See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/329674731

Bone Mineral Density Adult Age Estimation in Forensic Anthropology: A Test

of the DXAGE Application

Article  in  Journal of Forensic Sciences · December 2018

DOI: 10.1111/1556-4029.13987

CITATION READS

1 170

4 authors, including:

Jonathan Bethard Jacqueline Berger

University of South Florida University of South Florida
29 PUBLICATIONS   151 CITATIONS    3 PUBLICATIONS   3 CITATIONS   

SEE PROFILE SEE PROFILE

Ann H Ross
North Carolina State University
89 PUBLICATIONS   994 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Medieval Crisis Populations (Transylvania, Romania) - Adult and Juvenile Osteology Research View project

Carnivore Scavenging in the Southeast: Implications for Estimating Time-Since-Death View project

All content following this page was uploaded by Jonathan Bethard on 02 January 2019.

The user has requested enhancement of the downloaded file.

 J Forensic Sci, 2018
doi: 10.1111/1556-4029.13987
TECHNICAL NOTE Available online at: onlinelibrary.wiley.com

ANTHROPOLOGY

Jonathan D. Bethard ,1 Ph.D.; Jacqueline M. Berger ,1 M.S.; Justin Maiers,1 M.S.; and
Ann H. Ross ,2 Ph.D.

Bone Mineral Density Adult Age Estimation in

Forensic Anthropology: A Test of the DXAGE
Application*

ABSTRACT: Estimating age-at-death of individuals represented only by skeletonized human remains is a fundamental aspect of forensic
anthropological casework. Recently, several researchers have proposed that bone mineral density (BMD) is a useful predictor of age-at-death in
forensic contexts. Navega et al. (JFS 63(2):497–503) developed an online application called DXAGE for calculating age-at-death from BMD
parameters. This study tests the utility of DXAGE by utilizing data from the National Health and Nutrition Examination Survey (NHANES).
BMD data from a female subsample (n = 470) of the NHANES 2007–2008 dataset were analyzed, and the relationship between predicted age
and real age was examined. Inaccuracy was 14.25 years, and bias was
7.20 years. Results show that there is a weak correlation between pre-
dicted and actual age (r = 0.47) using the DXAGE application. While BMD data are potentially useful for predicting age age-at-death, the
DXAGE application should be used cautiously in forensic anthropological contexts.

KEYWORDS: forensic sciences, forensic anthropology, age-at-death estimation, bone mineral density, DXAGE, NHANES

Forensic anthropologists are limited in the ability to estimate
age-at-death (AAD) in skeletal remains of adult individuals, and in
particular, those over the age of fifty (1–3). Most of this limitation
has to do with the intrinsic nature of bone as a tissue itself, in addi-
tion to the many extrinsic variables surrounding the process of
human aging. A host of variables including but not limited to vari-
ation in habitual activity levels, the presence or absence of patho-
logical conditions such as osteoporosis, body mass, and parturition
are thought to impact the process of adult aging (4–10). In addi-
tion, age-at-death methodologies introduce error to the process
due to the biased nature of skeletal reference populations and the
statistical approaches used to generate age estimates (4, 11).
Although numerous published attempts to control the variables
associated with age-at-death estimation have been undertaken, it
is generally agreed upon that generating age-at-death estimates
for individuals over fifty years of age is tenuous at best and can-
not be performed with high precision or accuracy (12–14).
Given that approximately 34% of individuals in the United
States are aged 50 and older (www.census.gov) (15), the ina-
bility for forensic anthropologists to accurately and precisely
generate age-at-death estimates of individuals in this age coh-
ort is problematic. In addition, despite numerous approaches
1
Department of Anthropology, University of South Florida, 4202 East
Fowler Avenue, SOC107, Tampa, FL 33620.
2
Department of Biological Sciences, North Carolina State University, 100
Eugene Brooks Avenue, Raleigh, NC 27695.
Corresponding author: Jonathan D. Bethard, Ph.D.
E-mail: jbethard@usf.edu
*Funded by the National Institute of Justice (2018-DU-BX-0197).
Received 8 Oct. 2018; and in revised form 30 Nov. 2018; accepted 3 Dec.
2018.
generating age-at-death estimates, Garvin and Passalacqua (16)
have demonstrated that practicing forensic anthropologists pref-
erentially utilize few age-at-death estimation methodologies, the
majority of which involve macroscopic observation of a few
skeletal regions (e.g., pubic symphysis, auricular surface, ster-
nal rib ends). When these regions of the skeleton are missing
or damaged due to postmortem processes, estimating age-at-
death of adult individuals becomes increasingly difficult and
problematic.
Recently, several research teams have proposed that bone min-
eral density (BMD) is useful for estimating age-at-death in
forensic contexts (17–19). For many years, researchers working
outside medicolegal contexts have documented the decline of
BMD with increasing age (20–24). According to Navega et al.
(19), the process of aging affects bone health with both direct
and indirect effects on bone mass. Moreover, Navega et al. (19)
summarize the biology of age-related bone loss and indicate
numerous etiologies for decreased BMD. Among these, the
decrease in calcitriol (the molecule responsible for the absorption
of dietary calcium in the gut), secondary hyperparathyroidism,
and the reduction in osteocyte density have all been attributed to
the relationship between advanced age and reduced BMD.
Castillo and Ruiz (18) were the first to demonstrate that BMD
could be used as a predictor of age-at-death in forensic contexts,
utilizing a sample of 70 individuals drawn from a clinical hospi-
tal context in Spain. Individuals in the study sample were
patients who required BMD scanning via dual X-ray absorptiom-
etry (DXA), from which the researchers obtained significant cor-
relations between BMD and age. Castillo and Ruiz (18)
concluded that BMD, particularly in the region of interest (ROI)
of the Ward’s triangle of the proximal femur, was a useful
parameter for age-at-death estimation.

