Statistical Models for Predicting Automobile Driving Postures for Men and Women Including Effects of Age

Abstract

Background:
Previously published statistical models of driving posture have been effective for vehicle design but have not taken into account the effects of age.

Objective:
The present study developed new statistical models for predicting driving posture.

Methods:
Driving postures of 90 U.S. drivers with a wide range of age and body size were measured in laboratory mockup in nine package conditions. Posture-prediction models for female and male drivers were separately developed by employing a stepwise regression technique using age, body dimensions, vehicle package conditions, and two-way interactions, among other variables.

Results:
Driving posture was significantly associated with age, and the effects of other variables depended on age. A set of posture-prediction models is presented for women and men. The results are compared with a previously developed model.

Conclusion:
The present study is the first study of driver posture to include a large cohort of older drivers and the first to report a significant effect of age.

Application:
The posture-prediction models can be used to position computational human models or crash-test dummies for vehicle design and assessment.

Full text

Background: Previously published statistical mod-
els of driving posture have been effective for vehicle
design but have not taken into account the effects of
age.
Objective: The present study developed new sta-
tistical models for predicting driving posture.
Methods: Driving postures of 90 U.S. drivers with
a wide range of age and body size were measured
in laboratory mockup in nine package conditions.
Posture-prediction models for female and male driv-
ers were separately developed by employing a step-
wise regression technique using age, body dimensions,
vehicle package conditions, and two-way interactions,
among other variables.
Results: Driving posture was significantly asso-
ciated with age, and the effects of other variables
depended on age. A set of posture-prediction models
is presented for women and men. The results are com-
pared with a previously developed model.
Conclusion: The present study is the first study
of driver posture to include a large cohort of older
drivers and the first to report a significant effect of age.
Application: The posture-prediction models can
be used to position computational human models or
crash-test dummies for vehicle design and assessment.
Keywords: driver posture, hip location, eye location,
age, gender, regression
INTRODUCTION
Vehicle interior design relies on accurate
prediction of driving posture and position. For
some design applications, statistical tools that
predict the distributions of key variables, such
as driver-selected seat position and eye location,
are sufficient (Flannagan, Manary, Schneider,
& Reed, 1998; Manary, Reed, Flannagan, &
Schneider, 1998). Most new vehicle designs are
informed by population-based models of driver
spatial requirements, including design standards
and recommended practices developed by the
Society of Automotive Engineers (now SAE
International). These include statistical models
of driver reach capability (SAE J287), driver
seat position distributions (SAE J4004), and
driver eye location (SAE J941).
But when models of the human body are used
to do assessments, whether in software or physi-
cally, methods are needed for ensuring that the
models are accurately positioned. For example, if
a virtual driver representing a small woman is
used to assess the reachability of the parking
brake, the initial driving posture, including the
seat adjustments, must be correct for the assess-
ment to be useful. Similarly, assessments of exte-
rior vision using computer manikins require that
the drivers’ eye locations are accurately predicted.
Research has shown that driving posture is
influenced by the vehicle layout as well as the
driver’s body dimensions, but considerable
residual variance that is not attributable to either
of these sources remains (Reed, Manary, Flan-
nagan, & Schneider, 2000). This residual vari-
ance is important to quantify for use in vehicle
layout optimization (Parkinson, Reed, Kokko-
lars, & Papalambros, 2007).
Reed, Manary, Flannagan, and Schneider
(2002) published a set of regression models
developed from a laboratory study of driving
610249HFSXXX10.1177/0018720815610249Human FactorsModels for Predicting Driver Postures
Address correspondence to Jangwoon Park, University of
Michigan Transportation Research Institute, Ann Arbor,
MI 48109-2150, USA; e-mail: parkjang@umich.edu.
Statistical Models for Predicting Automobile Driving
Postures for Men and Women Including Effects of Age
Jangwoon Park, Sheila M. Ebert, Matthew P. Reed, University of Michigan
Transportation Research Institute, Ann Arbor, Michigan, and Jason J.
Hallman, Toyota Technical Center USA, Toyota Motor Engineering and
Manufacturing North America, Inc., Saline, Michigan
HUMAN FACTORS
Vol. XX, No. X, Month XXXX, pp. 1 –18
DOI: 10.1177/0018720815610249
Copyright © 2015, Human Factors and Ergonomics Society.
at TUFTS UNIV on December 2, 2015hfs.sagepub.comDownloaded from

2 Month XXXX - Human Factors
posture and validated against a set of in-vehicle
data. (To improve readability, this model is here-
after referred to as the Reed model.) This model,
which has been implemented in commercial
software, is based on a “cascade” approach, in
which the most important variables are predicted
first, followed by less important variables. Spe-
cifically, the model predicts hip location within
the vehicle, which reflects the effects of the fore-
aft seat position adjustment, the primary mecha-
nism by which drivers adapt the vehicle to their
body dimensions and preferences. The model
also predicts eye location, which reflects the
driver’s adjustment of the seat back recliner and
is critical for vision analyses. The remaining
degrees of freedom of interest, such as spine and
extremity postures, are then predicted using
inverse kinematics, with the redundancy in the
linkage addressed using data-based heuristics.
This approach has significant advantages over
other posture-prediction models discussed in the
literature. First, the entirety of the model has
been published; other methods for posturing
human figure models rely on unpublished algo-
rithms that cannot be independently verified.
Second, the modeling approach provides explicit
quantification of the residual variance, that is,
the extent to which posture can be expected to
vary after taking into account everything that is
known about the driver and the vehicle. Neglect-
ing this variance leads to underdesign and inac-
curate assessments (Parkinson & Reed, 2006;
Reed, Manary, Flannagan, & Schneider, 2001).
The Reed model has two important limita-
tions addressed in the current work. First, the
laboratory data were gathered using a seat lack-
ing vertical adjustability and with a fixed seat
cushion angle. Most contemporary vehicles are
designed for so-called six-way seats that incor-
porate fore-aft, vertical, and cushion angle
adjustment. Although the Reed model was vali-
dated against data from vehicles with height-
adjustable seats, the models do not provide guid-
ance for locating the vertical range of adjust-
ment. Second, the population used for the earlier
study included relatively few people over age
65. The postures and body shapes of older driv-
ers are of interest because of their increasing
numbers and because older individuals are at
greater risk of injury in crashes (Carter, Flanna-
gan, Reed, Cunninghan, & Rupp, 2014).
This paper presents a new set of posture-pre-
diction models for driving based on a laboratory
study of 90 women and men. The new models
address the shortcomings of the earlier work by
sampling a larger population, including a large
number of older and obese drivers, and using a
seat with greater adjustability. Separate models
are presented for women and men, enabling
more accurate prediction for each than would be
achieved with a pooled model.
METHOD
Participants
Ninety participants (47 women and 43 men)
were recruited through online advertisements,
newspaper, and word of mouth. The age of
participants ranged from 20 to 88 years, with a
mean age of 58.9 years (SD = 19.8). All partici-
pants were required to be licensed drivers free
of musculoskeletal ailments that would prevent
them from sitting for long periods and moving
safely around the laboratory. Participants were
required to be English speakers with a body
mass index (BMI) less than 40, with higher BMI
excluded due to the difficulty in measuring pel-
vis landmarks in that population.
Tables 1 and 2 summarize anthropometric
data obtained from the female and male partici-
pants, respectively.
Apparatus
The driver mockup (Figure 1) was constructed
using components from a 2010 Toyota High-
lander that were modified to achieve a high level
of adjustability. Testing was conducted in nine
package conditions (Table 3) manipulated by val-
ues of seat height (SAE H30) and steering-wheel
fore-aft position relative to the pedal reference
point (L6). The steering wheel height above the
heel surface (H17), steering wheel angle (A18),
and accelerator pedal plane angle (A47) were
fixed for each seat height. The three-dimensional
coordinate system in the present study was
defined following SAE J1100 (Society of Auto-
motive Engineers, 2009): x-axis runs positive
rearward and z-axis runs positive upward. The
origin for fore-aft coordinates is the pedal refer-
ence point (PRP) on the accelerator; the vertical
origin is the heel rest surface (accelerator heel
point [AHP]).
at TUFTS UNIV on December 2, 2015hfs.sagepub.comDownloaded from

