies suggest that Todd’s system imprecisely ages individuals, but Sinha
and Gupta (1995) found that it ages Indian males fairly accurately. When the McKernStewart system is applied, some elements of the Indian pubic bones develop one to three
years earlier than the method suggests for individuals of European ancestry; yet others,
like the dorsal plateau develop at later ages in the Indian population. The Hanihara and
Suzuki method tends to overestimate the age of the Indian bones by one to six years until
the age of 30. Between the ages of 31 and 39, the method then underestimates the ages
by one to four years (Sinha and Gupta, 1995). Other than these broad generalizations,
Sinha and Gupta (1995) do not present conclusions or discussions regarding their
findings. They do not address the significance of their results or what implications their
study has for applying Western methods to other populations. It would seem that
Western methods are fairly accurate (there were both overestimates and underestimates
of ages depending upon the method), but it is impossible to tell from such limited data.
The Suchey-Brooks Pubic Bone Aging Method
The ability to estimate age using the pubic symphyseal face gained more
widespread recognition with the introduction of the Suchey-Brooks method in 1990.
This method was developed using 1,225 pubic bones (739 males and 273 females)
obtained from Los Angeles County autopsies and can be used to age individuals between
14 and 99 years old (Brooks and Suchey, 1990; Suchey and Katz, 1998). The SucheyBrooks method is a revision of the modified Todd (1920) aging method developed by
Katz and Suchey in 1986. The Suchey-Brooks method focuses on the total pattern of
bone changes rather than individual elements or features and has six phases. Each of the
phases is represented by two bones demonstrating an “early pattern” and a “later pattern”

18

of development (Brooks and Suchey, 1990). Detailed phase descriptions, line drawings,
and plaster casts are available to assist observers in assigning an appropriate phase
(Suchey and Katz, 1998) (Figure 5).
Figure 5: Line drawings of the six male Suchey-Brooks pubic bone age phases
(Buikstra and Ubelaker, 1994).

In 1980, according to Brooks and Suchey (1990), the Workshop of European
Anthropologists recommended the Acsádi-Nemeskéri (1970) method as a system for age
estimation based on pubic bone changes. Suchey and Brooks evaluated the AcsádiNemeskéri system using their modern collection and recently established method. Since
casts for the Acsádi-Nemeskéri method were not available, Brooks and Suchey analyzed
the technique based upon the descriptions and photographs in the 1970 publication
(Brooks and Suchey, 1990). This method did not study many young individuals, as only
38% of the specimens were between the ages of 23 and 50 years old. Thus, the AcsádiNemeskéri study focused mainly on early and late morphological changes, with little
attention paid to the intermediate morphologies. The study also used samples from a

19

Hungarian cemetery, but the original publication does not explain how the ages of the
skeletons were obtained, or even if the skeletons were modern.
When the Suchey-Brooks collection was evaluated using the Acsádi-Nemeskéri
method, almost half of the males and females did not fit into a phase. Brooks and Suchey
(1990) believe that this is due in part to the lack of intermediate phases for the AcsádiNemeskéri method. This method focuses on the application of the symphyseal face for
“calculating whether an individual is under 50, about 50, or above 50 years” (Brooks and
Suchey, 1990:234). Brooks and Suchey conclude that the Acsádi-Nemeskéri method is
not supported by the modern Suchey-Brooks sample.
Today, particularly in the United States, the Suchey-Brooks system is the most
commonly used method for estimating age at death from the pubic bone (Baccino et al.,
1999; Djurić et al., 2007; Chen et al., 2008; Hens et al., 2008; Buikstra and Komar, 2008).
However, Suchey and Katz (1998) note that the successful use of the Suchey-Brooks
technique depends on a scientist’s ability to recognize key pubic bone features,
understand the terminology, and analyze pubic bones from varying states of preservation.
If these criteria are met, the use of the pubic bone for adult age estimation has been
shown to be of value (Suchey and Katz, 1998).
Tests of the Suchey-Brooks Pubic Bone Aging Method
Numerous studies have been conducted to test the accuracy of the Suchey-Brooks
(1990) method, especially for diverse populations. Saunders and colleagues (1992) tested
four morphological methods of adult age estimation including the Suchey-Brooks
technique. This study evaluated between 27 and 49 skeletons from a Canadian pioneer
cemetery in Ontario; however, the age distribution of this sample was older, with most of

20

the individuals being over 50 years old. The results indicate that the Suchey-Brooks
pubic bone method tends to underage the individuals by an average of 18 years, and has
high biases (up to 22.4 years for individuals older than 60 years). Saunders et al. (1992)
conclude that the Suchey-Brooks published age phases are so broad and often overlap
that only the youngest and oldest phases are mutually exclusive.
In 1996, Santos tested the Suchey-Brooks (1990) method on a sample from the
“Museu de Antropologia” of the University of Coimbra, Portugal. The study sample
consisted of 231 pelves of males and females between the ages of 16 and 95 years of age.
The results of this study are not encouraging. For the Suchey-Brooks method, 85.4% of
males and 76% of females were misclassified. Only Phase I (the youngest individuals)
was found to be distinct from the other five phases. Santos (1996) states that the SucheyBrooks pubic symphysis casts are useful for the application of the method, but the choice
of two extreme morphologies (young phase and old phase) is a possible source of error.
Santos feels that a third cast, representative of a middle phase, would be beneficial. She
concludes that this method is inaccurate for older individuals, and that the cut-off age for
using this method should be 40 years. She suggests that the main cause of error for this
method is its broad age ranges.
Baccino and colleagues conducted a study in 1999 evaluating four methods of
adult age estimation including a dental method, the sternal end of the fourth rib, the pubic
symphyseal surface, and femoral cortical remodeling. An autopsy sample of 19 French
individuals was assessed. With regards to the Suchey-Brooks pubic bone method,
Baccino et al. (1999) note that it has the largest standard error and very high mean bias
among different observers. Based upon this study they conclude that comprehensive

21

approaches are superior to individual aging methods when it comes to aging unknown
individuals.
In 2004, Schmitt tested two methods for estimating adult age at death: the
Suchey-Brooks (1990) pubic symphysis aging method and the Lovejoy et al. (1985b)
auricular surface method. This study was conducted on a Thai skeletal collection and
evaluated 37 males and 29 females, but the collection only had 8 individuals younger
than 40 years old. This was the first time that these aging techniques had been tested on
an Asian skeletal sample. The overall results for the Suchey-Brooks method indicate that
bias and inaccuracy increase with age and that the method tends to underage all
individuals, except for the youngest age group. Only 11 females and 13 males fell within
the Suchey-Brooks 95% confidence intervals. Because these confidence intervals are
wide, according to Schmitt (2004), one would expect that more individuals from the
sample would be correctly classified into an age phase. This study is not encouraging for
the application of these methods to Asian populations, and should therefore be applied
with caution. Based upon this information, the author encourages population-specific
standards for age at death estimation methods.
A study was conducted in 2007 by Djurić and colleagues to test whether the
Suchey-Brooks (1990) method could successfully age adult populations in the Balkans.
This method is often used for forensic identification of war victims in the former
Yugoslavia; thus, its accuracy needed to be demonstrated for this particular population.
A modern Serbian sample of known age, consisting of 33 female and 52 male pairs of
pubic bones collected from autopsy cases, was evaluated. Djurić et al. (2007)
demonstrate that the Suchey-Brooks method has an accuracy of 89.74% for males and

22

72.0% for females. Although this level of accuracy is fairly high, Djurić and colleagues
suggest that population-specific standards need to be established to avoid the problems
that arise between the reference sample and the population being studied. Based upon
this study, several adjustments to the original Suchey-Brooks method are proposed which
could be useful for aging Serbian populations in the future.
Martrille and colleagues evaluated four age estimation indicators for adults of
European and African ancestry in 2007. A sample of 218 American individuals ranging
in age from 25 to 90 years of age was studied from the Terry Collection. The sample was
chosen to represent a balanced distribution of sexes, ages, and ancestries. When the
sample is divided into age groups, the Suchey-Brooks method is the most accurate for the
youngest adults between the ages of 25 and 40 years, with an average error of 6.2 years.
After 60 years of age, or the oldest group, all four aging methods become highly
inaccurate (with average errors between 13.4 and 17.4 years), and all methods tend to
overage young individuals and underage older individuals (Martrille et al., 2007). For
young adults of African and European ancestry, Martrille et al. find that the SucheyBrooks method is the most reliable, with the least amount of error between 3 and 4.7
years, but this does not hold true for older adults.
In 2008, Hens and colleagues tested the Suchey-Brooks (1990) and Lovejoy et al.
(1985b) auricular surface aging methods on a known sample from the Sassari Collection
at the Museum of Anthropology at the University of Bologna, Italy. The Suchey-Brooks
method is quite accurate for the youngest adults, but bias and inaccuracy increase with
age. After the age of 40 years, for both sexes, this method underestimates the ages of the
individuals in the sample. This study demonstrates lower levels of inaccuracy and bias

23

than both Schmitt (2004) and Saunders (1992), but the trends of over-aging the young
and under-aging the old are similar. Overall, the results of this study suggest that the
auricular surface aging method performs better than the Suchey-Brooks method,
especially for older individuals (Hens et al., 2008).
Finally, Hartnett tested the accuracy of the Suchey-Brooks (1990) method on a
modern, documented, autopsy-based sample in 2010. A sample of 419 males and 211
females of known sex, age, and ancestry from the Forensic Science Center (FSC) in
Phoenix, Arizona were evaluated. Based upon the critiques of the Suchey-Brooks
method in anthropological literature, Hartnett establishes three goals for this research: 1)
to establish a new, documented sample for future education and research, 2) evaluate the
Suchey-Brooks standards on a modern, large and diverse skeletal sample, and 3) propose
revisions to the method that will increase its precision and accuracy (Hartnett, 2010).
The statistical analysis demonstrates that there are significant differences between
the actual and observed ages and that there is significant inter-observer error. Hartnett
describes a new phase 7, comprised of both males and females, for individuals over the
age of 70 years. With this additional phase, older individuals can be aged more
accurately because a phase and age range can be assigned, rather than stating that an
individual is 50+ years of age. To conclude, Hartnett states that the Suchey-Brooks
method is not an extremely accurate technique for estimating age at death.
The Chen et al. Pubic Bone Aging Method
Focusing on multiple morphological features of the male pubic bone, Chen and
colleagues developed a new method in 2008 for age estimation. For this study, 262 male
pubic bones from Chinese Han individuals were analyzed (Chen et al., 2008). Using

24

similar statistical analyses to Hanihara and Suzuki (1978), Chen and colleagues
developed four equations for male age estimation: multiple regression analysis (MRA)
and gradual regression analysis (GRA) were used to statistically analyze the nine features,
while quantification theory model-I (QMI) and GRA were applied to compare with the
MRA.
The Chen et al. (2008) study has two main goals. The first is to develop a method
for age-at-death estimation based upon morphological changes. The second is to improve
upon the aging methods of Hanihara and Suzuki (1978) and Suchey-Brooks (1990).
Chen and colleagues believe that the Todd (1920), Hanihara and Suzuki (1978), and
Suchey-Brooks (1990) methods have shortcomings. They note that their study
overcomes these limitations by using a large sample size, separating specimens by sex
(only used males), individually describing nine morphological indicators, and
subdividing the nine features with distinct scores (Chen et al., 2008). They state that with
the use of their statistical formulae, this method is highly reliable and that “age at death
can be quantitatively estimated in people of Chinese male ancestry with a fairly high
degree of accuracy” (Chen et al., 2008:42).
The Chen et al. (2008) pubic bone aging method employs nine indicators of
morphological change to evaluate age-at-death. These indicators are “ridges and furrows
on the symphyseal surface, ridge of pubic tubercle, lower extremity, ventral beveling,
ossific nodules, dorsal margin, ventral rampart, general macroscopic changes of
symphysial surface, and bone density of the symphysial surface” (Chen et al., 2008:36).
Each of these nine features is divided into three-, four-, or five-stage categories and
assigned a respective score of 0 through 2, 3, or 4. Each score has an accompanying

25

Figure 6: Example of ridges and furrows on the symphyseal surface: (0-4) (Chen et al.,
2008).

