Forensic Science International: Andrew S. Rozendaal, Shelby Scott, Tanya R. Peckmann, Susan Meek
Forensic Science International: Andrew S. Rozendaal, Shelby Scott, Tanya R. Peckmann, Susan Meek
Forensic Science International: Andrew S. Rozendaal, Shelby Scott, Tanya R. Peckmann, Susan Meek
A R T I C L E I N F O A B S T R A C T
Article history: Methods for estimating sex from a variety of skeletal elements is vital in forensic anthropology as many of the
Received 30 June 2019 traditionally studied bones are often fragmented upon recovery. The current study examined the seven
Received in revised form 14 November 2019 cervical vertebrae to establish an accurate sex estimation method for Europeans. Discriminant functions
Accepted 18 November 2019
were developed from the seven cervical vertebrae using the maximum body height (CHT), vertebral foramen
Available online 20 November 2019
anterior-posterior diameter (CAP), and vertebral foramen transverse diameter (CTR). To date, no study has
used these variables to estimate sex from human remains. Two hundred and ninety-five adult individuals
Keywords:
(157 males,138 females) from the University of Athens and the Luis Lopes skeletal collections were studied.
Forensic anthropology population data
Adults
No population differences were exhibited between the contemporary Greek and historic Portuguese
Sex estimation skeletons, indicating that discriminant functions for sex estimation from the cervical vertebrae may not be
Cervical vertebrae population- or temporally-specific. The results of this study indicate that only CHT and CTR exhibited
Discriminant functions statistically significant sexual dimorphism. Seven multivariate discriminant functions were developed with
European accuracy rates between 80.3% and 84.5%. Cross-validation studies showed that five of the seven functions
exhibited strong statistical algorithms. This study will assist with estimating the sex of unknown Europeans
from the cervical vertebrae especially in cases where more accurate predictors of sex, such as the skull or the
pelvis, are unavailable.
© 2019 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.forsciint.2019.110072
0379-0738/© 2019 Elsevier B.V. All rights reserved.
2 A.S. Rozendaal et al. / Forensic Science International 306 (2020) 110072
other vertebrae, making them easily identifiable; the unique from the Lopes collection (87 males and 73 females). The Athens
skeletal attributes of cervical vertebrae allow for rapid anatomical collection is comprised of individuals who lived the majority of
sequencing in comparison to other vertebral elements. their lives in the late 20th and early 21st centuries. It is an example
Previous research has demonstrated that the cervical vertebrae of a growing contemporary skeletal reference collection housing
exhibit sexual dimorphism suitable for accurate and reliable sex an estimated 225 individuals. The Athens collection is continu-
estimation [9,12,15–18]. Traditionally, research involving methods of ously accepting unclaimed remains from cemeteries around
sex estimation from cervical vertebrae have focused on three bones: Athens, Greece, and previous research has shown that it provides
atlas (C1), axis (C2), and the transitional seventh vertebra (C7) [9,12,15– the most accurate data available for sex estimation in a
17,19]. The unique morphology of C1, C2, and C7 has created an interest contemporary European population [23]. The Lopes collection is
in studying these bones as compared to the typical third through sixth comprised of historic individuals who lived the majority of their
cervical vertebrae (C3 to C6), which are less often investigated. In lives during the late 19th and early 20th centuries. Unclaimed
addition, previous research has, traditionally, utilized measurements remains are mostly acquired from three major cemeteries within
of the centrum, articular facets, and length and width of the vertebrae the city of Lisbon [22]. Sample selection criteria for the historic
as they have been shown to be highly dimorphic [9,15,17,20,21]. Lopes collection included individuals living the majority of their
However, these characteristics are likely to sustain taphonomic lives after 1900, the year that researchers have observed overall
damage and fragmentation upon recovery from a deposition site [9– positive secular changes (Albanese 2010). Cemetery registrations
11,14,20]. If vertebral preservation is poor, the spinous process and from both collections have resulted in a large range of biographic
features of the transverse processes are the most likely structures to be and demographic data for the individuals studied.
damaged, although the strong architectural construction increases the Three measurements were collected from each of the seven
potential for recovery of the complete vertebral column cervical vertebrae: maximum cervical vertebral body height (CHT),
[9–11,14,16]. Structural rigor make vertebrae ideal for sex estimation cervical anterior-posterior diameter (CAP), and cervical transverse
due to their preservation. The CAP and CTR diameters are enclosed and diameter (CTR); CHT was not collected for the first cervical
protected by the vertebral arches and the CHT is located on the dense vertebra, as this vertebra does not have a centrum. In total, 20
vertebral body, resulting in resiliency to mechanical, taphonomic, and measurements were taken for the purposes of this study. These
architectural stresses [9–11,14]. The CHT, CAP, and CTR morphometric variables are defined in Table 1 and are shown in Figs. 1 and 2. In
characteristics are, therefore, more likely to be present in forensic some instances, it was not possible to collect all three measure-
skeletal cases. Therefore, the current project aims to create a method ments from a single vertebra due to taphonomic damage or mild
for estimating sex from the cervical vertebrae using only the three osteoarthritis. Thus, all statistical analyses were performed using
vertebral measurements most resilient to mechanical, taphonomic, the total number of individuals utilized for each specific
and architectural stresses, and investigating their accuracy for the measurement (i.e. total number of individuals in which CHT,
estimation of sex using all cervical vertebrae, i.e. C1–C7. CAP, and CTR could be examined) rather than the total number of
Overall, while previous research has demonstrated accuracy individuals examined from each skeletal collection as outlined
rates greater than 80% for sex estimation from the cervical above. Measurements were taken using Vernier calipers and
vertebrae, they have cited these methods to be population-specific rounded to the nearest 0.01 mm, following the protocols of
[9,12,15–18]. However, the methods were created and tested on the previous researchers [15,17,18,22,24–29]. For inclusion in this
sample population used in the study and not cross referenced on study, individuals must have at least five of the seven vertebrae
other population samples. Currently, there are no studies that present to anatomically sequence the bones; if one or two
investigate the accuracy of cervical vertebrae for the estimation of vertebrae were missing, sequential ordering and numerical
sex in contemporary Greek and historic Portuguese populations. identification of the cervical bones was possible through anatomi-
Therefore, the aim of this research is to evaluate the relationship cal articulation. Vertebrae exhibiting taphonomic damage, trauma,
between sex and three morphometric characteristics (CHT, CAP, or pathological remodelling were excluded from this study.