Models for Predicting driver Postures
3
The seat was equipped with motorized adjust-
ments for fore-aft and vertical position, seat cush-
ion angle, and seat back angle that were used by
the drivers during testing. The head restraint was
removed to facilitate access to posterior landmarks
and to prevent interference with drivers’ preferred
head positions. The seat was mounted on a rail
system so that the overall position could be
adjusted fore-aft. The entire seat, with its adjust-
ment mechanisms, was moved to a more-forward
position for female drivers than for male drivers to
ensure that sufficient fore-aft travel was available
for all drivers.
A laboratory hardseat (Figure 2) was con-
structed to enable measurement of the posterior
spine and pelvis landmarks, such as posterior-
superior iliac spine (PSIS), that are inaccessible
on the driver mockup. The hardseat has a 14.5°
fixed cushion angle and a 23° fixed seatback
angle. Although the postures in the hardseat can
be expected to differ somewhat from those
obtained in a padded automotive seat, the data
are valuable for establishing a kinematic linkage
of the body. In both the driver mockup and hard-
seat, the three-dimensional coordinate measur-
ing machine (FARO Arm®, FARO Technology,
TABLE 1: Summary of Anthropometric Data: Women (n = 47)
Percentile
No. Anthropometric Dimension M SD Min Max 5th 50th 95th
1 Stature with shoes 1,628 69 1,452 1,788 1,508 1,628 1,743
2 Stature without shoes 1,606 69 1,435 1,780 1,487 1,607 1,721
3 Weight (kg) 68.0 13.5 44.1 101.8 49.3 65.7 89.7
4 Body mass index (kg/m2) 26.3 4.3 18.4 35.9 19.7 25.5 34.1
5 Biacromial breadth 348 29 285 415 302 352 390
6 Shoulder breadth 426 35 371 497 379 418 491
7 Chest depth scapula 273 38 201 363 217 263 340
8 Chest depth spine 216 29 162 295 176 210 263
9 Chest circumference 1,003 110 815 1,262 847 996 1,199
10 Waist circumference 929 158 635 1,243 695 928 1,209
11 Upper thigh circumference 588 56 477 692 490 592 678
12 Erect sitting height 845 43 744 948 772 844 908
13 Sitting eye height 743 46 621 865 668 744 812
14 Tragion to top of head 113 9 93 133 100 114 128
15 Head length 188 8 166 211 176 189 200
16 Head breadth 149 5 139 162 142 148 158
17 Acromial height 563 32 471 632 512 565 617
18 Shoulder-elbow length 339 19 303 379 310 338 371
19 Elbow-hand length 428 23 380 472 395 426 468
20 Knee height 499 27 435 565 457 499 542
21 Buttock-knee length 579 35 505 676 529 576 633
22 Buttock-popliteal length 488 33 416 561 430 482 537
23 Hip breadth 391 38 318 463 326 390 453
24 Hip circumference 1,047 101 870 1,270 891 1,037 1,206
25 BiASIS breadth 236 28 165 295 191 242 276
Note. In millimeters unless otherwise noted. ASIS = anterior superior iliac spine; anthropometric dimensions Nos.
1 to 11 and 12 to 24 were measured in standing and sitting posture, respectively; BiASIS breadth (No. 25) was
measured in either standing or sitting posture based on ease of locating the ASIS landmarks.
at TUFTS UNIV on December 2, 2015hfs.sagepub.comDownloaded from

4 Month XXXX - Human Factors
USA) was used to record the locations of body
surface landmarks and seat components.
Data Collection Procedures
When participants reported for the data col-
lection session, written informed consent was
obtained using procedures approved by the Uni-
versity of Michigan Institutional Review Board
for Health Behavior and Health Sciences. The
participants changed into test garments made of
thin material that provided good access to body
landmarks. The standard anthropometric mea-
sures were taken, followed by measurements in
the laboratory hardseat. The driver then entered
the vehicle mockup, and the investigator provided
scripted training on the use of the seat adjust-
ments, during which the participant demonstrated
use of the seat fore-aft and vertical adjustment,
the seat cushion angle adjustment, and the seat
back angle adjustment. The participant then exited
the mockup prior to the first trial. Test conditions
were presented in random sequence. For each test
condition, the investigator first set the steering
wheel and pedals to achieve the target values, set
the seat height and cushion angle to the middle of
the adjustment range, and set the seat back angle
TABLE 2: Summary of Anthropometric Data: Men (n = 43)
Percentile
No. Anthropometric Dimension M SD Min Max 5th 50th 95th
1 Stature with shoes 1,794 82 1,636 1,984 1,649 1,807 1,903
2 Stature without shoes 1,768 85 1,602 1,965 1,617 1,779 1,888
3 Weight (kg) 88.7 16.7 62.1 139.4 66.0 88.5 113.8
4 Body mass index (kg/m2) 28.4 4.9 19.6 40.2 21.5 28.3 38.2
5 Biacromial breadth 384 32 305 450 335 387 439
6 Shoulder breadth 482 32 430 562 437 483 535
7 Chest depth scapula 272 32 201 322 213 271 320
8 Chest depth spine 243 32 172 310 185 244 289
9 Chest circumference 1,084 106 874 1,288 918 1,105 1,230
10 Waist circumference 1,038 157 720 1,410 738 1,050 1,263
11 Upper thigh circumference 569 53 470 702 476 565 652
12 Erect sitting height 916 41 837 1,012 840 920 979
13 Sitting eye height 803 43 713 903 731 804 866
14 Tragion to top of head 125 8 106 140 110 125 137
15 Head length 198 11 151 216 183 199 210
16 Head breadth 156 9 141 192 143 154 165
17 Acromial height 613 35 544 668 558 615 666
18 Shoulder-elbow length 377 23 322 426 341 377 411
19 Elbow-hand length 484 26 425 545 444 482 526
20 Knee height 557 32 494 616 502 557 605
21 Buttock-knee length 619 37 539 680 556 618 675
22 Buttock-popliteal length 520 36 438 607 460 516 581
23 Hip breadth 381 33 305 450 334 376 433
24 Hip circumference 1,061 100 879 1,329 920 1,048 1,245
25 BiASIS breadth 240 21 205 290 207 240 274
Note. In millimeters unless otherwise noted. ASIS = anterior superior iliac spine; Anthropometric dimensions Nos.
1 to 11 and 12 to 24 were measured in standing and sitting posture, respectively; BiASIS breadth (No. 25) was
measured in either standing or sitting posture based on ease of locating the ASIS landmarks.
at TUFTS UNIV on December 2, 2015hfs.sagepub.comDownloaded from

