TECHNICAL NOTE

ANTHROPOLOGY
Claudia Garrido-Varas,
1
Ph.D.; Raveen Rathnasinghe,
1
B.Sc.; Tim Thompson,
1
Ph.D.; and
Yoland Savriama,
2
Ph.D.
A New Method to Pair-match Metacarpals
Using Bilateral Asymmetry and Shape
Analysis*
ABSTRACT: Matching pairs of skeletal elements is of fundamental importance when re-associating skeletons from commingled settings. This
study presents a new method that combines the use of traditional anthropometric size measurements and the analysis of shape, through geometric
morphometrics, to form pairs from commingled settings. We tested the hypothesis that shape asymmetry among individuals is smaller than
between individuals that share similar dimensions. The metacarpals were measured in their maximum length, and absolute means of asymmetry
between right and left metacarpals were calculated. To analyze the shape characteristics, two-dimensional landmarks were recorded from digi-
tal images of 111 metacarpals. The specimens were analyzed using generalized Procrustes analysis and multivariate statistics. The accuracy of
pair-matching bones using a combined method was of 100%, showing that the incorporation of geometric morphometrics techniquesin the
assessment of shape in physical anthropologyis of particular use when comparing the shape differences/similarities between individuals.
KEYWORDS: forensic science, forensic anthropology, commingled human remains, metacarpals, geometric morphometrics, Procrustes
superimposition, bilateral asymmetry
Matching pairs of skeletal elements is one of the approaches
used to re-associate skeletons, to estimate the number of individ-
uals, and determine the recovery probability in a sample of com-
mingled remains (13). Although at a first glance, the process of
pair matching might appear to be an easy task, there are factors
that increase its complexity, such as sample size, variability of
the asymmetry patterns and fragmentation which is often the
by-product of significant time in the ground, perimortem trauma,
and commingled events of decomposition and movement.
The accuracy of visually pairing tibiae, femora, and humeri
from human adults and subadults has been tested showing more
than 90% of accuracy (1). However, hand bones have not been
tested in this respect. The metacarpals were chosen for this study
because they have distinct anatomical features which make them
promptly recognizable with respect to digit and laterality. Meta-
carpals have a better preservational signature in comparison with
the other hand bones (4).
The aim of this study was to investigate whether shape is a
better discriminating feature than size when comparing bones
from different individuals when attempting to pair-match them.
First, the accuracy of visually matching pairs of human adult
metacarpals was tested. Subsequently, size and shape asymmetry
were calculated. Traditionally, pair matching has been performed
through osteometric sorting and gross morphology observation.
Although visual pair matching can result in high percentage of
correctly selected pairs, it can be thought as a subjective method
that might be difficult to assess in its reliability and repeatability.
Furthermore, it is very dependent on the experience of the obser-
ver (5,6). We propose a method to compare shape avoiding the
subjectivity of visual assessment. For this, fifteen left metacar-
pals were randomly selected, possible right pairs were then
identified using asymmetry ranges (derived from the asymmetry
evaluation of the maximum length of the metacarpals), subse-
quently, the left metacarpal with its possible right pairs was sub-
jected to shape analysis, and the right pair that had the smallest
shape difference (Procrustes distance), when compared to the left
target, was chosen to form a pair.
Materials and Methods
Test for the Accuracy of Matching Pairs of Metacarpals
To examine the reliability of matching pairs of metacarpals,
five experimental tests were performed using samples retrieved
from the Hayton Skeletal Collection housed at Teesside Univer-
sity, United Kingdom. This collection was archaeologically
recovered from a cemetery of the Anglo-Saxon period (6th to
9th century). It is composed of 42 individuals of whom 32 are
adults. The burials were primary and with no evidence of com-
mingling. The criterion used for pair matching was gross mor-
phology; parameters observed included overall symmetry and
robusticity, muscle attachment marks, diaphysis torsion, and
epiphyseal shape. Taphonomic indicators such as color changes
1
School of Science and Engineering, Teesside University, Borough Road,
Middlesbrough, TS1 3BA, UK.