and CTR) of the cervical vertebrae in a contemporary Greek and an Statistical analyses were performed using Minitab (version
historic Portuguese population. 17.0) and SPSS (version 21.0) statistical software following the
protocol of Tatarek [18,26]. As with any biological morphometric
2. Materials and methods study, it is important to minimize measurement error and to
understand its influence. Statistical significance levels were,
This research examined 1020 vertebrae utilizing 295 adult therefore, adjusted using a Bonferroni correction in order to
skeletonized individuals (157 males and 138 females) of European reduce the chance of errors associated with multiple comparisons
ancestry between 20 and 99 years of age. Two documented skeletal (Type-1 error). All data were tested for normality and all data were
cemetery collections were utilized: the University of Athens found to be normally distributed. Intra-observer error rates were
human skeletal reference collection (i.e. the Athens collection) in calculated by re-measuring 60 individuals (30 individuals from
Greece, and the Luis Lopes skeletal collection (i.e. the Lopes each collection) one week after the initial measurements were
collection) in Lisbon, Portugal. A total of 295 individuals were recorded. Inter-observer error rates were calculated with the aid of
examined for this research (n=295): 135 individuals from the two research assistants (i.e. one research assistant from the Athens
Athens collection (70 males and 65 females), and 160 individuals skeletal collection and one research assistant from the Lopes
Table 1
Measurement descriptions.
Measurements Descriptiona
Maximum Cervical Vertebral Body The maximum superior to inferior vertebral body height along the anterior border of each vertebra with the exception
Height (CHT) of C1. (Note: C2HT includes the odontoid process).
Cervical Anterior-Posterior The maximum mid-sagittal diameter from the anterior to posterior aspects of the vertebral foramen.
Diameter (CAP)
Cervical Transverse Diameter (CTR) The maximum medio-lateral diameter measured from the left to the right pedicles within the vertebral foramen.
a
Adapted from Clark (1985), Eisentstein (1983), Kibii and colleagues (2010), Taitz (1996), Tatarek (2005), and Verbiest (1955).
A.S. Rozendaal et al. / Forensic Science International 306 (2020) 110072 3
Table 2
Descriptive statistics for the Athens skeletal collection (n = 135), including assessment of sexual dimorphism.
Measurement n Mean (mm) Standard Minimum (mm) Maximum (mm) t-value p-value
Deviation (SD)
statistically significant differences between any cervical vertebral 3.2. Assessment of sexual dimorphism
measurements, with the exception of C4TR and C6TR. TEM values are Tables 2 and 3 show the results of the assessment of sexual
low, rTEM values range from 0.18% to 2.60%, and all vertebral dimorphism in the Athens and Lopes skeletal collections,
measurements have an R-value greater than 0.95 with the exception respectively. In the Athens collection, all CTR and CHT measure-
of C4TR (R = 0.94) and C5TR (R = 0.94). An examination of inter- ments demonstrate a statistically significant difference between
observer error shows statistically significant differences in vertebral males and females (p < 0.002). However, for the CAP measurement,
measurements C3AP, C1TR, C3TR, C5TR and C6TR. TEM values are high only the first cervical vertebra (C1) exhibits a statistically
(comparatively), rTEM values range from 4.87% to 15.42%, and all significant difference between males and females. Overall, the
vertebral measurements have an R-value of less than 0.50. CAP measurement does not demonstrate sexual dimorphism in the
Table 5 shows the results of intra-and inter-observer error in Athens collection. Similarly, in the Lopes collection, all CTR and
the Lopes collection. In the Lopes collection, intra-observer CHT measurements demonstrate a statistically significant differ-
error shows no statistically significant differences between ence between males and females (p < 0.002), with the exception of
vertebral measurements, with the exception of C 3 TR, C 5 TR, C1TR and C2TR, and only the first cervical vertebra (C1) exhibits a
C 6 TR, and C4 HT. TEM values are low, rTEM values range from statistically significant difference between males and females for
0.16% to 2.53%, and all vertebral measurements have an R-value the CAP measurement. These results indicate that the CAP
of 0.95 or greater, with the exception of C 3 HT (R = 0.92), C 5 HT measurement does not demonstrate sexual dimorphism in either
(R = 0.93), C6 HT (R = 0.94), C 7 HT (R = 0.94). An examination of the Athens or Lopes skeletal collections. As a result of these
inter-observer error shows no statistically significant differ- findings, males and females have been analysed separately for both
ences in vertebral measurements, with the exception of C1 TR the Athens and Lopes skeletal collections.
and C 3 HT. TEM values are also low, while rTEM values range
from 0.40% to 4.81%, and all vertebral measurements have an R- 3.3. Vertebral variation due to secular change and ancestry
value of an R-value of 0.95 or greater, with the exception of
C 5 AP (R = 0.93), C4 TR (R = 0.94), C 3 HT (R = 0.84), C 4 HT (R = 0.93), Table 6 shows the results of the assessment of secular change
C 5 HT (R = 0.92), and C 6 HT (R = 0.89). and ancestry in the Athens and Lopes collections. Overall, for both
A.S. Rozendaal et al. / Forensic Science International 306 (2020) 110072 5
Table 3
Descriptive statistics for the Lopes skeletal collection (n = 160), including assessment of sexual dimorphism.