Models for Predicting driver Postures
5
to the same midrange value. The fore-aft seat
position was placed in the mean predicted posi-
tion for men or women, as appropriate, based on
the Reed model. This procedure was intended to
minimize the bias that would have resulted if the
seat were set to an extreme initial position, while
also allowing participants to enter the mockup
comfortably.
The participant was asked to use all of the
seat adjustments to obtain a comfortable driving
posture, “as though for a long drive.” The par-
ticipant then donned the seat belt and held his or
her posture while looking straight ahead with the
hands at 10 and 2 o’clock on the steering wheel,
the right foot on the accelerator pedal, and the
left heel resting comfortably on the floor. The
investigator then recorded the posture and com-
ponent adjustments. After each condition, the
participant exited the mockup and moved behind
a screen while the investigator set the next con-
dition. The entire test session lasted approxi-
mately 2 hr.
Measurement of Driving Postures
The locations of 38 landmarks (Table 4) on
body surface and driver mockup were digitized
Figure 1. Driver mockup and recording a participant’s suprasternale landmark
location.
TABLE 3: Test Conditions
Condition
No.
Seat Height,
H30a (mm)
SW Center
From PRP,
L6 (mm)
Relative SW
Center,
L6re (mm)
SW Height,
H17 (mm)
SW Angle,
A18 (°)
Accelerator
Pedal Angle,
A47 (°)
1 180 650 50 578 23 71
2 600 0
3 550 –50
4 270 600 50 646 25 63
5 550 0
6 500 –50
7 360 550 50 715 27 52
8 500 0
9 450 –50
Note. AHP = accelerator heel point; PRP = pedal reference point; SW = steering wheel; L6re = SW center location
relative to the middle L6 value at each seat height.
aDimension definitions according to SAE J1100.
at TUFTS UNIV on December 2, 2015hfs.sagepub.comDownloaded from

6 Month XXXX - Human Factors
in each test condition. The locations of internal
joint centers defining a kinematic linkage were
estimated by using statistical models (Reed,
Ebert, & Hallman, 2013; Reed, Manary, &
Schneider, 1999) and optimization based on
the digitized data from the hardseat. The pel-
vis landmark locations (L5/S1, anterior supe-
rior iliac spine, PSIS, and hip joints) were
initially calculated in the hardseat data, then
repositioned in the vehicle-seat data using an
optimization algorithm (Park, Ebert, Reed, &
Hallman, 2015). Landmark definitions were
identical to those presented previously (see
Reed, Manary, Flannagan, et al., 2000; Reed,
Manary, & Schneider, 1999). Figure 3 illustrates
the kinematic linkage and the 16 joint and body
landmark locations relative to the PRP (defined
by SAE J4003; Society of Automotive Engi-
neers, 2008) and AHP in a sagittal plane.
Development of Driver
Posture-Prediction Models
Statistical analysis was performed in Minitab
Release 14 software (Minitab Inc.). A prelimi-
nary statistical analysis of landmark locations
revealed a substantial number of significant
interactions with gender, including three-way
interactions. To improve the tractability of both
the analysis and implementation, regression
models to predict the joint and body landmark
locations in Figure 3 were developed separately
for women and men. Additional models to pre-
dict a driver’s hip and eye locations relative to
H-point location were developed (see Appendix
Tables A1 and A2) to facilitate comparison with
the Reed model. Additional models to predict a
driver’s joint angles were also developed (see
Appendix Tables B1 and B2).
Stepwise regression (pin < .01 and pout > .05)
was applied with the potential regressors of age,
stature (S), sitting height divided by stature
(SHS), BMI, H30, L6re, and all two-way interac-
tions among the variables (e.g., S × SHS, BMI ×
Age, and H30 × L6re). Due to the correlation
between H30 and L6 in the current test condi-
tions (r = .710), L6re was used as a potential
regressor in the modeling process instead of L6.
(The correlation coefficients among the predic-
tors are summarized in Appendix Table C.) Fol-
lowing the automated stepwise process, any fac-
tors in the model with p > .01 were removed. If
some models had only two-way interactions
without the associated main effects due to the
stepwise process, we added the main effects in
the first step and deleted the interactions if they
were no longer significant. The residuals were
Figure 2. Hardseat and recording a participant’s right posterior-superior iliac spine
(PSIS) landmark location.
at TUFTS UNIV on December 2, 2015hfs.sagepub.comDownloaded from

Models for Predicting driver Postures
7
TABLE 4: Landmarks for Body Surface and Driver Mockup
Body Surface
C7 (seventh cervical vertebra) Elbow right (lateral humeral epicondyle)
Back of head max rearward Wrist right (ulnar styloid process, lateral)
Top of head max height Anterior-superior iliac spine (ASIS) right
Tragion right (near side) ASIS left
Corner eye right (ectoorbitale) Posterior-superior iliac spine (PSIS) righta
Center eye right (infraorbitale at pupil center) PSIS lefta
Glabella Supra-patella left
T8a (eighth thoracic vertebra) Infra-patella left
T12a (12th thoracic vertebra) Knee left (medial femoral epicondyle)
Lateral clavicle left Supra-patella right
Medial clavicle left Infra-patella right
Suprasternale Knee right (lateral femoral epicondyle)
Substernale Toe right (bottom edge of sole, longest shoe point)
Lateral clavicle right Ball of foot lateral right
Medial clavicle right Ankle right (lateral malleolus)
Acromion right (anterior) Heel right (bottom edge of sole at midline)
Driver Mockup
Platform Seat cushion aft
Origin Seat back bottom
Seat cushion fore Seat back top
aMeasured in the hardseat only.
Figure 3. Illustration of kinematic linkage in a sagittal plane. Note that
lower-extremity landmarks are calculated in 3D. *The center eye x is
defined as right infraorbitale at pupil center location, and the center eye z
is defined as right ectoorbitale location (see Table 3).
at TUFTS UNIV on December 2, 2015hfs.sagepub.comDownloaded from