© 2018 American Academy of Forensic Sciences 1

2 JOURNAL OF FORENSIC SCIENCES

Following the work of Castillo and Ruiz (18), Navega et al. TABLE 1––Total BMD values by age cohort.
(19) developed the framework that forms the basis of the present
validation study. In brief, Navega et al. analyzed femoral BMD BMD Total BMD Total BMD Total BMD Total
Age N Min Max Mean S.D
for 100 females aged 21–95 years, drawn from the Coimbra
Identified Skeletal Collection in Coimbra, Portugal. The sample 20–29 73 0.577 1.1319 0.97838 0.146702
was subdivided into seven age categories, which consisted of 30–39 80 0.667 1.225 0.94191 0.126764
40–49 87 0.670 1.190 0.96205 0.114294
either 14 or 15 individuals per category. The analysis was con- 50–59 79 0.600 1.165 0.88253 0.133035
ducted using a Hologic QDR 4500C Elite densitometer for three 60–69 89 0.605 1.202 0.87600 0.141262
common ROI’s: total femoral BMD, Ward’s triangle BMD, and 70–79 41 0.644 1.152 0.82585 0.132460
femoral neck BMD. Navega et al. applied artificial neural net- 80+ 21 0.492 0.992 0.74071 0.141157
works (ANN) to the Portuguese dataset. Often described as a Total 470
“black box” approach, ANN represent machine-learning tech-
nique best suited for examining the relationship of variables that TABLE 2––Neck BMD values by age cohort.
are not well understood or discerning complex patterns that are
difficult to identify using conventional methods (25). BMD Neck BMD Neck BMD Neck BMD Neck
Navega et al. created a user-friendly, web-based application Age N Min Max Mean S.D
called DXAGE (http://osteomics.com/DXAGE/). DXAGE per- 20–29 73 0.430 1.246 0.90567 0.130169
mits a practitioner to generate age-at-death estimates from one or 30–39 80 0.544 1.1218 0.84759 0.130169
more BMD variables. A predicted age, as well as a minimum 40–49 87 0.551 1.189 0.84593 0.11067
50–59 79 0.515 1.059 0.76872 0.129471
and maximum age-at-death estimate, is also calculated by the 60–69 89 0.518 1.127 0.76015 0.130041
application. The purpose of the present study was to test the 70–79 41 0.496 0.952 0.71334 0.113342
DXAGE application developed by Navega et al. (19) on a docu- 80+ 21 0.340 0.837 0.6343 0.120040
mented BMD dataset from the United States. Total 470