Measurement n Mean (mm) Standard Minimum (mm) Maximum (mm) t-value p-value
Deviation (SD)
sexes, there are no statistically significant differences between the 3.5. Discriminant functions for combinations of cervical vertebrae
collections (p > 0.002) for CHT, CAP and CTR. Therefore, the two
populations, i.e. Athens and Lopes collections, were combined into All 20 measurements were combined to create one discrimi-
one large independent sample “Europeans” for all statistical nant function to estimate sex using the enter method of regression.
analyses. This function resulted in overall accuracy rates of 84.1% (Function
2; Table 7). When a stepwise method was used, only seven
3.4. Discriminant functions for independent cervical vertebra measurements (C1AP, C2HT, C2TR, C3HT, C5TR, C5HT, C7TR)
exhibited large t-value coefficients, which indicate a high potential
Sex estimation from a single vertebra ranges from 66.9%–74% for for the estimation of sex. The resulting discriminant function
males of the combined European population and from 70.2%–79.5% resulted in an overall accuracy rate of 82.6% (Function 4, Table 7).
for females of the combined European population when all three Additional multivariate discriminant functions were created to
measurements (CHT, CAP, and CTR) are utilized. Discriminant investigate whether particular combinations of cervical vertebrae
functions were also created for each cervical vertebra using only may prove accurate for sex estimation. When only the first and
measurements of the vertebral foramen (CAP and CTR) and using second cervical vertebrae (C1 and C2) were utilized, the overall
only those vertebral measurements that were most sexually accuracy rate was 72.8% using all three measurements (CAP, CTR,
dimorphic (CTR and CHT). When measurements of the vertebral CHT). When C2 through C7 were utilized, the highest accuracy rates
foramen are used (CAP and CTR), analyses demonstrate accuracy were found using the two most dimorphic measurements (CTR and
rates between 61.2% and 70.6% for European males and between CHT). The overall accuracy rate was 84.5%. When the typical
58.7% and 70.1% for European females. When only those vertebral cervical vertebrae (C3 through C6) and the transitional C7 vertebra
measurements that were most sexually dimorphic (CTR and CHT) were utilized, overall accuracy was 83.3% using only the two most
were utilized, accuracy rates increased and ranged from 67.6%–73.3% sexually dimorphic measurements (CTR and CHT). Accuracy rates
for males and 70.4%–78.5% for females. Overall, sex estimation from a decreased when the transitional C7 vertebra was excluded. The
single cervical vertebra is most accurate when all three measure- results of this study indicate that second and fifth cervical
ments (CHT, CAP, an CTR) are utilized. vertebrae (C2 and C5) are the most sexually dimorphic. When
6 A.S. Rozendaal et al. / Forensic Science International 306 (2020) 110072
Table 4
Results of intra-observer error and inter-observer error in the Athens skeletal collection.
Table 5
Results of intra-observer error and inter-observer error in the Lopes skeletal collection.
only the second and fifth cervical vertebrae (C2 and C5) were addition, for the combined European population, the results of this
utilized, overall accuracy rates were 77%. study indicate that a high estimation of sex is achieved when all
Table 7 shows the results of all generated discriminant cervical vertebrae (C1 through C7) and all measurements (CAP, CTR,
functions that achieved an overall accuracy rate of 80% or higher. CHT) are utilized.
The results of this study demonstrate that combinations of
vertebrae provide a more accurate estimate of sex than single or 3.6. Biases in sex estimation
independent cervical vertebrae; this indicates that the accuracy of
sex estimation is enhanced when the most sexually dimorphic Table 7 shows that correct discriminant function classification
measurements (CTR and CHT) from all cervical vertebrae are by sex is greater in females in all functions, apart from Function 1,
utilized (excluding C1 due to an absence of the CHT variable). In whereas males are classified more correctly than females. The
A.S. Rozendaal et al. / Forensic Science International 306 (2020) 110072 7
Table 6
Results of the assessment of secular change/ancestry in the Athens and Lopes collections.
Males Females
Measurement Athens (n) Lopes (n) t-value p-value Athens (n) Lopes (n) t-value p-value
C1AP 59 77 2.47 0.015 57 70 1.03 0.305
C2AP 64 82 0.94 0.349 62 66 1.82 0.071
C3AP 59 81 1.59 0.115 56 70 2.00 0.048
C4AP 69 83 1.30 0.194 61 73 1.76 0.080
C5AP 63 80 0.41 0.682 61 72 0.95 0.342
C6AP 62 84 0.07 0.944 59 70 0.34 0.734
C7AP 65 81 0.97 0.336 61 64 1.47 0.144
C1TR 59 77 1.35 0.180 58 70 0.19 0.852
C2TR 64 84 4.90 0.000* 62 69 1.84 0.068
C3TR 59 80 2.78 0.006 56 71 2.25 0.026
C4TR 69 83 1.85 0.067 61 73 2.53 0.013
C5TR 63 80 1.76 0.080 61 72 2.04 0.043
C6TR 62 84 2.67 0.008 59 70 2.39 0.018
C7TR 65 81 1.77 0.079 61 64 2.22 0.029
C2HT 65 84 4.32 0.000* 62 68 0.32 0.746
C3HT 59 80 1.33 0.188 56 69 1.38 0.171
C4HT 69 81 3.39 0.001* 59 71 0.79 0.431
C5HT 62 79 1.65 0.100 60 70 0.56 0.578
C6HT 63 83 3.36 0.001* 57 68 1.94 0.055
C7HT 65 81 2.92 0.004 61 63 1.16 0.247
*
Significant difference at p-value 0.002.
