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

net/publication/247777746

Conditional Regression Model for Prediction of Anthropometric Variables

Conference Paper · June 2013

CITATIONS READS

2 921

4 authors, including:

Erik Brolin Dan Högberg

University of Skövde University of Skövde
37 PUBLICATIONS   103 CITATIONS    120 PUBLICATIONS   637 CITATIONS   

SEE PROFILE SEE PROFILE

Roland Örtengren
Chalmers University of Technology
111 PUBLICATIONS   3,370 CITATIONS   

SEE PROFILE

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

Predictive cognitive workload assessment View project

4D Ergonomics View project

All content following this page was uploaded by Dan Högberg on 02 June 2014.

The user has requested enhancement of the downloaded file.

Erik Brolin, Conditional Regression Model for Prediction of Anthropometric Variables

Conditional Regression Model for Prediction of

Anthropometric Variables
ERIK BROLIN*†‡, LARS HANSON ‡§, DAN HÖGBERG † and ROLAND ÖRTENGREN ‡

† Virtual Systems Research Centre, University of Skövde, Skövde, Sweden

‡ Department of Product and Production Development, Chalmers University of Technology, Gothenburg, Sweden
§ Industrial Development, Scania CV, Södertälje, Sweden

Abstract
In digital human modelling (DHM) systems consideration of anthropometry is central. Important functionality in
DHM tools is the regression model, i.e. the possibility to predict a complete set of measurements based on a
number of defined independent anthropometric variables. The accuracy of a regression model is measured by
how well the model predicts dependent variables based on independent variables, i.e. known key anthropometric
measurements. In literature, existing regression models often use stature and/or body weight as independent
variables in so-called flat regressions models which can produce estimations with large errors when there are low
correlations between the independent and dependent variables. This paper suggests a conditional regression
model that utilise all known measurements as independent variables when predicting each unknown dependent
variable. The conditional regression model is compared to a flat regression model, using stature and weight as
independent variables, and a hierarchical regression model that uses geometric and statistical relationships
between body measurements to create specific linear regression equations in a hierarchical structure. The
accuracy of the models is assessed by evaluating the coefficient of determination, R2 and the root-mean-square
deviation (RMSD). The results from the study show that using a conditional regression model that makes use of
all known variables to predict the values of unknown measurements is advantageous compared to the flat and
hierarchical regression models. Both the conditional linear regression model and the hierarchical regression
model have the advantage that when more measurements are included the models will give a better prediction of
the unknown measurements compared to the flat regression model based on stature and weight. A conditional
linear regression model has the additional advantage that any measurement can be used as independent variable.
This gives the possibility to only include measurements that have a direct connection to the design dimensions
being sought. Utilising the conditional regression model would create digital manikins with enhanced accuracy
that would produce more realistic and accurate simulations and evaluations when using DHM tools for the
design of products and workplaces.
Keywords: Anthropometry, Regression, Correlation, Multivariate, Prediction, Digital Human Modelling.

1. Introduction and the functionality of creating human models

based on a few predictive anthropometric
Digital human modelling (DHM) tools are used to
measurements. The known predictive
reduce the need for physical tests and to facilitate
measurements, seen as independent variables, are
proactive consideration of ergonomics in virtual
used in a regression model to calculate a complete
product and production development processes
set of anthropometric measurements which are used
(Chaffin et al. 2001; Duffy 2009). DHM tools
to create digital human models that facilitates
provide and facilitate rapid simulations,
accurate ergonomics simulation and analyses. The
visualisations and analyses in the design process
number of independent key variables varies from
when seeking feasible solutions on how the design
case to case and should be chosen based on
can meet set ergonomics requirements. DHM
relevance to the design problem (Dainoff et al.
software includes a digital human model, also
2004). Regression models can be seen as black
called a manikin, i.e. a changeable digital version of
boxes that use input, i.e. known key anthropometric
a human. An important part of DHM systems is
measurements, to produce output, i.e. a complete
anthropometry, the study of human measurements,
set of anthropometric measurements (Figure 1).