Materials and Methods

TABLE 3––Ward’s triangle BMD by age cohort.
Available femoral BMD data from the National Health and
Nutrition Examination Survey (NHANES) were utilized in this BMD Ward BMD Ward BMD Ward BMD Ward
study. NHANES BMD data from the femur were collected in Age N Min Max Mean S.D
2007–2008 and are freely available from the following website:
20–29 73 0.299 1.344 0.86751 0.175607
https://wwwn.cdc.gov/nchs/nhanes/2007-2008/DXXFEM_E.htm. 30–39 80 0.448 1.143 0.76099 0.146173
Additionally, the demographic parameters for each individual are 40–49 87 0.394 0.985 0.70860 0.127966
also available here: https://wwwn.cdc.gov/Nchs/Nhanes/2007- 50–59 79 0.267 0.930 0.60851 0.153155
2008/DEMO_E.htm. The samples have been widely utilized for 60–69 89 0.329 1.021 0.57248 0.144078
70–79 41 0.276 0.777 0.50729 0.122003
studies related to osteoporosis and other clinical conditions, and 80+ 21 0.119 0.657 0.41038 0.131169
were collected with a Hologic bone densitometer (26–28). It Total 470
must be noted that BMD acquisition and comparisons are manu-
facturer specific and comparisons of normative data should only
be compared to those acquired by the same manufacturer (29).
A randomly selected subsample composed of available female NHANES BMD data, age was calculated in DXAGE and these
BMD data from the 2007–2008 NHANES femoral dataset was predicted ages were compared to the known or actual ages from
utilized in this study (n = 470). The age range of individuals the NHANES dataset. The mean difference between predicted
included in the NHANES subsample was 20–80 years, with a and actual ages was assessed with the Matched Pair platform in
mean age of 49.15 years (SD = 16.77). Importantly, all individ- the statistical package JMP 13.1, which compares row-by-row
uals in the NHANES dataset aged 80 and over are top-coded at differences between two response columns (i.e., predicted age
80 years of age (30). and actual age) using a paired t-test. In addition, comparisons
Tables 1-3 present the number of individuals in each decade within age decades were made to explore those with the greatest
cohort and descriptive statistics of the three NHANES BMD difference. Predicted ages derived using Navega et al. “black
parameters utilized by Navega et al. (19): total femoral BMD, box” method were also compared to predicted ages attained
femoral neck BMD, and Ward’s triangle BMD. The DXAGE using the linear regression equation by Paschall and Ross which
application was utilized to calculate AAD estimates. When using was derived from a forensic sample (17):
the DXAGE application for the present study, alpha was set to
0.05 and the “Fast” method was selected. The predicted age pro- Age ¼ 82:1
ð39:9  femoral neck BMDÞ:
vided by the DXAGE application was used to calculate inaccu-
racy and bias following Merritt (8). Inaccuracy was calculated as Importantly, Paschall and Ross also utilized a Hologic
the average absolute error of the predicted age, disregarding QDR4500, the same manufacturer densitometer used by Navega
over- or underestimation: Σ|(predicted age – actual age)|/n. Bias et al. (19) and NHANES.
was calculated as the mean over- or underestimation of age:
Σ(predicted age – actual age)/n. In addition, the percentage of Results
individuals falling within the age range generated by DXAGE
(e.g., Min Age and Max Age) was calculated. Inaccuracy was 14.25 years, and bias was
7.20 years. In
A Pearson product-moment coefficient was used to examine addition, 78% of individuals (n = 367) fell within the age range
the relationship between the BMD variables and age. Using the calculated by the DXAGE software package. Age ranges varied