Table 7
Discriminant function equations (that achieved an accuracy rate of 80% or higher) for combinations of cervical vertebrae in the combined European population.
% Males % Females n %
1 y = 0.240(C2TR) + 0.197(C2HT) + 0.179(C3TR) + 0.327(C3HT) 0.232(C4TR) + 0.0625 0.809 0.669 85.0 83.0 158/ 84.5
0.251(C4HT) 0.322(C5TR) + 0.153(C5HT) + 0.234(C6TR) + 0.101(C6HT) + 0.358 187
(C7TR) 0.018(C7HT) 17.246
2 y = 0.201(C1AP) + 0.015(C1TR) 0.111(C2AP) 0.384(C2TR) + 0.151(C2HT) + 0.0485 1.004 0.708 83.7 84.6 138/ 84.1
0.222(C3AP) + 0.177(C3TR) + 0.325(C3HT) 0.447(C4AP) 0.180(C4TR) + 0.135 164
(C4HT) + 0.252(C5AP) 0.467(C5TR) + 0.338(C5HT) 0.105(C6AP) + 0.133(C6TR)
+ 0.068(C6HT) + 0.058(C7AP) + 0.508(C7TR) 0.058(C7HT) 15.219
3 y = 0.003(C3TR) + 0.377(C3HT) 0.130(C4TR) + 0.342(C4HT) 0.367(C5TR) + 0.05 0.692 0.639 82.9 83.9 165/ 83.3
0.107(C5HT) + 0.215(C6TR) + 0.104(C6HT) + 0.395(C7TR) + 0.133(C7HT) 16.568 198
4 y = 0.190(C1AP) 0.355(C2TR) + 0.175(C2HT) + 0.363(C3HT) 0.430(C5TR) + 0.0465 0.917 0.692 77.3 88.2 157/ 82.6
0.428(C5HT) + 0.565(C7TR) 16.994 190
5 y = 0.170(C3AP) + 0.012(C3TR) + 0.389(C3HT) 0.487(C4AP) 0.169(C4TR) + 0.052 0.747 0.654 81.9 82.8 163/ 82.3
0.375(C4HT) + 0.158(C5AP) 0.335(C5TR) + 0.108(C5HT) + 0.120(C6AP) + 0.189 198
(C6TR) + 0.037(C6HT) 0.005(C7AP) + 0.409(C7TR) + 0.119(C7HT) 15.536
6 y = 0.061(C3AP) + 0.103(C3TR) + 0.327(C3HT) 0.372(C4AP) 0.118(C4TR) + 0.0555 0.646 0.627 80.2 81.2 175/217 80.6
0.374(C4HT) + 0.138(C5AP) 0.271(C5TR) + 0.101(C5HT) + 0.209(C6AP) + 0.412
(C6TR) + 0.196(C6HT) 16.478
7 y = 0.089(C3TR) + 0.306(C3HT) 0.064(C4TR) + 0.373(C4HT) 0.367(C5TR) + 0.049 0.584 0.607 80.2 80.4 175/ 80.3
0.092(C5HT) + 0.498(C6TR) + 0.251(C6HT) 16.974 218
*
Females < Sectioning point value < Males.
variation between the correct classification of males and females is validation test, indicating that they are strong statistical algo-
within 2%. However, Function 4 utilizes the step-wise regression rithms. Functions 1 and 6 resulted in cross-validation accuracies
formula, in which the variation between male and female accuracy lower than 80%, indicating weaker algorithms for sex estimation.
rates is 10.9%, 77.3% and 88.2% respectively. Thus, functions 1 and 6 are not recommended for forensic
anthropological application.
3.7. Cross-validation
4. Discussion
For the Athens and Lopes skeletal collections, the seven
discriminant functions that achieved overall accuracy rates of 4.1. Intra- and inter-observer error rates
80% or greater were cross-validated on an independent sample of
32 individuals. This sample size is appropriate as previous studies The methods and techniques used in forensic anthropology
have shown that approximately 10%–20% of the total population must show high levels of reliability, accuracy, and precision [39].
size is required for testing an independent sample [30,31]. The Accuracy assesses the degree of correctness and, for estimation of
results of cross-validation are shown in Table 8. Functions 2, 3, 4, 5 sex in forensic anthropology, methods must show accuracy rates of
and 7 achieved accuracy rates greater than 80% in the cross- 80% or greater to be acceptable [5,9,33–40]. Precision is the ability
8 A.S. Rozendaal et al. / Forensic Science International 306 (2020) 110072
Table 8
Cross-validation classification accuracies for the combined European population (based on Table 7, discriminant function equations that achieved an accuracy rate of 80% or
higher).
