












































Forensic Anthropology Vol. 5, No. 3: 239–250
DOI: 10.5744/fa.2020.2023a

© 2022 University of Florida Press

Population variation can be explored using metric analysis of 
the human cranium (Algee- Hewitt 2016; Hefner et al. 2016; 
Howells 1973; Relethford 1994; 2010; Roseman & Weaver 
2004; Wrobel et al. 2018). In forensic anthropology, these analy-
ses are incorporated in the estimation of ancestry, or broad geo-
graphical origin (Algee- Hewitt 2017; Dunn et al. 2020; Hefner 
et al. 2016; Spradley 2014a; Stull et al. 2014). These studies rely 
on known data obtained from reference samples to create popu-
lation affinity estimation models generating probabilistic state-
ments for an unknown individual’s population affinity. The 
level of resolution in these models can range from broad to nar-
row, generally representing some form similar to the three-  or 
five- group ancestry, social constructs and peer- perceived classi-
fications, or even finer population levels (Hefner 2018). Apply-
ing these models expedites the identification process in forensic 
casework and increases our understanding of group similar-
ity within localized geographical arenas or larger geopolitical 

boundaries. Using these data, we investigate intraregional vari-
ation in a sample from Antioquia, Medellín, Colombia.

Colombians account for approximately 2%, or roughly 1.1 
million, documented immigrants to the U.S. each year (López 
2015). In 2015, the U.S. was the second most common destina-
tion for out- migration from Colombia (Carvajal 2017). Follow-
ing current U.S. governmental classifications, Colombians 
are included in the broad demographic category Hispanic. 
This paper reinforces the argument for population- specific 
methods beyond Hispanic and emphasizes the need for repre-
sentative comparative datasets. We agree with Ross et al. 
(2004) and Spradley (2014a, 2014b) that the term Hispanic is 
uninformative to the assessment of geographic origin in a 
forensic context. We present support for finer levels of resolution 
in po pulation affinity estimation, specifically for Hispanic 
groups. When we do use the term Hispanic in this paper, we 
apply it as it is used in the United States’ governmental classifi-
cations to distinguish persons originating from Latin America. 
We understand this term is not reflective of self- identification or 
ethnicity.

Materials and Methods

Data used in this study include craniometric measurements 
collected from skeletal material originating in Antioquia, 

aDepartment of Anthropology, Michigan State University, 
East Lansing, MI 48824, USA

bDepartamento de Antropología— FCSH, Universidad de Antioquia, 
Medellín, Colombia

*Correspondence to: Kelly R. Kamnikar, Department of 
Anthropology, 655 Auditorium Drive, Michigan State University, 
East Lansing, MI 48824, USA 

e- mail: kamnikar@msu . edu

Received 17 July 2019; Revised 29 April 2020;  
Accepted 02 May 2020

RESEARCH ARTICLE

Craniometric Variation in a Regional Sample from Antioquia, 
Medellín, Colombia: Implications for Forensic Work 
in the Americas

Kelly R. Kamnikara* ● Joseph T. Hefnera ● Timisay Monsalveb ● Liliana Maria Bernal Florezb

ABSTRACT: Population affinity estimation is frequently assessed from measurements of the cranium. Traditional models place indi-
viduals into discrete groups―such as Hispanic―that often encompass very diverse populations. Current research, including this study, 
challenges these assumptions using more refined population affinity estimation analyses. We examine craniometric data for a sample of 
individuals from different regions in Antioquia, Colombia. We first assessed the sample to understand intraregional variation in cranial 
shape as a function of birthplace or a culturally constructed social group label. Then, pooling the Colombian data, we compare cranial vari-
ation with global contemporary and prehistoric groups. Results did not indicate significant intraregional variation in Antioquia; classifica-
tion models performed poorly (28.6% for birthplace and 36.6% for social group). When compared to other groups (American Black, 
American White, Asian, modern Hispanic, and prehistoric Native American), our model correctly classified 75.5% of the samples. We 
further refined the model by separating the pooled Hispanic sample into Mexican and Guatemalan samples, which produced a correct 
classification rate of 74.4%. These results indicate significant differences in cranial form among groups commonly united under the classi-
fication “Hispanic” and bolster the addition of a refined approach to population affinity estimation using craniometric data.

KEYWORDS: forensic anthropology, craniometric analysis, population affinity estimation, human rights, biological distance



240 Cranial Variation in Antioquia, Colombia

TABLE 1—Interlandmark distances used in analysis.

Abbreviation Measurement Abbreviation Measurement

GOL
XCB
ZYB
BBH 

 
BNL
BPL 

 
MAB 

 
AUB
NLH
NLB
WFB* 

 
FOB*

cranial length
cranial breadth
bizygomatic breadth
basion- bregma 

height
cranial base length
basion- prosthion 

length
maxilla- alveolar 

breadth
biauricular breadth
nasal height
nasal breadth
minimum frontal 

breadth
foramen magnum 

breadth

OBB
OBH
EKB
DKB 

 
FRC
PAC
OCC
FOL 

 
MDH
MAL* 

 
UFBR* 

 
UFHT* 

 

orbital breadth
orbit height
biorbital breadth
interorbital 

breadth
frontal chord
parietal chord
occipital chord
foramen magnum 

length
mastoid height
maximum 

alveolar length
upper facial 

breadth
upper facial 

height

Adapted from Langley et al. (2016). (*) indicates a measurement used in 
the intraregional Colombian analysis.