from a minimum span of 5 years and a maximum span of
67 years. The average age range calculated by DXAGE was
38.3 years. The Pearson’s product-moment correlation coefficient
showed the strongest correlation between age and Ward’s trian-
gle BMD (r =
0.651) with moderate correlations with femoral
neck (r =
0.468) and total femoral BMD (r =
0.385), respec-
tively. Results show that there is a weak correlation between
predicted and actual age (r = 0.47) using Navega et al.’s appli-
cation. Results also show that there is a significant difference
between predicted age and actual age in the overall model
(Prob > │t│ < 0.0001; mean difference F ratio = 34.82) and
show that on average DXAGE under-ages individuals by 7 years
(mean difference =
7.2 years). In the youngest age category
(20–29 years), the mean difference is 7.47 years. DXAGE on
average over-ages members of this younger age cohort by
7.5 years. The age category with the least mean difference is the
30–39 age-group (mean difference = 0.65 years). DXAGE
under-ages individuals in the remaining age categories: by
5.3 years in the 40–49 age-group; by 9.7 years in the 50–59
age-group; by 15.45 years in the 60–69 age-group; by
24.44 years in the 70–79 age-group; and by 23.1 years in the
80+ age-group.
Predicted ages derived using the Paschall and Ross (17)
femoral neck BMD equation were then compared to actual ages,
and to Navega et al. predicted ages. No significant difference
between predicted age using linear regression and actual age in
the overall model was noted (Prob > │t│0.244; mean differ-
ence F ratio = 408.9). On average, linear regression under-ages
individuals by <1 year (mean difference =
0.804 years). In the
youngest age category (20–29 years), the mean difference is
21.15 years. In other words, DXAGE on average over-ages
members of this younger age cohort by 21 years. The age cate-
gory with the least mean difference is the 50–59 age-group
(mean difference =
2.329 years). The linear regression equa-
tion over-aged individuals by 21.153 years in the 20–29 age-
group; by 13.691 years in the 30–39 age-group; by 3.2 years in
the 40–49 age-group; and under-aged individuals by 11.74 years
in the 60.69 age-group; by 19.88 years in the 70–79 age-group
age; and by 23.24 years in the 80+ age-group.
Discussion
Linear regression has a tendency to over-age younger individ-
uals and under-age older individuals, a bias noted by other
researchers and is supported by the results of this study (31).
This study confirms the well-known relationship between
increasing age and decreasing BMD, especially in the Ward’s
triangle of the proximal femur. In addition, the NHANES data
utilized in this validation of the DXAGE application supports
what other researchers have documented in clinical contexts.
Notably, some individuals do not always attain peak BMD until
the third and fourth decade of life (31, 32). This reality is some-
what problematic for forensic anthropologists who might utilize
the DXAGE application to arrive at an age estimate of an indi-
vidual who is aged between 20 and 40 years, for example. In
addition, given the overall bias and inaccuracy scores generated
in this validation study, as well as the variable error rates associ-
ated with each age cohort, we caution practitioners from utilizing
the DXAGE software for active casework at the present time.
Though 78% of individuals fell within the estimated DXAGE
age range, the broad nature of the age ranges explains this clas-
sification rate.
BETHARD ET AL. . A TEST OF THE DXAGE APPLICATION 3
Currently, DXAGE is not recommended for forensic case-
work; however, the quantitative nature of BMD data eliminates
the problem of observer experience and interobserver subjectiv-
ity that has been documented by numerous researchers (33–35).
Moreover, DXA technology is widely available and BMD data
can be generated rapidly and at low cost (36). General practi-
tioner clinics, hospitals, and some veterinary hospitals routinely
use BMD data and procedures for rapidly obtaining BMD
parameters from isolated skeletal elements have been described
(36). We underscore that instrument manufacturer continuity
between reference and sample data remains compulsory and
should be explicitly described in future research involving BMD
in forensic contexts. In this study, given that Hologic densitome-
ters were utilized to collect all of the BMD parameters herein,
