Gait & Posture 27 (2008) 622–627

www.elsevier.com/locate/gaitpost

The symmetry angle: A novel, robust method of

quantifying asymmetry
Rebecca Avrin Zifchock a,*, Irene Davis b,c,1,
Jill Higginson d,2, Todd Royer e,3
a
510 E. 73rd Street, The Leon Root, MD Motion Analysis Laboratory, Hospital for Special Surgery, New York, NY 10021, United States
b
305 McKinly Laboratory, University of Delaware, Newark, DE 19716, United States
c
Drayer Physical Therapy Institute, Hummelstown, PA 17036, United States
d
201A Spencer Laboratory, University of Delaware, Newark, DE 19716, United States
e
144 Fred Rust Ice Arena, University of Delaware, Newark, DE 19716, United States

Received 26 January 2007; received in revised form 19 August 2007; accepted 21 August 2007

Abstract

Quantification of asymmetry is a common objective in both research and clinical settings. The most common method for quantification of
asymmetry of discrete variables is calculation of the symmetry index. Essentially a measure of the percent difference between sides, the
symmetry index requires the choice of a reference value. This is a limitation as the choice of value is not always clear, and can lead to
inconsistent results and artificially inflated values. Therefore, the purposes of the current study were to examine the limitations of the
symmetry index in depth, define a new method of quantifying symmetry that is robust to those limitations (the symmetry angle), and compute
the correlations between the two measures. The results showed that, when using the symmetry index, the interpretation of asymmetry can be
highly affected by the choice of reference value. The symmetry angle does not require the choice of a reference value. Therefore, it is not prone
to the same limitations. While symmetry angle values tend to be smaller than symmetry index values, the measures are very highly correlated.
This suggests that the symmetry angle is a good substitute for the symmetry index. Future studies of asymmetry may benefit from the use of
the symmetry angle, as it is equally effective for identifying intra-limb differences as the symmetry index, but is not prone to problems due to
normalization and provides a standard scale (100%) to interpret results.
# 2007 Elsevier B.V. All rights reserved.

Keywords: Symmetry; Gait; Strength; Structure; Methods

1. Introduction levels of asymmetry in healthy individuals [1,2]. Addition-

Quantifying differences in gait mechanics, strength, and
anthropometrics between limbs is a common clinical and
research objective. A single discrete measure describing the
symmetry between sides is useful for characterizing the
functional imbalance between limbs of an individual. Such
measures have been used in the past to characterize normal
* Corresponding author. Tel.: +1 212 606 1215.
E-mail addresses: zifchockr@hss.edu (R.A. Zifchock),
mcclay@udel.edu (I. Davis), higginso@udel.edu (J. Higginson),
royer@udel.edu (T. Royer).
1
Tel.: +1 302 831 4263.
2
Tel.: +1 302 831 6622.
3
Tel.: +1 302 831 4351.
ally, since asymmetry is often associated with pathology,
comparisons of symmetry levels have been drawn between
groups [3] and in the same group prior to and following an
intervention [4].
The symmetry index (SI) [4] is one of the most common
methods of quantifying asymmetry between discrete
measures [1,5–7]. A measurement of the percent difference
between two limbs, the SI ascribes a single value to the level
of asymmetry between two sides. Therefore, it quantifies the
imbalance between the sides of an individual, or permits
comparisons of asymmetry levels between groups. A general
formula for SI is: SI = (Xside1  Xside2)/(reference value)
 100%. Unfortunately, there are disadvantages to this
measure. First, the SI must be normalized to a reference