8 Month XXXX - Human Factors
examined; in no case was evidence found that
would indicate overfitting. Factors included in
the models (all p < .01) were considered signifi-
cant. The overall performance of the models was
evaluated by adjusted R2 and root mean square
error (RMSE).
For the comparison between the Reed model
and the current models, model predictions were
obtained in three steps. First, seat positions
(translated H-points) for each female and male
driver were predicted using the existing models
(Flannagan et al., 1998; Reed, 1998) and the cur-
rent models (see Appendix Tables A1 and A2),
respectively. Second, mid-hip joint relative to
the H-point, center eye relative to the mid-hip,
and ankle joint locations were sequentially pre-
dicted. Last, inverse-kinematics submodels from
Reed (1998) were employed to predict C7/T1,
T12/L1, and L5/S1 locations for the Reed model.
RESULTS
Tables 5 and 6 show the posture-prediction
models for female and male drivers, respec-
tively. The average adjusted R2 and RMSE val-
ues for the models are 0.72 (range = 0.13~0.95)
and 26.9 mm (range = 10.2 mm~48.1 mm),
respectively. Figures 4, 5, and 6 show model
predictions using stick-figure representations of
the body linkage.
Figure 4 shows the effect of age on driving
posture, and differences with gender, using
young and old female and male drivers with
median stature and BMI for U.S. adults (Fryar,
Gu, & Ogden, 2012; female: 1,621 mm and 27.3
kg/m2; male: 1,761 mm and 27.8 kg/m2, respec-
tively). SHS of each female and male driver was
fixed as 0.52, and predictions were generated at
H30 = 270 mm and L6 = 500 mm and 600 mm
(L6re = –50 mm and 50 mm).
For female drivers at the middle package con-
dition (H30 = 270 mm, L6 = 550 mm), an
increase in age from 20 to 80 years was associ-
ated with L5/S1 joint 13 mm rearward and 14
mm lower, mid-hip joint 14 mm rearward and 25
mm lower, right knee joint 24 mm lower, and
right ankle joint locations 23 mm more forward.
For male drivers, an increase in age from 20 to
80 years was associated with a tragion 12 mm
lower, T12/L1 joint 44 mm more rearward, L5/
S1 22 mm more rearward, mid-hip joint 24 mm
rearward and 11 mm downward, and right ankle
joint locations 7 mm higher. Note that eye loca-
tions in package coordinates were not signifi-
cantly affected by age for men or women after
accounting for body size. Rather, age effects
were observed primarily in the torso for men,
whereby increased age was associated with
small differences in spine flexion, and in the pel-
vis for women, with increased flexion with age.
Figure 5 shows the comparison between the
Reed model and the current models using the
predictions at center eye, mid-hip joint loca-
tions, and seat position (translated H-point) for
5th, 50th, and 95th percentile stature of U.S.
adults (Fryar et al., 2012; female: 1,507 mm,
1,621 mm, and 1,737 mm; male: 1,632 mm,
1,761 mm, and 1,882 mm, respectively). The
SHS, BMI, and age of each female and male
driver were fixed as 0.52, 27.0 kg/m2, and 45
years, respectively; the vehicle package condi-
tion was set as H30 = 270 mm, L6 = 550 mm
(L6re = 0 mm), and cushion angle = 14.5°. Table
7 shows the discrepancies between the model
predictions at center eye, mid-hip joint loca-
tions, and seat position.
Figure 6 shows another comparison between
the two model predictions at the nine package
conditions for female and male drivers with
median stature and BMI for U.S. adults (Fryar
et al., 2012; female: 1,621 mm and 27.3 kg/m2;
male: 1,761 mm and 27.8 kg/m2, respectively).
SHS of each female and male driver was fixed
as 0.52. Figure 6 demonstrates that male and
female driving postures are quite similar overall,
except that due to the difference in overall body
size, female drivers sit farther forward, on aver-
age.
DISCUSSION
The results of the regression analysis showed
significant two-way interactions between driv-
er’s age and anthropometric dimensions (stat-
ure, SHS, and BMI). For example, the effect of
age on mid-hip joint location was significantly
different across statures, BMI, and/or SHS for
women and men, as shown in Tables 5 and 6.
This is one of the most important distinguish-
able features of the current models compared to
the previous models, which did not include any
interactions.
at TUFTS UNIV on December 2, 2015hfs.sagepub.comDownloaded from

Models for Predicting driver Postures
9
Effects of age and interaction were identified
(see Figure 4):
•Both older men and older women tend to sit
slightly lower in the seat than younger drivers
with similar body dimensions. This difference
may be related to reduced buttock muscle volume
with increased age, but a more detailed analysis of
anatomy would be necessary to assess this hypoth-
esis.
•For female drivers, an increase in age from 20 to
80 years was associated with more extended knee
posture than for younger drivers without signifi-
cant different upper-body posture. On the other
hand, older male drivers had more-rearward hip
and thorax locations, but similar eye locations,
indicating greater slouching (more spine flexion).
•Overall, although statistically significant differ-
ences in driving posture due to age were noted,
these effects are small compared with the residual
TABLE 5: Female Driver Posture-Prediction Models
Dependent Variable Regression Model Adjusted R2RMSE
Center eye x340.0 + (0.355 × S) + (2.820 × BMI) – (0.413 × H30) +
(0.550 × L6re)
.50 48.1
Center eye z–432.0 + (0.347 × S) + (942.0 × SHS) + (0.923 × H30) .94 19.0
Tragion s426.0 + (0.348 × S) + (3.130 × BMI) – (0.420 × H30) +
(0.550 × L6re)
.51 47.7
Tragion z–538.0 + (0.369 × S) + (1054 × SHS) + (0.927 × H30) .94 18.5
C7/T1 x430.0 + (0.356 × S) + (2.990 × BMI) – (0.394 × H30) +
(0.535 × L6re)
.56 41.9
C7/T1 z–399.0 + (0.299 × S) + (699.0 × SHS) + (1.070 × BMI) +
(0.930 × H30)
.94 18.9
T12/L1 x8674 – (4.710×S) – (15609×SHS) + (3.090×BMI) –
(0.370×H30) + (0.491×L6re) + (9.570×S×SHS)
.66 32.4
T12/L1 z81.30 + (8.930 × 10–2 × S) + (0.939 × H30) .90 23.4
L5/S1 x9853 – (5.360 × S) – (17602 × SHS) – (5.400 × BMI)
– (2.840 × Age) – (0.388 × H30) + (0.467 × L6re) +
(10.80 × S × SHS) + (0.112 × BMI × Age)
.71 29.0
L5/S1 z407.0 – (673.0 × SHS) + (1.290 × BMI) – (7.410 × Age)
+ (0.939 × H30) + (13.80 × SHS × Age)
.93 20.2
Mid-hip x9446 – (5.190 × S) – (5.060 × BMI) – (16970 × SHS)
– (2.750 × Age) – (0.365 × H30) + (0.465 × L6re) +
(10.50 × S × SHS) + (0.109 × BMI × Age)
.66 31.5
Mid-hip z276.0 – (680.0 × SHS) + (4.540 × BMI) – (5.550 × Age)
+ (0.906 × H30) – (5.760 × 10–2 × BMI × Age) +
(12.90 × SHS × Age)
.94 17.5
Right knee x5036 – (2.730 × S) – (8853 × SHS) – (0.429 × H30) +
(0.346 × L6re) + (5.400 × S × SHS)
.71 23.4
Right knee z600.0 + (7.890 × 10–2 × S) – (1049 × SHS) – (0.405 ×
Age) + (0.748 × H30) – (0.233 × L6re)
.82 27.6
Right ankle x409.0 – (0.137 × S) – (1.230 × BMI) – (0.382 × Age) +
(0.115 × H30) + (0.142 × L6re)
.33 21.8
Right ankle z63.40 + (0.479 × BMI) + (0.115 × H30) .29 14.1
Note. RMSE = root mean square error. Age is in years. BMI = body mass index (kg/m2); H30 = seat height (mm);
L6re = relative steering wheel center with respect to the middle location at each seat height (mm); S = stature
(mm); SHS = sitting height/stature.
pin < .01, pout > .05; all terms p < .01.
at TUFTS UNIV on December 2, 2015hfs.sagepub.comDownloaded from