TABLE 2—Colombian sample by birthplace 
and social group.

Birthplace Social Group n

Uraba A 6
Occidente A 17
Oriente B 29
Suroeste B 44
Nordeste C 6
Norte D 25
Valle de Aburra E 89

Total:  216
*Adapted from Monsalve & Hefner (2016).

Medellín, Colombia; a sample from the Forensic Databank 
(FDB; Jantz & Moore- Jansen 1998); and a sample of prehis-
toric Native American data from the Howells dataset (1973) 
and the National Museum of Natural History (JTH). A sep-
arate discussion for each sample follows.

Colombian Sample

The craniometric data in the Colombian sample comprises 
19 interlandmark distance measures (Table 1) from 243 indi-
viduals (males = 172; females = 70; unknown = 1). This 
sample— referred to hereafter as the Colombian sample―
originates from the osteological reference collection (N = 317) 
housed at the University of Antioquia, Medellín, Colombia 
(Monsalve & Hefner 2016). This collection contains individ-
uals with known birth years between the 1950s and the cur-
rent century, with age- at- death values ranging from birth to 
99 years. Many of the individuals in the Colombian sample 
originate from the San Pedro Cemetery Museum and the Uni-
versity Cemetery in the city of Medellín, but have birth loca-
tions from multiple municipalities within the department of 
Antioquia (Table 2; Figure 1). To ensure ethical use of data, 
we only included those individuals who donated their remains 
or were donated responsibly by next of kin for skeletal 
research. Information on phylogeographical social classifica-
tions were included in our analysis, although the nature of 
self- assignment or assignment after death is not known (see 
Table 2). As social group is used to distinguish living peo-
ples in Colombia, we also test the classificatory power of those 
labels derived from the osteological collection. Monsalve 
and Hefner (2016) provide a discussion on the roles social 
groupings play in Antioquia. Lastly, we pooled the Colom-
bian sample for comparison to other groups from the FDB, the 
Howells data, and the NMNH.

Comparative Samples

A subset of individuals from the FDB (n = 654), the Howells 
dataset (n = 268), and the NMNH (n = 60) were compared to 
the Colombian sample (see Table 3). The FDB is a repository 
for cranial and postcranial measurements from identified 
skeletal remains (Moore- Jansen & Jantz 1998). Our analysis 
includes samples representing American Black, American 
White, Asian, and individuals identified as Hispanic in the 
FDB. These samples were selected to reflect the current demo-
graphic structure of the U.S. Data for the American Black, 
American White, and Asian samples in the FDB include sam-
ples from the Terry, Bass, and Hamman- Todd Skeletal Collec-
tions and identified casework submitted to the FDB by forensic 
practitioners across the U.S. The FDB Hispanic sample 
broadly reflects Hispanic individuals as defined by U.S. prac-
tices, and includes individual with origins from a number of 
countries (e.g., Costa Rica, El Salvador, Guatemala, Hondu-
ras, Mexico, Puerto Rico, and Panamá) (Jantz & Ousley 2005). 
Due to small sample sizes (n < 2) for some countries in the 
FDB Hispanic sample, we limited our samples to the Mexican 
and Guatemalan individuals. We also compiled a third com-
parative sample representing prehistoric Native Americans 
derived from the Howells dataset and the NMNH. This sam-
ple was included to attempt to reflect groups with  similar pop-
ulation histories to indigenous populations in the Americas.

Statistical Analyses

All analyses were conducted using R Statistical Software 
(R Core Team 2018), a free software environment for statisti-
cal computing and graphics. To remove the influence of sex on 
the data, all observations were first scaled and centered (set-
ting mean to 0 and standard deviation to 1). For the majority 
of the sample, complete observation was the norm. However, 
some of the samples did have missing data. This includes 
(percent missing): Asian (9%), Colombian (33%), and prehis-
toric Native American (8%). Individuals with ten or more 
missing ILDs were removed entirely from the analysis. Miss-
ing values were imputed, by variable, using the mice 



Kamnikar et al. 241

package (van Buuren & Groothuis- Oudshoorn 2011). This 
method is appropriate for imputation of several data types, 
including continuous data. We used the predictive mean 
matching approach to select a random observation from the 
pool of observed values (by variable) to replace a missing 
value (van Buuren & Groothuis- Oudshoorn 2011).

Outliers were identified using Cook’s distance. Birth-
place and social group were tested among the intraregional 
sample for associations with craniometric measurements 
using a multivariate analysis of variance (MANOVA) and 

pairwise analysis of variance (ANOVA). The MANOVA 
was used to examine the effect of social group and birth-
place on cranial morphology, while the pairwise ANOVAs 
examined the individual relationships between the ILDs and 
the grouping variables.