0966-6362/$ – see front matter # 2007 Elsevier B.V. All rights reserved.
doi:10.1016/j.gaitpost.2007.08.006
 R.A. Zifchock et al. / Gait & Posture 27 (2008) 622–627 623

value. The choice of reference value is typically based upon 2. Methods

the question being asked. For example, when comparing
between injured and uninjured sides, the uninjured side An a priori power analysis (a = 0.05, b = 0.8, expected differ-
may be the logical choice. However, if the injured side ence = 15%) was conducted using variability measures from pre-
vious literature and pilot data [3]. Results of the analysis suggested
were chosen instead, the value of the SI could be quite
that 50 subjects would be needed to adequately power the study.
different. Assessing asymmetry in a healthy population can
Therefore, 52 volunteers (28 male, 24 female) were recruited to
be difficult where there is no obvious side to use as a take part in the study. All participants were rearfoot strikers, 18–45
reference; thus, the average of the two sides is often used. years of age, and were running at least 20 miles/week. All subjects
Zifchock et al. [3] compared asymmetry levels between were free from injury at the time of the study. Written consent was
subjects with a history of tibial stress fracture and healthy obtained prior to participation in the study, and all procedures were
controls. These authors calculated the SI values of both approved by the University Human Subjects Review Board.
groups using the average of their left and right sides as the An instrumented gait analysis was conducted to obtain the gait
reference value. While it may have been useful to calculate kinetics and kinematics. All subjects wore a neutral running shoe
the SI values of the injured group using the uninjured side as (Nike Air Pegasus; Beaverton, OR). Anatomical and tracking
the reference value, there was no clear choice for the healthy markers were placed on the pelvis, thigh, shank, and rearfoot of
both sides (Fig. 1). The rearfoot markers were attached directly to
group. Averaging values can filter out differences between
the heel, through holes in the heel counter of the shoe. In addition,
sides and may have limited the ability to compare
light (2.4 g), uniaxial accelerometers (PCB Piezotronics; Buffalo,
asymmetry levels between two groups. NY) were mounted on the distal, anterio-medial aspect of both
Another limitation of the SI is its potential for artificial tibiae. Following a standing calibration trial, the anatomical mar-
inflation. This was highlighted in a paper by Herzog et al. [1] kers on the forefeet, malleoli, knees, and hips were removed.
who described typical asymmetry values for ground reaction Subjects ran along a 25 m runway at a speed of 3.7 m/s 5%,
force data during walking. They reported SI values ranging striking the centers of two forceplates (Bertec Corp.; Worthington,
from 4% to 13,000%. These inflated values can occur when a OH) with consecutive footstrikes. Kinematic data were collected
clinically irrelevant difference between sides is divided by a with a 6-camera, VICON motion analysis system (Oxford Metrics
much smaller reference value. This was exemplified by Ltd.; Oxford, UK). Kinetic and kinematic data were collected at
Herzog et al. who showed that the difference between a 1080 and 120 Hz, respectively. Five trials of two consecutive
footstrikes were collected for each subject. Kinetic and kinematic
positive value on one side and a negative value on the other
may be referenced to the average of the two values, which
will be close to zero.
Recently, Crenshaw and Richards [8] proposed a new
method for assessing symmetry between two waveforms.
This method is very useful for comparing curves over a
period of time. However, it does not allow for the
comparison of discrete values such as peak knee flexion
or hip abduction strength. While the SI is a valuable method
of assessing symmetry of discrete values, the previously
discussed limitations underscore the need for a more robust
method of quantification.
The current study proposes the use of the symmetry angle
(SA) and compares its performance with the traditionally
used SI. The objectives were to examine the limitations of
the SI, define the SA, and compute the correlations between
the two measures. Therefore, the first purpose was to
identify whether the use of different reference values
significantly alters the SI value. It was hypothesized that SI
values would be significantly different when referenced to
the right versus the left side. In addition, SI values
referenced to the average of the right and left sides were
expected to be smaller than values referenced to a given
side. The second purpose was to assess the correlation
between values calculated using the SA and the SI. These
values were expected to be significantly correlated. To
assess the applicability of these results to the types of
Fig. 1. Marker set for the static calibration trial. Anatomic tracking markers
biomechanical variables that are assessed in studies of gait, were removed from the greater trochanters, knees, malleoli, and forefeet for