0

1

2

3

4

Ridges and furrows on the symphysial surface: The ridges and furrows on the symphysial
surface vary from individual to individual, and change with increasing age. Ridges and
photograph of symphysial surface were alterations and almost all
furrows on the the visible morphological classified into five stages. have written
0: ridges and furrows alternate distinctly;
1: the furrows fill in and ridges and furrows alternate indistinctly;
2: the bone substance has a granular look with low, blunt ridges and shallow furrows;
3: the surface becomes flat and fine-textured, and/or again becomes more granular;
4: ridges and furrows disappear entirely and the surface becomes pitted and eroded.

descriptions. Figure 6 is an example of one of the Chen et al. photographs and scores.
These descriptions are provided to assist observers in assigning a score.
The Chen et al. method examines both the right and left pubic bones, but does not
elucidate whether an average is taken for the pair. The majority of literature on pubic
bone aging methods does not explicitly state whether the right or left pubic bones have
been evaluated; if both sides are assessed, most literature does not specify if an average is
derived from both bones (Hanihara and Suzuki, 1978; Meindl et al., 1985; Brooks and
Suchey, 1990). Only a few studies address this issue (Buckberry and Chamberlain, 2002;
Hens et al. 2008). Hens et al. (2008) did not find significant differences when evaluating
the right or left pubic bone. Although Buckberry and Chamberlain (2002) did not
investigate the pubic bone, their results show no statistical difference between evaluating
the right and left ilia. Subsequently, both Buckberry and Chamerlain (2002) and Hens et

26

al. (2008) choose to analyze the left side when both right and left are available, and the
side that is available for all other individuals.
As described in the Introduction chapter, Chen et al. (2008) state, via a forensic
case report example, that their method can accurately age a Chinese male within one year
from actual age at death. This is the only independent test of the method that they report.
Additionally, Chen and colleagues do not present accuracy values, rates of error, or levels
of significance for their method. These are distinct weaknesses of this method.
As this review has established, there are numerous methods which can be used to
estimate the age of an individual from the pubic bone. The morphological changes of the
pubis were recognized as early as the mid 18th century, but they were not attributed to the
progression of age. As described above, many different systems have been developed to
address age related changes in the pubic symphysis. However, there is still a discrepancy
as to the upper age limits of pubic symphysis aging methods; some studies suggest that
the upper limit is 40 years old (Todd, 1920; Hanihara and Suzuki, 1978; Suchey, Wisely,
and Katz, 1986; Katz and Suchey, 1986; Santos, 1996; Hens et al., 2008), while others
suggest that the upper limit is around 60 to 70 years old (Martrille et al., 2007; Chen et al.,
2008; Hartnett, 2010).
All of these studies reveal that the pubic bone is the most frequently used
anatomical feature for estimating the ages of unknown individuals. They also
demonstrate that the assessment of age-at-death for adult human skeletal remains is
central to both forensic and bioarchaeological analysis. Therefore, understanding the
morphological development and remodeling of the pubic bone is crucial for all
osteologists.

27

STUDY OBJECTIVES
As the previous chapter indicates, the anthropological literature is not lacking for
adult age estimation methods which utilize the pubic bone. Although small, one gap in
the literature is an independent test of the Chen et al. (2008) pubic aging method. The
purpose of this thesis project, therefore, is to evaluate the Chen et al. (2008) pubic bone
aging study, and to determine if the method can accurately estimate the age of individuals
outside of the Chinese Han population. Despite the statements by Chen et al. that their
method is highly accurate and reliable, their publication does not provide rates of error.
This project will addresses this weakness, and contribute to the adult age estimation
literature, by generating statistical data regarding error, significance, and accuracy via an
independent evaluation of the Chen et al. method.
Furthermore, the Chen et al. (2008) publication does not discuss how accurate the
method is for different age groups. This study will test three different age groups to
assess the accuracy of the Chen et al. method for young, middle, and old individuals.
Bias will also be evaluated for the three age groups.
The overall results of this project are expected to support the hypothesis that the
Chen et al. pubic aging method can accurately assess age-at-death for the males of
European ancestry in the Maricopa County Forensic Science Center collection. These
results are expected because discontinuities in the aging process of the pubic bone are not
common among different ancestries (Todd, 1921; Meindl et al., 1985)
An additional aspect of this project is to compare the accuracy of the Chen et al.
method to the Suchey-Brooks method, which is currently the most accepted technique for
estimating age from the pubic bones (Baccino et al., 1999). The purpose of this

28

comparison is to demonstrate that the Chen et al. technique is not simply another method
for estimating age; rather, it is an improved and more accurate method for estimating ageat-death using male pubic bones.
Research Questions and Hypotheses
Research Questions
The objective of this research is to address the following questions:
1. Will the Chen et al. (2008) method accurately assess age-at-death for males of
European ancestry?
2. Will a revised Chen et al. method accurately assess age-at-death for males of
European ancestry?
3. Will a revised Chen et al. method be more accurate than the Suchey-Brooks pubic
aging method?
4. Which morphological features of the Chen et al. method are most predictive of
age?
5. How accurate is the revised method for each age group?
6. What are the rates of error between and within the observers?
Research Hypotheses
1. The Chen et al. (2008) method will accurately assess age-at-death for nonChinese males.
2. The revised method will be more accurate at assessing age-at-death for males of
European ancestry than the Original Chen et al. method.
3. There will be no significant differences between the rates of accuracy and the R
values of the revised Chen et al. method and the Suchey-Brooks method.

29

2

4. Based upon the data presented by Chen et al. (2008), it is hypothesized that
ventral beveling, bone density of the symphysial surface, and the ventral rampart
will be the morphological features most predictive of age.
5. The revised method will be most accurate for young adults (18-34 years) and
middle adults (35-49 years). The revised method will be less accurate for old
adults (50 years and above).
6. Both the inter- and intra-observer rates of error will be low.

30

MATERIALS AND METHODS
Materials
This research is based on a modern pubic bone and sternal fourth rib end
collection curated at the Maricopa County Forensic Science Center (FSC) in Phoenix,
Arizona. This skeletal sample was obtained between January 2005 and June 2006 from
individuals of known age, ancestry, and sex. The specimens were acquired from
autopsies conducted at the FSC and from donated cadavers at Barrow Neurological
Institute in Phoenix, Arizona (Hartnett, 2007; Hartnett, 2010). This new skeletal sample,
as indicated by Hartnett (2010:6), is an important, autopsy-based sample for all
anthropologists to study, and “will provide a means for independent testing and reevaluation of rib and pubic bone techniques.” This research has utilized the FSC
collection for just such an evaluation.
In order to directly replicate the Chen et al. method, only males were evaluated
for this study. To age females, a new method would be required to address the greater
variability seen in the female pubis (Katz and Suchey, 1986; Brooks and Suchey, 1990;
Klepinger, 2006). Although females were not evaluated for this project, they are
described here to provide a comprehensive understanding of the FSC collection (Table 1).
The collection of 630 individuals is composed of 419 males and 211 females between the
ages of 18 and 99
years old. The
males range in
age from 18 to 97

Table 1: Summary Statistics of the FSC Pubic Bone Sample
from Hartnett (2007)
N
Mean
SD
Age Range
Total Male Sample
419
52.6
19.0
18-97
Total Female Sample
211
59.2
21.4
18-99
Sample Total
630
54.8
20.1
18-99

years old with a mean age of 52.6 years, and the females range from 18 to 99 years old

31

with a mean age of 59.2 years (Figure 7). However, due to preservation issues,
insufficient antemortem data, or pathological conditions, four males were eliminated
from the sample; also, four males and two females are exclusively represented by rib
ends (which are not pertinent to this project) reducing the available pubic bone sample to
620 individuals (411 males and 209 females) (Hartnett, 2007).
Figure 7:

The FSC collection is represented by Asians (n = 4), African Americans (n = 20),
European ancestry (white) (n = 598), and Native Americans (n = 8). Individuals who
were self-identified as Hispanic (n = 40) were included in the European ancestry category
according to the FSC protocols (Hartnett, 2007; Hartnett, 2010). This project focuses
32

exclusively on males of European ancestry between the ages of 18 and 70 years. Only
individuals of European ancestry are evaluated to directly correspond with the
homogeneous Chinese Han sample used by Chen et al. (2008). After females, individuals
of non-European ancestry, individuals represented exclusively by ribs or with
pathological pubic bones, and individuals over the age of 70 were excluded from the FSC
collection, a sample of 296 male pubic bones was selected for analysis.
Methods
The objective of this research is to statistically evaluate the Chen et al. (2008) and
Suchey-Brooks (1990) aging methods. The Chen et al. (2008) method is evaluated to
determine if it is a valuable technique for use in the United States and whether it is more
or less accurate than the Suchey-Brooks (1990) method.
The Chen et al. Method
To estimate age, the Chen et al. (2008) method uses four statistical equations:
multiple regression analysis (MRA) and its gradual regression analysis (GRA) to
statistically analyze the multiple features, and quantification theory Model-I (QMI) and
GRA are applied simultaneously to compare with MRA. For this project’s statistical
analyses the equations are abbreviated as follows: MRA, MRA+GRA, QMI, and
QMI+GRA.
Based upon the discussion of evaluating the right or left pubic bone in the
Background chapter, a decision was made for this project to evaluate the left pubic bone
of each individual. This decision reduces ambiguity for future research and eliminates
one potential source of error. In the event that the left side is damaged or can not be
scored, which occurred for ten individuals in this sample, the right side is scored.

33

However, the use of only the left pubic bone might present an inaccurate age assessment,
especially if one bone of the pubic symphyses sustained trauma during life, or appears
younger or older than the opposite side.
The Chen et al. method has an attenuated age distribution of 14 to 70 years.
Therefore, only individuals between the ages of 18 and 70 (18 being the youngest
individuals in the FSC collection, and 70 being the oldest age which the Chen et al.
method can estimate) from the FSC collection are evaluated for this study. Based upon
the age groups established by Buikstra and
Ubelaker (1994), the FSC male pubic bones
are divided as follows: Young adults (18-34
years), Middle adults (35-49 years) and Old

Table 2: Age Groups for FSC Sample
Age Groups
Age Range
n
Young Adult (1)
18-34 years
69
Middle Adult (2) 35-49 years
95
Old Adult (3)
50+ years
132

adults (50 years and above) (Table 2). The sample is divided as such to gauge accuracy
for younger and older individuals. Although not addressed in the original Chen et al.
(2008) study, other researchers have established that older individuals are more difficult
to age and produce more inaccurate age estimates (Todd, 1920; Hanihara and Suzuki,
1978; Katz and Suchey, 1986; Suchey and Katz, 1998; Baccino et al., 1999; Mulhern and
Jones, 2005).
The Suchey-Brooks Method
The Suchey-Brooks pubic bone aging method, as discussed previously, is derived
from an extensive autopsy sample of identified modern pubic bones. To apply this
method, an observer must examine the unknown pubic symphysis and match it to the
descriptions, line drawings, and/or casts depicting one of six morphological phases.

34

There are unisex descriptions, but separate casts, drawings, and descriptions for males
and females are preferable (Brooks and Suchey, 1990).
Intra- and Inter-Observer Error
Each of the 296 pubic bones for this study was evaluated in accordance with the
Chen et al. (2008) and Suchey-Brooks (1990) methods. The bones were examined by
four Observers all with osteological experience: 1) the author, a forensic anthropology
graduate student, 2) a professional forensic anthropologist, 3) a recent college graduate
with a degree in anthropology, and 4) an undergraduate anthropology student. The
author was self-trained in the Chen et al. (2008) method after having read the article and
observing numerous bones to understand the features. All Observers were given tutorials
by the author on how to use and apply the Chen et al. method before they began scoring
the bones. All Observers were previously familiar with the use of the Suchey-Brooks
method, having learned and applied it during their education and/or professional
experiences. The Observers recorded their Chen et al. scores and Suchey-Brooks phase
for each pubic bone on individual data-collection sheets. The actual age of each
specimen was unknown to all Observers during the scoring process. The impetus for
incorporating less experienced observers is to demonstrate the ease-of-use of this aging
method. If Observers Three and Four are found to have lower accuracy rates, it is an
indication that the method requires practice and training before proficiency is achieved—
it does not imply that the method is ineffective for professional anthropologists.
Prior to evaluating the accuracy of the Chen et al. (2008) and Suchey-Brooks
(1990) methods, inter- and intra-observer errors were tested. To measure intra-observer
error, a random sample of 100 previously scored pubic bones was reassessed by the

35

author; the bones were scored again for both the Chen et al. and the Suchey-Brooks
methods. These data were evaluated by calculating Cohen’s Kappa measurement of
agreement and Pearson’s correlation coefficient.
Inter-observer error was also evaluated for both the Chen et al. (2008) and the
Suchey-Brooks (1990) methods. Observer Two scored 200 pubic bones via the Chen et
al. method and 75 bones via the Suchey-Brooks method. Observers Three and Four
scored the same 50 pubic bones using both methods. These data were evaluated using
Pearson’s correlation coefficient.
Statistical Analysis Part 1: Evaluation and Revision of the Chen et al. (2008) Method
This project generated statistical data concerning the accuracy, rates of error, and
significance of the Chen et al. (2008) model’s utility for aging male populations of
European ancestry. The statistical analysis for this project was multifaceted; the Chen et
al. model was tested and then revised based on the FSC sample, and later the Chen et al.
and Suchey-Brooks methods were compared. All statistical analyses were performed
using SPSS statistical software version 18.
To construct their four statistical models (MRA, MRA+GRA, QMI, and
QMI+GRA), Chen and colleagues evaluated 262 male pubic bones from the Chinese Han
population (Chen et al., 2008). Unlike the homogenous Chen et al. sample, the FSC
collection is composed of multiple ancestral groups. In order to directly evaluate the
Chen et al. model, and to determine its utility for aging additional populations, 296 male
pubic bones of European ancestry from the FSC collection were scored as specified by
the Chen et al. (2008) method.