*Corresponding author. Email: [email protected] 1

Erik Brolin, Conditional Regression Model for Prediction of Anthropometric Variables

be constructed for each dependent variable. This

Input: Output:
Known key Regression Complete set of multivariate statistical analysis is based on the
anthropometric model anthropometric
assumption that anthropometric measurements can
measurements measurements
be approximated with a normal distribution, which
holds true in most cases (Pheasant and Haslegrave
Figure 1 The regression model seen as a black box that
uses input to produce output 2006). However, the conditional linear regression
model predicts the dependent variables with the
The accuracy of a regression model should smallest mean square error even if the normality
therefore be measured by how good the model assumption is not valid (Johnson and Wichern
predicts the unknown measurements, i.e. dependent 1992). The statistical and mathematical analysis
variables, based on the known key anthropometric was done using MATLAB (MathWorks 2010) and
measurements, i.e. independent variables. In ANSUR (Gordon et al. 1989) anthropometric data
literature, concerning anthropometry, existing with measurements from 1774 males and 2208
regression models often use stature and/or body females.
weight as independent variables in linear regression 2.1. Mathematical procedure of the conditional
equations (Drillis et al. 1966; Pheasant 1982; regression model
Gannon et al. 1998; Peacock et al. 2012). However,
these so-called flat regressions models can make A multivariate regression model uses k number of
estimations with large errors when there are low independent variables 𝐙 = [𝑍1 , 𝑍2 , … , 𝑍𝑘 ]𝑇 for the
correlations between the independent and prediction of j number of dependent variables
𝑇
dependent variables (You and Ryu 2005). To 𝐘 = �𝑌1 , 𝑌2 , … , 𝑌𝑗 � as
reduce this problem You and Ryu (2005) presented
a hierarchical regression model that uses geometric 𝐘 = 𝜷𝒐 + 𝜷𝐙 =
and statistical relationships between body 𝑌1 𝛽𝑜 1 + 𝛽11 ∙ 𝑧1 + 𝛽12 ∙ 𝑧2 + ⋯ + 𝛽1𝑘 ∙ 𝑧𝑘
⎡ ⎤
measurements to create specific linear regression 𝑌 ⎢𝛽𝑜 + 𝛽21 ∙ 𝑧1 + 𝛽22 ∙ 𝑧2 + ⋯ + 𝛽2𝑘 ∙ 𝑧𝑘 ⎥
equations in a hierarchical structure. Using a � ⋮2 � = ⎢ 2 ⎥.
⋮
hierarchical regression model gives better estimates ⎢ ⎥
𝑌𝑗 ⎣ 𝛽𝑜 𝑗 + 𝛽𝑗1 ∙ 𝑧1 + 𝛽𝑗2 ∙ 𝑧2 + ⋯ + 𝛽𝑗𝑘 ∙ 𝑧𝑘 ⎦
of predicted measurements if more measurements
are known and used as input. Still, the hierarchical
system requires measurements higher up in the When combining Y and Z the regression model
hierarchy, i.e. stature and body weight, to be gives a complete set of anthropometric
𝑇
included in the analyses even if they do not measurements 𝐗 = �𝑋1 , 𝑋2 , … , 𝑋𝑗+𝑘 � which is later
necessarily have a direct connection to the design used to describe joint centre positions and link
dimensions being sought (Bertilsson et al. 2011). lengths of a biomechanical model.

Using a conditional linear regression model that Suppose

uses all known measurements to predict all 𝐘
⎡ (𝑗 × 1) ⎤
unknown measurements would give better ⎢ − − − ⎥ is distributed as 𝑁 (𝝁, 𝚺)
predictions and at the same time give the possibility ⎢ ⎥ 𝑗+𝑘
to choose more freely which anthropometric ⎢ 𝐙 ⎥
⎣(𝑘 × 1)⎦
measurements that should be used as input. It is
with
possible to calculate the regression coefficients for
𝝁𝒀 𝚺𝒀𝒀 𝚺𝒀𝒁
a linear regression model through analysis of the
⎡ (𝑗 × 1) ⎤ ⎡ (𝑗 (𝑗 × 𝑘) ⎤
× 𝑗)
correlation or covariance between known and ⎢ ⎥
𝝁 = ⎢ − − − ⎥ and 𝚺 = ⎢ ⎥.
unknown measurements (Johnson and Wichern ⎢ 𝝁𝒁 ⎥ ⎢ 𝚺𝒁𝒀 𝚺𝒁𝒁 ⎥
1992). This paper presents a conditional linear ⎣(𝑘 × 1)⎦
regression model and compares its predicted results ⎣(𝑘 × 𝑗) (𝑘 × 𝑘)⎦
with the results of a flat regression model based on
stature and weight and a hierarchical regression The conditional expectation of Y, given the fixed
model based on the method presented by You and values Z of the independent variables, is
Ryu (2005).
𝐸[𝐘|𝑍1 , 𝑍2 , … , 𝑍𝑘 ] = 𝜷𝒐 + 𝜷𝒛
−1 (𝒛
2. Materials and Methods = 𝝁𝒀 + 𝚺𝒀𝒁 𝚺𝒁𝒁 − 𝝁𝒛 )