two kinetic, two kinematic, two structural, and two strength the dynamic trials. Rearfoot markers were mounted directly on the skin
variables were analyzed for each aim. (inset).
624 R.A. Zifchock et al. / Gait & Posture 27 (2008) 622–627
data were low-pass filtered at 50 and 8 Hz, respectively, with a
fourth-order zero-lag Butterworth filter. Data were processed using
Visual 3D (C-Motion, Inc., Rockville, MD) and custom code
written in LabVIEW (National Instruments, Austin, TX) in order
to identify the variables of interest for each trial.
Two kinetic, two kinematic, two structural, and two strength
variables were measured. The kinetic variables of interest were the
impact peak of the ground reaction force directed along the shank in
three-dimensional space (SHfz) and peak value of the shock
attenuated in the tibia, as measured by the accelerometer (PPA).
Measurements of loading in the shank coordinate system were
chosen due to the prevalence of overuse tibial injuries in running
[9]. The kinematic variables of interest were the average velocities
of hip internal rotation (HIRv) and rearfoot eversion (REVv) from
heel strike to the impact peak of the vertical ground reaction force.
Joint velocities at this period of stance were chosen as they appear
to be particularly useful for predicting injury risk in runners [10].
Alignment and strength measures were then assessed. Each sub-
ject’s quadriceps angle (QANG) was measured as they lay supine.
This is the angle between a line from the anterior superior iliac
spine to the center of the patella, and a line from the center of the
patella to the tibial tuberosity. Hip internal rotation range of motion
(HIRr) was measured as they lay prone, with the knee flexed to 908.
Additionally, hip abduction (HABs) and hip external rotation
(HERs) strength were measured isometrically. This was done using
a hand-held dynamometer (Lafayette Instruments; Lafayette, IN)
as described by Leetun et al. [11]. The average of three strength
trials was used for analysis. Strength was normalized to body mass
and multiplied by the perpendicular distance from the hip to the
application point of the dynamometer to convert the measure to
torque. The distance was approximated as the femur length from
ASIS to lateral femoral condyle for HABs. Since HERs was
measured in a seated position, the hip was approximately level
with the knee. Therefore, the distance was approximated as the
shank length from lateral femoral condyle to lateral malleolus for
HERs.
SI values were calculated for each variable using three separate
formulae:
jðX left  X right Þj
SIleft ¼  100% (1)
X left
jðX left  X right Þj
SIright ¼  100% (2)
X right
ðX left  X right Þ
SIaverage ¼  100% (3)
avgðX left ; X right Þ
For the kinetic and kinematic variables, the symmetry values were
calculated from consecutive footstrikes, and then averaged across
the five trials.
To assess the differences in SI when referenced to the left or
right sides, j(SIleft  SIright)j was calculated. A difference of 15%
was considered to be a clinically relevant difference, such that it
could lead to a change in the interpretation of the data. For
each variable, the number of subjects with a difference of more
than 15% was compared to the number whose difference was less.
An exact binomial test for goodness-of-fit was used to statistically
compare the ratio of subjects, NUMBER<15%:NUMBER>15%, to
an expected ratio of 1:1. A simple comparison between SIleft and
SIright, such as a t-test, could not be drawn since individual
differences were of interest and there was no reason to assume
that either value would be consistently lower. Conversely, the SI
values obtained using the average of the two sides as a reference
value were expected to be consistently lower than the values
obtained using a single side as a reference value. SIleft was
arbitrarily chosen to describe values obtained using a single side
as a reference value. Therefore, SIleft was compared to SIaverage
using a two-tailed, paired t-test. To eliminate the effect of the
direction of the asymmetry (indicated by positive and negative
values), comparisons were made between absolute values.
The SA is a measure of the relationship between discrete values
obtained from the left and right sides. It is related to the angle
formed when a right-side value is plotted against a left-side value:
(Xright, Xleft). Any set of values will create a vector that forms some
angle, a, with respect to the x-axis and can be quantified as
a = arctan(Xleft/Xright). Two identical values would create a 458
angle with respect to the x-axis, forming an axis of perfect
symmetry. Any deviation, d, from 458 indicates some asymmetry:
d = 458  a (see Fig. 2). The positive and negative lines of sym-
metry can be considered to lie along the axis of perfect symmetry.
The positive line of symmetry identifies symmetry between two
positive values, while the negative line of symmetry identifies
symmetry between two negative values. The SA can be quantified
as long as the values can be graphed, and any set of values will
deviate a maximum of 908 from either line of symmetry. There-
fore, values that fall in the shaded region shown in Fig. 2 will