n %
1 y = 0.240(C2TR) + 0.197(C2HT) + 0.179(C3TR) + 0.327(C3HT) 0.232(C4TR) + 0.251(C4HT) 23 (13/10) 18/23 78.3
0.322(C5TR) + 0.153(C5HT) + 0.234(C6TR) + 0.101(C6HT) + 0.358(C7TR) 0.018(C7HT) 17.246
2 y = 0.201(C1AP) + 0.015(C1TR) 0.111(C2AP) 0.384(C2TR) + 0.151(C2HT) + 0.222(C3AP) + 0.177 19 (10/9) 16/19 84.2
(C3TR) + 0.325(C3HT) 0.447(C4AP) 0.180(C4TR) + 0.135(C4HT) + 0.252(C5AP) 0.467(C5TR) +
0.338(C5HT) 0.105(C6AP) + 0.133(C6TR) + 0.068(C6HT) + 0.058(C7AP) + 0.508(C7TR) 0.058
(C7HT) 15.219
3 y = 0.003(C3TR) + 0.377(C3HT) 0.130(C4TR) + 0.342(C4HT) 0.367(C5TR) + 0.107(C5HT) + 23 (10/13) 19/23 82.6
0.215(C6TR) + 0.104(C6HT) + 0.395(C7TR) + 0.133(C7HT) 16.568
4 y = 0.190(C1AP) 0.355(C2TR) + 0.175(C2HT) + 0.363(C3HT) 0.430(C5TR) + 0.428(C5HT) + 0.565 23 (9/14) 20/23 87.0
(C7TR) 16.994
5 y = 0.170(C3AP) + 0.012(C3TR) + 0.389(C3HT) 0.487(C4AP) 0.169(C4TR) + 0.375(C4HT) + 0.158 23 (11/12) 19/23 82.6
(C5AP) 0.335(C5TR) + 0.108(C5HT) + 0.120(C6AP) + 0.189(C6TR) + 0.037(C6HT) 0.005(C7AP) +
0.409(C7TR) + 0.119(C7HT) 15.536
6 y = 0.061(C3AP) + 0.103(C3TR) + 0.327(C3HT) 0.372(C4AP) 0.118(C4TR) + 0.374(C4HT) + 0.138 24 (9/15) 19/24 79.2
(C5AP) 0.271(C5TR) + 0.101(C5HT) + 0.209(C6AP) + 0.412(C6TR) + 0.196(C6HT) 16.478
7 y = 0.089(C3TR) + 0.306(C3HT) 0.064(C4TR) + 0.373(C4HT) 0.367(C5TR) + 0.092(C5HT) + 24 (9/15) 21/24 87.5
0.498(C6TR) + 0.251(C6HT) 16.974
for a method to consistently produce repeatable results and, for vertebra. However, with the exception of the first cervical vertebra
the estimation of sex in forensic anthropology, must show less than (C1), CAP diameters did not exhibit statistically significant sexual
10% error to be accepted [5,9,33–40]. Accuracy rates and precision dimorphism. The etiology of the C1AP dimorphism may be attributed
levels that fall within the acceptable standards determine the to sexual variation between the male and female cranium [9]. Studies
reliability of the method. In forensic anthropological research, by Holland [41] and Gapert and colleagues [42] have shown that the
accuracy and precision error rates must be presented to demon- cranial base (i.e. occipital condyles and foramen magnum) exhibits
strate that the techniques utilized for the estimation of sex are sexual dimorphism. The C1 cradles the weight of the skull and
created using scientifically accepted principles that produce consequently shares a functional relationship with the cranial base.
results that are statistically greater than chance [39,38]. The dimorphic structures of the cranial base will, therefore, influence
Intra- and inter-observer error tests measure the precision levels the morphological structure of C1, such as the articular facets and the
of measurement reproducibility. In the Athens collection, the vertebral foramen [9,12].
evaluation of intra- and inter-observer error demonstrated error Predictive development models of vertebral growth by previous
variations that were less than 10% (Table 4). Error variations that are researchers have also shown that, of the three currently studied
less than 10% indicate acceptable precision standards [9,20,33–37]. characteristics (CAP, CTR, CHT), the CTR and CHT measurements
However, the results of inter-observer error (i.e. TEM, rTEM, and develop more slowly and for a longer period of time than CAP and
R) indicate issues with replicability of vertebral measurements. are therefore influenced by secondary sexual development and
Given the large raw value measurement discrepancies, this is environmental factors [24,26,51]. The results of vertebral growth
likely the result of misunderstanding or misinterpretation by the indicate that, by six years of age, the vertebral foramen has reached
research assistant as English was not their first language. In the approximately 90% of its complete adult size. During this
Luis Lopes collection, the evaluation of intra- and inter-observer developmental time, biomechanical forces will change the
error also demonstrated error variations that were less than 10%, vertebral foramen from a circular to a triangular shape to reflect
with the exception of C1TR (Table 4). However, the inter- the necessary range of neck motions that will morphologically
observer error for C1TR exhibited a 25.18% error (Table 5). This protect the spinal cord [24]. The majority of dimensional increases
may have also resulted from a misinterpretation of the C1TR occur in the CTR diameter, which reaches its adult size by
definition by the research assistant as English was not their approximately 10 years of age [24,26,51]. Therefore, the develop-
first language. Overall, the CTR variable exhibits the least ment of CTR extends into early puberty. The height of the vertebral
consistency with measurement repeatability as compared to bodies (CHT) has the longest period of growth reaching its full
CAP and CHT. Imprecision is likely decreased if the morpho- length at approximately 20 years of age [26,52,53], which may
metrics are performed by trained individuals. Therefore, experi- attribute to its greater sexual dimorphism compared to the
ence in morphometrics is essential and the authors suggest extra vertebral foramen measurements (CAP, CTR).