2
Department of Biological Sciences, St. Johns University, St. Albert Hall,
Rm 272, 8000 Utopia Parkway, New York, NY 11439.
*Financial support provided by Becas Chiles and Servicio Medico Legal,
Chile.
Received 14 June 2013; and in revised form 13 Nov. 2013; accepted 17
Nov. 2013.
1 2014 American Academy of Forensic Sciences
J Forensic Sci, 2014
doi: 10.1111/1556-4029.12577
Available online at: onlinelibrary.wiley.com
and weathering patterns were not considered as significant as the
general morphology attributes when reaching a conclusion.
The selection of bones included all the complete adult metacar-
pals of the collection. Bones that were incomplete or fragmented,
presented erosion, healed fractures, osteoporosis, osteomyelitis,
osteoarthritis, or osteolytic lesions were excluded. These requi-
sites resulted in 111 metacarpals derived from 17 individuals
(65.2%). The sample size for each metacarpal pair-matching test,
number of pairs, and unpaired metacarpals are detailed in
Table 1.
All five metacarpals were independently tested. The inventory
numbers were covered with a small auto-adhesive label, and the
bones commingled and left on five trays on the laboratory table.
The sample selection and the preparation for the tests were per-
formed by the first author (CGV), and the test was taken by two
trained physical anthropologists with broad experience in osteol-
ogy. The observers did not know the number of individuals of
the sample or the recovery percentage that they represented. The
observers were instructed to start the tests using any digit, with
no time constraints. A minimum of 10-min break was installed
between tests. After completing the tests, the observers were
allowed to go back to the previous tests if desired. Results of
the tests were recorded as the number of correct pairs, incorrect
pairs, and missed pairs.
Elaboration of Ranges of Asymmetry of Size
The maximum length of the paired metacarpals (41 pairs) was
measured with a sliding calliper (0.1 mm). To minimize error,
measurements were taken on three different occasions, and the
mean of them used in the analyses. The mean error (0.1 mm)
and percentage error (0.11%) (7) were calculated and considered
negligible. To evaluate the mean asymmetry between sides, the
absolute value of right minus left, of each pair, was calculated.
This was carried out independently for each set of metacarpals
(first to fifth).
A
RjR Lj
number of pairs
A, mean asymmetry; R, right; L, left.
This formula is equivalent to the fluctuating asymmetry index
FA1 (8). In this study, it was chosen to quantify the variation
between sides because it is an unbiased estimator of the sample
standard deviation (9) and not to establish the type of pattern of
bilateral variation. The data were then checked for normality
with the ShapiroWilk test using SPSS version 19 software
package (10). The ranges of asymmetry for each metacarpal
were calculated as the mean absolute asymmetry plus/minus 2
standard deviations, to consider 95% of inclusion.
R A 2 SD
R, range of asymmetry, SD, standard deviation.
Analysis of Shape
Shape was analyzed through geometric morphometrics; this
approach allows to extract shape information from the specimens
under study and to analyze the asymmetry patterns. To retrieve
such information, each specimen is defined as a configuration of
landmarks, which are anatomical points that can be found among
all the specimens included in the analysis. In the field of geo-
metric morphometrics, shape is defined as all the geometric
information about a configuration of landmarks except for its
size, position, and orientation (11). Due to the general morphol-
ogy of the metacarpals, type 2 landmarks were used, defined as
maxima of curvatures or other morphogenetic processes (12).
The metacarpals were photographed in a standardized manner,
which consisted in the bone resting on its palmar surface, avoid-
ing rotation of the diaphysis while keeping the dorsal surface of
the bone on a plane parallel to the supporting table. A digital
camera was mounted on a tripod with a distance of the lens to
the supporting table of 50 cm. The lens was set at 90 degrees
with respect to the table, and the bone situated in the center of
the visual field to avoid distortion. Each bone was photographed
twice with a one-day interval between pictures, and these pic-
tures duplicated, resulting in four pictures per specimen. Pictures
of the left metacarpals were reflected in their horizontal axis to
standardize the digitising process of the landmarks. Landmarks
were digitized using the software tpsDIG2 (13). All images were
TABLE 1Sample size of the visual pair-matching test.