10 Month XXXX - Human Factors
error after accounting for body dimensions and
vehicle factors.
Some differences between the Reed model
and the current models were identified on the
predictions of center eye and mid-hip joint
locations (see Figures 5 and 6). However, those
differences were generally smaller than the
RMSE values of the current models (see Table
7 and Appendix Table D). Overall, the results
are quite similar to those from the preceding
study, providing additional evidence that the
predictions are reasonable. We recommend that
the new models be used in place of the Reed
models, because they provide gender-specific
predictions, include age effects, and are based
on data from a height-adjustable seat.
In general, the vertical (z) coordinates of most
landmarks are better predicted than the fore-aft
(x) coordinates because the vertical positions of
TABLE 6: Male Driver Posture-Prediction Models
Dependent Variable Regression Model Adjusted R2RMSE
Center eye x291.0 + (0.436 × S) – (0.482 × H30) + (0.591 × L6re) .62 44.1
Center eye z–413.0 + (0.313 × S) + (878.0 × SHS) + (2.240 × BMI)
+ (0.968 × H30)
.94 19.0
Tragion s357.0 + (0.447 × S) – (0.482 × H30) + (0.597 × L6re) .63 44.1
Tragion z–365.0 + (0.305 × S) + (830.0 × SHS) + (1.990 × BMI)
– (0.198 × Age) + (0.974 × H30)
.94 18.9
C7/T1 x367.0 + (0.452 × S) – (0.454 × H30) + (0.552 × L6re) .68 37.7
C7/T1 z–290.0 + (0.209 × S) + (659.0 × SHS) + (3.180 × BMI)
+ (0.972 × H30)
.95 16.5
T12/L1 x–420.0 + (0.862 × S) + (11.50 × Age) – (0.414 × H30)
+ (0.541 × L6re) – (6.110 × 10–3 × S × Age)
.74 33.4
T12/L1 z62.50 + (8.400 × 10–2 × S) + (0.806 × BMI) + (0.965 ×
H30) + (3.210 × 10–2 × L6re)
.91 23.4
L5/S1 x1619 + (0.380 × S) – (2286 × SHS) – (2.340 × BMI)
– (15.50 × Age) – (0.410 × H30) + (0.524 × L6re) +
(30.50 × SHS × Age)
.75 30.8
L5/S1 z11.20 + (2.140 × BMI) + (0.960 × H30) .93 20.0
Mid-hip x214.0 + (0.616 × S) – (680.0 × SHS) + (7.210 × Age)
– (0.392 × H30) + (0.546 × L6re) – (3.860 × 10–3 × S
× Age)
.69 35.1
Mid-hip z533.0 – (1051 × SHS) – (25.10 × BMI) – (0.182 × Age)
+ (0.935 × H30) + (50.80 × SHS × BMI)
.95 17.0
Right knee x4082 – (1.850 × S) – (6958 × SHS) + (0.838 × BMI) –
(0.474 × H30) + (0.391 × L6re) + (3.680 × S × SHS)
.74 24.0
Right knee z–68.90 + (0.256 × S) – (361.0 × SHS) + (0.731 × H30)
– (0.272 × L6re)
.81 29.3
Right ankle x131.0 + (0.725 × BMI) + (7.530 × 10–2 × H30) +
(0.141 × L6re)
.13 21.8
Right ankle z–34.50 + (6.900 × 10–2 × S) + (0.107 × Age) + (9.740
× 10–2 × H30)
.44 10.2
Note. RMSE = root mean square error. Age is in years. BMI = body mass index (kg/m2); H30 = seat height (mm);
L6re = relative steering wheel center with respect to the middle location at each seat height (mm); S = stature
(mm); SHS = sitting height/stature.
pin < .01, pout > 0.05; all terms p < .01.
at TUFTS UNIV on December 2, 2015hfs.sagepub.comDownloaded from

Models for Predicting driver Postures
11
the landmarks of interest are closely related to
design variables and available anthropometric
measures due to the nature of the task and the
design of the study. For example, the test condi-
tions included a wide range of steering wheel
heights that produced a similarly large range of
vertical hip locations. Consequently a large
portion of the variance in the vertical hip loca-
tion can be attributed to H17 (steering wheel
height), yielding a high R2 value. In contrast,
fore-aft position hip location is related more to
driver preference, with consequently weaker
relationships with predictors and greater resid-
ual variance.
This research is limited by the laboratory
context. Although a wide range of vehicle pack-
age conditions were used, the mockup was
equipped with a single vehicle seat that may
have affected the results. However, previous
studies have shown good correlation between
laboratory and in-vehicle data (Manary et al.,
1998; Reed, Manary, Flannagan, & Schneider,
1999). The mockup did not include a headliner,
although a realistic instrument panel was used.
The participants were instructed to look forward
as though they were driving, but the visual task
was not realistic. However, a previous study in
the laboratory and in vehicles showed that
restrictions to the visual field have minimal
effects on driver postures (Reed, Manary, &
Figure 4. Effects of fore-aft steering wheel locations on driver posture for female
and male and young and old drivers. Predictions are shown for midsize female
and male drivers for U.S. adults age 20 (dotted line) and age 80 (solid line) years.
H30 = 270 mm, L6 = 500 mm and 600 mm (L6re = –50 mm and 50 mm). The
more-forward postures in each case correspond to the L6 = 500 mm condition.
Figure 5. Comparison between the Reed model
(dotted line) and the current models (solid line).
Predictions for 5th, 50th, and 95th percentile stature
of female and male drivers for U.S. adults age 45
years. H30 = 270 mm, L6 = 550 mm (L6re = 0 mm). The
axis lengths of the ellipses over the joint locations
equal to ±1 root mean square error.
at TUFTS UNIV on December 2, 2015hfs.sagepub.comDownloaded from

12 Month XXXX - Human Factors
TABLE 7: Discrepancies With the Reed Model (Reed Model – Current Predictions) for U.S. Female and
Male Drivers
Discrepancy (mm) Discrepancy/RMSEa
Stature
Center Eye
Location
Mid-Hip
Location
Seat
Position
Center Eye
Location
Mid-Hip
Location
Seat
Position
x z x z x z x z x z x z
Female
5th percentile –43.0 30.8 –44.5 23.4 –9.3 1.0 –1.02 1.77 –1.82 1.61 –0.37 0.08
50th percentile –6.3 24.3 –21.4 18.5 8.4 3.1 –0.15 1.39 –0.88 1.28 0.34 0.25
95th percentile 30.1 18.7 1.3 14.7 25.4 6.3 0.71 1.08 0.05 1.01 1.02 0.51
Male
5th percentile 28.8 9.7 –0.7 18.6 15.4 –5.5 0.78 0.52 –0.03 1.37 0.61 –0.52
50th percentile 35.6 7.7 2.8 14.0 22.3 –3.0 0.96 0.41 0.11 1.03 0.88 –0.28
95th percentile 42.5 5.6 6.6 9.4 29.2 –0.9 1.15 0.30 0.27 0.69 1.15 –0.08
Note. RMSE = root mean square error. Seat position = translated H-point location; H30 = 270 mm and L6 = 550
mm (L6re = 0 mm).
aRMSE values from Appendix Tables A1 and A2.
Figure 6. Comparison between the Reed model (dotted line) and the current models
(solid line) on the nine package conditions. Predictions (see Appendix Table D) for
midsize female (more-forward location) and male (more-rearward location) drivers
for U.S. adults age 45 years. The axis lengths of the ellipses over the joint locations
equal to ±1 root mean square error.
at TUFTS UNIV on December 2, 2015hfs.sagepub.comDownloaded from