36

The nine scores for each of the 296 left male pubic bones were input into the
Original (published) Chen et al. MRA+GRA and QMI+GRA statistical models. These
two models, as opposed to all four, were evaluated because the Chen et al. study favors
the inclusion of the gradual regression analysis (GRA) when estimating an individual’s
age (Chen et al., 2008). Furthermore, when statistical analyses were run on the four
equations using the FSC data, the inclusion of GRA produced slightly more accurate age
estimates.
Based upon their popularity in forensic and bioarchaeological literature, bias and
absolute mean error (referred to as “inaccuracy” in some articles) were applied to
demonstrate the accuracy of the Chen et al. (2008) and Suchey-Brooks (1990) aging
methods (Lovejoy et al., 1985a; Meindl et al., 1985; Meindl and Lovejoy; 1989; Schmitt,
2004; Mulhern and Jones, 2005; Martrille et al., 2007; Hens et al., 2008; Passalacqua,
2010). Both absolute mean error (AME) and bias were calculated for each pubic bone
and for both Chen et al. Original statistical models (MRA+GRA and QMI+GRA).
Absolute mean error, or Σ |estimated age – actual age|/N, is the absolute
difference between the estimated and actual ages. Bias, or Σ (estimated age – actual
age)/N, is the difference between the estimated and actual ages. Bias demonstrates the
average over- or under-aging of the individual’s actual age, while absolute mean error
demonstrates the average error from the individual’s actual age (Lovejoy et al., 1985a;
Meindl et al., 1985; Meindl and Lovejoy; 1989; Schmitt, 2004; Mulhern and Jones, 2005;
Martrille et al., 2007; Hens et al., 2008; Passalacqua, 2010).
The second step in evaluating the Chen et al. (2008) method was to create revised
statistical models based exclusively on the FSC sample. Using the raw data from the 296

37

left male pubic bones of European ancestry, new coefficients were generated for the nine
Chen et al. morphological features. The coefficients were then used to produce four
Revised statistical models (MRA, MRA+GRA, QMI, and QMI+GRA), which could be
directly compared to the Original Chen et al. (2008) models (see Table 3). The MRA and
QMI equations were entered into SPSS without modifications. The MRA+GRA and
QMI+GRA equations required the use of a backwards stepwise regression with an F
criterion of removal = 0.10 to replicate their complementary equations in the Chen et al.
(2008) publication.
A stepwise regression adds or removes variables that do not make a significant
contribution to the model (Agresti and Finlay, 2009). A backwards stepwise regression
begins with all data and removes variables that do not significantly explain the variance
of the model, while a forward stepwise regression begins with no data and adds variables
that significantly explain variance. A backwards stepwise regression was applied here
because it was more conservative and left more variables in place to explain the models.
Again, the two Revised models which include the gradual regression analysis
(MRA+GRA and QMI+GRA) were the focus of statistical analysis. To test the models
the data of all 296 previously scored pubic bones were input into the new models.
Absolute mean error and bias were calculated for each pubic bone and for both Revised
statistical models to demonstrate accuracy (Lovejoy et al., 1985a; Meindl et al., 1985;
Meindl and Lovejoy; 1989; Schmitt, 2004; Mulhern and Jones, 2005; Hens et al., 2008;
Passalacqua, 2010).
For both the Original and Revised Chen et al. models, frequencies of AME were
generated. These brackets demonstrate how closely the estimated ages are to the actual

38

ages. Frequency brackets were developed rather than using standard deviations because
the Chen et al. method does not produce standard deviations—it just provides a point age
estimate.
The frequencies were evaluated in brackets of one, five, ten, and fifteen years
from the actual age values. Since these are absolute values, they represent 1, 5, 10 or 15
years either above or below an individual’s actual age. They do not represent a range.
The level of accuracy for each model was based upon the percentage (or frequency)
within each of the brackets; the frequency represents the number of individuals who were
accurately aged within that bracket. A high percentage indicates that more individuals
fell within that bracket, and thus, is more accurate than models with lower percentages.
For example, if 10% of the sample’s AME derived from the Original Chen et al. model
fell within one year of the actual age, while only 7% of the sample’s AME derived from
the Revised Chen et al. model fell within one year of the actual age, then the Original
Chen et al. model had a higher level of accuracy for that one year bracket. The same is
true for the five, ten, and fifteen year brackets.
To further evaluate the Revised Chen et al. models (MRA+GRA and QMI+GRA),
statistical analyses were run on all 296 pubic bones. A forward stepwise regression was
employed to determine which morphological features contributed most to the prediction
of age (Buckberry and Chamberlain, 2002). A forward stepwise regression was applied
here to obtain the fewest number of features that would significantly predict age (Agresti
and Finlay, 2009; Field, 2009). To assess the relationships between each of the nine
morphological features and age, Pearson’s correlation coefficient was applied. To
collectively assess the relationships between each of the nine features, Pearson’s

39

correlation coefficient was again applied. Finally, the accuracy of the Revised models
was tested for each of the three age groups (young, middle, and old adults) in the FSC
sample by employing bias and AME.
The accuracy of the Original and Revised Chen et al. models for each of the three
age groups was assessed by calculating bias and absolute mean error. In addition,
analysis of variance (ANOVA) was computed for each age group to determine if there
was a significant difference between the mean biases and errors for each of the Chen et al.
models. These values were derived from the ANOVA post hoc test. When significant fvalues are found using ANOVA, a post hoc test allows for a more comprehensive
exploration of the differences among the means, and to determine the significance of
these differences (Huck, 2008).
Statistical Analysis Part 2: Comparison of the Chen et al. (2008) and Suchey-Brooks
(1990) Models
In order to assess whether the Original and Revised Chen et al. models were more
or less accurate than the Suchey-Brooks (1990) model, all 296 left male pubic bones of
European ancestry were evaluated. Each of the 296 bones, which were previously scored
in accordance with the Chen et al. method, were later assigned a phase in accordance
with the Suchey-Brooks method. Absolute mean error and bias were calculated for the
Suchey-Brooks model as a whole and for each of the three age groups.
Accuracy was evaluated by comparing the biases and frequencies of AME for the
Original Chen et al. (2008) model, the Revised Chen et al. model and the Suchey-Brooks
(1990) model. The frequencies were again evaluated in brackets of one, five, ten, and
fifteen years from the actual age values; the model with the highest percentage of
absolute mean error values for each of the brackets was the most accurate. As stated
40

previously, the absolute mean error for the Chen et al. model is the sum of the absolute
difference between the estimated and actual age divided by the sample size. For the
Suchey-Brooks model, the absolute mean error is the sum of the absolute difference
between the mean and actual age divided by the sample size (Σ |mean age – actual
age|/N), which is in keeping with the forensic and bioarchaeological literature (Martrille
et al., 2007; Passalacqua, 2010).
2

To support this analysis of accuracy, R values were also evaluated for the Chen
et al. models and the Suchey-Brooks model. Derived from the data of the 296 left male
pubic bones of European ancestry, linear multiple regressions were run for both the
Original and Revised Chen et al. MRA+GRA and QMI+GRA models and a linear
2

regression was run for the Suchey-Brooks model. The R values, or the coefficients of
determination, provide the proportion of variance in Y explained by X (Hamilton, 1992;
Agresti and Finlay, 2009). In this case X is the aging model (Chen et al. and SucheyBrooks) while Y is actual age at death. These values were compared and the higher the
2

R the more accurate the model. Pearson’s correlation coefficients were calculated to
determine the relationship between the aging models and actual age.
The accuracy of the Suchey-Brooks method was also assessed for each of the
three age groups. In addition to bias and absolute mean error, analysis of variance
(ANOVA) was calculated to determine if there was a significant difference between the
mean errors and biases produced by the Suchey-Brooks method and the Original and
Revised Chen et al. methods. These values were derived from the ANOVA post hoc test.

41

RESULTS
Test of the Original Chen et al. Models
The scores for each of the 296 FSC pubic bones were input into the published
Original Chen et al. (2008) multiple regression analysis (MRA and MRA+GRA) and the
quantification theory model-I equations (QMI and QMI+GRA):
MRA: Y = 16.97 + 0.42X1 + 1.48X2 + 1.88X3 + 2.51X4 – 0.43X5 + 1.76X6 + 3.25X7 –
0.66X8 + 7.31X9
MRA+GRA: Y = 16.79 + 1.76X2 + 1.71X3 + 2.47X4 + 1.68X6 +3.03X7 + 7.30X9
QMI: Y = 15.93 + 1.43X1-1 + 2.22X1-2 + 2.02X1-3 + 1.43X1-4 + 1.72X2-1 + 2.87X2-2 +
1.49X3-1 + 2.99X3-2 + 5.64X3-3 + 1.35X4-1 + 4.36X4-2 + 8.83X4-3 + 0.48X5-1 +
2.29X5-2 + 1.01X6-1 + 3.92X6-2 + 5.46X6-3 + 1.92X7-1 + 4.40X7-2 – 0.34X8-1 +
0.66X8-2 + 5.61X9-1 + 9.62X9-2 + 19.45X9-3
QMI+GRA: Y = 16.45 + 0.89X1-2 + 2.56X2-1 + 3.99X2-2 + 1.32X3-1 + 3.10X3-2 +
5.75X3-3 + 1.44X4-1 + 4.70X4-2 + 9.18X4-3 + 1.49X5-2 + 1.32X6-1 + 4.21X6-2 +
5.76X6-3 + 2.37X7-1 + 4.86X7-2 + 5.62X9-1 + 9.62X9-2 + 19.45X9-3
Absolute mean error and bias were calculated. As the histograms in Figure 8
demonstrate, the Original Chen et al. models have high frequencies of low absolute mean
error, with the average error (AME) being approximately 9 years from the specimens’
actual ages (AME = 9.18 and 9.19, respectively). The histograms in Figure 9 illustrate
that the average biases for the Original Chen et al. models are positive (higher
distribution of bars above the 0 line), which indicates an average over-estimation of the
specimens’ actual age. The average biases are also low (MRA+GRA Bias = 1.82 and
QMI+GRA Bias = 1.51).