Metacarpal test n Pairs Isolated
First 24 9 6
Second 17 7 3
Third 25 8 9
Fourth 24 9 6
Fifth 21 8 5
FIG. 1First right metacarpal. Dorsal view, scale in cm. Localization of
landmarks: 1: most medial point of the base, 2: most proximal point of the
base, 3: most lateral point of the base, 4: most lateral point of the head, 5:
most distal point of the head, and 6: most medial point of the head.
2 JOURNAL OF FORENSIC SCIENCES
digitized by the first author. The definitions of the landmarks
and their locations on the metacarpals are detailed in Figs 15.
Taking two pictures and duplicating them allow to partition out
the error due the photographic technique and the digitizing pro-
cess when quantifying asymmetry with a two-factor analysis of
variance (ANOVA) (8,14).
To evaluate the error due to digitizing, one randomly selected
picture from each set of metacarpals was digitized independently
by two of the authors (CGV and RR) twenty times and superim-
posed using Generalized Procrustes superimposition. The intra-
observer error was evaluated by the displacement of the
landmarks between the twenty pictures. For the interobserver
error, one of the twenty observations of each set of metacarpals
was randomly selected. Then, the configuration digitized by
CGV was superimposed to the one digitized by RR.
Shape and statistical analyses were performed with MorphoJ
(15). Each group of metacarpals was subjected to Procrustes
superimposition. To visualize the shape variation among indi-
viduals, a covariance matrix was generated from the Procrustes
coordinates and subjected to a principal component analysis
(PCA). Shape asymmetry was assessed with a Procrustes
analysis of variance (ANOVA); this is a two-way mixed model
that accounts for variation among individuals (symmetric vari-
ation), directional asymmetry (mean asymmetry), fluctuating
asymmetry (variation within individuals), and measurement
errors due to the image acquisition and the digitizing process
(1518).
Pair-matching Experiments
Because of the existence of asymmetry in the maximum
length, bones that share similar dimensions cannot be paired or
excluded as belonging to the same individual based only on this
one attribute. We hypothesized that shape differences should be
smaller in bones belonging to the same individual than bones
belonging to different individuals. To test this hypothesis, 3 left
metacarpals from each set (first to fifth) were randomly selected.
The possible pairs of these metacarpals were selected using the
asymmetry ranges derived from the size asymmetry assessment.
They were subjected to Procrustes superimposition, and the Pro-
crustes distances calculated. As a manner of visualization of the
relative position between individuals, a graph of coordinates
PC1 and PC2 was generated for each experiment.
Results
Visual Pair-matching Tests
From the total of 111 metacarpals selected for the test, there
were 41 pairs and 29 unpaired metacarpals. Considering the five
independent tests together, observer A correctly identified 31
pairs (75.6%) and missed 10 pairs (24.4%); observer B correctly
identified 34 (82.9%) and missed 7 (17.1%). Results showed that
incorrect pairs were also formed; observer A formed 5 wrong
pairs and observer 2, 2 pairs. Results of the pair-matching tests
are shown in Table 2.
FIG. 2Second right metacarpal. Dorsal view, scale in cm. Localization
of landmarks: 1: most medial point of the base, 2: most proximal and medial
point of the base, 3: most distal point of the base, 4: most proximal and lat-
eral point of the base, 5: most lateral point of the head, 6: most distal point
of the head, and 7: most medial point of the head.
FIG. 3Third right metacarpal. Dorsal view, scale in cm. Localization of
landmarks: 1: most medial point of the base, 2: most distal point of the base,
3: most proximal point of the stylus process, 4: most lateral point of the
base, 5: most lateral point of the head, 6: most distal point of the head, and
7: most medial point of the head.
GARRIDO-VARAS ET AL
.
