You are on page 1of 6


discussions, stats, and author profiles for this publication at:




Yosuke Nakayama

Kazutoshi Kudo


The University of Tokyo





Tatsuyuki Ohtsuki
The University of Tokyo

All in-text references underlined in blue are linked to publications on ResearchGate,

letting you access and read them immediately.

Available from: Yosuke Nakayama

Retrieved on: 11 November 2015

Gait & Posture 31 (2010) 331335

Contents lists available at ScienceDirect

Gait & Posture

journal homepage:

Variability and uctuation in running gait cycle of trained runners and

Yosuke Nakayama, Kazutoshi Kudo *, Tatsuyuki Ohtsuki
Department of Life Sciences, Graduate School of Arts and Sciences, The University of Tokyo, Meguro-ku, Komaba 3-8-1, Tokyo 153-8902, Japan



Article history:
Received 21 July 2009
Received in revised form 3 December 2009
Accepted 10 December 2009

The current study examined variability and uctuation in the running gait cycle, focusing on differences
between trained distance runners and non-runners. The two groups of participants performed treadmill
running at 80%, 100%, and 120% of their preferred speed for 10 min. Stride-interval time-series were
recorded during running using footswitches. The average preferred speed was signicantly higher for the
trained runners than for the non-runners. The trained runners showed signicantly smaller variability of
stride interval than did the non-runners, and at the same time the scaling exponent a evaluated by
detrended uctuation analysis tended to be smaller for the trained runners. These results suggest that
expert runners can reduce variability in the trained movement without loosing dynamical degrees of
freedom for spatiotemporal organization of the gait pattern.
2009 Elsevier B.V. All rights reserved.

Stride interval
Detrended uctuation analysis (DFA)
Trained distance runners
Nonlinear time-series analysis

1. Introduction
Trained distance runners have highly adaptive cardiovascular
and musculoskeletal properties for endurance performance
compared with untrained non-runners [1]. Long-term running
practice can not only promote endurance but also produce a stable
and consistent running gait cycle because variability in interlimb
coordination has been reported to decrease with practice [2,3]. In
the early stage of research on human motor control and learning,
variability was predominantly quantied by static measures such
as standard deviation, and changes in movement variability were
discussed predominantly in relation to neuromotor noise [4].
However, the development of time-series analyses has enabled
researchers to reveal dynamic or time-dependent uctuating
properties embedded in complex data records [5]. In this study we
examined differences in the variability and uctuation of the
running gait cycle between trained distance runners and nonrunners.
Studies of running movement have shown that the strideinterval time-series tends to uctuate in a self-similar fractal
manner and exhibits long-range correlations [6]. For example, using
detrended uctuation analysis (DFA), Jordan et al. [6] showed that
the stride-interval time-series of running at a preferred running
speed (PRS) exhibits persistent long-range correlations that are
typically characterized as a 1/f uctuation. Because this type of

* Corresponding author. Tel.: +81 03 5454 6133; fax: +81 03 5454 4317.
E-mail address: (K. Kudo).
0966-6362/$ see front matter 2009 Elsevier B.V. All rights reserved.

uctuation has been observed not only in different types of

locomotion such as walking [7,8] but also in a variety of human
movements such as persistent standing [9], rhythmic nger tapping
[10,11], and repetitive reaching [12], these properties reect a basic
principle of spatiotemporal organization for human motor system to
produce persistent, cyclic, or repetitive movements.
Recent developments in nonlinear time-series analyses of
human movements have found that the spatiotemporal variability
of repeated movements cannot be solely explained by neuromotor
noise, and more importantly, that the structure of the uctuation
reects signicant characteristics of motor control. For example,
Hausdorff et al. [13] found that long-range correlations in the
walking stride interval diminished in aged subjects and subjects
with Huntingtons disease compared with healthy controls,
suggesting that uctuation properties in the stride interval have
practical utility for neurological diagnostics. Environmental
information such as an auditory pacing signal also weakens
long-range correlations of stride interval during walking [14] while
reducing spatiotemporal variability of rhythmic interlimb coordination [15]. In addition, rate-dependent changes in uctuation
properties have been reported for human walking [16], running
[6], and tapping [11].
These studies suggest that trained distance runners have
different uctuation properties in their running movements,
because they are expected to show less variability in their
movement kinematics and a higher running speed at their
preferred pace than do non-runners. In this study, we examined
whether differences in gait cycle variability and uctuations exist
between trained runners and untrained non-runners. We also