The conditional linear regression model analyses This conditional expected value, considered as a
the covariance between the independent and function of Z is called the multivariate regression
dependent variables to calculate the regression of the vector Y on Z. It is composed of j univariate
coefficients. Based on the regression coefficients regressions. The j×k matrix
and the mean values, for both the independent and
−1
dependent variables, linear regression equations can 𝜷 = 𝚺𝒀𝒁 𝚺𝒁𝒁

2
Erik Brolin, Conditional Regression Model for Prediction of Anthropometric Variables

is called the matrix of regression coefficients and Table 1 The 56 anthropometric measurements included
the j×1 vector in the evaluation of regression models

−1
𝜷𝒐 = 𝝁𝒀 − 𝚺𝒀𝒁 𝚺𝒁𝒁 𝝁𝒛 # Anthropometric measurement
1 Stature
2 Weight
is the vector containing the intersection point for 3 Acromial height
each regression equation. 4 Knee height at midpatella
5 Trochanterion height
2.2. Description of comparison procedure 6 Thumb-tip reach
7 Waist circumference at omphalion
The described conditional regression model was 8 Buttock circumference
compared to a flat regression model based on 9 Chest circumference
stature and weight and a hierarchical regression 10 Elbow circumference
11 Forearm-hand length
model based on the method presented by You and 12 Functional leg length
Ryu (2005). In the analyses gender was treated 13 Hand circumference at metacarpale
separately by creating specific regression equations 14 Hand length
for each gender for the flat and hierarchical 15 Head circumference
16 Thigh circumference, proximal
regression models and letting the conditional 17 Wrist circumference, stylion
regression model analyse both female and male 18 Ankle circumference
data. 56 anthropometric measurements (Table 1) 19 Axilla height
were included in the analysis and four comparative 20 Arm circumference at axillar
21 Foot circumference
tests were done where the number of independent 22 Biacromial breadth
variables varied for each test. The first test was 23 Bideltoid breadth
done with stature and weight as independent 24 Buttock depth
variables which are the measurements that are 25 Buttock-knee length
26 Buttock-popliteal length
necessary in the flat and hierarchical regression 27 Calf circumference
models. The second and third test was done using 28 Cervicale height
the first 7 and 17 measurements respectively 29 Chest breadth
according to Table 1. These measurements were 30 Chest depth
31 Crotch height
chosen as they could be found high up in the 32 Eye height (sitting)
hierarchal model described by You and Ryu (2005). 33 Foot breadth
The last and final test was done using the last three 34 Foot length
measurements in Table 1, hip breadth (sitting), 35 Forearm circumference, flexed
36 Gluteal furrow height
popliteal height and radiale-stylion length. These 37 Hand breadth at metacarpale
measurements are found further down in the 38 Head breadth
hierarchy of the hierarchical model, but could still 39 Head length
be interesting to use, for example in the design of 40 Heel breadth
41 Hip breadth
an office chair. The last test was not possible to 42 Interpupillary distance
perform with the flat and hierarchical model, since 43 Knee circumference
these models require stature and weight as 44 Knee height (sitting)
independent variables, but was useful to show the 45 Lateral malleolus height
46 Neck circumference over larynx
capability of the conditional regression model. The 47 Shoulder-elbow length
three regression models were compared by 48 Sitting height
assessing the coefficient of determination, R2, as 49 Thigh clearance
50 Waist breadth at omphalion
51 Waist depth at omphalion
∑𝑛𝑗=1(𝑦�𝑗 − 𝑦�)2
𝑅2 = 52 Waist height at omphalion
∑𝑛𝑗=1(𝑦𝑗 − 𝑦�)2 53 Wrist to centre-of-grip length
54 Hip breadth (sitting)
55 Popliteal height
and the root-mean-square deviation (RMSD) as 56 Radiale-Stylion length

∑𝑛𝑗=1(𝑦𝑗 − 𝑦�𝑗 )2
𝑅𝑀𝑆𝐷 = � 3. Results
𝑛
In the regression models gender was treated
where 𝑦𝑗 is the measured value, 𝑦� the mean value, separately and the coefficient of determination and
the root-mean-square deviation was calculated for
𝑦�𝑗 the predicted value and 𝑛 the number of
each dependent variable for each test. However,
measured individuals (1774 males and 2208 only the combined average results, for both genders
females). and the dependent variables for each test, are
presented in the following text and figures (Table 2,
Figure 2 and Figure 3).