A canonical (linear discriminant function [LDFA]) analy-
sis quantified the relationship between samples and measured 
variability within and between each to develop prediction equa-
tions. For intraregional tests, the dependent variables were 
birthplace and social group. In the more comprehensive com-
parative analysis, geographic origin was the dependent variable. 
In both LDFAs, prior probabilities were set to simulate equal 
probability of group membership. One analysis explored classi-
fication accuracies for the American Black, American White, 
Asian, pooled Colombian, prehistoric Native American, and 
pooled Hispanic samples; a second analysis separated the His-
panic sample into Colombian,  Guatemalan, and Mexican 
groups. All models were cross- validated using a leave- one- out 
(LOOCV) procedure. Finally, we calculated Mahalanobis dis-
tances (D2) using the HDMD package (McFerrin 2013) to assess 
similarity/dissimilarity and explore potential group relatedness.

Results

Intraregional Variability within Antioquia

Cook’s distance identified six potential outliers; however, four 
were from the Uraba (n = 6) region and one from the Nordeste 
(n = 6) region. Due to these already small samples, the outli-
ers were not removed.

The MANOVA results indicate significant effects (α = 0.05) 
for both birthplace and social group. The pairwise ANOVAs iden-
tified significant interaction between (1) birthplace and BBH, 

FIG. 1—Map of Antioquia, Colombia (Monsalve & Hefner 2016).

TABLE 3—Sample distribution.

Sample Size

males females unknown total

Colombian 172 70 1 243
American White 126 74 200
American Black 101 99 200
Asian 151 8 1 160
Native American 146 175 7 328

Arikara 42 27 69 ―
Santa Cruz 51 51 102 ―
Blackfeet 25 45 70 ―
Indian Knoll 11 16 27 ―
Smithsonian (JTH) 17 36 7 60 ―

Hispanic 94
Mexican 74 9 83 ―
Guatemalan 6 5 11 ―

Total: 1,198
*The subset of individuals in the American Black, American White, and Asian groups were 
randomly selected using the rand() function in Excel.



242 Cranial Variation in Antioquia, Colombia

BNL, BPL, MAB, and NLH, and (2) social group and 
BBH, BPL, MAB, and NLH.

The LDFA tested the influence of birthplace on cranial 
variability. Figure 2 illustrates the group centroids and their 
relationships. The overall correct classification rate by birth-
place using LOOCV was 28.6%; an accuracy above random 
allocation (~14%). Correct classifications range from 16.7% 
(Nordeste, Uraba) to 37.1% (Valle de Aburra) (Table 4). Here, 
the regions with the highest accuracies were Valle de Aburra 
followed by Occidente (28.6%) and Norte (28.0%). Uraba and 
Nordeste had the lowest classification accuracy at 16.7%, just 
above random, which could indicate disparate populations 
or issues related to sample size (n = 6) for both groups 
(Figure 2).

To understand if social designations influence craniomet-
ric variability (i.e., gene flow), a second LDFA was performed 
using social group categories. Performance rates were not 
high, although group classification increased to 36.6%. The 
highest classification accuracy was 43.8% (Table 5). Figure 3 
shows the centroids for each social group.

In order to understand overall similarity, a Mahalanobis 
distance (D2) matrix was calculated using birthplace and 
social group (Tables 6 and 7). Here, Suroeste and Oriente are 

similar, while the rest of the groups appear as separate 
branches. Uraba is most unlike other groups (Figure 4). Social 
groups ‘A’ and ‘D’ are most similar to each other, as are ‘B’ 
and ‘E’; however, social group ‘C’ is distinct from all other 
groups (Figure 5).

Variability among the Colombian Sample and 
Comparative Samples

The LDFA of the Colombian and the comparative samples 
perform well (Figure 6). The American Black, American 
White, and Colombian sample exhibit slight overlap, but over-
all good separation. There is significant overlap between the 
Hispanic, prehistoric Native American, and Asian samples. 
Table 8 provides the classification accuracies for each group, 
with an overall model accuracy of 74.7%. The lowest classi-
fication rate is for the combined Hispanic group (50.0%), 
while all other groups had classification accuracies higher 
than 71.0%.

An additional LDFA on the refined Hispanic dataset has 
a similar overall classification rate (71.4%). Figure 7 illustrates 
the separation of the LDFA on the test sample. However, 
group classification rates vary considerably (Table 9). The 

FIG. 2—The first two discriminant axes, by birthplace.



Kamnikar et al. 243

TABLE 4—Classification rate of Colombian sample by birthplace (LOOCV).

Nordeste Norte Occidente Oriente Suroeste Uraba
Valle de 
Aburra CCR (%)

Nordeste 1 0 0 2 2 0 1 16.7
Norte 3 7 6 3 2 1 3 28.0
Occidente 0 2 4 4 2 1 3 28.6
Oriente 5 5 2 6 7 2 2 20.7
Suroeste 6 7 1 7 9 6 8 20.5
Uraba 0 1 0 1 2 1 1 16.7
Valle de Aburra 10 13 7 8 12 6 33 37.1

Overall 28.6

TABLE 5—Classification rate of the Colombian sample by social group (LOOCV).