Y. Nakayama et al. / Gait & Posture 31 (2010) 331335


investigated the inuence of running speed on gait cycle variability

and uctuation to examine whether these properties are speeddependent for both trained runners and non-runners.
2. Methods
2.1. Participants
Seven male trained distance runners (mean age = 23.9  1.9 years) and seven
male non-runners (mean age = 23.9  0.7 years) participated in the study. The mean
height and mass was 170.6  6.4 cm and 58.0  7.0 kg for the trained runners, and
173.4  5.0 cm and 68.4  7.1 kg for the non-runners, respectively. The trained
runners practiced distance running on university track and eld teams and the best
time (between-participant standard deviation) for a 5000-m run was 14 min 50 s
(23 s). The non-runners were recruited from among graduate and undergraduate
students. They had not undergone training specic to distance running. In accordance
with the Declaration of Helsinki, all participants submitted written informed consent
after receiving a clear explanation of the experimental procedure. The experimental
procedure was approved by the Ethics Committee of the Graduate School of Arts and
Sciences of the University of Tokyo.
2.2. Apparatus
Each participant ran on a treadmill apparatus (Takei, Japan). The treadmill had a
speed range of 0.436 km/h, and the smallest increment in speed was 0.1 km/h.
Footswitches (force sensing resistance sensors) (Biometrics, USA) placed on the rst
metatarsal area and heel of the sole of each foot recorded the moments of initial foot
contacts and foot-offs. In these footswitches, the output voltage decreased from
baseline value when external force was applied to the heel side, and increased when
the force was applied to the toe side. Data-recording software (Biopac Systems,
USA) collected the gait data from the footswitches sampled at a rate of 1000 Hz. All
of the participants used the same type of running shoes (ASICS, Japan).

2.6. Statistics
Preferred running speeds differ between runners and non-runners. Because
running speed can affect spatiotemporal variability and the uctuation in running
movement [6,17], we conducted a regression analysis for the relationship between
speed and stride-related variables (i.e., mean stride interval, CV of stride interval,
DFA exponent a). If the running speed covaried with the stride-related variables, we
conducted an analysis of covariance (ANCOVA) to examine the effect of training on
the stride-related variables after removing the variance for which the speed
accounts. Otherwise, we conducted a two-way analysis of variance (ANOVA) to
examine the effect of training (runner/non-runner, between-subject factor) and
%PRS (80/100/120%PRS, within-subject factor) on these dependent variables. Since
a preliminary analysis revealed that there was no signicant difference in any of the
dependent variables between the left and right legs, we pooled the data obtained
from the both legs in the subsequent analyses.

3. Results
All the participants were able to complete all of the trials;
however, footswitch data from one participant in the non-runner
group was not recorded due to a technical problem. Therefore, the
remaining data (i.e., seven trained runners and six non-runners)
were analyzed.
3.1. Preferred running speed
Fig. 1 shows the PRS of the trained runners and non-runners.
The average PRS for trained runners (10.7 km/h) was signicantly
faster than for non-runners (8.8 km/h), t(12) = 2.27, p < 0.05, twotailed.

2.3. Task and procedures

Preceding the data collection, participants performed warm-up exercises for 30
60 min to familiarize themselves with treadmill running. During this warm-up and
familiarization session, the participants were able to walk and run on the treadmill
at variable speeds. The PRS determined following the procedure of Jordan et al. [6].
Briey, each participant reported whether running speeds were comfortable as the
treadmill speed was increased or decreased by 0.1 km/h from relatively slow and
fast speeds, respectively. The PRS was dened as the average speed.
After the PRS was established, the participants performed 10 min trials at 80%,
100%, and 120% of the PRS in a random order. The participants were given between 3
and 12 min to recover between trials. They were permitted to rest until their heart
rate returned to the warm-up level (the heart rate was measured by palpation of a
radial pulse).
2.4. Analysis of footswitches

3.2. Mean stride interval

Fig. 2 shows representative example of a stride-interval timeseries for the trained runners and non-runners running at the PRS.
Fig. 3A and B shows the mean stride interval as a function of the
percentage of PRS and running speed, respectively. We conducted
an ANCOVA to examine the effect of training on stride interval after
removing the variance due to speed for the following reasons: (1)
previous studies showed that mean stride interval decreases
linearly with increasing running speed [6]; (2) the correlation
coefcients between running speed and stride interval averaged
across subjects following a Fisher Z transformation were sufciently high (i.e., r = 0.98 for the runners and r = 0.99 for the