42

Figure 8:

Figure 9:

43

Revision of Original Chen et al. Models
Based upon the FSC data for each of the 296 bones, Revised Chen et al. models
were developed. New regression constants, coefficients, and standard deviations were
produced for the multiple regression analysis and its gradual regression analysis (MRA
and MRA+GRA) as well as the quantification theory model-I and its gradual regression
analysis (QMI and QMI+GRA) (see Table 3):
MRA: Y = 12.69 + 2.57X1 + 2.83X2 + 0.36X3 + 0.60X4 – 1.59X5 + 0.49X6 + 7.78X7 +
0.07X8 + 2.04X9
MRA+GRA: Y = 15.27 + 2.88X1 + 3.37X6 + 8.25X7 + 2.13X9
QMI: Y = 19.93 – 0.65X1-1 + 3.90X1-2 + 10.55X1-3 + 10.70X1-4 + 0.14X2-1 + 4.05X2-2 –
0.53X3-1 + 3.56X3-2 + 3.25X3-3 + 2.46X4-1 + 2.56X4-2 + 2.40X4-3 – 4.25X5-1 –
6.63X5-2 + 0.79X6-1 + 6.97X6-2 + 9.46X6-3 + 4.25X7-1 + 11.77X7-2 – 2.43X8-1 –
5.52X8-2 + 3.55X9-1 + 6.24X9-2 + 6.58X9-3
QMI+GRA: Y = 21.27 + 5.74X1-3 + 7.29X1-4 + 7.91X6-2 + 10.61X6-3 + 8.39X7-1 +
16.22X7-2 + 2.57X9-2
To test the accuracy of these Revised models, bias and AME were calculated. The results
of the Revised models, as well as the Original Chen et al. and Suchey-Brooks models, are
described in the sections that follow.
Accuracy of the Models
Frequencies of Absolute Mean Error
As previously stated in the Methods chapter, frequencies of absolute mean error
(AME) were generated for all three models: Original Chen et al., Revised Chen et al., and
Suchey-Brooks. The frequencies were evaluated in brackets of one, five, ten, and fifteen
years from the actual age values. The level of accuracy for each model was based upon
the percentage (or frequency) of individuals that fell within each of the brackets. A

44

Table 3: Evaluation Criteria of Male Pubic Symphyses and Regression Coefficients for Original and Revised Chen et al. Multiple
Regression Analysis (MRA), Gradual Regression Analysis, and Quantification Theory Model-I (QMI)
Morphological Feature Scores and Descriptions
Revised Chen et al. Model
Original Chen et al. Model
MRA MRA+GRA QMI QMI+GRA MRA MRA+GRA QMI QMI+GRA
X1 Ridges and furrows
0.42 0
0
0
2.57 2.88
0
0
0: Alternate distinctly
1: Furrows fill in,
alternate indistinctly
1.43 0
-0.7 0
2: Granular, blunt ridges,
shallow furrows
2.22 0.89
3.90 0
3: Flat and fine-textured,
may be granular
2.02 0
10.55 5.74
4: Disappearance, pitting
and erosion
1.43 0
10.70 7.29
X2 Ridges of pubic tubercle 0: Tubercle completed
1: Tubercle almost gone
2: Tubercle completely
disappeared
X3 Lower extremity

X4 Ventral rampart

1.48

1.76

2.83

0

2.87 3.99

0: No appearance of lower 1.88
extremity
1: Appearance of dividing
line
2: Presence of “V” angle
3: Atrophy of “V” angle

1.71

2.51

2.47

0: No appearance
1: Local ventral rampart
2: Fully developed

0
0
1.72 2.56

0

0

4.05 0
0.36

0

1.49 1.32
2.99 3.10
5.64 5.75

45

0
0
1.35 1.44
4.36 4.70

0
0
0.14 0

0

0

-0.53 0
3.56 0
3.25 0
0.60

0

0
0
2.46 0
2.56 0

Table 2 (cont’d)
Morphological Feature

Scores and Descriptions

Original Chen et al. Model
MRA MRA+GRA QMI QMI+GRA

2: Fully developed
3: Wider or nodular in
superior portion
X5

Ossific nodule

0: No appearance

4.36 4.70
8.83 9.18

Dorsal margin

0: No appearance

Ventral beveling

0: No appearance

0
0
0.48 0
2.29 1.49

-1.59

0

0
0
-4.25 0
-6.63 0

1.76

1.68

0

0.49

3.73

0

0

1.01 1.32

0.79

0

3.92 4.21
5.46 5.76

6.97
9.46

7.91
10.61

0
4.25

0
8.39

3.25

3.03

1: Clear edged margin
2: Flattening or
disappearance
Macroscopic changes
X8

0: Prominence of
symphyseal surface

0
0

0

1: Edged margin without
plateau
2: Plateau and lipping of
dorsal margin
3: Middle destruction,
atrophy
X7

2.56
2.40

-0.43

1: Appearance
2: Fusion and disappearance
X6

Revised Chen et al. Model
MRA MRA+GRA QMI QMI+GRA

0

0
0
1.92 2.37

7.78

8.25

4.40 4.86

-0.66

0

0

46

0

11.77 16.22

0.07

0

0

0

Table 2 (cont’d)

Morphological Feature

Scores and Descriptions
1: Irregular surface
2: Flatness or fovea, clear
periphery

Original Chen et al. Model
MRA MRA+GRA QMI QMI+GRA
-0.34 0

Revised Chen et al. Model
MRA MRA+GRA QMI QMI+GRA
-2.43 0

0.66 0

0: Ridges, or rough with
7.31 7.30
no ridges
1: Smooth, dense and
solid
2: Concavo-convex or
dense porosity
3: Big pits and/or loss of
density
Original Chen et al. Models Constants and Standard Deviations (SD)
MRA: Constant = 16.97; SD = 2.13
MRA+GRA: Constant = 16.79; SD = 2.14
QMI: Constant = 15.93; SD = 1.96
QMI+GRA: Constant = 16.45; SD = 1.97

-5.52 0

Bone density

X9

Revised Chen et al. Models Constants and Standard Deviations (SD)
MRA: Constant = 12.69; SD = 9.54
MRA+GRA: Constant = 15.27; SD = 9.52
QMI: Constant = 19.93; SD = 9.78
QMI+GRA: Constant = 21.27; SD = 9.63
47

0

0

2.04

2.13

0

0

5.61 5.62

3.55 0

9.62 9.62

6.24 2.57

19.45 19.45

6.58 0

higher percentage per bracket represents higher accuracy. As mentioned before, these
values do not represent a range; rather, they represent the number of years either above or
below actual age.
The most accurate model for aging this sample varies by frequency bracket, as
shown in Table 4 and Figures 10 through 14. The most accurate model for one year from
actual age is the Original Chen et al. MRA+GRA (10.8%). This percentage indicates that
if the Original Chen MRA+GRA model is used, almost 11% of individuals will be
assigned an estimated age within one year from their actual age at death. Figure 10
provides a visual assessment of this same information; vertical lines are placed at 1, 5, 10,
and 15 years from actual age and the percentages are derived from the “amount” of the
histogram which is to the left of the lines. In this figure, 10.8% of the histogram value is
to the left of the one year line.
The most accurate model for five years from actual age is Suchey-Brooks (38.9%)
(Figure 14), and for both ten and fifteen years it is the Revised MRA+GRA (65.7% and
87.3%) (Figure 12). As both Table 4 and Figures 10 through 14 demonstrate, age at
death will be accurately estimated within 15 years for between 79% and 87% of
individuals depending upon the model used. The most accurate model for 15 years from
actual age is the Revised Chen et al. MRA+GRA at 87.3% (Figure 12); thus, when this
model is applied, 87% of individuals will be assigned an estimated aged within 15 years
Table 4: Frequencies of Absolute Mean Error (AME)
1 year* 5 years*
Revised Chen et al. MRA+GRA Model 7.8%
37.3%
Revised Chen et al. QMI+GRA Model 5.9%
38.6%
Suchey-Brooks Model
9.5%
38.9%
Original Chen et al. MRA+GRA Model 10.8%
35.3%
Original Chen et al. QMI+GRA Model 9.8%
33.3%
* Within 1, 5, 10, or 15 year(s) of actual age

48

10 years*
65.7%
63.4%
63.9%
59.8%
62.7%

15 years*
87.3%
85.1%
78.7%
85.3%
85.3%

Figure 10:

Figure 11:

49

Figure 12:

Figure 13:

50

Figure 14:

from their actual age at death. However, the Suchey-Brooks model is far less accurate at
15 years with only 79% of individuals being assessed correctly within 15 years from their
actual age at death.
Absolute Mean Error for All Models
Table 5 and Figure 15 represent the overall error for each model. The Original
Chen models have an average error of approximately 9 years from the individual’s actual
age. The Revised Chen models

Table 5: Absolute Mean Error for each Model

are slightly more accurate, with an

Revised Chen et al. MRA+GRA Model
Revised Chen et al. QMI+GRA Model
Suchey-Brooks Model
Original Chen et al. MRA+GRA Model
Original Chen et al. QMI+GRA Model
n = 296

average error between 8.48 and
8.63 years from actual age. The

51

8.628
8.483
8.946
9.182
9.185

Suchey-Brooks model has an average error in between the Chen et al. models at 8.95
years from actual age; however, this is derived from the Suchey-Brooks mean age per
phase. Thus, for the FSC sample, the Revised Chen et al. models have the least amount
of error from an individual’s actual age.
Figure 15:

Figure 5:

Bias for All Models
Table 6 and Figure 16 demonstrate the over-estimation and under-estimation of
age for each model. The Original
Chen et al. models over-age
individuals by approximately 1.5
to 2 years while the Revised Chen

Table 6: Bias for Each Model
Revised Chen et al. MRA+GRA Model
Revised Chen et al. QMI+GRA Model
Suchey-Brooks Model
Original Chen et al. MRA+GRA Model
Original Chen et al. QMI+GRA Model
n = 296

52

0.000
0.000
-0.217
1.824
1.516

et al. models have no bias. The Revised models equally over-age young individuals and
Figure 16:

under-age old individuals which negates their influence and results in a lack of bias, as
seen in the scatterplots (Figure 17 and 18). The Suchey-Brooks model consistently
under-ages the sample resulting in a negative bias (-0.22); again, this is from the mean
age and does not consider the full 95% range. For the FSC sample, the Revised Chen et
al. models have the lowest bias, although all models have biases of less than 2 years.
Variance and Correlation
2

Variance, designated as R , was generated from regressions of each model and
actual age. These values provide the proportion of variance of age explained by the aging
53

models (Hamilton, 1992). As demonstrated in Figure 19, the Revised Chen et al.
MRA+GRA model explains

Figure 17:

2

almost 50% (R = 0.491) of the
variance in actual age. The
remaining 50% of the variance is
unexplained by the aging models,
but could be explained by
factors not addressed in these
regressions (e.g. an individual’s
lifestyle, health, environment,
etc.) (Hamilton, 1992). The Revised Chen et al. models explain the most variance in age
2

2

(MRA+GRA R = 0.491; QMI+GRA R = 0.487) followed by the Suchey-Brooks model
2

(R = 0.454). The Original

Figure 18:

Chen et al. models explain the
smallest amount of variance,
with the Original QMI+GRA
explaining the least at 42%.
Despite the differences
between the models, they are
all within a fairly close range
2

of R = 0.42 – 0.49; thus, they
explain between 42% and almost 50% of the variance in actual age.

54

Figure 19:

Pearson’s correlation coefficient was calculated to determine the relationship
between each of the aging models and actual age. Pearson’s correlation coefficient
measures the linear relationship between two variables and assumes normal distribution
of the sample (Hamilton, 1992; Field, 2009).
All correlations are positive and statistically significant at the 0.01 level (Table 7).
The strongest correlation is derived from the Suchey-Brooks model (r = 0.674) which
indicates that for every one standard deviation increase in the Suchey-Brooks model the
predicted value of actual age will increase by 0.674 standard deviations. The weakest
correlation with actual age is the Original Chen et al. QMI+GRA model (r = 0.603)
2

which also explains the least amount of variance (R ) as discussed above. Collectively,
55

these correlations are fairly strong signifying

Table 7: Pearson's Correlations (r)
between Models and Actual Age

that there is a positive relationship between

Correlation Coefficients
Revised Chen
et al.
0.662
MRA+GRA JRJRJRJRJRJRJRJRJ

the models and actual age.
The Morphological Features Most
Predictive of Age
A forward stepwise regression (with a
criterion of F to enter = 0.05) was calculated
to determine which of the nine Chen et al.
morphological features were most predictive
on an individual’s age at death. The
dependent variable was actual age and the
independent variables were each of the nine

Revised Chen
et al.
0.670
QMI+GRA JRJRJRJRJRJRJRJRJ
Original
Chen et al.
MRA+GRA

0.615
JRJRJRJRJRJRJRJRJ

Original
Chen et al.
QMI+GRA

0.603
JRJRJRJRJRJRJRJRJ

Suchey0.674
Brooks
JRJRJRJRJRJRJRJRJ
p < 0.01; 2-tailed test; n = 296

features. As stated in the Methods section, a stepwise regression adds or removes
variables that do not make a significant contribution to the model (Agresti and Finlay,
2009). A forward stepwise regression begins with only the constant (β0), and adds
variables that significantly predict the outcome of the model. One variable is added at a
time until the most variance is predicted/explained (Field, 2009).
Based upon Observer One’s scores for all nine morphological features, the
following explain the most variance in the Chen et al. (2008) model: features 7, 1, 6, and
9—ventral beveling, ridges and furrows of the symphyseal surface, dorsal margin, and
2

bone density of the symphyseal surface. As the R values in Table 8 demonstrate, feature
7 alone explains 35% of the variance in actual age. Only 9% more variance in age is

56

explained when all four features (7, 1, 6, and 9) are included in the regression (Step 4 R

2

2

= 0.44, or 44%) (Field, 2009). These R values are all significant at the 0.05 level.
The standardized coefficients in Table 8 represent the conversion of each feature
into a standard unit of measurement (standard deviation). This allows for a direct
comparison of all features (Field, 2009). All of the coefficients are positive, so as the
Table 8: Forward Stepwise Regression Analysis of the Chen et al. Morphological
Features and Actual Age
Constant (β0) Standardized Coefficients (β 1)
Step 1