Models for Predicting driver Postures
13
Schneider, 2000). The removal of the head
restraint from the seat prevented head restraint
interference from affecting driver posture, but as
a result, the current models may not be accurate
when used to predict the postures of drivers who
would experience head restraint interference in a
particular vehicle.
The estimates of age effects are limited by
the characteristics of the current sample. The
older individuals in the sample may have been
less representative of the older driver popula-
tion than was the case for younger drivers due to
the exclusion of drivers with musculoskeletal
impairment, which is more common among
older drivers. The extent to which those impair-
ments and other health issues affect driving pos-
ture is not addressed by the current study, but
authors of future work should consider these
issues, particularly in the context of the older
driver population.
Model quality was evaluated using adjusted
R2 values, but it is important to consider that
these values include the effects of the experi-
mental manipulations, such as seat height (H30).
For example, the high adjusted R2 value for mid-
hip z (0.95) is due almost entirely to the large
range of H30 included in the experiment. Conse-
quently, the RMSE values are a more meaning-
ful measure of model performance, particularly
in relation to the effects of anthropometric fac-
tors. The RMSE values are also important for
considering the uncertainty associated with any
particular model prediction. For example, the
RMSE for the fore-aft eye location for a male
driver is 44.1 mm (the comparable value was
50.9 mm in Reed et al., 2002). Under the
assumption of normally distributed residuals,
95% of eye locations for men with identical
body dimensions in a particular vehicle lay in a
range of 2 × 1.64 × 44.1 = 145 mm. This finding
indicates that the notion of a “midsize male” or
“small female” eye location is not very mean-
ingful, because body dimensions and vehicle
layout do not precisely predict eye location.
More research is needed to determine if these
results are applicable in other markets. The U.S.
and non-U.S. drivers have some differences in
their anthropometry (Lee, Jeong, et al., 2013;
Lee, Jung, et al., 2013). Therefore, the differ-
ences between driving postures of U.S. and non-
U.S. drivers need to be identified for designing
more accommodating vehicle interiors in target
markets.
CONCLUSION
A laboratory study of driving posture pro-
duced the first posture-prediction model for men
and women to include the effects of age. The
large sample size and broad range of experimental
conditions enabled consideration of interactions
between age, body dimensions, and vehicle
layout, which were not included in earlier mod-
els. These models, which can be implemented
directly and completely from the information
in this paper, should be used in preference to
all previously published models of driving pos-
ture for vehicles with automatic transmissions
intended for the U.S. market.
at TUFTS UNIV on December 2, 2015hfs.sagepub.comDownloaded from

14 Month XXXX - Human Factors
APPENDIX
TABLE A1: Female Driver Posture-Prediction Models for the Cascade Modeling Approach
Dependent Variable Regression Model (mm) Adjusted R2RMSE
Center eye x re mid-hip 1.900 + (2.320 × BMI) + (9.020 × 10–2 × L6re) .06 42.3
Center eye z re mid-hip –914.0 + (0.348 × S) + (1942 × SHS) – (3.190 ×
BMI) + (8.430 × Age) – (17.60 × SHS × Age) +
(4.420 × 10–2 × BMI × Age)
.68 17.4
Center eye x re H-point –364.0 + (7.640 × 10–2 × S) + (540.0 × SHS) +
(0.124 × L6re)
.07 37.7
Center eye z re H-point –461.0 + (0.163 × S) + (1445 × SHS) – (15.10 ×
BMI) + (3.610 × Age) + (9.930 × 10–3 × S ×
BMI) – (6.540 × SHS × Age)
.86 12.7
Mid-hip x re H-point –162.3 + (358.3 × SHS) – (1.757 × BMI) .14 24.4
Mid-hip z re H-point 2238 – (1.260 × S) – (4719 × SHS) + (4.660 × BMI)
– (6.000 × Age) – (5.300 × 10–2 × BMI × Age) +
(13.80 × SHS × Age) + (2.540 × S × SHS)
.44 14.5
H-point x re PRP 678.0 + (0.284 × S) – (494.0 × SHS) + (2.330 ×
BMI) – (0.388 × H30) + (0.426 × L6re)
.75 24.9
H-point z re AHP 129.0 – (6.200 × 10–2 × S) + (0.909 × H30) –
(4.650 × 10–2 × L6re)
.97 12.4
Note. RMSE = root mean square error. Age is in years. AHP = accelerator heel point; BMI = body mass index (kg/
m2); H30 = seat height (mm); L6re = relative steering wheel center with respect to the middle location at each seat
height (mm); S = stature (mm); SHS = sitting height/stature; PRP = pedal reference point.
pin < .01, pout > .05; all terms p < .01.
TABLE A2: Male Driver Posture-Prediction Models for the Cascade Modeling Approach
Dependent Variable Regression Model (mm) Adjusted R2RMSE
Center eye x re mid-hip –696.0 + (9.320 × 10–2 × S) + (1193 × SHS) –
(9.290 × 10–2 × H30)
.16 37.1
Center eye z re mid-hip –241.0 + (0.313 × S) + (587.0 × SHS) + (1.420 ×
BMI)
.62 18.7
Center eye x re H-point –578.0 + (0.174 × S) + (706.0 × SHS) – (1.550 ×
BMI) – (6.480 × 10–2 × H30)
.22 32.1
Center eye z re H-point –498.0 + (0.361 × S) + (845.0 × SHS) + (2.760 ×
BMI) – (0.175 × Age)
.80 14.4
Mid-hip x re H-point 196.0 + (7.380 × 10–2 × S) – (552.0 × SHS) –
(2.440 × BMI)
.26 24.9
Mid-hip z re H-point 2269 – (1.390 × S) – (4615 × SHS) + (1.420 × BMI)
– (0.222 × Age) + (2.780 × S × SHS)
.24 13.6
H-point x re PRP –48.00 + (0.560 × S) + (1.390 × BMI) + (7.530 ×
Age) – (0.420 × H30) + (0.505 × L6re)
– (3.98 × 10–3 × S × Age)
.77 25.3
H-point z re AHP 123.1 – (5.864 × 10–2 × S) + (0.945 × H30) .98 10.6
Note. RMSE = root mean square error. Age is in years. AHP = accelerator heel point; BMI = body mass index (kg/
m2); H30 = seat height (mm); L6re = relative steering wheel center with respect to the middle location at each seat
height (mm); S = stature (mm); SHS = sitting height/stature; PRP = pedal reference point.
pin < .01, pout > .05; all terms p < .01.
at TUFTS UNIV on December 2, 2015hfs.sagepub.comDownloaded from