We detected the moment of initial foot contacts and foot-offs from the
footswitch signals using custom-written MATLAB software. Because the initial foot
contacts were from the heel in all of the participants, stride interval was dened as
the time between falling edges of the time-series.
2.5. Data analysis
The mean, standard deviation (SD), and coefcient of variation (CV: SD/
mean  100%) for the stride interval and the strength of long-range correlations
were calculated from 512 strides of data in the middle 10 min of running.
The strength of long-range correlations was calculated by DFA [7], that yields the
DFA exponent a. The stride-interval time-series of the total length N was rst

t i  ht ii

where t(i) is the ith stride interval and ht(i)i is the mean stride interval. The
integrated time-series was divided into windows of equal length n. For each
window, the variance contributed by the local trend was eliminated by using a

linear least-squares t yk.

The detrended uctuation represented as the rootmean-square was then calculated over every window:
u N
u1 X
Fn t
yk  yk
N k1
The DFA scaling exponent a was determined as the slope of a line relating log F(n)
to log n. An a greater than 0.5 and less than (or equal to) 1.0 indicates persistent
long-range correlations. We calculated F(n) over n window sizes ranging between 4
and 128.

Fig. 1. Preferred running speed (PRS) of the runners and non-runners.

Y. Nakayama et al. / Gait & Posture 31 (2010) 331335


regression model cannot explain the relationship between running

speed and the %CV of the both groups, we performed a two-way
ANOVA. It revealed a signicant main effect of training,
F(1,11) = 8.19, p < 0.05 but either the main effect of %PRS or the
interaction of training  %PRS did not reach signicance,
Fs(2,22) < 1.40, ps > 0.05.
3.4. Long-range correlations
Fig. 3E and F shows the DFA exponent a of the stride interval as
a function of %PRS and running speed, respectively. Because a
single regression model cannot explain the relationship between
running speed and the %CV of the both groups, we performed a
two-way ANOVA. It revealed a signicant tendency of the main
effect of training, F(1,11) = 4.60, p = 0.055, but either the main
effect of %PRS or the interaction of training  %PRS did not reach
signicance, Fs(2,22) < 0.46, ps > 0.05.
Fig. 2. Representative example of stride-interval time-series for the runners and

non-runners); and (3) the average slope of the speed-interval

regression line was not different between the runners and nonrunners, t(12) = 0.54, p > 0.05, two-tailed. The ANCOVA revealed
no signicant effect of training on stride interval after controlling
for the effect of speed, F(1,11) = 0.01, p > 0.05.
3.3. Variability of the stride interval
Fig. 3C and D shows the %CV of the stride interval as a function
of %PRS and running speed, respectively. Because a single

4. Discussion
In this study, trained distance runners and non-runners
performed treadmill running at 80%, 100%, and 120% of their
preferred speed for 10 min. The PRS of the trained runners was
signicantly faster than was that of the non-runners (Fig. 1),
suggesting the trained runners adapted their cardiovascular and
musculoskeletal properties for running [1]. Mean stride interval
was smaller for the runners than for the non-runners when it was
plotted against %PRS (Fig. 3A). However, when the variance due to
running speed was removed, no signicant effect of training on
stride interval was found (Fig. 3B). This suggests that the difference
in stride interval between the two groups of participants was due

Fig. 3. Dependent variables calculated from the stride-interval time-series averaged over participants: (A) mean stride interval as a function of percentage of preferred
running speed (%PRS); (B) mean stride interval as a function of running speed; (C) coefcient of variation (CV) as a function of %PRS; (D) CV as a function of running speed; (E)
detrended uctuation analysis (DFA) exponent a as a function of %PRS; (F) DFA exponent a as a function of running speed.