deviate around the positive line of symmetry and those in the
unshaded region will deviate around the negative line of symmetry.
Since the maximum deviation from either line of symmetry is
908, the following equation for SA, converted to percent of the
Fig. 2. Quantification of the symmetry angle. Any set of left and right-side
values will form a vector at an angle, a, from the x-axis. Asymmetry is
represented by a deviation, d, of the vector from the vector of perfect
symmetry. Positive (Xright1, Yleft1) and negative (Xright2, Yleft2) values deviate
from their respective lines of symmetry. A set of values that falls within the
shaded region will deviate about the positive line of symmetry. Otherwise,
they will deviate about the negative line of symmetry. An alternative SA
equation is used to quantify asymmetry for those values that fall within the
hatched region.
 R.A. Zifchock et al. / Gait & Posture 27 (2008) 622–627
maximum is:
ð45  arctanðX left =X right ÞÞ
SA ¼  100%
90
An SA value of 0% indicates perfect symmetry, while 100%
indicates that the two values are equal and opposite in magnitude.
This can only occur when a negative value can be measured for a
given variable. It should be noted that there are two regions, shown
hatched in Fig. 2, for which the quantity (458  arctan(Xleft/Xright))
will be greater than 908. A set of values that falls within those
regions will deviate about the wrong line of symmetry. It is
important that the set deviates about the correct line of symmetry
since a given pair of left and right-side values should yield an SA
value of the same magnitude despite the side to which each value is
attributed. In such a case, only the sign of the SA value should differ
(see Fig. 2). Therefore, if (458  arctan(Xleft/Xright)) > 908, the
following equation should be substituted:
ð45  arctanðX left =X right Þ  180 Þ
SA ¼  100%
90
These equations were used to calculate SA values from the data for
each variable identified previously. To assess the relationship
between SA and SIaverage, Pearson correlation coefficients were
calculated for each variable. A value of p < 0.05 was considered
significant for all analyses. All statistical tests were run using SPSS
14.0 (Chicago, IL).
3. Results
The first analysis identified the proportion of subjects
who exhibited at least a 15% difference between SI values
referenced to either the left or right side. For all variables,
the majority of subjects exhibited a difference of at least
15%. The proportion was significant for the kinetic and
kinematic variables, as well as HIRr (Fig. 3). The second
analysis compared SIleft to SIaverage. For six of the eight
Fig. 3. Ratio of subjects whose SI values were 15% different based upon the
choice of reference (left or right side). A ‘‘*’’ indicates that the ratio was
significantly different from 1:1, where HIRv, REVv, SHfz, and PPA
( p < 0.01); and HIRr ( p = 0.04).
625
(4)
Fig. 4. Comparison between SIleft, SIaverage, and SA values. A ‘‘*’’ indicates
that SIleft was statistically larger than SIaverage, where SHfz ( p < 0.01);
QANG ( p = 0.01); HIRr ( p < 0.01); and HABs ( p = 0.04). Error bars
represent the standard deviation.
(5)
variables, SIaverage tended to be lower than SIleft (Fig. 4). For
the two variables in which SIaverage was higher, the
difference was not significant.
SA and SIaverage values were perfectly correlated for the
kinetic, structural, and strength variables (r = 1.00). How-
ever, correlations for HIRv and REVv were much lower:
r = 0.63 and 0.58, respectively. To determine whether the
correlations were affected by some spurious, artificially
inflated SIaverage values, an outlier analysis was carried out
using boxplots. For each variable, extreme outliers were
identified as SI values >3  interquartile range. Three
outliers were identified for HIRv and two were identified for
REVv. When these outliers were removed from the analyses,
HIRv and REVv exhibited correlations of r = 0.78 and 0.94,
respectively.
4. Discussion
Assessment of biomechanical asymmetry is useful in
both clinical and research settings. However, it is important
that the measurement tool is robust to problems of reference
value and inflation. The purpose of this study was to examine
the effect of reference value selection on SI values, and to
compare SI values to a new method of quantifying
asymmetry, the SA. The results of the first aim of this
study suggested that quantification of gait asymmetry is
influenced by the side chosen as the reference value in the
ratio. This is of particular importance when there is not a
clear reason for choosing a particular side as the reference,
such as in a healthy population. When faced with this
situation, an arbitrarily chosen side can lead to inconsistent
results. For example, consider the case of two individuals
who both exhibit a 108 difference in knee flexion angle,
where the right side is larger in one subject and the left side is
larger in the other. If the SI is calculated using the right side
as the reference value for both cases the following would
626 R.A. Zifchock et al. / Gait & Posture 27 (2008) 622–627