care and training when measuring and explaining the CTR Overall, the results of the current study are consistent with the
measurement variable, both in research and practical contexts. majority of previous research, which has also found minimal
Future research should focus on clarification and standardization dimorphism in the CAP measurement [18,24,43–46]. However,
of measurements of the cervical vertebrae. some studies have demonstrated conflicting results, noting the
existence of dimorphism in CAP diameters among various
4.2. Assessment of sexual dimorphism populations [9,15,17,21,25,26,42]. Human morphometrics can vary
between populations due to differences in ancestral groups,
The results of the current research indicate that the CHT genetic distance, environmental factors, and socioeconomic status,
measurement is the most dimorphic variable of the cervical which can influence the size and shape of skeletal characteristics
vertebrae, followed by the CTR measurement. These results are [9,22,24,25,47]. These influences may account for the differences
consistent with previous research that has also found CHT and CTR to displayed between the current research study and other popula-
be more dimorphic within other populations [9,15–17,19,24,25]. In tion studies that examined cervical vertebrae.
both the Athens and Lopes skeletal collections, mean CAP diameter In addition, variations in the methodological approaches and
values were larger in males than in females for every cervical statistical tools applied by the researchers in each study may be
A.S. Rozendaal et al. / Forensic Science International 306 (2020) 110072 9
related to the non-biological variability [48]. Sex estimation explore population-specificity of these discriminant functions
methods accept a type-1 error rate of 5% to identify whether using larger sample sizes.
relationships exist between measurements when a p-value 0.05 The European discriminant function equations achieved
is used to evaluate statistical significance [48–50]. Type-1 error is accuracy rates up to 84.5%. The results of this study indicate that
the misclassification of a relationship between testable variables, the accuracy of sex estimation is enhanced when all measurements
which is not biologically present, caused by procedural multiplicity from all cervical vertebrae are utilized; the more variables utilized
error rather than a quantitative error in the data set. When a in sex estimation methods from the cervical vertebrae, the more
p-value of 0.05 is selected, a ‘false positive’ is likely to occur five accurate the result. Therefore, when available, all measurements
times out of every 100 tests when identifying variable relation- and all cervical vertebrae should be utilized, as they are better
ships; Type-1 error results in a ‘false positive’ when examining predictors of sex in the European population. While the cervical
relationships between variables. To reduce the chances of a’ false vertebrae have proved useful for the estimation of sex from
positive’, a Bonferroni correction adjusts the significance level, skeletonized remains, the discriminant functions derived from the
since there is a greater likelihood that an error appears in the present study should be utilized only in cases in which other, more
statistical outcome as the number of testable variables increases. In accurate predictors of sex (i.e. the skull and the pelvis) are not
the current research, this adjustment made statistical analyses available for analyses.
more rigorous in identifying statistical significance.
CRediT authorship contribution statement
4.3. Discriminant functions for cervical vertebrae
Andrew S. Rozendaal: Conceptualization, Data curation,
A single vertebra did not accurately estimate sex at, or above, the
Formal analysis, Methodology, Project administration, Resources,
80% accuracy threshold. Studies have shown that with every increase
Software, Validation, Visualization, Writing - original draft,
in the number of measurements used in a discriminant function, the
Writing - review & editing. Shelby Scott: Formal analysis,
accuracy for estimating the sex also increases [15,16]. This is
Validation, Writing - original draft, Writing - review & editing.
consistent with the current research, which has found that utilizing a
Tanya R. Peckmann: Supervision, Writing - review & editing.
greater number of measurements and a greater number of cervical
Susan Meek: Writing - review & editing.
vertebrae achieves higher accuracies than utilizing fewer measure-
ments and fewer bones. For instance, when two cervical vertebrae
(C1 and C2) and all three measurements (CHT, CAP, and CTR) were Acknowledgments
used to estimate sex, the function achieved an overall accuracy rate of
72.8%. When the number of cervical vertebrae increased to four Special thanks are extended to Saint Mary’s University for
(C3–C6), the accuracy of the discriminant function utilizing all three awarding this research the Canadian Governor General’s Gold
measurements (CHT, CAP, and CTR) increased to 80.6%. When all Academic Medal for 2016. Thanks are extended to the University of
seven cervical vertebrae (C1 to C7) and all three measurements (CHT, Athens, Greece, Dr. Sortiris Manolis and Dr. Constantine Moun-
CAP, and CTR) were used, the discriminant function achieved an even trakis for allowing access to their skeletal collection. Also, thanks to
higher accuracy rate of 84.1% in the combined European population. the National Museum of Natural History and Science, Portugal, and
This is consistent with the findings of other researchers; with every Dr. Susana Garcia for allowing access to their skeletal collection.
increase in the number of measurements used in a discriminant Thanks are extended to Maria-Eleni Chovalopoulou and Natasha
function, the accuracy for estimating the sex also increases [15,16]. Carvalho for volunteering as research assistants in Greece and
Only seven discriminant functions, from the current study, Portugal, respectively.
achieved predictive accuracies at, or above, 80%. A cross-validation
study showed that four of the seven functions (Functions 2, 4, 5, and 7) References
achieved accuracies equal to or greater than their predictive
accuracies, which indicates they are strong statistical algorithms. [1] T.R. Peckmann, K. Orr, S. Meek, S.K. Manolis, Sex estimation from the calcaneus
in a 20th century Greek population using discriminant function analysis, Sci.