SD

28.77*
Feature 7

Step 2

0.67

16.83*

Step 3

0.41
0.46*
0.28*

0.74

13.42*
Feature 7
Feature 1
Feature 6

Step 4

0.43
0.41*
0.22*
0.17*

0.65

15.27*
Feature
Feature
Feature
Feature

7
1
6
9

2

0.35
0.59*

Feature 7
Feature 1

R

0.44
0.38*
0.15**
0.17*
0.13**

0.86

SE
1.53
1.01
2.66
1.10
1.00
2.81
1.13
1.05
1.16
2.92
1.15
1.21
1.16
1.00

* p < 0.01 p < 0.05
**

score for each feature increases, actual age will also increase. For example, if the scores
for feature 7 increase by one standard deviation (0.67 units), actual age will increase by
0.59 standard deviations, holding all other variables constant.
Relationships between the Morphological Features and Age
To establish the relationship between each of the nine Chen et al. (2008)
morphological features and age, both Pearson’s (r) and Spearman’s (ρ) correlations were
calculated. Pearson’s correlation coefficients were stronger than Spearman’s, so the
Pearson’s data are presented in Table 9.
57

Table 9: Pearson's Correlations (r)
between Morphological Features
and Actual Age

The correlations between the nine
features and actual age are not very strong
(Table 9). The strongest correlation with actual
age is feature 7 (ventral beveling; r = 0.592).
This indicates that if the score of feature 7
increases by one standard deviation, the actual
age of an individual will also increase by 0.592
standard deviations. Conversely, the weakest
correlation with actual age is feature 5 (ossific

Correlation Coefficients
Feature
0.592
7
4
0.532
3
0.522
1
0.499
9
0.472
6
0.462
8
0.406
2
0.402
5
0.397
p < 0.01; 2-tailed test; n = 296

nodule; r = 0.397). If the score of feature 5 increases by one standard deviation, the
actual age of an individual will only increase by approximately 0.4 standard deviations.
All correlations are positive and significant at the 0.01 level.
Relationships between All Morphological Features
Pearson’s correlation coefficient was calculated to determine the relationship
between each of the nine Chen et al. (2008) morphological features. All correlations are
positive and are statistically significant at the 0.01 level. The strongest correlation, as
shown in Table 10, is between feature 5 (ossific nodule) and feature 2 (ridge of the pubic
tubercle) (r = 0.824). Thus, if the score of feature 5 increases by one standard deviation,
the score of feature 2 will increase by 0.824 standard deviations, and vice versa. The
weakest correlation is between features 9 (bone density of the symphyseal surface) and 6
(dorsal margin) (r = 0.391). This correlation is positive, which indicates that both values
will increase concurrently, but only by 0.391 standard deviations.

58

Table 10: Pearson's Correlation Coefficients (r) for All Nine Chen et al. Morphological Features
Feature 1
2
3
4
Feature 1
1
0.565
0.597
0.628
2
0.565
1
0.558
0.571
3
0.597
0.558
1
0.726
4
0.628
0.571
0.726
1
5
0.563
0.824
0.569
0.589
6
0.483
0.493
0.573
0.550
7
0.485
0.449
6.664
0.672
8
0.587
0.559
0.593
0.597
9
0.652
0.369
0.551
0.568
All p-values significant at the 0.01 level; 2-tailed; n = 296
strongest correlation
weakest correlation

59

5
0.563
0.824
0.569
0.589
1
0.520
0.464
0.562
0.393

6
0.483
0.493
0.573
0.550
0.520
1
0.448
0.479
0.391

7
0.485
0.449
6.664
0.672
0.464
0.448
1
0.469
0.476

8
0.587
0.559
0.593
0.597
0.562
0.479
0.469
1
0.412

9
0.652
0.369
0.551
0.568
0.393
0.391
0.476
0.412
1

Accuracy for Age Groups
Bias and Absolute Mean Error
As stated previously, the FSC sample is divided into three age groups to gauge
accuracy for younger and older individuals. Table 11 provides the summary statistics for
each age group.
Accuracy was evaluated
by measures of bias and
absolute mean error for
each model (Tables 12a

Table 11: Summary Statistics for FSC Sample by Age Group*
Young Adult
Middle Adult
Old Adult
N
69
95
132
Mean
25.33
43.03
59.03
SD
4.87
4.36
6.08
*Young Adult: 18-34 years; Middle Adult 35-49 years;
Old Adult: 50+ years

and 12b). For all models except Suchey-Brooks, absolute mean error is lowest for the
middle and old adults. The highest error is derived from the Original Chen et al.
QMI+GRA model for the young adult group (AME = 10.96 years).
Bias is both positive and negative for each model which indicates that the models
both under- and over-estimate actual age. The highest bias (farthest from an individual’s
actual age) is again derived from the Original Chen et al. QMI+GRA model for the young
adult group (Bias = over-estimated age by 10.50 years). The lowest bias is derived from
the Suchey-Brooks model for the middle adult group (Bias = over-estimated age by 3.38
years).
As Figure 20 demonstrates, the young adults have higher levels of error when
compared to the middle and old adults (with the exception of the Suchey-Brooks model).
However, this could be a reflection of the unbalanced sample sizes. Collectively the
middle adult group shows the least amount of error. For the young adult group, the
Suchey-Brooks model has the least error. For the middle adult group, the Revised

60

†

Table 12a: Absolute Mean Error (AME) and Bias for each Model by Age Group
Original Chen et al. MRA+GRA Original Chen et al. QMI+GRA
Age Group
n
AME*
Bias
AME*
Bias
Young Adult (1)
69
10.862 10.485
10.957
10.500
Middle Adult (2)
95
8.285
6.998
7.772
6.668
Old Adult (3)
132
8.949
-6.427
9.275
-6.888
*Absolute Mean Error: Σ |estimated age – actual age|/N
**Absolute Mean Error: (Σ |mean age – actual age|/N)
†Bias: Σ (estimated age – actual age)/N

†

Table 12b: Absolute Mean Error (AME) and Bias for each Model by Age Group
Revised Chen et al. MRA+GRA Revised Chen et al. QMI+GRA
Age Group
n
AME*
Bias
AME*
Bias
10.500 Young Adult (1)
69
9.896
9.514
9.609
9.105
6.668 Middle Adult (2) 95
6.728
4.546
6.889
4.923
-6.888 Old Adult (3)
132
9.332
-8.244
9.041
-8.301
*Absolute Mean Error: Σ |estimated age – actual age|/N
**Absolute Mean Error: (Σ |mean age – actual age|/N)
†Bias: Σ (estimated age -- actual age)/N

61

Suchey-Brooks
AME**
Bias
7.194
6.168
8.768
3.375
9.990
-6.186

Figure 20:

Chen et al. MRA+GRA model has the least error and for the old adult group, the Original
Chen et al. MRA+GRA model has the least error. Overall, the Original Chen et al.
models have the highest error for both the young and middle adult groups. The Revised
Chen et al. models have the lowest error for the middle adult group and the second
highest error for both the young and old adult groups. The Suchey-Brooks model follows
a linear trend with the least amount of error for the young adults and the most error for
the old adults.
Figure 21 shows that the young and middle adults tend to be over-aged while the
old adults tend to be under-aged. The Suchey-Brooks model has the least amount of bias
62

Figure 21:

(either over-estimation or under-estimation of actual age) for each of the three age groups.
The middle adult group shows the least bias overall, which corresponds to the absolute
mean error results discussed above. The Original Chen et al. models have the highest
bias for both the young and middle adult groups, but have lower biases for the old adult
group than the Revised Chen et al. models. The Revised Chen et al. models have the
highest bias for the old adults and the second highest biases for the young and middle
adults.
Analysis of Variance (ANOVA)
Analysis of variance was calculated to determine if there was a difference
63

between the mean biases and
errors for each of the three age
groups. As Figures 20 and 21
demonstrate, there are
differences between each of the

Table 13: Analysis of Variance (ANOVA) of Absolute
Mean Error and Bias for Each Model by Age Group
Bias
Absolute Mean Error
F
F
Young Adults (1)
2.53*
3.06*
Middle Adults (2)
2.15
3.17*
Old Adults (3)
0.46
1.56
*p<0.05; 2-tailed test

models, but ANOVA was
applied to establish if these differences were statistically significant. For ANOVA, the
null hypothesis states that all means are equal, while the alternative hypothesis states that
at least two of the means are unequal. If the null hypothesis is false, the f-value will be
larger than 1.0 (Agresti and Finlay, 2009). As Table 13 shows, the f-values are larger
than 1.0 for both the young and middle adults which indicates that the null hypothesis is
rejected and the alternative hypothesis is supported. Three of the four f-values for the
young and middle adults are significant at the 0.05 level. The f-values for the old adults
are not larger than 1.0, but they are also not significant at the 0.05 level. Because
significant f-values were found using ANOVA, a post hoc test was evaluated. This post
hoc test also allowed for comparisons between each of the five models for each of the
three age groups.
As Tables 14 and 15 demonstrate, there are significant differences between some
of the models for the young adult and middle adult groups. For both bias and absolute
mean error, it is only the Suchey-Brooks model which is different from the Chen et al.
models. Table 14 shows that the young adult group has significant absolute mean error
differences between the Suchey-Brooks model and both Original Chen et al. models as
well as the Revised MRA+GRA model. For the middle adult group, there are significant

64

Table 14: Analysis of Variance (ANOVA) Post Hoc Comparisons for Absolute Mean Error for Each Age Group
Young Adults (1)
Middle Adults (2)
Models
Significance
Models
Significance
Original MRA+GRA Original QMI+GRA 0.982
Original MRA+GRA Original QMI+GRA 0.444
Revised MRA+GRA 0.472
Revised MRA+GRA 0.066
Revised QMI+GRA 0.350
Revised QMI+GRA 0.100
Suchey-Brooks
0.007*
Suchey-Brooks
0.568
Original QMI+GRA Revised MRA+GRA 0.457
Revised QMI+GRA 0.339
Suchey-Brooks
0.006*

Original QMI+GRA Revised MRA+GRA 0.284
Revised QMI+GRA 0.377
Suchey-Brooks
0.181

Revised MRA+GRA Revised QMI+GRA 0.830
Suchey-Brooks
0.045*

Revised MRA+GRA Revised QMI+GRA 0.850
Suchey-Brooks
0.016*

Revised QMI+GRA Suchey-Brooks

Revised QMI+GRA Suchey-Brooks

0.073

*p<0.05; 2-tailed test

65

0.027*

Table 14 (cont’d)

Old Adults (3)
Models
Significance
Original MRA+GRA Original QMI+GRA 0.567
Revised MRA+GRA 0.654
Revised QMI+GRA 0.914
Suchey-Brooks
0.223
Original QMI+GRA Revised MRA+GRA 0.901
Revised QMI+GRA 0.642
Suchey-Brooks
0.518
Revised MRA+GRA Revised QMI+GRA 0.733
Suchey-Brooks
0.441
Revised QMI+GRA Suchey-Brooks

66

0.261

Table 15: Analysis of Variance (ANOVA) Post Hoc Comparisons for Bias for Each Age Group
Young Adults (1)
Middle Adults (2)
Models
Significance
Models
Significance
Original MRA+GRA Original QMI+GRA 0.985
Original MRA+GRA Original QMI+GRA 0.444
Revised MRA+GRA 0.501
Revised MRA+GRA 0.066
Revised QMI+GRA 0.338
Revised QMI+GRA 0.100
Suchey-Brooks
0.003*
Suchey-Brooks
0.568
Original QMI+GRA Revised MRA+GRA 0.489
Revised QMI+GRA 0.329
Suchey-Brooks
0.003*

Original QMI+GRA Revised MRA+GRA 0.284
Revised QMI+GRA 0.377
Suchey-Brooks
0.181

Revised MRA+GRA Revised QMI+GRA 0.776
Suchey-Brooks
0.021*

Revised MRA+GRA Revised QMI+GRA 0.850
Suchey-Brooks
0.016*

Revised QMI+GRA Suchey-Brooks
*p<0.05; 2-tailed test

Revised QMI+GRA Suchey-Brooks

0.042*

67

0.027*

Table 15 (cont’d)

Old Adults (3)
Models
Significance
Original MRA+GRA Original QMI+GRA 0.546
Revised MRA+GRA 0.105
Revised QMI+GRA 0.095
Suchey-Brooks
0.829
Original QMI+GRA Revised MRA+GRA
Revised QMI+GRA
Suchey-Brooks

0.308
0.285
0.414

Revised MRA+GRA Revised QMI+GRA
Suchey-Brooks

0.959
0.067

Revised QMI+GRA

0.060

Suchey-Brooks

68

error differences between the Suchey-Brooks model and both of the Revised Chen et al.
models. Table 15 indicates that for the young adult group, the bias for the SucheyBrooks model is different from all of the other models, but for the middle adults, SucheyBrooks is only different from the Revised Chen et al. models. There are no significant
differences between any of the aging models for the old adult group.
Intra- and Inter-Observer Error
Intra-Observer Error
Cohen’s Kappa measure of agreement and Pearson’s correlation coefficient were
used to measure intra-observer error for a random sample of 100 previously scored pubic
bones. “Cohen’s kappa measures
the agreement between two
observations, while taking into
account any agreement that would
occur by chance” (Hefner,
2009:991). The closer the resulting
value is to 1 the higher the
agreement. Cohen’s Kappa was
calculated for each of the nine Chen
et al. (2008) morphological features

Table 16: Intra-Observer Cohen's Kappa Measure of
Agreement (κ) for the Chen et al. Morphological
Features and the Suchey-Brooks Method*
κ
SE
Feature 1 vs. Feature 1
Feature 2 vs. Feature 2
Feature 3 vs. Feature 3
Feature 4 vs. Feature 4
Feature 5 vs. Feature 5
Feature 6 vs. Feature 6
Feature 7 vs. Feature 7
Feature 8 vs. Feature 8
Feature 9 vs. Feature 9
Suchey-Brooks Method
* n = 100

0.505
0.885
0.624
0.615
0.858
0.621
0.662
0.666
0.631
0.628

0.073
0.114
0.070
0.073
0.092
0.074
0.070
0.077
0.063
0.061

and for the Suchey-Brooks (1990) model. Feature 2 (ridges of the pubic tubercle, κ =
0.885) has the highest intra-observer agreement, while Feature 1 (ridges and furrows of
the symphyseal surface, κ = 0.505) has the lowest inter-observer agreement (Table 16).
As such, Feature 1 has the highest intra-observer error rate.