result:

jðX left  X right Þj

Right larger : SI ¼  100%
X right
jð30  45 Þj
¼  100% ¼ 22:2%
45

jðX left  X right Þj

Left larger : SI ¼  100%
X right
jð45  35 Þj
¼  100% ¼ 28:6%
35
Therefore, the SI values are different despite an identical
imbalance in the two subjects. It is interesting to note that, in
general, the structural and strength variables did not exhibit
the same sensitivity to the choice of reference side. The
reason for this is not completely clear, however further
research on a more comprehensive data set may elucidate
the types of variables that are more prone to differences due
to normalization.
When there is not a clear choice for the reference value
for the calculation of the SI, some studies have used the
average of both sides [1,4,6,12]. The results of the current
study show that using this method tends to produce
significantly lower SI values. This suggests that those
studies using that choice of reference value may be
underestimating asymmetry.
The final aim of this study was to compare the newly
proposed measure of asymmetry, the SA, to the SI Fig. 5. Illustrative plot of the non-linear nature of the SA and SIaverage
method. Since the SA does not require a reference value, calculated where left-side values were held constant at 1 and right-side
it is not prone to the problems of reference value and values were varied. (A) SA values are a maximum (100%) between values
inflation. However, it appears that the SA is generally a that are equal and opposite in magnitude, and asymptotically approach 50%
good substitute for the SI. Since SA values tend to be (dashed line) to represent the difference with respect to the relative
magnitudes. (B) SIaverage values do not have a maximum, but asymptotically
much lower than SI values (see Fig. 4), it may be approach 200% (dashed line). Note: an SIaverage value could not be
necessary to adjust the minimum SA value that is calculated when the right side was 1, as the denominator of the equation
considered clinically relevant. Further, direct comparisons was equal to zero.
should not be drawn between absolute SA and SI values.
The lower correlation values for the kinematic variables
were likely due to the problem of inflation, resulting in a nature (see Fig. 5). Therefore, a difference of twice the
few very large SI values. When those values were magnitude between sides will not result in an SA or SI
removed from the analysis, the correlations were greatly value that is twice as large. Thus, asymmetry data should
improved. The kinematic variables may be particularly not be interpreted in this manner, but rather as a relative
prone to artificial inflation as they can have a positive measure. The clinical relevance of the actual difference
value on one side and a negative value on the other. The between sides should always be assessed in a separate
average of these values could be very small or close to analysis.
zero. Therefore, normalizing a large difference between This study has helped to elucidate some of the limitations
sides to such a small value can result in an excessively of the commonly used SI. This work raised the level of
inflated SI value. Conversely, the new method handles awareness of the consequences of choosing reference values.