the overall weak correlation between actual and predicted age
may be more of an artifact of the small reference sample built
into the DXAGE software, particularly given the overlapping
values of the BMD total and BMD neck variables across numer-
ous age categories.
While BMD parameters have been utilized frequently in the
clinical literature, they have received little attention for AAD
estimation in forensic anthropological contexts until recently. In
all of the studies investigating BMD and age estimation in foren-
sic anthropology to date, sample sizes are small and do not
encompass sufficient variation among and between age cohorts
to be able to estimate age in a forensic context which has evi-
dentiary ramifications. The NHANES dataset eliminates the issue
of small sample sizes and provides a powerful tool for future
work investigating the link between BMD and age-related
changes. In addition, other large BMD datasets such as the
Korea National Health and Nutrition Examination Survey
(KHANES) are available and could be utilized to examine the
broad applicability of this approach in forensic contexts outside
of the United States (37). Moreover, while taphonomic variables
are known to affect BMD (38), documenting BMD parameters
of the NHANES dataset will enable researchers to better under-
stand how the postmortem interval affects skeletal attrition.
These results also underscore the importance of performing vali-
dation studies on web-based applications for admissibility in
forensic casework. Furthermore, additional studies utilizing
intrinsic bone biology parameters (e.g., BMD) may prove useful
for age estimation in the future.
References
1. Boldsen J, Milner GR, Konigsberg LW, Wood JW. Transition analysis: a
new method for estimating age from skeletons. In: Hoppa RD, Vaupel
JW, editors. Paleodemography: age distributions from skeletal samples.
Cambridge, U.K.: Cambridge University Press, 2002.
2. Brennaman AL, Love KR, Bethard JD, Pokines JT. A Bayesian approach
to age-at-death estimation from osteoarthritis of the shoulder in modern
North Americans. J Forensic Sci 2017;62(3):573–84.
3. Cappella A, Cummaudo M, Arrigoni E, Collini F, Cattaneo C. The issue
of age estimation in a modern skeletal population: are even the more
modern current aging methods satisfactory for the elderly? J Forensic Sci
2017;62(1):12–7.
4. Hoppa RD, Vaupel JW. Paleodemography: age distributions from skele-
tal samples. Cambridge, U.K.: Cambridge University Press, 2002.
5. Rissech C, Appleby J, Cosso A, Reina F, Carrera A, Thomas R. The
influence of bone loss on the three adult age markers of the innominate.
Int J Legal Med 2018;132(1):289–300.
6. Calce SE, Kurki HK, Weston DA, Gould L. Effects of osteoarthritis on
age-at-death estimates from the human pelvis. Am J Phys Anthropol
2018;167(1):3–19.
7. Merritt CE. The influence of body size on adult skeletal age estimation
methods. Am J Phys Anthropol 2015;156(1):35–57.
4 JOURNAL OF FORENSIC SCIENCES
8. Merritt CE. Inaccuracy and bias in adult skeletal age estimation: assess-
ing the reliability of eight methods on individuals of varying body sizes.
Forensic Sci Int 2017;275:315.e1–11.
9. Wescott DJ, Drew JL. Effect of obesity on the reliability of age-at-death
indicators of the pelvis. Am J Phys Anthropol 2015;156(4):595–605.
10. Bongiovanni R. Effects of parturition on pelvic age indicators. J Forensic
Sci 2016;61(4):1034–40.
11. Algee-Hewitt BFB. Age estimation in modern forensic anthropology. In:
Tersigni-Tarrant M, Shirley N, editors. Forensic anthropology: an intro-
duction. Boca Raton, FL: CRC Press, 2013;181–230.
12. Scientific Working Group for Forensic Anthropology (SWGANTH). Age
estimation, 2013[updated 2013]; https://www.nist.gov/sites/default/files/docu
ments/2018/03/13/swganth_age_estimation.pdf (accessed September 29, 2018).
13. Alves-Cardoso F, Assis S. Can osteophytes be used as age at death esti-
mators? Testing correlations in skeletonized human remains with known
age-at-death. Forensic Sci Int 2018;288:59–66.
14. Rouge-Maillart C, Vielle B, Jousset N, Chappard D, Telmon N, Cunha
E. Development of a method to estimate skeletal age at death in adults
using the acetabulum and the auricular surface on a Portuguese popula-
tion. Forensic Sci Int 2009;188(1–3):91–5.
15. United States Census Bureau; https://www.census.gov (accessed October
1, 2018).
16. Garvin HM, Passalacqua NV. Current practices by forensic anthropolo-