69

Based upon the Pearson’s correlation calculations, the Suchey-Brooks method is
more correlated between the first and second
rounds of scoring than is the published Chen
et al. method (Table 17). The correlations
between the scoring rounds of Observer One
are positive and fairly strong (r=0.796 and

Table 17: Intra-Obsersver Pearson's
Correlations (r) for the Chen et al.
and Suchey-Brooks Methods
Correlation Coefficients
Methods
0.796
Chen et al.
Suchey-Brooks
0.906
p < 0.01; 2-tailed test; n = 100

0.906). This indicates that if Observer One’s
scoring improves by one standard deviation during the first round, Observer One’s
scoring will improve during the second round by 0.796 and 0.906 standard deviations for
each of the methods. Both correlations are significant at the 0.01 level. Based upon
these correlations, the Chen et al. method has a higher intra-observer error rate than does
the Suchey-Brooks method.
Inter-Observer Error
Inter-observer error was also evaluated for both the Chen et al. and the SucheyBrooks methods. Observer Two scored 200 pubic bones via the Chen et al. method and a
subsample of 75 of those 200 bones via the Suchey-Brooks method. Observers Three
and Four scored the same 50 pubic bones for both methods. Pearson’s correlation
coefficients were calculated to compare Observer One with Observers Two, Three, and
Four.
As Table 18 indicates, the highest correlation for the Chen et al. method is
between Observers One and Four (r = 0.694), while the highest correlation for the
Suchey-Brooks method is between Observers One and Three (r = 0.885). All of the
correlations are positive between Observer One and the additional Observers. All

70

correlations are significant at either the 0.01 level or the 0.05 level. Overall, the SucheyBrooks method has higher r-values which suggests that this method has lower interobserver error than the Chen et al. method.

Table 18: Inter-Observer Pearson's Correlations ( r )
n Chen et al. Method Suchey-Brooks Method
0.640
0.790
Observer 1 vs. Observer 2* 200
Observer 1 vs. Observer 3* 50
0.693
0.885
Observer 1 vs. Observer 4** 50
0.694
0.746
*p < 0.01; 2-tailed test
**p < 0.05; 2-tailed test

71

DISCUSSION
As the previous chapter demonstrates, the Chen et al. method can accurately age
males of European ancestry. Based upon frequencies of mean error, almost 90% of
individuals’ estimated ages will fall within 15 years of their actual age at death.
Therefore, this method has the potential to be a useful aging technique in the United
States. In particular, the most accurate model for estimating age for this sample is the
Revised QMI+GRA, based upon the lowest error, no bias, and a high correlation with
actual age.
The Revised Chen et al. Model Compared to the Suchey-Brooks Model
One of the most significant research questions in this thesis asks whether a
Revised Chen et al. method will be more accurate for estimating age than the SucheyBrooks method. While it is important to evaluate whether the Chen et al. method is
capable of accurately estimating age for males of European ancestry, it is imperative to
test the claim by Chen et al.—that their method is accurate within one year of actual
age—with that of the standard currently used in the United States—the Suchey-Brooks
method.
When the data are evaluated for the whole sample the Revised Chen et al. models
(both MRA+GRA and QMI+GRA collectively) are more accurate than the SucheyBrooks model. The Revised models are the most accurate for the 1 year frequency
bracket; have the lowest amounts of error; demonstrate no bias; and explain the highest
amount of variance in age.
When the absolute mean error data are considered for the Revised Chen and
Suchey-Brooks models, the Revised models have an average error of approximately 8

72

years while the Suchey-Brooks model has an average error of approximately 9 years.
While this does not initially suggest a large difference, it must briefly be considered that
for the Suchey-Brooks model this only represents the mean age at death. If the full 95%
age range is to be considered, the Revised Chen models are in fact much more accurate.
However, the results for the Suchey-Brooks model are interesting. It is the only
model whose average bias is negative, which results in under-aging of the specimens by
0.20 years. This tendency to under-estimate the age of the sample is consistent with
previous research (Schmitt, 2004; Djurić et al., 2007; Hens et al., 2008). Even more
interesting is that this model has the highest correlation with actual age (r = 0.674). Even
though the data in this study only represent the mean Suchey-Brooks ages, Figure 22
indicates that the

Figure 22:

majority of the
samples’
estimated ages do
fall within the
Suchey-Brooks
phase ranges.
Analysis
of the age group
data indicates that
the SucheyBrooks model is
distinct as it has a steady pattern of increasing absolute mean error. As Figure 20 shows,

73

the Suchey-Brooks model has the lowest error for the young adults and steadily increases
across the middle and old adults. This is consistent with other evaluations of the SucheyBrooks method (Suchey et al., 1986; Djurić et al., 2007; Martrille et al., 2007). This
model also has the least amount of bias for each age group which indicates that it is more
consistent than the Chen et al. models with regards to over- and under-estimation of
actual age.
2

When variance (R values) is compared between the Revised Chen et al. models
2

and the Suchey-Brooks model they are virtually the same—Suchey-Brooks R = 0.454
2

2

while the Revised MRA+GRA R = 0.491 and Revised QMI+GRA R = 0.487.
Therefore, these models all explain almost 50% of the variance in age for this sample.
This measure of model comparison suggests that the Suchey-Brooks and Revised Chen et
al. models are equally accurate for estimating age for this sample.
Review of the intra- and inter-observer error correlation data indicate that the
Suchey-Brooks model is more consistent and reliable within and between observers than
the Chen et al. model (Tables 17 and 18). For intra-observer error the correlation scores
(r) are 0.796 for the Chen et al. model and 0.906 for the Suchey-Brooks model. The
Suchey-Brooks model demonstrates a 12% higher correlation than the Chen et al. model.
For inter-observer error, Suchey-Brooks is 19% more correlated than Chen et al. between
Observer One and Observer Two, while between Observer One and Observer Three,
Suchey-Brooks is 22% more correlated. Finally, the Suchey-Brooks model has a 7%
higher correlation than the Chen et al. model between Observer One and Observer Four.
Accuracy for Older Ages

74

Most studies have found that pubic aging methods are more accurate for younger
individuals and are of little use for older individuals (Lovejoy et al., 1985a; Suchey et al.,
1986; Brooks and Suchey, 1990; Komar, 2003; Martrille et al., 2007; Hens, 2008). As
Hartnett (2010:6) states, “anthropologists have generally accepted that pubic bone
morphology cannot be used to determine the age of older individuals accurately.” In fact,
when statistically analyzing the male pubic bone sample which would be used to develop
the Suchey-Brooks method, Suchey and colleagues removed individuals over the age of
40 which improved their statistical results (Suchey et al., 1986; Katz and Suchey, 1986).
In contrast to these previous studies, this research found that the middle adult
group (ages 35 – 49) is the most accurately aged. All models except Suchey-Brooks have
the lowest absolute mean errors for this group, and all models except for the Original
MRA+GRA have the least amount of bias. This suggests that the Chen et al. method (in
particular the Revised models) is unique among pubic bone age estimation techniques,
since it is most accurate for the middle adult age category. As noted above, this is not a
typical result for adult age estimation methods, and is thus, distinctive. Further research
is required, especially for additional ancestral groups and for females, but the data from
this study indicate that the Chen et al. method can be used to age adults between 35 and
49 years old.
Variance and Correlation
2

Compared to the variance (R ) results from other age estimation studies, both the
Chen et al. models and the Suchey-Brooks models from this study do not explain as much
2

variance in actual age. The R values from this study are as follows: Original Chen
MRA+GRA = 0.440; Original Chen QMI+GRA = 0.420; Revised Chen MRA+GRA =

75

0.491; Revised Chen QMI+GRA = 0.487; and Suchey-Brooks = 0.454 (Figure 19).
When evaluating research based on the pubic bone, Hanihara and Suzuki (1978) report an
2

2

R of 0.85 while Katz and Suchey (1986) report an R of 0.83. This means that Hanihara
and Suzuki developed a model which can explain 85% of the variance in actual age and
Katz and Suchey’s model explains 83% of the variance in actual age. Đşcan, Loth, and
Wright (1984; 1985) also report high variance values in their studies on age estimation
2

from the sternal rib. For males they report an R of 0.85, but for females it was slightly
2

less at an R of 0.76.
2

The lower R values presented in this study simply indicate that these models are
not incorporating all variables which explain actual age. As noted in the Results chapter,
the remaining 50% of the variance may be explained by factors which are not addressed
in this study, such as an individual’s lifestyle, health, or environment.
Correlation data (r) between this study (Table 7) and other aging methods also
indicate that the Chen et al. and Suchey-Brooks models are not extremely well correlated
with actual age. For example, original research evaluating the pubic bone by Hanihara
and Suzuki (1978), McKern and Stewart (1957), and Suchey et al. (1986) report
correlation values of 0.92, 0.90, and 0.72, respectively. After Suchey et al. (1986)
removed individuals over the age of 40 years, their correlation increased to 0.78. One
original study by Meindl and Lovejoy (1985) on ectocranial suture age estimation did
present much lower correlations at 0.50 for the vault sutures and 0.57 for the lateralanterior cranial sutures. This study does not evaluate the pubic bone, but it is an
interesting comparison for correlation values from different adult age estimation methods.
When correlations are compared between the present study and other age

76

estimation studies conducted on samples other than the reference sample, however, the
data are much more consistent. Lovejoy et al. (1985a) tested numerous aging methods,
but specifically for Todd’s (1920) pubic method, they report a correlation of 0.57. After
revisions of Todd’s phases, the authors report a correlation of 0.78 (Lovejoy et al., 1985a).
Bedford et al. (1993) later tested Lovejoy et al.’s (1985a) multifactorial age estimation
method and produce correlations between 0.27 and 0.66. The correlations for the pubic
bones alone are between 0.53 and 0.60. Osborne et al. (2004) and Falys et al. (2006) both
tested auricular surface aging methods and report correlations between composite scores
and actual ages of 0.59. The correlation values from studies applying age estimation
methods to different samples than those from the original research are all much closer to
the correlations reported for this study. Thus, this research (with correlation values
between 0.60 and 0.67) demonstrates comparable correlations between the aging models
and actual ages when the reference sample is not used.
Intra- and Inter-Observer Error
When the Chen et al. morphological features are assessed for intra-observer error,
Feature 2 (ridges of the pubic tubercle) has the highest intra-observer agreement, and thus,
the lowest error, while Feature 1 (ridges and furrows of the symphyseal surface) has the
highest intra-observer error rate. When the Chen et al. and Suchey-Brooks models are
compared, Chen et al. has a higher intra-observer error rate than does the Suchey-Brooks
method. As stated above, the Suchey-Brooks model demonstrates a 12% higher
correlation than the Chen et al. model for Observer One. This indicates that the SucheyBrooks method is more consistent during different intervals of age assessment.