3
Erik Brolin, Conditional Regression Model for Prediction of Anthropometric Variables

Table 2 Average R2 value and the root-mean-square model by 58.2% compared to the flat model and
deviation for the dependent variables for each test 16.4% compared to the hierarchical model.

Predictive Regression In test 4, when the last three measurements in Table

R2 value RMSD
variables model
1 were used as independent variables, results could
Flat 58.2% 13.88
Test 1: only be attained from the conditional regression
Hierarchical 54.9% 14.27
#1,2 model as it can use any variables as input.
Conditional 58.2% 13.88
Test 2:
Flat 53.9% 15.29 However, the results from test 4 show a decrease in
Hierarchical 59.6% 11.32 accuracy of predicting dependent variables.
#1-7
Conditional 66.5% 10.07
Flat 42.1% 19.22
Compared to test 1, with stature and weight as
Test 3:
Hierarchical 68.0% 9.60 independent variables, test 4 shows a decrease in R2
#1-17
Conditional 75.0% 8.02 value by 8.70% and an increase in root-mean-
Flat N/A N/A square error of 16.4%.
Test 4:
Hierarchical N/A N/A
#54-56
Conditional 53.1% 16.16 25
Flat model
In test 1, when stature and weight was used as Hierarchical model

Rootmean-square deviation
independent variables the resulting R2 value and 20
Conditional model
root-mean-square deviation were approximately the
same for all three regression models. However 15
when the number of independent variables
increases the accuracy of the flat regression model
decreases compared to the hierarchical and 10
conditional regression models. In test 2, when 7
measurements were used as independent variables,
5
the conditional model had an average R2 value that
was 23.3% higher than that of the flat model and
11.6% higher than that of the hierarchical model. 0
Analysis of root-mean-square deviation showed a #1,2 #1-7 #1-17 #54-56
decrease for the conditional model by 34.2% Predictive variables (according to Table 1)
compared to the flat model and 11.1% compared to Figure 3 Graph illustrating the evaluation of the root-
the hierarchical model. mean-square deviation for the dependent variables based
on the results from the four different tests (Lower value
100%
indicates higher accuracy)
90% Flat model
Coefficient of determination, R2

Hierarchical model
80% In total the conditional regression model shows the
Conditional model
70%
highest accuracy when predicting unknown
variables. For the first three tests the conditional
60% model had, on average, an accuracy that was 31.7%
50% higher than that of the flat model and 9.3% higher
40% than that of the hierarchical model (depending on if
the coefficient of determination or the root-mean-
30%
square deviation was assessed). In test 4 the
20% conditional model was the only model that could
10% produce any results.
0%
#1,2 #1-7 #1-17 #54-56 4. Discussion
Predictive variables (according to Table 1)
The results from the study shows that using a
Figure 2 Graph illustrating the evaluation of the R2 value conditional regression model that makes use of all
for the dependent variables based on the results from the known variables to predict the values of unknown
four different tests (Higher value indicates higher measurements is advantageous compared to the flat
accuracy) and hierarchical regression models. Both the
hierarchical regression model and the conditional
In test 3, when 17 measurements were used as linear regression model have the advantage that
independent variables, the conditional model had an when more measurements are included the models
average R2 value that was 78.0% higher than that of will give a better prediction of the unknown
the flat model and 10.2% higher than that of the measurements compared to the flat regression
hierarchical model. Analysis of root-mean-square model based on two variables, stature and weight.
deviation showed a decrease for the conditional A conditional linear regression model has the
additional advantage that any measurement can be