gists in adult skeletal age estimation. J Forensic Sci 2012;57(2):427–33.
17. Paschall A, Ross AH. Biological sex variation in bone mineral density in
the cranium and femur. Sci Justice 2018;58(4):287–91.
18. Castillo RF, Ruiz MdCL. Assessment of age and sex by means of DXA
bone densitometry: application in forensic anthropology. Forensic Sci Int
2011;3:53–8.
19. Navega D, Coelho JdO, Cunha E, Curate F. DXAGE: a new method for
age at death estimation based on femoral bone mineral density and artifi-
cial neural networks. J Forensic Sci 2018;63(2):497–503.
20. Dequeker J, Burssens A, Creytens G, Bouillon R. Ageing of bone: its
relation to osteoporosis and osteoarthrosis in post-menopausal women.
In: Lauritzen C, van Keep P, editors. Estrogens in the post-menopause.
Basel, Switzerland: Karger Publishers, 1975;116–30.
21. Riggs B, Wahner H, Dunn W, Mazess R, Offord K, Melton L. Differen-
tial changes in bone mineral density of the appendicular and axial skele-
ton with aging: relationship to spinal osteoporosis. J Clin Invest 1981;67
(2):328–35.
22. Steiger P, Cummings SR, Black DM, Spencer NE, Genant HK. Age-
related decrements in bone mineral density in women over 65. J Bone
Miner Res 1992;7(6):625–32.
23. Warming L, Hassager C, Christiansen C. Changes in bone mineral den-
sity with age in men and women: a longitudinal study. Osteoporos Int
2002;13(2):105–12.
24. Lee EY, Kim D, Kim KM, Kim KJ, Choi HS, Rhee Y, et al. Age-related
bone mineral density patterns in Koreans (KNHANES IV). J Clin Endo-
crinol Metab 2012;97(9):3310–8.
View publication stats
25. Olden JD, Jackson DA. Illuminating the “black box”: a randomization
approach for understanding variable contributions in artificial neural net-
works. Ecol Modell 2002;154(1–2):135–50.
26. Wright NC, Looker AC, Saag KG, Curtis JR, Delzell ES, Randall S,
et al. The recent prevalence of osteoporosis and low bone mass in the
United States based on bone mineral density at the femoral neck or lum-
bar spine. J Bone Miner Res 2014;29(11):2520–6.
27. Cosman F, Krege J, Looker A, Schousboe J, Fan B, Isfahani NS, et al.
Spine fracture prevalence in a nationally representative sample of US
women and men aged ≥40 years: results from the National Health and
Nutrition Examination Survey (NHANES) 2013–2014. Osteoporos Int
2017;28(6):1857–66.
28. Noel SE, Mangano KM, Griffith JL, Wright NC, Dawson-Hughes B,
Tucker KL. Prevalence of osteoporosis and low bone mass among Puerto
Rican older adults. J Bone Miner Res 2018;33(3):396–403.
29. Shepherd JA, Morgan SL, Lu Y. Comparing BMD results between two
similar DXA systems using the generalized least significant change. J
Clin Densitom 2008;11(2):237–42.
30. CDC. https://wwwn.cdc.gov/Nchs/Nhanes/2007-2008/DEMO_E.htm#RI
DAGEYR (accessed September 29, 2018).
31. Slosman DO, Rizzoli R, Pichard C, Donath A, Bonjour J-P. Longitudinal
measurement of regional and whole body bone mass in young healthy
adults. Osteoporos Int 1994;4(4):185–90.
32. Henry YM, Fatayerji D, Eastell R. Attainment of peak bone mass at the
lumbar spine, femoral neck and radius in men and women: relative con-
tributions of bone size and volumetric bone mineral density. Osteoporos
Int 2004;15(4):263–73.
33. Slice DE, Algee-Hewitt BF. Modeling bone surface morphology: a fully
quantitative method for age-at-death estimation using the pubic symph-
ysis. J Forensic Sci 2015;60(4):835–43.
34. Stoyanova D, Algee-Hewitt BF, Slice DE. An enhanced computational
method for age-at-death estimation based on the pubic symphysis using
3D laser scans and thin plate splines. Am J Phys Anthropol 2015;158
(3):431–40.
35. Stoyanova DK, Algee-Hewitt BF, Kim J, Slice DE. A computational
framework for age-at-death estimation from the skeleton: surface and
outline analysis of 3D laser scans of the adult pubic symphysis. J Foren-
sic Sci 2017;62(6):1434–44.
36. Hale AR, Ross AH. Scanning skeletal remains for bone mineral density
in forensic contexts. J Vis Exp 2018;131:e56713.
37. Chang H, Chang D-G, Myong J-P, Kim J-H, Lee S-J, Lee Y, et al.
Bone mineral density among Korean females aged 20–50 years:
influence of age at menarche (The Korea National Health and Nutri-
tion Examination Survey 2008–2011). Osteoporos Int 2017;28
(7):2129–36.
38. Lyman RL. Bone density and bone attrition. In: Pokines JT, Symes SA,
editors. Manual of forensic taphonomy. Boca Raton, FL: CRC Press,
2014;51–72.