77

With regards to inter-observer error, the Suchey-Brooks method has lower error
than does the Chen et al. method. However, all r-values for both methods are above 0.60
which indicates fairly strong correlations. This data reveal that the Chen et al. method
can be learned with relative ease and can be applied by individuals with limited
osteological experience.
Limitations
There are a few limitations which should be noted: limitations of the Chen et al.
method and limitations of this study. The primary limitation of the Chen et al. method is
subjectivity. The evaluation of the nine morphological features requires an observer to be
familiar with the features, their developmental stages, and associated scores. Time and
practice are required before an observer can appropriately assign a score for each feature.
The scoring system is also problematic and challenging due to translation issues in the
published article. This may have influenced the raw data scores, and is the reason why
the author modified the language of the scoring criteria (see Appendix 2 and 3).
Additionally, the pubic bone must be intact for the Chen et al. method to be employed. A
completely intact pubic bone may not always be available in a forensic case or
bioarchaeological material, which is a considerable limitation for this method.
A limitation of this study, rather than the Chen et al. method, involves defining
accuracy for the Chen et al. and Suchey-Brooks techniques. The data derived from these
methods are inherently different; the Chen et al. method provides a point age estimate,
while the Suchey-Books method produces a mean and a 95% confidence range. Each of
these methods was developed with different goals in mind; the Chen et al. method
intended to provide a specific age estimate based upon nine features while the Suchey-

78

Brooks method offered a wide age range which encompassed changes from a large
sample (however, this wide age range is one of the limitations of the Suchey-Brooks
model).
Calculating absolute mean errors based upon two different forms of data (point
age estimate and a mean) does raise the question of the internal validity of this study, but
these data are what each method provides. For example, when applying the SucheyBrooks method to an unknown male, practitioners will state that the individual is a Phase
4; he is approximately 35.2 years old (the mean), but is within a range of 23-57 years of
age. Thus, the mean is what practitioners use as the estimated age even though it is not
the age of every male assigned to Phase 4. As such, absolute mean error (from a point
estimate or a mean) is the best way to define and assess accuracy for both methods, even
if they are innately different forms of data.
Finally, the data for this research may be slightly skewed due to the age
distribution of the study collection. Many of the individuals in the sample are above the
age of 49 years (n = 132) which could have influenced the age group results.
Furthermore, the Chen et al. method (2008) is not generalizable to both sexes; it must
only be applied to males which eliminates its potential utility for half of the population.

79

CONCLUSION
The purpose of this thesis project was to evaluate the Chen et al. pubic bone aging
study, and to determine if the method could accurately estimate the age of individuals
outside of the Chinese Han population. The overall results of this project were expected
to support the hypothesis that the Chen et al. method could accurately assess age at death
for the males of European ancestry in the Maricopa County Forensic Science Center
sample. Furthermore, this project compared the accuracy of the Chen et al. method to the
Suchey-Brooks method, which is currently the most accepted technique for estimating
age from the pubic bones. This comparison was made in order to test whether the Chen
et al. technique is an improved and more accurate method for estimating age at death
using male pubic bones.
The majority of the hypotheses for this research are supported by the data. The
first two hypotheses state that the Chen et al. method will be able to accurately estimate
age for males of European ancestry and that a revision of the method will be the most
accurate. Both the Original and Revised Chen et al. methods are able to accurately age
the males in this sample and both methods have an average error of approximately 8 to 9
years from an individual’s actual age, which is fairly low for adult aging methods.
The next hypothesis states that there will be no significant differences between the
2

rates of accuracy and the R values of the Revised Chen et al. and the Suchey-Brooks
models. As the previous chapter discusses, there are some interesting differences
between the models, but with regard to accuracy and variance they are virtually
indistinguishable. The Revised Chen and Suchey-Brooks models have almost identical

80

variance values and also have very similar correlation values with actual age. However,
these models are significantly different when the sample is distributed into age groups.
Based upon the data presented by Chen et al., it was hypothesized that ventral
beveling, bone density of the symphysial surface, and the ventral rampart would be the
morphological features most predictive of age. Assessment of the sample reveals that
ventral beveling and bone density are among the most predictive of age. Ridges and
furrows of the symphyseal surface and the dorsal margin are included based upon the
forward stepwise regression analysis. The ventral rampart is not among the features most
predictive of age in this sample.
The fifth hypothesis states that the Revised method will be most accurate for
young adults (18-34 years) and middle adults (35-49 years), and less accurate for old
adults (50 years and above). This was partially refuted by the data as the Revised Chen et
al. models are most accurate for the middle adults followed by the old adults. It is the
Suchey-Brooks model which is most accurate for aging the young adults. The accuracy
of aging the middle adult group is an important aspect of this research, and one which is
not often seen with adult aging models.
Finally inter- and intra-observer error rates are addressed in the last hypothesis
which states that both levels of error will be low. For both intra- and inter-observer tests
the correlations are high indicating low levels of error. Among all observers, the SucheyBrooks model has the highest level of correlation, and thus, the least amount of error.
However, the r-values for the Chen et al. method are all moderately high, indicating fairly
strong correlations. This signifies that the Chen et al. method can be learned with relative

81

ease and can be applied by individuals with limited osteological experience, as well as
professional anthropologists.
Most authors who have developed or tested aging methods have noted that using
only one skeletal element in isolation is not advisable for estimating adult age at death
(Brooks, 1955; Acsádi and Nemeskéri, 1970; Meindl and Lovejoy, 1985; Lovejoy et al.,
1985a; Krogman and Đşcan, 1986; Đşcan and Loth, 1989; and Baccino et al., 1999;
Buikstra and Komar, 2008). They fervently argue that multiple skeletal elements and/or
methods should be applied to properly age an individual. The Chen et al. method does
not use multiple skeletal elements, but it does employ multiple features of one bone to
estimate age. Additionally, the Suchey-Brooks method only uses one bone.
I strongly support the use of multiple bones and multiple methods to estimate
adult age at death. Especially if the remains are damaged or missing elements, multiple
techniques should be utilized to accurately estimate age. Skeletal remains should not be
viewed as disjointed bones but should be thought of, and analyzed, as the whole
individual that they once were. Bones age at different rates due to variable stresses on
the body, so when available, all bones useful for age estimation should be evaluated.
Although it has been demonstrated that the Chen et al. method is accurate, I do
have some reservations regarding the technique. When applied to a North American
sample, the method is not nearly as accurate as it is in China. For this sample, ages are
rarely (between 5.9% and 10.8%) correctly estimated within one year of actual age; thus,
the claim by Chen et al. that their method can accurately estimate age within one year of
an individual’s actual age at death is not substantiated by this research. This outcome is
to be expected, however, since this study is not testing the original reference sample, but

82

it may reduce the method’s utility in the United States. This method can be challenging
to implement and the published descriptions and photographs may be ambiguous. Based
upon these limitations, and until more research is conducted to test this aging technique, I
believe that the Chen et al. method should be applied in conjunction with other, more
rigorously studied, adult aging methods.
This research does not imply that the Suchey-Brooks method should be
supplanted by the Chen et al. method. If the Chen et al. method was significantly more
accurate than the Suchey-Brooks method, further consideration for abandoning SucheyBoroks would be warranted. This is not the case. This study does, however, suggest that
the Chen et al. method is comparable, with regards to accuracy, to the Suchey-Brooks
method for estimating age for this particular male sample.
If specific age groups are being considered, however, the Revised Chen methods
are more accurate than the Suchey-Brooks method for males between the ages of 35 and
49 years. The ability of the Revised Chen et al. models to accurately age individuals in
the middle adult group is the precise reason why this method should be used.
The ability of the Chen et al. method to accurately age this group is a significant
contribution to the forensic and bioarchaeological disciplines. Aging this older group is
often challenging, so the Chen et al. method is a unique tool for physical anthropologists.
To conclude, I believe that if an anthropologist is willing to take the time to learn the
technique and apply it to complete, modern pubic bones, the Chen et al. method can be
used to accurately assess age at death for males of European ancestry.

83

APPENDICES

84

Appendix 1: Table 19: Nine Morphological Features of the Pubic Bone and Associated
Scores (Chen et al., 2008)
Feature Name

Score and Description

1. Ridges and furrows of
symphyseal surface

0: ridges and furrows alternate distinctly
1: the furrows fill in and ridges and furrows
alternate indistinctly
2: the bone substance has a granular look with
low, blunt ridges and shallow furrows
3: the surface becomes flat and fine-textured,
and/or again becomes more granular
4: ridges and furrows disappear entirely and the
surface becomes pitted and eroded

2. Ridges of the pubic tubercle

0: ridges on pubic tubercle completed
1: ridges on pubic tubercle almost gone
2: ridges on pubic tubercle completely
disappeared

3. Lower extremity of the
symphysial surface

0: no appearance of lower extremity
1: appearance of dividing line between
symphysial surface and inferior ramus of pubis
2: forming a “V” angle
3: atrophy or disappearance of “V” angle

4. Ventral rampart

0: no appearance of ventral rampart
1: local ventral rampart
2: fully developed ventral rampart
3: ventral rampart becomes wider or nodular in
superior portion

5. Ossific nodule

0: no appearance of ossific nodule
1: appearance of ossific nodule
2: fusion and disappearance of ossific nodule

6. Dorsal margin

0: no appearance of dorsal margin
1: edged margin without a plateau
2: forming a plateau and lip-like thickness and
extension in superior part of dorsal margin
3: middle destruction or generalized atrophy of
dorsal margin

7. Ventral beveling

0: no appearance of ventral beveling
1: clear-edged margin with a right angle
between ventral beveling and the symphyseal
surface
85

Table 18 cont’d
2: ventral beveling becomes flat in the lower
portion or disappears
8. General macroscopic changes of
the symphyseal surface

0: prominence of symphyseal surface
1: irregular surface
2: flatness, or fovea inferior with clear periphery

9. Bone density of the symphyseal
surface

0: ridged or rough with no ridge
1: smooth, dense and solid
2: concavo-convex or with dense pores on
surface
3: big pits and/or loss of density

86

Appendix 2: Revised Descriptions for the Nine Morphological Features of the Pubic
Bone
Feature 1: Ridges and Furrows of the Symphyseal Surface
The ridges and furrows in young individuals are very prominent and have an order of
alternating in a distinct pattern. With age, the furrows begin to fill in and the pattern is
less orderly and clear eventually resulting in blunt ridges and shallow furrows. The
surface may begin to appear granular. The ridge and furrow pattern disappears and the
surface becomes flat and fine-textured. With advanced age the surface can become pitted
and eroded.
Feature 2: Ridges of the Pubic Tubercle
These are horizontal ridges that connect the superior surface of the symphyseal face with
the ventral side of the pubis. They are prominent and well formed in the young but begin
to fade and disappear with age. The final stage is the disappearance of the ridges and/or
a completed pubic tubercle.
Feature 3: Lower Extremity of the Symphysial Surface
This refers to the dividing line between the inferior symphyseal face and the superior
ischiopubic ramus. There is no evidence of this line in the young. A line or ridge begins
to develop and eventually forms a “V” shape (2a in Appendix 4) or “U” shape (2b)
separating the face from the ramus. With age the “V” or “U” begins to atrophy (3a) and
disappear (3b) resulting in a flat surface.
Feature 4: Ventral Rampart
The ventral rampart develops to form the ventral margin of the symphysial face (the
ventral half of the symphyseal rim). The ventral rampart begins to develop locally on the
inferior aspect of the face, or the superior aspect of the face (or sometimes both) and
grows towards the midline. A hiatus in the superior ventral aspect can be present even in
mature adults. With advanced age the ventral rampart may become wider or nodular in
the middle and/or superior aspect(s).
Feature 5: Ossific Nodule
This is a nodule of bone found on the superior aspect of the symphyseal face which aids
in the formation of the superior portion of the ventral rampart and upper extremity of the
symphyseal rim. The nodule is not present in the young. It appears in the mid-twenties,
according to Scheuer and Black (2004), and later becomes incorporated with the ventral
rampart and symphyseal face (2a in Appendix 4) and disappears (2b).
Feature 6: Dorsal Margin
The dorsal margin develops to form the dorsal half of the symphyseal rim. In the young
there is not a true distinction separating the dorsal symphyseal face from the dorsal side
of the pubis. With age a slight edge develops in this region that is palpable. The ridge
later extends backwards and flattens forming a plateau. With advanced age the dorsal
margins atrophies, and in some cases a cavitation or localized breakdown occurs in the
middle or superior aspect.