PAIR-MATCHING METACARPALS 3
Absolute Asymmetry of the Maximum Length
Mean values of absolute asymmetry varied from 0.4 mm to
0.9 mm. The ShapiroWilk test of normality showed that the
data were normally distributed, for all of the groups. The mean,
standard deviation, and ranges of asymmetry, for each set of
metacarpals, are presented in Table 3.
Shape Analysis
Regarding the intra-observer error in the digitizing process,
the displacement of the landmarks for both the observers was
minimal, demonstrated by a Procrustes distance <0.001 for both
the observers. The interobserver error again showed minimal dis-
placement of the landmarks with a Procrustes distance <0.001.
The PCA of the shape variables showed that the principal com-
ponents 1 and 2 accounted for 66, 57, 63, 69, and 70% of the
variance for metacarpal 15, respectively. All five metacarpals
showed significant shape asymmetry among and between indi-
viduals; directional shape asymmetry was only observed in the
second metacarpal. Results are presented in Table 4.
Pair-matching Experiments
When comparing the Procrustes distances between the left tar-
get metacarpals and their possible pairs, in all the 15 experi-
ments, the smallest distance corresponded to the true pair of the
left target. Results are presented in Table 5. When visualizing
the relative position in the PCA of shape variables graphs, in all
the cases the right metacarpal that was closer to the target left
FIG. 4Fourth right metacarpal. Dorsal view, scale in cm. Localization
of landmarks: 1: most proximal and medial point of the base, 2: most proxi-
mal point of the base, 3: most lateral point of the base, 4: most lateral point
of the head, 5: most distal point of the head, and 6: most medial point of the
head.
FIG. 5Fifth right metacarpal. Dorsal view, scale in cm. Localization of
landmarks: 1: most proximal point of the base, 2: most distal point of the
base, 3: most lateral point of the base, 4: most lateral point of the head, 5:
most distal point of the head, and 6: most medial point of the head.
TABLE 2Results of the pair-marching test.
Metacarpal Observer Correct Pairs Incorrect Pairs Missed Pairs
1 A 5 4 4
1 B 9 0 0
2 A 5 1 2
2 B 4 1 3
3 A 8 0 0
3 B 7 1 1
4 A 6 0 3
4 B 8 0 1
5 A 7 0 1
5 B 6 0 2
TABLE 3Absolute asymmetry of the maximum length of the metacarpals.
Metacarpal
Absolute Asymmetry
Mean SD Range
1 0.8 0.6 2.0
2 0.9 0.6 2.1
3 0.7 0.6 1.9
4 0.4 0.3 1.0
5 0.4 0.2 0.8
4 JOURNAL OF FORENSIC SCIENCES
metacarpal was the correct one. Although these graphs do not
show all the variation due to shape difference, they are a
good graphical mean of visualization of the variation, as PC1
and PC2 accounted for more than 57% of the total shape varia-
tion. Figures 6 and 7 are shown as examples.
Discussion and Conclusions
The visual pair-matching test showed that matching pairs of
metacarpals only through gross morphology is not error-free.
Although one observer performed better than the other, which
could be attributed to experience, both of them over-looked pairs
and also made incorrect pairings. Osteometric sorting (19) and
the use of the tables of asymmetry generated through the statistic
M (20) are techniques that provide powerful means to re-associ-
ate commingled remains. As they both utilize size measure-
ments, they are especially useful in determining whether two
bones derive from different individuals. However, they rely on
additional analyses to conclude that the bones are from a same
individual, for example, visual pair matching. The new method
presented here can complement the metric assessments to reach
the conclusion that two bones belong to the same individual.
This method is not as sensitive to the experience of the observer
and is mathematically defined, lacking subjectivity.
The combination of metric asymmetry ranges and the compar-
ison of pairs through morphometric analysis resulted in 100% of
correct pair matching. These results have to be seen under the
light of the experimental approach taken in this study where the
true pairs of the left bones were present among the possible
pairs. The method presented here will find the closest pair
through the GM analysis; if the true pair is not in the sample,
TABLE 4Results of the Procrustes ANOVA for the shape effect.
Effect
Metacarpals
1 2 3 4 5
F P F P F P F P F P
Individual 5.44 <0.0001 5.61 <0.0001 3.08 <0.0001 8.39 <0.0001 7.56 <0.0001
Side 1.38 0.2238 2.58 0.01 0.25 0.8677 0.71 0.68 1.1 0.3813
Individual/side 11.16 <0.0001 3.78 <0.0001 30.9 <0.0001 10.72 <0.0001 4.58 <0.0001
Error 1 3.09 <0.0001 2.44 <0.0001 1 0.4885 1.83 <0.0001 6.36 <0.0001
TABLE 5Possible pairs for left metacarpals.
Metacarpal Left Possible Pairs Procrustes Distance Closest Pair
1 204 204204 0.001256 204
204216 0.002686
254 254254 7.098991*10E-4 254
254783 0.002587
783 783783 0.00209 783
783204 0.00526
2 254 254254 6.593696*10E-4 254
254216 0.002037
254783 9.392391*10E-4
783 783783 7.582978*10E-4 783
783216 0.001701
870 870870 4.786860*10E-4 870
870883 4.920022*10E-4
3 216 216216 9.918945*10E-4 216
216240 0.001438
254 254254 2.319882*10E-4 254
254889 8.411495*10E-4
870 870870 9.738292*10E-4 870
870254 0.001219
4 212 212212 1.363570*10E-4 212
212204 8.068839*10E-4
216 216216 7.846951*10E-4 216
216254 0.001463
783 783783 1.744199*10E-4 783
783216 9.879588*10E-4
5 207 207207 4.541933*10E-4 207
207204 0.02821
207216 0.001877
216 216216 1.878048*10E-4 216
216204 0.004626
254 254254 7.861387*10E-4 254
254783 0.001389
FIG. 6Principal components 1 and 2 for the pair-matching experiment
254 left- 254 right- 216 right- 783 right. Black dots indicate bones of a same
individual; grays are unpaired bones.
FIG. 7Principal components 1 and 2 for the pair-matching experiment
207 left- 207 right- 204 right- 216 right. Black dots indicate bones of a same
individual; grays are unpaired bones.
GARRIDO-VARAS ET AL
.
PAIR-MATCHING METACARPALS 5
this methodology will lead to error. Nevertheless, the scenarios
in forensic and archaeological contexts that mimic this setting
(where the true pair is present in the sample) are not uncommon.
For example, CGV has found this situation in burials belonging
to the period of the civil war in Spain, where two or more indi-
viduals were buried in the same pit. These burials were primary
and undisturbed; nevertheless, the hand bones were found com-
mingled. Other cases where this method can be applied are the
ones where two skeletons have been reburied together after
skeletisation and posteriorly re-exhumed for analysis.
With regard to the anatomical unit, the method presented has
been applied to pair other elements of the appendicular skeleton
(21). In theory, this method can be applied to any paired element
of the skeleton, although further research is underway that inves-
tigates the accuracy on other bones and means of improving this.
For example, other approaches in GM can be tested such as out-
lines and the use of three-dimensional data, while 3D scanners
can also be used as an alternative to the traditional 2D image
data. Also, the size component of the configuration of landmarks
(centroid size) could be incorporated into the analysis. Although
the methodology presented here is not appropriate for all field
situations, once familiar with the process of acquiring the pic-
tures on a standard position, the method is not time-consuming
and will prove effective for particular commingled cases. The
experimental hypothesis that stated that the shape difference
among individuals is smaller than between individuals was pro-
ven correct. Analyzing shape with geometric morphometrics with
the purpose of pair-matching bones has never been attempted
before in physical anthropology. This study demonstrated the
great applicability that GM has and it is of major relevance in
physical anthropology, where shape has been traditionally
assessed from a visual perspective which can be subjective and
difficult to document.
Acknowledgments
We thank Dr. David Gonc alves and the anonymous reviewers
for their constructive comments, David Errickson for editing the
manuscript, and Dr. Francisco Etxeberria G. for giving access to
his laboratory in the Universidad del Pas Vasco.