However, the cross-validation accuracies for Functions 1, 3, and 6 were
Justice 55 (2015) 377–382.
lower than their predicted accuracies. Functions 1 and 6 also exhibit [2] S. Scott, S. Ruengdit, T.R. Peckmann, P. Mahakkanukurah, Sex estimation from
less than 80% accuracy in the cross-validation testing. Christensen and measurements of the calcaneus: applications for personal identification in
Crowder [39] have acknowledged that although some methods may Thailand, Forensic Sci. Int. 278 (2017) 405.e1–405.e8.
[3] M.A. Tersigni-Tarrant, N.R. Shirley (Eds.), Forensic Anthropology: An Introduc-
achieve accuracy rates less than 80% they may still be used in a forensic tion, CRC Press, Boca Rotan, London, New York, 2013.
context if the available skeletal material does not allow for other [4] P.N. Gonzalez, V. Bernal, S.I. Perez, Geometric morphometric approach to sex
methods to be employed, e.g. fragmented remains. estimation of the human pelvis, Forensic Sci. Int. 189 (2009) 68–74.
[5] D.A. Komar, J.E. Buikstra, Forensic Anthropology Contemporary Theory and
Practice, Oxford University Press, Oxford, New York, 2008.
5. Conclusions [6] K.M. Spradley, R.L. Jantz, Sex estimation in forensic anthropology: skull versus
postcranial elements, J. Forensic Sci. 56 (2) (2011) 289–296.
[7] M.A. Bidmos, Metrical and non-metrical assessment of population affinity
To date, no study has utilized the combination of vertebral body from the calcaneus, Forensic Sci. Int. 159 (2006) 6–13.
height (CHT) and measurements of the vertebral foramen (CAP and [8] D.I. Kim, S.S. Lee, Y.S. Kim, Statistical analysis of bone elements excavated from
CTR) from the cervical vertebrae to estimate sex. This research the forensic context, Korean J. Phys. Anthropol. 23 (2010) 1–8.
[9] E.J. Marlow, R.F. Pastor, Sex determination using the second cervical vertebra –
demonstrates that the CHT and CTR measurements of the cervical a test of the method, J. Forensic Sci. 56 (1) (2011) 165–169.
vertebrae are the most sexually dimorphic, and thus make the [10] J. Dittrick, J.M. Suchey, Sex determination of prehistoric central California
most accurate contribution to biological sex differentiation. In skeletal remains using discriminant analysis of the femur and humerus, Am. J.
Phys. Anthropol. 70 (1) (1986) 3–9.
addition, this research shows no significant differences between
[11] T. Waldron, The relative survival of the human skeleton: implications for
the contemporary Athens (Greek) and historic Lopes (Portuguese) paleopathology, in: A. Boddington, A.N. Garland, R.C. Janaway (Eds.), Death,
skeletal collections and therefore the populations were combined Decay and Reconstruction: Approaches to Archaeology and Forensic Science,
for analyses. While these results indicate that discriminant Manchester University Press, Manchester, 1987, pp. 55–64.
[12] E.A. Marino, Sex estimation using the first cervical vertebra, Am. J. Phys.
functions for sex estimation from the cervical vertebrae are not Anthropol. 97 (2) (1995) 127–133.
affected by secular change or ancestry, future research should [13] M.J. Hollis, S. Kolakanuru, Robotic biomechanical testing of cervical spine
structures, Int. Fed. Med. Biol. Eng. Proc. 24 (2009) 211–212.
10 A.S. Rozendaal et al. / Forensic Science International 306 (2020) 110072
[14] M.D. Voisin, Sexual Dimorphism in the 12th Thoracic Vertebra and Its Potential [33] J.E. Molto, The assessment and meaning of interobserver in population studies
for Sex Estimation of Human Skeletal Remains, Wichita State University, 2011 based on discontinuous cranial traits, Am. J. Phys. Anthropol. 51 (3) (1979)
(Ph.D. Dissertation). 333–344.
[15] D.J. Wescott, Sex variation in the second cervical vertebra, J. Forensic Sci. 45 (2) [34] C.R. Nichol, C.G. Turner, Intra- and interobserver concordance in classifying
(2000) 462–466. dental morphology, Am. J. Phys. Anthropol. 69 (3) (1986) 299–315.
[16] J.D. Bethard, B.L. Seet, Sex determination from the second cervical vertebra: a [35] V. Novotny, M.Y. Işcan, S.R. Loth, Morphologic and osteometric assessment of
test of Wescott’s method on a modern American sample, J. Forensic Sci. 58 (1) age, sex, and Race from the skull, in: M.Y. Işcan, R.P. Helmer (Eds.), Forensic
(2013) 101–103. Analysis of the Skull: Craniofacial Analysis, Reconstruction, and Identification,
[17] J.M. Kibii, R. Pan, P. Tobias, Morphometric variations of the 7th cervical Wiley-Liss, New York, 1993, pp. 71–88.
vertebrae of Zulu, White, and Colored South Africans, Clin. Anat. 23 (4) (2010) [36] T.L. Rogers, A visual method of determining the sex of skeletal remains using
399–406. the distal humerus, J. Forensic Sci. 44 (1) (1999) 57–60.
[18] N.E. Tatarek, Variation in the human cervical neural canal, Spine J. 5 (2005) [37] T.L. Rogers, S.R. Saunders, Accuracy of sex determination using morphological
623–631. traits of the human pelvis, J. Forensic Sci. 39 (1994) 1047–1056.