Models for Predicting driver Postures
15
TABLE B1: Female Driver Joint-Angle-Prediction Models
Joint Angle Regression Model (°) Adjusted R2RMSE
Hip-to-eye vector 0.320 + (0.214 × BMI) .05 4.1
Head 72.30 – (1.830 × 10–2 × S) – (81.80 ×
SHS) + (0.267 × BMI)
.13 5.3
Neck 111.0 – (7.360 × 10–2 × S) – (2.150 ×
Age) – (1.240 × 10–2 × H30) + (1.370
× 10–3 × S × Age)
.15 5.7
Thorax –39.73 + (81.29 × SHS) .03 7.0
Abdomen 5.093 + (0.8034 × BMI) .13 9.4
Pelvis 129.0 – (4.540 × 10–2 × S) – (0.204 ×
Age)
.08 13.9
Thigh right –93.20 + (228.0 × SHS) + (7.060 × BMI)
– (2.620 × 10–2 × H30) – (3.580 × 10–2
× L6re) – (13.90 × SHS × BMI)
.40 4.5
Knee-included right 372.0 – (449.0 × SHS) – (12.90 × BMI) +
(7.410 × 10–2 × Age) – (0.102 × H30)
+ (8.190 × 10–2 × L6re) + (25.50 × SHS
× BMI)
.55 8.7
Note. RMSE = root mean square error. Age is in years. BMI = body mass index (kg/m2); H30 = seat height (mm);
L6re = relative steering wheel center with respect to the middle location at each seat height (mm); S = stature
(mm); SHS = sitting height/stature.
pin < .01, pout > .05; all terms p < .01.
TABLE B2: Male Driver Joint-Angle-Prediction Models
Joint Angle Regression Model (°) Adjusted R2RMSE
Hip-to-eye vector –36.80 + (86.00 × SHS) – (8.430 × 10–3
× H30)
.14 3.3
Head –90.80 + (162.0 × SHS) + (0.327 × BMI)
+ (1.650 × Age) – (3.110 × SHS ×
Age)
.12 5.0
Neck –7.830 – (5.240 × 10–2 × Age) .02 6.5
Thorax –49.00 + (109.0 × SHS) – (6.710 × 10–2
× Age) – (9.460 × 10–3 × H30)
.21 4.3
Abdomen –71.60 + (140.0 × SHS) + (1.110 × BMI) .16 11.4
Pelvis –124.0 + (0.120 × S) – (1.500 × BMI) +
(3.310 × Age) – (1.890 × 10–3 × S ×
Age)
.24 13.2
Thigh right 35.20 + (1.900 × 10–2 × S) – (72.50 ×
SHS) – (0.162 × BMI) – (3.040 × 10–2 ×
H30) – (4.360 × 10–2 × L6re)
.39 4.4
Knee-included right 204.0 – (3.48 × 10–2 × S) – (8.570 × 10–2
× H30) + (9.330 × 10–2 × L6re)
.44 8.9
Note. RMSE = root mean square error. Age is in years. BMI = body mass index (kg/m2); H30 = seat height (mm);
L6re = relative steering wheel center with respect to the middle location at each seat height (mm); S = stature
(mm); SHS = sitting height/stature.
pin < .01, pout > .05; all terms p < .01.
at TUFTS UNIV on December 2, 2015hfs.sagepub.comDownloaded from

16 Month XXXX - Human Factors
TABLE C: Pearson’s Correlation Coefficients Among the Potential Predictors
Predictor S SHS BMI Age H30
Female
SHS –.120
BMI .096 –.046
Age –.343 –.387 –.058
H30 –.022 .001 .017 .033
L6re .013 –.017 –.011 .007 .001
Male
SHS –.363
BMI –.045 –.176
Age –.241 –.170 .425
H30 –.018 .007 .001 .014
L6re .006 –.025 –.033 –.012 .005
Note. Age is in years. BMI = body mass index (kg/m2), H30 = seat height (mm), L6re = relative steering wheel
location (mm), S = stature (mm), SHS = sitting height/stature.
TABLE D: Discrepancies Between the Model Predictions (Reed Model – Current Predictions) for
Midsize U.S. Female and Male Drivers at the Nine Package Conditions
Discrepancy (mm) Discrepancy/RMSEa
Center Eye
Location
Mid-Hip
Location
Seat
Position
Center Eye
Location
Mid-Hip
Location
Seat
Position
Gender
H30
(mm)
L6
(mm) xzxzxzxzxzxz
Female 180 650 10.8 18.4 –15.3 18.1 14.4 –4.8 0.26 1.06 –0.63 1.25 0.58 –0.39
600 8.6 13.9 –14.3 10.9 15.4 –5.1 0.20 0.80 –0.59 0.75 0.62 –0.41
550 5.6 9.4 –14.0 3.7 15.7 –5.4 0.13 0.54 –0.57 0.26 0.63 –0.44
270 600 –4.3 29.7 –22.7 26.7 7.1 4.4 –0.10 1.71 –0.93 1.84 0.28 0.35
550 –6.3 24.3 –21.4 18.5 8.4 3.1 –0.15 1.39 –0.88 1.28 0.34 0.25
500 –9.3 19.8 –21.1 11.3 8.7 2.7 –0.22 1.14 –0.86 0.78 0.35 0.22
360 550 –19.2 40.1 –29.7 34.3 0.0 12.6 –0.45 2.30 –1.22 2.37 0.00 1.02
500 –21.2 35.6 –28.4 27.1 1.3 12.3 –0.50 2.05 –1.17 1.87 0.05 0.99
450 –24.2 31.1 –28.1 19.9 1.6 11.9 –0.57 1.79 –1.15 1.38 0.06 0.96
Male 180 650 42.8 2.8 2.8 14.5 22.2 –9.9 1.15 0.15 0.11 1.07 0.88 –0.94
600 39.3 –0.4 7.0 8.7 26.5 –8.9 1.06 –0.02 0.28 0.64 1.05 –0.84
550 36.7 –2.5 12.3 3.8 31.7 –6.9 0.99 –0.14 0.49 0.28 1.25 –0.65
270 600 39.2 9.9 –1.4 18.9 18.0 –5.0 1.06 0.53 –0.06 1.39 0.71 –0.47
550 35.6 7.7 2.8 14.0 22.3 –3.0 0.96 0.41 0.11 1.03 0.88 –0.28
500 33.1 5.6 8.1 9.2 27.5 –1.0 0.89 0.30 0.32 0.67 1.09 –0.09
360 550 35.5 18.0 –5.6 24.3 13.8 1.0 0.96 0.96 –0.23 1.79 0.55 0.09
500 32.0 15.8 –1.4 19.4 18.1 3.0 0.86 0.85 –0.06 1.43 0.71 0.28
450 29.4 12.7 3.9 13.5 23.3 4.0 0.79 0.68 0.16 0.99 0.92 0.37
Note. RMSE = root mean square error; seat position = translated H-point location.
aRMSE values from Appendix Tables A1 and A2.
at TUFTS UNIV on December 2, 2015hfs.sagepub.comDownloaded from