Y. Nakayama et al. / Gait & Posture 31 (2010) 331335

to a difference in running speed, but not due to a difference in

general running strategy involving the selection of an average
stride interval (and probably step length) over a given range of
running speed.
Of particular interest from the viewpoint of motor control and
learning was whether variability and uctuation in the running
gait cycle changed with specialized long-term training. The
results showed that variability in stride interval evaluated by the
%CV was signicantly smaller for the trained runners than for the
non-runners (Figs. 2 and 3C and D). Because variation in the
stride interval, %CV, could not be explained by running speed, the
difference can be considered to occur due to the specialized
training for distance running. These ndings agree with previous
studies showing that in addition to performance outcome,
movement patterns become consistent and stable with practice
of a particular task [2,3,18,19]. Because stabilizing or reducing
variability in the stride interval itself is not a task requirement for
distance running, this change can be thought of as a byproduct of
task-related optimization such as an increase in running
economy [20]. Learning-related stabilization of stride interval
may also improve due to exible coupling among the neural
system, musculoskeletal system, and environment [21,22], as
well as compensatory relationships among task-related parameters [18,23] or utilization/compensation of interaction
torques [24].
We also found that the long-range correlations evaluated by the
DFA exponent a tended to be smaller for the trained runners than
for the non-runners (Fig. 3E and F). Because variation in the DFA
exponent a could not be explained by running speed, it could be
attributed to specialized training. The long-range correlations in
human repetitive movements have reportedly been weakened by
receiving additional information such as auditory pacing signals
[14] or visual feedback about the moving limb [12]. For example,
Hausdorff et al. [14] showed that a decrease in gait cycle variability
is accompanied by a decrease in long-range correlations when
walking with a rhythmic pacing tone is compared to walking
without the tone. Miyazaki et al. [12] also showed that in repetitive
reaching movements, both variability and long-range correlations
of movement amplitude decrease when visual feedback is
provided during the movement. These studies suggest that the
availability or utilization of perceptual information can contribute
to reducing the variability in the movements by altering
uctuation or dynamical properties of a given system. Indeed,
simple reduction in white noise process from the nal motoroutput stage do not change long-range correlation properties.
Because specialized training in a specic motor skill improves the
ability to utilize and integrate sensory and proprioceptive
information [25], it is possible that the decrease in the runners
a reects their ability to perceive movement-related information
such as visual, tactile, or proprioceptive feedback.
Weakened long-range correlations in the stride interval of
running movements indicate the decreased dependence of any
given stride interval upon previous stride interval and thus reect
the increase in the combinations of stride length and stride time
that are available for maintaining the speed [6]. Therefore,
decrease in the runners a, as compared with those of nonrunners, can also reect their increased (or at least maintained)
dynamical degrees of freedom in the spatiotemporal organization
of running gait pattern within a given range of running speed.
However, the ndings that long-range correlations can be
decreased as a result of exible and adaptive motor control
utilizing rich information and many degrees of freedom and at the
same time as a result of less exible control due to pathological
states or aging as indicated by Hausdorff et al. [13] seem confusing.
It is clear that further investigation is needed to examine the
physiological mechanisms, theoretical principles, and functional

factors underlying alterations in long-range correlations. At this

stage, we must emphasize that both cases are plausible.
In terms of the effect of running speed on variability and
uctuation properties of the gait cycle, although Jordan et al. [6]
reported a U-shaped relationship between running speed and DFA
exponent with the lowest a at the preferred speed, we did not nd
such a relationship in our experiment. This difference might have
been caused by the relatively large individual differences, the small
sample size, and the minimum variation in running speeds in our
experiment. Because the %CV of stride interval also did not change
as a function of running speed of the trained runners, it is possible
that 20% range of the preferred speed was not sufciently wide to
reveal changes in the uctuation property that derive from the
reorganization of the gait cycle in the trained runners. Therefore,
future studies should examine whether the U-shaped relationship
emerges in the context of a wider range of running speeds than was
examined in the present study.
Finally, the application of nonlinear time-series analyses has
provided us with important information about motor control
under various states or constraints relating to disease [13],
development [26], aging [13], task requirements such as changes
in running/walking speed [14] or movement rate [11]. The results
of this study add additional evidence that it reects signicant
characteristics in motor control of trained movement. Nonlinear
time-series analyses, together with other nonlinear methods
[15,27] will provide us useful information on the spatiotemporal
organization of human gaits.
This study was partly supported by a Grant-in-Aid for Scientic
Research (A) (No. 20240060) from Japan Society for the Promotion
of Science (JSPS) awarded to K. Kudo and a Grant-in-Aid for
Scientic Research (B) (No. 19300216) from JSPS awarded to T.
Ohstuki. We thank Dr. Hiroshi Kadota for his assistance with data
Conict of interest

The authors have declared that no conict of interests exists.