Group A Group B Group C Group D Group E CCR (%)

Group A 5 5 1 5 4 25.0
Group B 11 23 13 14 12 31.5
Group C 0 4 1 0 1 16.7
Group D 6 3 3 10 3 40.0
Group E 11 14 10 15 39 43.8

Overall 36.6

FIG. 3—The first two discriminant axes, by social group.



244 Cranial Variation in Antioquia, Colombia

Colombian sample classifies moderately well (84.7%) with 
eight individuals misclassifying as Mexican and two indi-
viduals misclassifying as Guatemalan, the two other His-
panic reference samples (Table 9).

The Mahalanobis distances among the refined sample 
are presented in Table 10. The D2 values for the Colombian 
group was closest to American Whites and Mexicans, fol-
lowed by the Prehistoric Native Americans and Guatema-
lans, then the Mexican and Asian and American Black 
samples (Figure 8).

FIG. 4—Unrooted dendrograms projecting the Mahalanobis distances.

TABLE 6—Mahalanobis distance (D2) matrix on birthplace.

Nordeste Norte Occidente Oriente Suroeste Uraba

Nordeste —
Norte 12.41 —
Occidente 12.67 6.39 —
Oriente 9.61 8.21 9.21 —
Suroeste 10.96 8.99 9.54 7.52 —
Uraba 19.70 14.32 13.32 14.52 10.94 —
Valle de Aburra 11.18 10.07 10.52 9.81 7.02 12.30

TABLE 7—Mahalanobis distance (D2) matrix 
on social group.

A B C D

A —
B 7.62 —
C 13.84 9.78 —
D 7.27 7.87 12.41 —
E 9.26 7.37 11.18 10.07

FIG. 5—Unrooted dendrograms projecting the Mahalanobis distances; 
see Table 2 for relationship between social group and birthplace.

Discussion

Colombian Cranial Variation

To understand cranial variation and its relationship to social 
and geographical categories in Antioquia, several statistical 
models explored the interaction of cranial morphology to a 
priori group labels. The models did not perform well when 
birthplace or social group were used to ‘identify’ potential 
subsamples within the Antioquian sample. Heavy overlap of 
all groups in the LDFA suggest relative homogeneity across 
the groups in Antioquia. Interestingly, when plotting birth-
place (see Figure 2), the ellipses for the Uraba and Nordeste 
cohorts exhibit the least amount of overlap compared to the 
other groups. This may indicate differences in cranial shape 
that make Uraba and Nordeste more unique, but the small 
sample size (n = 6) and the inclusion of outliers may play a 



Kamnikar et al. 245

role in these distributions. Socially defined category ‘B’ (Ori-
ente and Suroeste) and ‘E’ (Valle de Aburra), are geograph-
ically proximate and exhibit heavy overlap in the LDFA plot 
and are likewise similarly positioned in the D2 plot (see 
Figure 5). Social groups ‘A’ (Uraba and Occidente) and ‘D’ 
(Norte) did not exhibit much overlap in the LDFA, despite 
geographical proximity; however, these two groups were 
more similar on the D2 plot (see Figure 5). Meaningful pat-
terns did not emerge to relate craniometric variables to birth-
place or social group categories. Additional study with a larger 
sample size could clarify or support our results.

Classifying individuals by birthplace or a peer- perceived 
social category using cranial measurements does not work 

well in Antioquia. While this system can be used with some 
success in places like the U.S. and South Africa (Stull et al. 
2014) where social race is used by the legal and law enforce-
ment community as a ‘culturally constructed labeling system’ 
(Sauer 1992:109), it cannot be applied in Colombia, despite 
distinguishing individuals by social group or ethnicity (Mon-
salve & Hefner 2016). Using MMS data, Monsalve and Hef-
ner (2016) found no significant differences in trait expression 
by birthplace, but more intraregional variation (46.0%) could 
be explained with MMS trait data compared to the cranio-
metric data (26.0%) used in this study. Monsalve and Hefner 
(2016) identified three main clusters (Occidente- A, Sur-
oeste- B, and Uraba- A; Valle de Aburra- E and Nordeste- C; 

FIG. 6—The first two discriminant axes for the six- group analysis.

TABLE 8—Classification rate of Colombian and comparative groups (LOOCV).

American Black American White Asian Colombian Hispanic

Prehistoric 
Native 
American CCR (%)

American Black 142 14 3 6 22 10 72.1
American White 10 148 3 16 11 6 76.3
Asian 4 5 113 2 24 11 71.0
Colombian 7 19 2 205 8 2 84.4
Hispanic 9 9 11 2 45 14 50.0
Prehistoric Native American 15 9 31 1 20 251 76.7

Overall: 74.7%



246 Cranial Variation in Antioquia, Colombia

FIG. 7—The first two discriminant axes for the seven- group analysis.

TABLE 9—Classification rate of Colombian and comparative groups; refining the Hispanic group (LOOCV).

American Black American White Asian Colombian Guatemalan Mexican

Prehistoric 
Native 
American CCR (%)

American Black 139 14 3 6 9 18 8 70.1
American White 10 145 3 16 4 9 6 75.1
Asian 3 5 110 1 10 20 10 69.2
Colombian 6 18 2 206 2 8 1 84.7
Guatemalan 0 1 1 0 5 3 3 38.5
Mexican 10 8 6 2 13 31 9 39.2
Prehistoric Native American 15 10 27 0 33 13 229 70.0

Overall: 71.4%

TABLE 10—Mahalanobis distance (D2) matrix on the refined sample.

American Black American White Asian Colombian Guatemalan Mexican

American Black —
American White 11.64 —
Asian 14.92 13.85 —
Colombian 15.85 13.56 15.34 —
Guatemalan 16.34 17.17 13.60 15.12 —
Mexican 11.65 11.78 7.82 13.86 10.62 —
Prehistoric Native American 13.26 15.27 12.41 14.80 10.65 9.52



Kamnikar et al. 247

and Oriente- B and Norte- D) based on mid- facial and vault 
characteristics. These clusters are not present within our 
results, indicating craniometric variation is not measurably 
different across groups or adequately captured by the ILDs 
used in this study. Further attempts to classify based on 
socially defined groups did not improve classifications with 
MMS data, which is similar to results reported here.

Craniometric analysis of the Colombian sample as a sin-
gle dataset produced promising results. While other research-
ers noted heavy overlap with Hispanic craniometric data 
(namely Mexican and Guatemalan samples) (Dudzik & Jantz 
2016; Hughes et al. 2019), the Colombian sample, e.g., “His-
panic”, did not overlap or misclassify considerably with these 
groups. This pattern was evident in the LDFA classification 
models, where Colombians most frequently misclassified as 
American Whites. Despite these results, classification accu-
racies were high for all groups except Mexican and Guate-
malan samples. D2 scores for the Colombian sample were 
closest to the American White and Mexican samples, followed 
by the Prehistoric Native American, Guatemala, Asian, then 
American Black samples. Proximity to and misclassifications 
in the American White sample suggest a significant European 
ancestral component in Antioquia. Morphological and genetic 
research in Colombia suggests a diverse population structure 
corresponding to African, European, Mesoamerican, and 
South American origins (Bryc et al. 2010; Lopez et al. 2012). 

FIG. 8—Unrooted dendrograms projecting the Mahalanobis distances 
(Pre NA = Prehistoric Native American; Colom = Colombian; Black = 
American Black; White = American White; Guate = Guatemalan).

In Antioquia, geographical barriers, most notably the Andes 
mountain range, have contributed to population isolation and 
development of the paisa, a distinct regional identity based 
on socio- political processes and ancestral origins. The paisa 
includes a racial hierarchical component (Posada 2003), cor-
responding to ‘whiteness’ or ‘white’ ethnic groups (Álvarez 
1996). Genetic studies in Antioquia demonstrate a significant 
European component, with smaller contributions from Afri-
can and Native American groups (Bravo et al. 1996; Carvajal- 
Carmona et al. 2000; Sandoval et al. 1993). A significant 
genomic contribution from a male, Spanish founder popula-
tion, persists in the Y- chromosome of individuals in the region 
today (Carvajal- Carmona et al. 2000). Interestingly, the 
majority of mitochondrial DNA contributions came from four 
Amerindian founder linages (Carvajal- Carmona et al. 2000).

Because craniometric data is highly heritable (Adhikari 
et al. 2016; Relethford & Harpending 1994; Roseman & 
Weaver 2004; Šešelj et al. 2015), and genetic structure in the 
Antioquian region indicates a large European contribution 
through migration (Bravo et al. 1996; Carvajal- Carmona et al. 
2000), it is not surprising that the cranial phenotype is com-
parable to the American White sample of the groups tested. 
Both populations are genetically descended from European 
populations (Bryc et al. 2010, 2015), and therefore, are more 
similar to each other than other groups used in this analysis. 
Migration and restricted gene flow in Antioquia could explain 
the separation between the Colombian sample and the Asian- 
derived samples (Asian, Guatemalan, Mexican and prehis-
toric Native American). Genetic studies identify a mix of 
African, European, and Native American genetic contribu-
tions to populations in Antioquia (Bryc et al. 2010; Wang et al. 
2008), which could explain these results.

Comparing our results to Monsalve and Hefner’s (2016) 
conclusions, we found that craniometric data has a higher 
classification accuracy (74.7%: pooled Hispanic data; and 
71.4%: separate Hispanic groups) when compared with results 
derived with Artificial Neural Network analysis used on 
cranial MMS traits (48.0%). In their study, Colombians mis-
classified as American Black (10.0%), American White 
(16.6%), and Hispanic (10.0%) individuals (Monsalve & Hef-
ner 2016). Our craniometric analysis shows less frequent mis-
classification for the same groups: American Black (2.5%), 
American White (7.4%), Asian (0.8%), prehistoric Native 
American (0.4%), and the two Hispanic groups: Mexican 
(3.2%) and Guatemalan (0.8%). Further comparison of the 
spatial distribution using the two approaches illustrates sim-
ilar patterning. The PCA using cranial MMS data identified 
four clusters: 1) Hispanic and American White, 2) Pacific 
Islander and Asian, 3) American Black, and 4) Colombian. 
The D2 results from this metric- based study identified three 
clusters: 1) Asian and Mexican, 2) Guatemalan and prehis-
toric Native American 3) American Black, American White, 
and Colombian. Again, the separation of the Colombian sample 



248 Cranial Variation in Antioquia, Colombia

from the Hispanic samples in both analyses bolsters calls for 
the refinement of this category in population affinity estima-
tion. We recommend that Colombians from Antioquia not be 
included under the broad heading Hispanic, but as a sepa-
rate sample in comparative analysis.

Implications for Future Work in the Region

There are several implications for future forensic work in the 
Americas and within Colombia. While we could not distin-
guish individuals by birthplace or social group using cranio-
metric data, cranial variation is not necessarily homogeneous 
in Colombia. These results warrant additional testing of the 
Antioquian sample against other regions in the country. For 
example, mountainous and costal populations may differ from 
each other, which may be true for individuals born in cities 
like Barranquilla or Cartagena compared with Bogotá or Cali. 
Identification of larger, intraregional variation within Colom-
bia may have direct implications for current migration events 
from Venezuela (Faiola 2018; Miami Herald 2018), potentially 
providing an option for estimating geographic origin in the 
region. Additionally, regional studies in Colombia may pro-
vide useful for identification efforts for unidentified individ-
uals from Colombia’s armed conflict, which potentially 
number into the 100,000s (Fondebrider 2016). We hope our 
results stimulate this type of research in the region and other 
countries, especially those involved in mass migration events, 
as a possible avenue for estimating geographic origin.

All three Hispanic samples in this study (Mexico, Gua-
temala, and Colombia) are grouped under one classificatory 
group within traditional population affinity estimation mod-
els. However, the level of separation between the Colombian 
sample from the Mexican and Guatemalan samples war-
rants a rethinking of population affinity estimation for this 
group. Our results support previous findings by Spradley 
et al. (Figueroa- Soto & Spradley 2012; Spradley 2014b; 
Tise et al. 2014) and Ross et al. (Ross et al. 2014; Humphries 
et al. 2015) suggesting populations in Latin America, while 
geographically proximate, show considerable cranial varia-
tion. We suggest practitioners consider the origin of His-
panic samples and the use of pooled data in three-  or 
five- group models. Hispanic samples, such as the subsample 
from the FDB, comprise individuals from neighboring 
countries like Costa Rica, Cuba, El Salvador, Guatemala, 
Mexico, Puerto Rico, and Panamá (Anderson 2008). Differ-
ences among these groups are certainly apparent at a finer 
level of refinement since each was subjected to a unique 
historical migration and evolutionary events.

While the debate over appropriate and necessary levels 
of population affinity refinement for forensic casework is 
ongoing, this research bolsters support toward a more refined 
approach. The addition of this regional sample to modern, 
reference datasets allows for population specific models for 

Colombians (Hefner & Spradley 2018; Spradley 2016). 
Future research using this and other diverse samples from 
Latin America is poised to address current issues in foren-
sic anthropology research and practice.

Conclusion

This study aimed to assess if craniometric analysis can be 
used to identify intraregional variation within Antioquia, 
Medellín, Colombia. While we failed to demonstrate sepa-
ration based on birthplace or social labels, the pooled Colom-
bian sample separated well from other comparative groups, 
including other populations traditionally classified under the 
term Hispanic. This study demonstrates heterogeneity within 
Latin American populations, offering further support to the 
call for refinement of the Hispanic category.

Following the suggestions outlined in Hefner and Sprad-
ley (2018), we advocate a broad- level analysis, followed by 
further refinement, especially within the Hispanic group des-
ignation. As more data are collected on this diverse group and 
incorporated into reference databanks, researchers will iden-
tify patterned differences and nuances within those samples. 
Understanding and investigating these patterns could have 
profound implications for diasporic events in the Americas 
and contribute to identification in forensic casework in Latin 
America and the U.S.

In this study, the regional Antioquian sample was most 
useful when grouped together as a single Colombian dataset, 
and compared to other Hispanic and non- Hispanic groups 
encountered in forensic casework. This approach could be 
useful when incorporating reference data for new Hispanic 
populations. Further studies should incorporate cranial MMS 
and craniometric data together, as previous research demon-
strates more accurate classification results when combined 
(Hefner et al. 2014).

References

Adhikari K, Fuentes- Guajardo M, Quinto- Sánchez M, Mendoza- 
Revilla J, Chacón- Duque JC, Acuña- Alonzo V, et al. A 
genome- wide association scan implicates DCHS2, RUNX2, 
GLI3, PAX1 and EDAR in human facial variation. Nature 
Communications 2016;7:11616.

Algee- Hewitt BFB. Population inference from contemporary Amer-
ican craniometrics. American Journal of Physical Anthropol-
ogy 2016;160(4):604– 624.

Algee- Hewitt BFB. Temporal trends in craniometric estimates of 
admixture for a modern American sample. American Journal 
of Physical Anthropology 2017;163(4):729– 740.

Álvarez V. Poblamiento y población en el valle de Aburrá y Medellín, 
1541– 1951. In: Melo JO, ed. Historia de Medellín. Medellín: 
Suramericana;1996:57– 84.

Anderson BE. Identifying the dead: Methods utilized by the 
Pima County (Arizona) Office of the Medical Examiner for 



Kamnikar et al. 249

undocumented border crossers, 2001– 2006. Journal of Foren-
sic Sciences 2008;53(1):8– 15.

Bravo ML, Valenzuela CY, Arcos- Burgos OM. Polymorphisms and 
phyletic relationships of the Paisa community from Antioquia 
(Colombia). Gene Geography: A Computerized Bulletin on 
Human Gene Frequencies 1996;10(1):11– 17.

Bryc K, Durand EY, Macpherson M, Reich D, Mountain JL. The 
genetic ancestry of African Americans, Latinos, and European 
Americans across the United States. The American Journal of 
Human Genetics 2015;96:37– 53.

Bryc K, Velez C, Karafet T, Moreno- Estrada A, Reynolds A, Auton 
A, Hammer M, Bustamante CD, Ostrer H. Genome- wide pat-
terns of population structure and admixture among Hispanic/
Latino populations. PNAS 2010;107(Supplement 2):8954– 8961.

Carvajal D. As Colombia emerges from decades of war, migration 
challenges mount. Migration Policy Institute. 2017. https:// 
www . migrationpolicy . org / article / colombia - emerges - decades 
- war - migration - challenges - mount. Accessed January 8, 2020.

Carvajal- Carmona LG, Soto ID, Pineda N, Ortíz- Barrientos D, 
Duque C, Ospina- Duque J, et al. Strong Amerind/White sex 
bias and possible Sephardic contribution among the founders 
of a population in Northwest Colombia. American Journal of 
Human Genetics 2000;67:1287– 1295.

Dudzik B, Jantz RL. Misclassifications of Hispanics using Fordisc 
3.1: Comparing cranial morphology in Asian and Hispanic pop-
ulations. Journal of Forensic Sciences 2016;61(5):1311– 1318.

Dunn RR, Spiros MC, Kamnikar KR, Plemons AM, Hefner JT. 
Ancestry estimation in forensic anthropology: A review. 
WIREs Forensic Science; 2020. doi: https:// onlinelibrary . wiley 
. com / doi / 10 . 1002 / wfs2 . 1369.

Faiola A. From riches to rags: Venezuelans become Latin America’s 
new underclass. Washington Post. July 27, 2018. https:// www 
. washingtonpost . com / news / world / wp / 2018 / 07 / 27 / feature / as 
- venezuela - crumbles - its - fleeing - citizens - are - beco ming - latin 
- americas - new - underclass /  ? noredirect=on & utm _term= 
. 103b2782766f.

Figueroa- Soto C, Spradley MK. Cranial variation among three 
regional groups in Mexico. In: Proceedings of the 82nd Annual 
Meeting of the American Association of Physical Anthropol-
ogists, April 9– 13, 2013; Knoxville, TN.

Fondebrider L. The application of forensic anthropology to the 
investigation of cases of political violence: Perspectives from 
South America. In: Blau S, Ubelaker DH, eds. Handbook of 
Forensic Anthropology and Archaeology. 2nd ed. New York: 
Routledge; 2016:65– 74.

Hefner JT. The macromorphoscopic databank. American Journal 
of Physical Anthropology 2018;166(4):994- 1004.

Hefner JT, Pilloud MA, Buikstra JE, Vogelsberg CCM. A brief his-
tory of biological distance analysis. In: Pilloud MA, Hefner 
JT, eds. Biological Distance Analysis: Forensic and Bio archae-
ol o gi cal Perspectives. New York: Elsevier; 2016:3– 22.

Hefner JT, Spradley MK. Ancestry (forensic applications). In: Tre-
vathan W, ed. The International Encyclopedia of Biological 
Anthropology. Hoboken, NJ: John Wiley & Sons, Inc.; 2018: 
1– 3.

Hefner JT, Spradley MK, Anderson BE. Ancestry assessment 
using random forest modeling. Journal of Forensic Sciences 
2014;59(3):583– 589.

Howells WW. Cranial Variation in Man: A Study by Multivariate 
Analysis of Patterns of Differences among Recent Human Pop-
ulations. Cambridge, MA: The Museum; 1973.

Hughes CE, Dudzik B, Algee- Hewitt BFB, Jones A, Anderson BE. 
Undertanding (mis)classification trends of Latin Americans in 
Fordisc 3.1: Incorporating cranial morphology, microgeo-
graphic origin, and admixture proportions for interpretation. 
Journal of Forensic Sciences 2019;64(2):353– 366.

Humphries AL, Maxwell AB, Ross AH, Ubelaker DH. A geomet-
ric morphometric study of regional craniofacial variation 
in Mexico. International Journal of Osteoarchaeology 
2015;25(6):795– 804.

Jantz RL, Moore- Jansen PH. A database for forensic anthropology: 
Structure, content, and analysis. Knoxville: Department of 
Anthropology, University of Tennessee; 1988.