positive and negative values by ascribing SA values that It also proposed a new method to assess asymmetry that
approach 100% as the two sides approach values that are eliminates the need to choose a reference value, but is highly
equal and opposite in magnitude. The SA and SI were correlated to the widely accepted SI. Future studies of
highly correlated since they both quantify asymmetry as a asymmetry may benefit from the use of the SA, as it is
proportion. Therefore, for variables measured on different equally effective for identifying intra-limb differences as the
scales, a similar absolute difference between sides could SI, but is not prone to problems due to the requirement of a
yield different SA, or SI, values. It is important to note reference value and provides a standard scale (100%) to
that the equations for both SA and SI are not linear in interpret results.
 R.A. Zifchock et al. / Gait & Posture 27 (2008) 622–627
Acknowledgements
This work was supported by the American Society of
Biomechanics Student Grant-In-Aid and the International
Society of Biomechanics Matching Dissertation Grant.
Conflict of interest
None.
References
[1] Herzog W, Nigg BM, Read LJ, Olsson E. Asymmetries in ground
reaction force patterns in normal human gait. Med Sci Sports Exerc
1989;21(1):110–4.
[2] Gundersen LA, Valle DR, Barr AE, Danoff JV, Stanhope SJ, Snyder-
Mackler L. Bilateral analysis of the knee and ankle during gait: an
examination of the relationship between lateral dominance and sym-
metry. Phys Ther 1989;69(8):640–50.
[3] Zifchock RA, Davis I, Hamill J. Kinetic asymmetry in female runners
with and without retrospective tibial stress fractures. J Biomech
2006;39(15):2792–7.
627
[4] Robinson RO, Herzog W, Nigg BM. Use of force platform variables to
quantify the effects of chiropractic manipulation on gait symmetry. J
Manipulative Physiol Ther 1987;10(4):172–6.
[5] Becker HP, Rosenbaum D, Kriese T, Gerngross H, Claes L. Gait
asymmetry following successful surgical treatment of ankle fractures
in young adults. Clin Orthop 1995;(311):262–9.
[6] Nolan L, Wit A, Dudzinski K, Lees A, Lake M, Wychowanski M.
Adjustments in gait symmetry with walking speed in trans-femoral
and trans-tibial amputees. Gait Posture 2003;17(2):142–51.
[7] Karamanidis K, Arampatzis A, Bruggemann GP. Symmetry and
reproducibility of kinematic parameters during various running tech-
niques. Med Sci Sports Exerc 2003;35(6):1009–16.
[8] Crenshaw SJ, Richards JG. A method for analyzing joint symmetry
and normalcy, with an application to analyzing gait. Gait Posture
2006;24(4):515–21.
[9] Taunton JE, Ryan MB, Clement DB, McKenzie DC, Lloyd-Smith DR,
Zumbo BD. A retrospective case-control analysis of 2002 running
injuries. Br J Sports Med 2002;36(2):95–101.
[10] Zifchock RA. Hip, knee, and ankle velocities may predict injury risk in
female distance runners. In: Proceedings of the Annual Meeting of the
American College of Sports Medicine; 2006.
[11] Leetun DT, Ireland ML, Willson JD, Ballantyne BT, Davis IM. Core
stability measures as risk factors for lower extremity injury in athletes.
Med Sci Sports Exerc 2004;36(6):926–34.
[12] Bennell KL, Crossley K, Wrigley TV, Nitschke A. Test–retest relia-
bility of selected ground reaction force parameters and their symmetry
during running. J Appl Biomech 1999;15(3):330–6.