References
1. Adams BJ, Konigsberg LW. Estimation of the most likely number of
individuals from commingled human skeletal remains. Am J Phys
Anthropol 2004;125(2):13851.
2. Nikita E, Lahr MM. Simple algorithms for the estimation of the initial
number of individuals in commingled skeletal remains. Am J Phys
Anthropol 2011;146(4):62936.
3. White TD, Folkens PA. The human bone manual. London: Elsevier Aca-
demic Press, 2005.
4. Cox M, Bell L. Recovery of human skeletal elements from a recent UK
murder inquiry: preservational signatures. J Forensic Sci 1999;44(5):94550.
5. Kerley ER. Special observations in skeletal identification. J Forensic Sci
1972;17(3):34957.
6. Ubelaker D. Approaches to the studies of commingling in human skeletal
biology. In: Haglund WD, Sorg MH, editors. Advances in forensic
taphonomy: method, theory, and archaeological perspectives. Boca
Raton, FL: CRC Press, 2010.
7. White TD, Folkens PA. Human osteology, 2nd edn. San Diego, CA:
Academic Press, 2000.
8. Palmer AR, Strobeck C. Fluctuating asymmetry: measurement, analysis,
patterns. Annu Rev Ecol Syst 1986;17:391421.
9. Palmer AR. Fluctuating asymmetry analyses: a primer. In: Markow TA,
editor. Developmental instability-its origins and evolutionary implica-
tions: Proceedings of the International Conference; 1993 June 1415;
Tempe, AZ. Norwell, MA: Kluwer Academic Publishers, 1994;33564.
10. IBM Corp. IBM SPSS statistics for Windows, Version 19.0. Armonk,
NY: IBM Corp., 2010.
11. Dryden IL, Mardia KV. Multivariate shape analysis. Sankhya 1993;55
(A):46080.
12. Bookstein FL. Morphometric tools for landmark data. Geometry and
biology. Cambridge: Cambridge University Press, 1991.
13. Rohlf FJ. tpsDig, digitize landmarks and outlines, version 2.05. Stony
Brook, NY: Department of Ecology and Evolution, State University of
New York at Stony Brook, 2005.
14. Palmer AR, Strobeck C. Fluctuating asymmetry analyses revisited. In:
Polak M, editor. Developmental instability: causes and consequences.
Oxford: Oxford University Press, 2003;279319.
15. Klingenberg CP. MorphoJ: an integrated software package for geometric
morphometrics. Mol Ecol Resour 2011;11(2):3537.
16. Savriama Y, Klingenberg CP. Beyond bilateral symmetry: geometric
morphometric methods for any type of symmetry. BMC Evol Biol
2011;11(1):280.
17. Klingenberg CP, Barluenga M, Meyer A. Shape analysis of symmetric
structures: quantifying variation among individuals and asymmetry. Evo-
lution 2002;56(10):190920.
18. Mardia KV, Bookstein FL, Moreton IJ. Statistical assessment of bilateral
symmetry of shapes. Biometrika 2000;87(2):285300.
19. Byrd JE, Adams BJ. Osteometric sorting of commingled human remains.
J Forensic Sci 2003;48(4):71724.
20. Thomas RM, Ubelaker DH, Byrd JE. Tables for the metric evaluation of
pair-matching of human skeletal elements. J Forensic Sci 2013;58
(4):9526.
21. Garrido-Varas CE. An investigation into bilateral asymmetry of the
appendicular skeleton of the adult human and its use in physical and
forensic anthropology [doctoral thesis]. Middlesbrough: Teesside Univer-
sity, 2013.
Additional information and reprint requests:
Claudia Garrido-Varas, Ph.D.
Willie Arthur Aranguiz 2019
Departamento 502
Providencia Santiago
Chile
E-mails: c.garrido-varas@tees.ac.uk and dragarrido1@gmail.com
6 JOURNAL OF FORENSIC SCIENCES