Jantz RL, Ousley SD. Fordisc 3.0: Personal computer forensic dis-
criminant functions. Knoxville: University of Tennessee; 2005.

Langley NR, Meadows Jantz L, Ousley SD, Jantz RL. Data collec-
tion procedures for forensic skeletal material 2.0. Knoxville: 
University of Tennessee; 2016

López, G. Hispanics of Colombian Origin in the United States, 2013. 
Pew Research Center. September 15, 2015. https:// www 
. pewresearch . org / hispanic / 2015 / 09 / 15 / hispanics - of - colombian 
- origin - in - the - united - states - 2013 / . Accessed January 8, 2020.

Lopez MA, Casallas DA, Castellanos D, Soto FV, Bethard JD. 
Unveiling ancestry in Colombia through morphometric anal-
ysis. In: Proceedings of the 64th Annual Meeting of the Amer-
ican Academy of Forensic Sciences, February 20– 25, 2012; 
Atlanta, GA.

McFerrin L. Statistical analysis tools for high dimension molecu-
lar data (HDMD). Created 2013. https:// www . rdocumentation 
. org / packages / HDMD / versions / 1 . 2. Accessed January 14, 2020.

Monsalve T. Hefner JT. Macromorphoscopic trait expression in a 
cranial sample from Medellín, Colombia. Forensic Science 
International 2016;266:574.e1– 574.e8.

Posada C. La raza negra en Colombia: Antioqueños y chocoanos. 
Estudios de Literatura Colombiana 2003;12.

R Core Team. R: A language and environment for statistical com-
puting. R Foundation for Statistical Computing, Vienna, 
Austria, 2018. http:www . R - project . org / . Accessed October 14, 
2018.

Relethford JH. Craniometric variation among modern human 
populations. American Journal of Physical Anthropology 
1994;95(1): 53– 62.

Relethford JH. Population- specific deviations of global human cra-
niometrics variation from a neutral model. American Journal 
of Physical Anthropology 2010;142(1):105– 111.

Relethford JH, Harpending HC. Craniometric variation, genetic the-
ory, and modern human origins. American Journal of Physi-
cal Anthropology 1994;95(3):249– 270.

Roseman CC, Weaver TD. Multivariate apportionment of global 
human craniumetric diversity. American Journal of Physical 
Anthropology 2004;125(3):257– 263.

Ross AH, Slice DE, Ubelaker DH. Population affinities of Hispanic 
crania. In: Berg GD, Ta’ala SC, eds. Biological Affinity in 
Forensic Identification of Human Skeletal Remains: Beyond 
Black and White. Boca Raton, FL: CRC Press; 2014:155– 164.

Ross AH, Slice D, Ubelaker DH, Falsetti AB. Population affinities 
of 19th century Cuban crania: Implications for identification 
criteria in South Florida Cuban Americans. Journal of Foren-
sic Sciences 2004;49:11– 16.

Sandoval C, de la Hoz A, Yunis E. Estructura genética de la 
población Colombiana. Revista de la Facultad de Medicina– 
Universidad Nacional de Colombia 1993;41(1):3– 14.

Sauer NJ. Forensic anthropology and the concept of race: If races 
don’t exist, why are forensic anthropologists so good at 
 identifying them? Journal of Social Science and Medicine 
1992;34:107– 111.

Šešelj M, Duren DL, Sherwood RJ. Heritability of the human cranio-
facial complex. The Anatomical Record 2015;298: 1535– 1547.

Spradley MK. Metric ancestry estimation from the postcranial 
skeleton. In: Berg GE, Ta’ala SC, eds. Biological Affinity in 
Forensic Identification of Human Skeletal Remains: Beyond 
Black and White. Boca Raton, FL: CRC Press; 2014a: 83– 94.



250 Cranial Variation in Antioquia, Colombia

Spradley MK. Metric methods for the biological profile in forensic 
anthropoology: Sex, ancestry, and stature. Academic Foren-
sic Pathology 2016;6(3):391– 399.

Spradley MK. Toward estimating geographic origin of migrant 
remains along the United States– Mexico border. Annals of 
Anthropological Practice 2014b;38:101– 110.

Stull KE, Kenyhercz MW, L’Abbe EN. Ancestry estimation in South 
Africa using craniometrics and geometric morphometrics. 
Forensic Science International 2014;245:206- e1.

Tise ME, Kimmerle EH, Spradley MK. Craniometric variation of 
diverse populations in Florida: Identification challenges within 

a border state. Annals of Anthropological Practice 2014;38: 
111– 123.

van Buuren S, Groothuis- Oudshoorn K. Multiple imputation by 
chained equations (MICE) v.2.5; 2011.

Wang S, Ray N, Rojas W, Parra MV, Bedoya G, Gallo C, et al. Geo-
graphic patterns of genome admixture in Latin American mes-
tizos. PLoS Genetics 2008;4(3):e1000037.

Wrobel G, Hair A, Biggs J. Introducing the Maya Crania Photo-
grammetry Initiative. In: Proceedings from the 16th Annual 
Meeting of the Belize Archaeology Symposium, June 2018; 
San Ignacio, Belize.