87

Feature 7: Ventral Beveling
The ventral bevel is one aspect of the ventral margin. It forms a palpable right angle
between the symphyseal face and the ventral aspect of the pubis. The sharpness of this
bevel later becomes blunt, and after the ventral margin is fully developed this bevel
becomes flat and can disappear.
Feature 8: General Macroscopic Changes of the Symphyseal Surface
This feature addresses the macroscopic topography of the symphyseal surface. The
surface is initially raised above the surrounding pubis when observed from the ventral or
dorsal side. It is billowy due to the ridges and furrows. As the ridges and furrows fill in
the surface becomes irregular and more flat. The final stage is a flat surface, or one
depressed below the symphyseal rim when viewed from the ventral or dorsal aspect.
Feature 9: Bone Density of the Symphyseal Surface
Bone density describes the quality of the bone on the symphyseal surface. The bone is
dense with ridges and furrows (0a in Appendix 4), or it is rough with no distinct ridges
(0b). The surface then becomes smooth, but is still dense and solid. The surface later
becomes irregular and may show evidence of microporosity and/or macroporosity (2a
and 2b). In advanced age the bone looses density and may have large pits on the surface.

88

Appendix 3: Table 20: Revised Nine Morphological Features of the Pubic Bone and
Associated Scores
Feature Name

Score and Description

1. Ridges and furrows of
symphyseal surface

0: ridges and furrows alternate in a regular
pattern
1: the furrows fill in and ridges and furrows
alternate irregularly
2: the bone substance has a granular look with
low, blunt ridges and shallow furrows
3: the surface becomes flat and fine-textured,
and/or again becomes more granular
4: ridges and furrows disappear entirely and the
surface may become pitted and eroded

2. Ridges of the pubic tubercle

0: ridges are present
1: ridges are fading or are almost gone
2: ridges completely disappeared and /or pubic
tubercle

3. Lower extremity of the
symphysial surface

0: no appearance of lower extremity
1: appearance of dividing line between
symphysial surface and inferior ramus of pubis
2: presence of a “V” angle
3: atrophy or disappearance of “V” angle

4. Ventral rampart

0: no appearance of ventral rampart
1: local ventral rampart
2: fully developed ventral rampart, may have
hiatus
3: ventral rampart becomes wider or nodular in
middle and/or superior portion(s)

5. Ossific nodule

0: no appearance of ossific nodule
1: presence of ossific nodule
2: incorporation and disappearance of ossific
nodule

6. Dorsal margin

0: no appearance of dorsal margin
1: edged margin without a plateau
2: forming a plateau with lipping
3: middle destruction or generalized atrophy of
dorsal margin

7. Ventral beveling

0: no appearance of ventral beveling

89

Table 19 cont’d
1: palpable right angle between ventral and
symphyseal surfaces
2: right angle becomes flat in the lower portion
or disappears
8. General macroscopic changes of
the symphyseal surface

0: symphyseal surface is raised and billowy
1: irregular surface
2: flat or depressed with clear periphery

9. Bone density of the symphyseal
surface

0: ridges and furrows, or rough with no distinct
ridges and furrows
1: smooth, dense and solid
2: irregular, some microporosity and/or
macroporosity
3: big pits and/or loss of density

90

Appendix 4: Revised Images of the Nine Chen et al. Morphological Features of the
Pubic Bone

Figure 23: Feature 1: Ridges and Furrows of the Symphyseal Surface

Figure 24: Feature 2: Ridges of the Pubic Tubercle

Figure 25: Feature 3: Lower Extremity of the Symphysial Surface

91

Figure 26: Feature 4: Ventral Rampart

Figure 27: Feature 5: Ossific Nodule

Figure 28: Feature 6: Dorsal Margin

92

Figure 29: Feature 7: Ventral Beveling

Figure 30: Feature 8: General Macroscopic Changes of the Symphyseal Surface

Figure 31: Feature 9: Bone Density of the Symphyseal Surface

93

REFERENCES

94

REFERENCES
Agresti A, and Finlay B. 2009. Statistical methods for the social sciences (4th ed.). Upper
Saddle River: Prentice Hall, Inc.
Acsádi G, and Nemeskéri J. 1970. History of human life span and mortality. Balás K,
translator. Akadémiai Kiadó:Budapest.
Baccino E, Ubelaker DH, Hayek LAC, Zerilli A. 1999. Evaluation of seven methods of
estimating age at death from mature human skeletal remains. Journal of Forensic Science
44:931-936.
Bedford ME, Russell KF, Lovejoy CO, Meindl RS, Simpson SW, and Stuart-Macadam
PL. 1993. Test of the multifactorial aging method using skeletons with known ages-atdeath from the Grant Collection. American Journal of Physical Anthropology 91:287-297.
Brooks ST. 1955. Skeletal age at death: the reliability of cranial and pubic age indicators.
American Journal of Physical Anthropology 13:567-597.
Brooks ST, and Suchey JM. 1990. Skeletal age determination based on the os pubis: a
comparison of the Acsádi-Nemeskéri and Suchey-Brooks methods. Human Evolution
5:227-238.
Buckberry JL, Chamberlain AT. 2002. Age estimation from the auricular surface of the
ilium: a revised method. American Journal of Physical Anthropology 119:231-239.
Buikstra, JE, Ubelaker, DH (eds.). 1994. Standards for Data Collection from Human
Skeletal Remains. Fayetteville, Arkansas Archaeological Survey.
Byers SN. 2008. Introduction to forensic anthropology. Third edition. Boston: Pearson
Education, Inc.
Chen X, Zhang Z, Tao L. 2008. Determination of male age at death in Chinese Han
population: using quantitative variables statistical analysis from pubic bones. Forensic
Science International 175:36-43.
Djurić M, Djonić D, Nikolić S, Popović D, and Marinković J. 2007. Evaluation of the
Suchey-Brooks method for aging skeletons in the Balkans. Journal of Forensic Sciences
52:21-23.
Falys CG, Schutkowski H, and Weston DA. 2006. Auricular surface aging: worse than
expected? A test of the revised method on a documented historic skeletal assemblage.
American Journal of Physical Anthropology 130:508-513.
Field A. 2009. Discovering statistics using SPSS (3rd edition). London: Sage Publications.

95

Hamilton LC. 1992. Regression with graphics: a second course in applied statistics.
Belmont: Duxbury Press.
Hanihara K, and Suzuki T. 1978. Estimation of age from the pubic symphysis by means
of multiple regression analysis. American Journal of Physical Anthropology 48:233-240.
Hartnett KM. 2007. A re-evaluation and revision of pubic symphysis and fourth rib aging
techniques. Ph.D. Dissertation, Arizona State University.
Hartnett KM. 2010. Analysis of age-at-death estimation using data from a new, modern
autopsy sample part I: pubic bone. Journal of Forensic Sciences 55:1145-1151.
Hefner JT. 2009. Cranial nonmetric variation and estimating ancestry. Journal of Forensic
Sciences 54:985-995.
Hens M, Rastelli E, and Belcastro G. 2008. Age estimation from the human os coxa: a
test on a documented Italian collection. Journal of Forensic Sciences 53:1040-1043.
Huck SW. 2008. Reading statistics and research. 5th edition. New York: Pearson.
Igarashi Y, Uesu K, Wakebe T, and Kanazawa E. 2005. New method for estimation of
adult skeletal age at death from the morphology of the auricular surface of the ilium.
American Journal of Physical Anthropology 123:324-339.
Đşcan MY. 1989. Assessment of age at death in the human skeleton. In: Đşcan MY, editor.
Age markers in the human skeleton. Springfield: Charles C. Thomas. p 5-18.
Đşcan MY, Loth SR, and Wright RK. 1984. Age estimation from the rib by phase analysis:
White males. Journal of Forensic Sciences 29:1094-1104.
Đşcan MY, Loth SR, and Wright RK. 1985. Age estimation from the rib by phase analysis:
White females. Journal of Forensic Sciences 30:853-863.
Katz D, and Suchey JM. 1986. Age determination of the male os pubis. American Journal
of Physical Anthropology 69:427-435.
Klepinger LL. 2006. Fundamentals of forensic anthropology. Hoboken. John Wiley&
Sons, Inc.
Komar DA. 2003. Lessons from Srebrenica: The contributions and limitations of physical
anthropology in identifying victims of war crimes. Journal of Forensic Sciences 48:713716.
Komar DA, and Buikstra JE. 2008. Forensic anthropology: contemporary theory and
practice. New York: Oxford University Press.

96

Krogman WM, and Đşcan MY. 1986. The human skeleton in forensic medicine. 2nd
edition. Springfield: Charles C. Thomas.
Lovejoy CO, Meindl RS, Mensforth RP, Barton TJ. 1985a. Multifactorial determination
of skeletal age at death: a method of blind tests of its accuracy. American Journal of
Physical Anthropology 68:1-14.
Lovejoy CO, Meindl RS, Pryzbeck TR, Mensforth RP. 1985b. Chronological
metamorphosis of the auricular surface of the ilium: a new method for the determination
of adult skeletal age at death. American Journal of Physical Anthropology 68:15-28.
Martrille L, Ubelaker DH, Cattaneo C, Seguret F, Tremblay M, and Baccino E. 2007.
Comparison of four skeletal methods for the estimation of age at death on white and
black adults. Journal of Forensic Sciences 52:302-307.
McKern TW, Stewart TD. 1957. Skeletal age changes in young American males.
Headquarters Quartermaster Research and Development Command, Technical Report EP
45: Natick Massachusetts. p 53-88.
Meindl RS, and Lovejoy CO. 1985. Ectocranial suture closure: A revised method for the
determination of skeletal age at death based on the lateral-anterior sutures. American
Journal of Physical Anthropology 68:57-66.
Meindl RS, and Lovejoy CO. 1989. Age changes in the pelvis: implications for
paleodemography. In: Đşcan MY, editor. Age markers in the human skeleton. Springfield:
Charles C. Thomas. p 137-168.
Meindl RS, Lovejoy CO, Mensforth RP, Walker RA. 1985. A revised method of age
determination using the os pubis, with a review and tests of accuracy of other current
methods of pubic symphyseal aging. American Journal of Physical Anthropology 68:2945.
Mulhern DM, Jones EB. 2005. Test of revised method of age estimation from the
auricular surface of the ilium. American Journal of Physical Anthropology 126:61-65.
Osborne DL, Simmons TL, and Nawrocki SP. 2004. Reconsidering the auricular surface
as an indicator of age at death. Journal of Forensic Sciences 49:1-7.
Passalacqua NV. 2010. The utility of the Samworth and Gowland age-at death “look-up”
tables in forensic anthropology. Journal of Forensic Sciences 55:482-487.
Santos AL. 1996. How old is this pelvis? A comparison of age at death estimation using
the auricular surface of the ilium and the os pubis. In: Aspects of African Archaeology.
Papers from the 10th Congress of the Pan African Association for prehistory and related
studies, Pwiti G, Soper R, editors. Harare: University of Zimbabwe Publications. p 29-36.

97

Saunders SR, Fitzgerald C, Rogers T, Dudar C, and McKillop H. 1992. A test of several
methods of skeletal age estimation using a documented archaeological sample. Canadian
Society of Forensic Science 25:97-118.
Schmitt A. 2004. Age-at-death assessment using the os pubis and the auricular surface of
the ilium: a test on an identified Asian sample. International Journal of Osteoarchaeology
14:1-6.
Schmitt A, Murail P, Cunha E, and Rougé D. 2002. Variability of the pattern of aging on
the human skeleton: evidence from bone indicators and implications on age at death
estimation. Journal of Forensic Sciences 47:1-7.
Sinha A, and Gupta V. 1995. A study on estimation of age from pubic symphysis.
Forensic Science International 75:73-78.
Scheuer L, and Black S. 2000. Developmental juvenile osteology. San Diego: Academic
Press.
Scheuer L, and Black S. 2004. The juvenile skeleton. San Diego: Academic Press.
Stewart TD. 1979. Essentials of forensic anthropology. Springfield: Charles C. Thomas.
Suchey JM, and Katz D. 1998. Applications of pubic age determination in a forensic
setting. In: Reichs KJ, editor. Forensic osteology: advances in the identification of human
remains (2nd edition). Springfield: Charles C. Thomas. p 204-36.
Suchey JM, Wiseley DV, and Katz D. 1986. Evaluation of the Todd and McKern-Stewart
methods for aging the male os pubis. In: Reichs KJ, editor. Forensic osteology: advances
in the identification of human remains. Springfield: Charles C. Thomas. p 33-67.
Todd TW. 1920. Age changes in the pubic bone. I. The male white pubis. American
Journal of Physical Anthropology 3:285-339.
Todd TW. 1921. Age changes in the pubic bone. II. The pubis of the male negro-white
hybrid. III. The pubis of the white female. IV. The pubis of the female negro-white
hybrid. American Journal of Physical Anthropology 4:1-77.
Ubelaker DH. 2008. Forensic anthropology: methodology and diversity of applications.
In: Katzenberg M, and Saunders SR, editors. Biological anthropology of the human
skeleton (2nd ed.). Chichester: Wiley-Liss. p 41-69.

98