Models for Predicting driver Postures
17
ACKNOWLEDGMENTS
This research was sponsored by the Toyota Col-
laborative Safety Research Center (CSRC). We thank
our collaborators at the CSRC who contributed signifi-
cantly to this work, including Chuck Gulash, Megan
Mackenzie, Jason Hallman, and Palani Palaniappan.
Many people at the University of Michigan Transporta-
tion Research Institute contributed to the success of this
project, including Brian Eby, Charlie Bradley, Steven
Thomas, and Stewart Simonett, who developed the
mockups and fixtures. Laura Malik and Jamie Moore
led the data collection, assisted by numerous student
research assistants, including Alexis Baker, Olivia
DeTroyer, Tiffany Fredrick, Mollie Pozolo, Rachel
Palmer, Sarah Scholten, and Lindsay Youngren. These
students were assisted in data processing and scan land-
mark extraction by Christian Calyore, David Hayashi,
Danielle Hedden, Jordan MacDonald, Huibin Hu,
Ryan Warner, and Mikhail Wise.
KEY POINTS
•Driving postures of 90 U.S. drivers with a wide
range of age and body size were measured in nine
package conditions using a laboratory vehicle
mockup and hardseat.
•A set of new posture-prediction models for driv-
ing was separately developed for women and men
with inputs including driver’s age, anthropometric
dimensions, vehicle package dimensions, and the
two-way interactions among the variables.
•The models show good prediction performances
in terms of adjusted R2 and root mean square error
(RMSE) (adjusted R2: M = 0.72, range = 0.13~0.95;
RMSE: M = 26.9 mm, range = 10.2~48.1 mm).
•The models can be used to accurately position
computational human models or crash-test dum-
mies for older drivers in a certain vehicle package
condition.
•The quantification of the residual variance in pos-
ture that is not attributable to vehicle or driver
variables is useful for stochastic simulation.
REFERENCES
Carter, P. M., Flannagan, C. A. C., Reed, M. P., Cunninghan, R.
M., & Rupp, J. D. (2014). Comparing the effects of age, BMI,
and gender on severe injury (AIS 3+) in motor-vehicle crashes.
Accident Analysis & Prevention, 72, 146–160.
Flannagan, C. A. C., Manary, M. A., Schneider, L. W., & Reed,
M. P. (1998). An improved seating accommodation model with
application to different user populations (SAE Technical Paper
980651). Warrendale, PA: Society of Automotive Engineers.
Fryar, C. D., Gu, Q., & Ogden, C. L. (2012). Anthropometric refer-
ence data for children and adults: United States, 2007–2010.
National Center for Health Statistics. Vital and Health Statis-
tics, 11(252), 1–48.
Lee, W., Jeong, J., Park, J., Jeon, E., Kim, H., Jung, D., Park, S., &
You, H. (2013). Analysis of the facial measurements of Korean
Air Force pilots for oxygen mask design. Ergonomics, 56,
1451–1464.
Lee, W., Jung, K., Jeong, J., Park, J., Cho, J., Kim, H., Park, S., &
You, H. (2013). An anthropometric analysis of Korean male
helicopter pilots for helicopter cockpit design. Ergonomics, 56,
879–887.
Manary, M. A., Reed, M. P., Flannagan, C. A. C., & Schneider,
L. W. (1998). ATD positioning based on driver posture and
position (SAE Technical Paper 983163). Warrendale, PA: SAE
International.
Park, J., Ebert, S. M., Reed, M. P., & Hallman, J. J. (2015, July).
Development of an optimization method for locating the pelvis
in an automobile seat. Paper presented at the 4th International
Digital Human Modeling Conference, Las Vegas, NV.
Parkinson, M. B., & Reed, M. P. (2006). Optimizing vehicle occu-
pant packaging (SAE Technical Paper 2006-01-0961). Warren-
dale, PA: SAE International.
Parkinson, M. B., Reed, M. P., Kokkolars, M., & Papalambros, P.
Y. (2007). Optimizing truck cab layout for driver accommoda-
tion. ASME Journal of Mechanical Design, 129, 1110–1117.
Reed, M. P. (1998). Statistical and biomechanical prediction of
automobile driving posture (Unpublished doctoral disserta-
tion). University of Michigan, Ann Arbor.
Reed, M. P., Ebert, S. M., & Hallman, J. J. (2013). Effects of driver
characteristics on seat belt fit. Stapp Car Crash Journal, 57, 43–57.
Reed, M. P., Manary, M. A., Flannagan, C. A. C., & Schneider,
L. W. (1999). Automobile occupant posture prediction for use
with human models (SAE Technical Paper 1999-01-0966).
Warrendale, PA: SAE International.
Reed, M. P., Manary, M. A., Flannagan, C. A. C., & Schneider, L.
W. (2000). Effects of vehicle interior geometry and anthropo-
metric variables on automobile driving posture. Human Fac-
tors, 42, 541–552.
Reed, M. P., Manary, M. A., Flannagan, C. A. C., & Schneider, L.
W. (2001). New tools for vehicle interior design. In Proceed-
ings of the Human Factors and Ergonomics 45th Annual Meet-
ing (pp. 1138–1140). Santa Monica, CA: Human Factors and
Ergonomics Society.
Reed, M. P., Manary, M. A., Flannagan, C. A. C., & Schneider, L.
W. (2002). A statistical method for predicting automobile driv-
ing posture. Human Factors, 44, 557–568.
Reed, M. P., Manary, M. A., & Schneider, L. W. (1999). Methods
for measuring and representing automobile occupant pos-
ture (SAE Technical Paper 990959). Warrendale, PA: SAE
International.
Reed, M. P., Manary, M. A., & Schneider, L. W. (2000). The effects
of forward vision restriction on automobile driver posture.
Transportation Human Factors, 2, 173–179.
Society of Automotive Engineers. (2008). J4003: H-Point Machine
(HPM-II). Procedure for H-point determination—Benchmark-
ing vehicle seats. Warrendale, PA: Author.
Society of Automotive Engineers. (2009). J1100: Motor vehicle
dimensions. Warrendale, PA: Author.
Jangwoon Park is a postdoctoral research fellow in
the Biosciences Group of the University of Michigan
at TUFTS UNIV on December 2, 2015hfs.sagepub.comDownloaded from

18 Month XXXX - Human Factors
Transportation Research Institute. He received a
PhD in industrial and management engineering from
Pohang University of Science and Technology in
2013.
Sheila M. Ebert is a senior research specialist in the
Biosciences Group of the University of Michigan
Transportation Research Institute. She received an
MS from Michigan State University in 1999.
Matthew P. Reed is a research professor and head of
the Biosciences Group of the University of Michigan
Transportation Research Institute. He received a
PhD in industrial and operations engineering from
the University of Michigan in 1998.
Jason J. Hallman is a senior engineer in the Toyota
Technical Center at Toyota Motor Engineering and
Manufacturing North America, Inc. He received a
PhD in biomedical engineering from Marquette Uni-
versity in 2010.
Date received: May 17, 2015
Date accepted: September 12, 2015
at TUFTS UNIV on December 2, 2015hfs.sagepub.comDownloaded from