4
Erik Brolin, Conditional Regression Model for Prediction of Anthropometric Variables

used as independent variable. This gives the 5. Conclusion

possibility to only include measurements that have
Results from the study shows that the conditional
a direct connection to the design dimensions being
model produces more accurate predictions
sought. For example, in the case of creating
compared to a flat regression model based on
multidimensional boundary cases it is of interest to
stature and weight, and also to a hierarchical
reduce the number of anthropometric measurements
regression model that uses geometric and statistical
that are used as input, i.e. it is advantageous to not
relationships between body measurements to make
always be forced to include stature and weight
predictions. Utilising the conditional regression
(Brolin et al. 2012). In other cases, stature and
model would create digital manikins with enhanced
weight might not be of interest to include in the
accuracy that would produce more realistic and
analysis, e.g. when designing a shoe, helmet or a
accurate simulations and evaluations when using
hand control. However, using a DHM tool to
DHM tools for the design of products and
evaluate the design of the product could still be of
workplaces.
interest and thus requiring the functionality of the
conditional regression model.
Acknowledgement
Sorting the mean vector and the covariance matrix This work has been made possible with the support
by independent and dependent variables and by from the Swedish Foundation for Strategic
using matrix algebra as described in section 2.1 Research/ProViking and by the participating
eases the process of defining regression equations organisations. This support is gratefully
for the dependent variables. The comparison of the acknowledged.
different regression models was done using
ANSUR anthropometric data (Gordon et al. 1989). References
This data is not representative for any civilian
population today (since it was measured 20 years Bertilsson, E, Hanson, L, Högberg, D, Rhén, I-M,
ago on army personnel) but considered relevant 2011. Creation of the IMMA manikin with
here in that it covers large data set of both consideration of anthropometric diversity.
measurements and individuals. The presented Proceedings of the 21st International Conference on
conditional regression model could easily be Production Research (ICPR). Fraunhofer Verlag.
applied to any anthropometric data, which can be Brolin, E, Högberg, D, Hanson, L, 2012.
approximated by a normal distribution and for Description of boundary case methodology for
which the mean value and covariance matrix are anthropometric diversity consideration.
known or can be calculated. Even though the International Journal of Human Factors Modelling
described method makes good prediction of and Simulation, 3, 204-223.
dependent variables the diversity variance of Chaffin, D B, Thompson, D, Nelson, C, Ianni, J D,
anthropometric data is not considered. A manikin Punte, P A, Bowman, D, 2001. Digital human
based on a number of specific measurements will, modeling for vehicle and workplace design,
by using the presented model, always look the Warrendale, PA, Society of Automotive Engineers.
same. This is not the case in human populations, Dainoff, M, Gordon, C, Robinette, K M, Strauss,
e.g. persons of a specific stature will most likely M, 2004. Guidelines for using anthropometric data
have different weights and proportions. Parkinson in product design. HFES Institute Best Practices
and Reed (2010) proposes a model for creating Series. Santa Monica: Human Factors and
virtual user populations which also incorporates a Ergonomics Society.
stochastic component retaining relevant variance of Drillis, R, Contini, R, Bluestein, M, 1966. Body
the anthropometric data. The presented conditional Segment Parameters. Technical Report No.
model would be possible to extend to calculate and 1166.03. NTIS No. PB 174 945. New York: New
incorporate the variance for each dependent York University, School of Engineering and
variable based on the independent variables. This Science, Research Division.
would give digital human models with Duffy, V G, 2009. Handbook of Digital Human
anthropometry that better resembles the variance Modeling, Boca Raton, CRC Press.
and diversity that exist within human populations. Gannon, A J, Moroney, W F, Biers, D W, 1998.
This would in turn produce more realistic and The validity of anthropometric predictions derived
accurate simulations and evaluations and thus give from proportional multipliers of stature.
better assistance to engineers and designers using Proceedings of the Human Factors and Ergonomics
DHM tools when developing products and Society Annual Meeting. SAGE Publications.
workplaces. Gordon, C C, Churchill, T, Clauser, C E,
Bradtmiller, B, Mcconville, J T, Tebbetts, I,
Walker, R A, 1989. 1988 Anthropometric Survey of
US Army Personnel: Methods and Summary
Statistics. Technical Report Natick/TR-89-044.

5
 Erik Brolin, Conditional Regression Model for Prediction of Anthropometric Variables

Natick, MA.: U.S. Army Natick Research,

Development and Engineering Center.
Johnson, R A, Wichern, D W, 1992. Applied
multivariate statistical analysis, Englewood Cliffs,
NJ, Prentice-Hall, Inc.
Mathworks, 2010. MATLAB version 7.10.0.499
(R2010a). Natick, Massachusetts: The MathWorks
Inc.
Parkinson, M B, Reed, M P, 2010. Creating virtual
user populations by analysis of anthropometric
data. International Journal of Industrial
Ergonomics, 40, 106-111.
Peacock, J B, Manoharan, A, Tong, X, Ping, C Y,
Ping, L W, Ding, F, Chuan, T K, Hartono, M,
Stella, N Y, 2012. Drillis and Contini Revisited. In:
REBELO, F. & SOARES, M. M. (eds.) Advances
in Usability Evaluation, Part II. Boca Raton: CRC
Press.
Pheasant, S, 1982. A technique for estimating
anthropometric data from the parameters of the
distribution of stature. Ergonomics, 25, 981-992.
Pheasant, S, Haslegrave, C M, 2006. Bodyspace:
Anthropometry, Ergonomics and the Design of
Work, Boca Raton, CRC Press.
You, H, Ryu, T, 2005. Development of a
hierarchical estimation method for anthropometric
variables. International Journal of Industrial
Ergonomics, 35, 331-343.

View publication stats