[19] A. Amores, M.C. Botella, I. Aleman, Sexual dimorphism in the 7th cervical and [38] B.A. Williams, T.L. Rogers, Evaluating the accuracy and precision of cranial
12th thoracic vertebrae from a Mediterranean population, J. Forensic Sci. 59 (2) morphological traits for sex determination, J. Forensic Sci. 51 (4) (2006) 729–735.
(2014) 301–305. [39] A.M. Christensen, C.M. Crowder, Evidentiary standards for forensic anthro-
[20] I. Gama, D. Navega, E. Cunha, Sex estimation using the second cervical pology, J. Forensic Sci. 54 (6) (2009) 1211–1216.
vertebra: a morphometric analysis in a documented Portuguese skeletal [40] D. Dirkmaat, L.L. Cabo, S.D. Ousley, S.A. Symes, New perspectives in forensic
sample, Int. J. Leg. Med. 129 (2015) 365–372. anthropology, Yearbook Phys. Anthrop. 51 (2008) 33–52.
[21] J.E. Buikstra, D.H. Ubelaker, Standards for Data Collection from Human Skeletal [41] T.D. Holland, Sex determination of fragmentary Crania by analysis of the
Remains: Proceedings of a Seminar at the Field Museum of Natural History, cranial base, Am. J. Phys. Anthrop. 70 (2) (1986) 203–208.
Arkansas Archeological Report Research Series, 1994. [42] R. Gapert, S. Black, J. Last, Sex determination from the occipital condyle:
[22] H.F.C. Cardoso, The collection of identified human skeletons housed at the discriminant function analysis is an eighteenth and nineteenth century British
Bocage Museum (National Museum of Natural History), Lisbon, Portugal. Am. J. sample, Am. J. Phys. Anthrop. 138 (2009) 384–394.
Phys. Anthropol. 129 (2) (2005) 173–176. [43] B. Grave, T. Brown, G. Townsend, Comparison of cervicovertebral dimensions
[23] C. Eliopoulos, M. Ember, A modern, documented human skeletal collection in Australian aborigines and caucasians, South Eur. J. Orthod. Dentofac. Res. 21
from Greece, Homo. 58 (3) (2007) 221–228. (1999) 127–135.
[24] G.A. Clark, Heterochrony, Allometry, and Canalization in the Human Vertebral [44] V.M. Swenson, Ancestral and Sex Estimation Using E.A. Marino’s Analysis of
Column: Examples from Prehistoric American Populations, University of the First Cervical Vertebra Applied to Three Modern Groups, The University of
Massachusetts, 1985 (Ph.D. Dissertation). Montana, 2013 (Ph.D. Dissertation).
[25] C. Taitz, Anatomical observations of the developmental and spondylotic [45] B.S. Epstein, The Spine. A Radiological Text and Atlas, Lea and Febiger,
cervical spinal canal in South African blacks and whites, Clin. Anat. 9 (1996) Philadelphia, 1976.
395–400. [46] S. Singh, M. Balakrishnan, Morphometric study of cervical spine vertebrae in
[26] N.E. Tatarek, Changing Views of Health: Spinal Stenosis and Its Impact on Well- eastern region Nepalese population, Health Ren. 11 (3) (2013) 224–228.
Being, The Ohio State University, 1999 (Ph.D. Dissertation). [47] S.K. Gupta, R.C. Roy, A. Srivastava, Sagittal diameter of the cervical canal in
[27] H. Verbiest, Further experiences on the pathological influence of a normal Indian adults, Clin. Rad. 33 (1982) 681–685.
developmental narrowness of the bony lumbar vertebral canal, J. Bone Joint [48] J. Albanese, A metric method for sex determination using the hipbone and the
Surg. 37B (1955) 576–583. femur, J. Forensic Sci. 48 (2) (2003) 1–11.
[28] G. Fully, Une Nouvelle Méthode De Détermination De La Taille, Ann. De Med. [49] H.V. Chandrakanth, T. Kanchan, K. Krishan, Osteometric analysis for sexing of modern
Leg. 35 (1956) 266–273. sternum: an autopsy study from South India, Leg. Med. 16 (6) (2014) 350–356.
[29] M.H. Raxter, B.M. Auerback, C.B. Ruff, Revision of the fully technique for [50] L.J. Chase, L.R. Chase, R.K. Tucker, Statistical power in physical anthropology: a
estimating stature, Am. J. Phys. Anthropol. 130 (2006) 374–384. technical report, Am. J. Phys. Anthrop. (1978) 133–138.
[30] T.D. White, M.T. Black, P.A. Folkens, Human Osteology, 3rd ed, Elsevier [51] R. Porter, D. Pavitt, The vertebral canal: I. Nutrition and development, an
Academic Press, Burlington, California, Oxford, 2012. archaeological study, Spine 12 (9) (1987) 901–906.
[31] C. Bernau, M. Riester, A. Boulesteix, C. Huttenhower, L. Waldron, L. Trippa, [52] H.F. Cardoso, L. Rios, Age estimation from stages of Epiphyseal Union in the
Cross-study validation for the assessment of prediction algorithms, Bioinfor- presacral vertebrae, Am. J. Phys. Anthrop. 144 (2011) 238–247.
matics 30 (2014) i105–i112. [53] M. Albert, D. Mulhern, M.A. Torpey, E. Boone, Age estimation using thoracic
[32] P. Refaeilzadeh, L. Tang, H. Liu, Cross validation, in: M.T. Özsu, L. Liu (Eds.), and first two lumbar vertebral ring epiphyseal union, J. Forensic Sci. 55 (2010)
Encyclopedia of Database Systems, Springer, 2009. 287–294.