[1] Joyner MJ, Coyle EF. Endurance exercise performance: the physiology of
champions. J Physiol Lond 2008;586:3544.
[2] Fujii S, Kudo K, Ohtsuki T, Oda S. Tapping performance and underlying wrist
muscle activity of non-drummers, drummers, and the worlds fastest drummer. Neurosci Lett 2009;459:6973.
[3] Lee TD, Swinnen SP, Verschueren S. Relative phase alterations during bimanual
skill acquisition. J Mot Behav 1995;27:26374.
[4] Schmidt RA, Zelaznik H, Hawkins B, Frank JS, Quinn JT. Motor output variability: theory for the accuracy of rapid motor acts. Psychol Rev 1979;86:41551.
[5] Newell KM, Corcos DM, editors. Variability and motor control. Champaign, IL:
Human Kinetics; 1993.
[6] Jordan K, Challis JH, Newell KM. Long-range correlations in the stride interval
of running. Gait Posture 2006;24:1205.
[7] Hausdorff JM, Peng CK, Ladin Z, Wei JY, Goldberger AL. Is walking a random
walk? Evidence for long-range correlations in stride interval of human gait. J
Appl Physiol 1995;78:34958.
[8] West BJ, Grifn L. Allometric control, inverse power laws and human gait.
Chaos Solit Fractal 1999;10:151927.
[9] Duarte M, Zatsiorsky VM. Long-range correlations in human standing. Phys
Lett A 2001;283:1248.
[10] Chen Y, Ding M, Kelso JAS. Long term memory processes (1/f type) in human
coordination. Phys Rev Lett 1997;79:45014.
[11] Kadota H, Kudo K, Ohtsuki T. Time-series pattern changes related to movement rate in synchronized human tapping. Neurosci Lett 2004;370:97101.
[12] Miyazaki M, Nakajima Y, Kadota H, Chitose K, Ohtsuki T, Kudo K. 1/f-type
uctuation in human visuomotor transformation. NeuroReport 2004;15:
[13] Hausdorff JM, Mitchell SL, Firtion R, Peng CK, Cudkowicz ME, Wei JY, et al.
Altered fractal dynamics of gait: reduced stride-interval correlations with
aging and Huntingtons disease. J Appl Physiol 1997;82:2629.

Y. Nakayama et al. / Gait & Posture 31 (2010) 331335

[14] Hausdorff JM, Purdon PL, Peng CK, Ladin Z, Wei JY, Goldberger AL. Fractal
dynamics of human gait: stability of long-range correlations in stride interval
uctuations. J Appl Physiol 1996;80:144857.
[15] Kudo K, Park H, Kay BA, Turvey MT. Environmental coupling modulates the
attractors of rhythmic coordination. J Exp Psychol Hum Percept Perform
[16] Jordan K, Challis JH, Newell KM. Walking speed inuences on gait cycle
variability. Gait Posture 2007;26:12834.
[17] West BJ, Scafetta N. Nonlinear dynamical model of human gait. Phys Rev E
[18] Kudo K, Ito T, Tsutsui S, Yamamoto Y, Ishikura T. Compensatory coordination of
release parameters in a throwing task. J Mot Behav 2000;32:33745.
[19] Tyldesley D, Whiting HTA. Operational timing. J Hum Mov Stud 1975;1:
[20] Belli A, Lacour JR, Komi PV, Candau R, Denis C. Mechanical step variability
during treadmill running. Eur J Appl Physiol Occup Physiol 1995;70:


[21] Ohgane K, Ei S, Kudo K, Ohtsuki T. Emergence of adaptability to time delay in

bipedal locomotion. Biol Cybern 2004;90:12532.
[22] Taga G, Yamaguchi Y, Shimizu H. Self-organized control of bipedal locomotion by
neural oscillators in unpredictable environment. Biol Cybern 1991;65:14759.
[23] Latash ML, Scholz JP, Schoner G. Toward a new theory of motor synergies. Mot
Contr 2007;11:276308.
[24] Hirashima M, Kudo K, Ohtsuki T. Utilization and compensation of interaction
torques during ball-throwing movements. J Neurophysiol 2003;89:178496.
[25] Vuillerme N, Teasdale N, Nougier V. The effect of expertise in gymnastics on
proprioceptive sensory integration in human subjects. Neurosci Lett
[26] Hausdorff JM, Zemany L, Peng C, Goldberger AL. Maturation of gait dynamics:
stride-to-stride variability and its temporal organization in children. J Appl
Physiol 1999;86:10407.
[27] Hamill J, van Emmerik REA, Heiderscheit BC, Li L. A dynamical systems
approach to lower extremity running injuries. Clin Biomech 1999;14: