Physical Fitness and Performance

Dynamic Pacing Strategies during the Cycle

Phase of an Ironman Triathlon
CHRIS R. ABBISS1, MARC J. QUOD1,2, DAVID T. MARTIN2, KEVIN J. NETTO1, KAZUNORI NOSAKA1,
HAMILTON LEE2, ROB SURIANO3, DAVID BISHOP3, and PAUL B. LAURSEN1
1

School of Exercise, Biomedical and Health Sciences, Edith Cowan University, Joondalup, Western Australia, AUSTRALIA;
Department of Physiology, Australian Institute of Sport, Canberra, Australian Capital Territory, AUSTRALIA; and
3
School of Human Movement and Exercise Science, The University of Western Australia, Perth, Western
Australia, AUSTRALIA
2

ABSTRACT
ABBISS C. R., M. J. QUOD, D. T. MARTIN, K. J. NETTO, K. NOSAKA, H. LEE, R. SURIANO, D. BISHOP, and P. B.
LAURSEN. Dynamic Pacing Strategies during the Cycle Phase of an Ironman Triathlon. Med. Sci. Sports Exerc.,Vol. 38, No. 4, pp.
726734, 2006. Introduction: A nonlinear dynamic systems model has previously been proposed to explain pacing strategies
employed during exercise. Purpose: This study was conducted to examine the pacing strategies used under varying conditions during
the cycle phase of an Ironman triathlon. Methods: The bicycles of six well-trained male triathletes were equipped with SRM power
meters set to record power output, cadence, speed, and heart rate. The flat, three-lap, out-and-back cycle course, coupled with
relatively consistent wind conditions (1730 kmIhj1), enabled comparisons to be made between three consecutive 60-km laps and
relative wind direction (headwind vs tailwind). Results: Participants finished the cycle phase (180 km) with consistently fast
performance times (5 h, 11 T 2 min; top 10% of all finishers). Average power output (239 T 25 to 203 T 20 W), cadence (89 T
6 to 82 T 8 rpm), and speed (36.5 T 0.8 to 33.1 T 0.8 kmIhj1) all significantly decreased with increasing number of laps (P G
0.05). These variables, however, were not significantly different between headwind and tailwind sections. The deviation (SD) in
power output and cadence did not change with increasing number of laps; however, the deviations in torque (6.8 T 1.6 and 5.8 T 1.3
NIm) and speed (2.1 T 0.5 and 1.6 T 0.3 kmIhj1) were significantly greater under headwind compared with tailwind conditions,
respectively. The median power frequency tended to be lower in headwind (0.0480 T 0.0083 Hz) compared with tailwind (0.0531 T
0.0101 Hz) sections. Conclusion: These data show evidence that a nonlinear dynamic pacing strategy is used by well-trained triathletes
throughout various segments and conditions of the Ironman cycle phase. Moreover, an increased variation in torque and speed was found
in the headwind versus the tailwind condition. Key Words: NONLINEAR COMPLEX SYSTEM, SRM, FATIGUE, WIND,
FOURIER TRANSFORMATION, QUADRANT ANALYSIS

event progresses (18,19). Currently, however, it is unclear

what specific physiological, cognitive, and environmental
factors may control the detailed distribution of work during
exercise performance (1,2). Consequently, little is known
about the optimal pacing strategies to delay the development of fatigue and ensure the best possible performance
outcome under a variety of conditions that may be
experienced by endurance athletes.
The results of prior research into exercise-induced
fatigue have suggested that exercise performance may be
directly influenced through a number of linear cause-andeffect models (1). For instance, it has been suggested that
fatigue may be caused by the attenuation of high internal
body temperatures (15,36), the depletion of muscle
glycogen stores (9), or possibly accumulation of metabolites, such as hydrogen (11) or potassium (8) ions. Within
these models, work or power output is dictated by the
inability of an organism to maintain homeostasis of an
individual physiological system (1). More recently, however, it has been suggested that exercise performance is
controlled in a more complex, nonlinear, and dynamic

n individuals pacing strategy is defined as the

distribution of work or energy expenditure that
occurs throughout an exercise task (6,10). Generally speaking, an even distribution of work or energy
throughout an endurance event (92 min) is considered to be
optimal (2,3,13,14,23,27). However, during self-paced
ultraendurance events, evidence suggests that athletes
may actually self-select a positive pacing strategy, whereby
the energy expenditure or work gradually decreases as the

Address for correspondence: Chris R. Abbiss, School of Exercise,

Biomedical and Health Sciences, Edith Cowan University, Building 19,
Room 127, 100 Joondalup Drive, Joondalup, WA, Australia 6027; E-mail:
c.abbiss@ecu.edu.au.
Submitted for publication September 2005.
Accepted for publication November 2005.
0195-9131/06/3804-0726/0
MEDICINE & SCIENCE IN SPORTS & EXERCISE
Copyright 2006 by the American College of Sports Medicine
DOI: 10.1249/01.mss.0000210202.33070.55

726
Copyright @ 2006 by the American College of Sports Medicine. Unauthorized reproduction of this article is prohibited.

manner to ensure that the body is able to maintain

homeostasis (17,32). Within this model, muscle activation
is constantly manipulated in response to intrinsic (i.e.,
physiological, biomechanical, and cognitive) and extrinsic
(i.e., environmental) sensory signals (17,32). Furthermore,
it is suggested that the end point of the exercise task offers a
reference point to which athletes continually calculate the
work required to complete the exercise task and, as a result,
continually alter the distribution of work accordingly (32).
Evidence for this model stems from the variant nature of
heart rate experienced during prolonged cycle racing (even
on relatively flat courses) (25,26,32). Through power
spectral analysis (fast and short-time Fourier analysis), it
has recently been shown that an increase in exercise load
can result in a shift in both time and frequency domains of
heart rate variability (28). Although such uses of heart rate
monitoring may give an accurate indication of physiological stress on the body (25), it has been suggested that the
ability of heart rate monitors (as well as rating of perceived
exertion) to measure minor changes in power output and
thus pacing are limited (5). To date, few researchers have
looked at the periodic variations and spectral analysis of
power output during prolonged cycling events and examined how these fluctuations may be related to pacing
strategy (5,34).
Previous research into pacing strategies used during
cycling events has been complicated by a number of external
factors, including wind direction, course geography, temperature, team work, and drafting (6,20,26). As a result, most
research has been performed in velodromes (10,24) or under
simulated environmental conditions (laboratory-based time
trials) (5,13,22,27). Constant environmental conditions and
flat-course profiles are extremely unlikely to occur during
an outdoor road cycling time trial (5,20). Atkinson and
Brunskill (5) examined the effects of a simulated headwind
(first 8.05 km) and tailwind (second 8.05 km) on selfselected and enforced (constant and variable (5% T mean))
pacing strategies during a 16.1-km laboratory-based time
trial. It was found that during the self-paced trial, participants rode the first 20% of the headwind section with a
power output of 267 T 56 W, which was 14% above the
overall trial mean (235 T 41 W) (5). The authors also found
that performance time was approximately 10 and 12 s faster
when pacing strategy was enforced in the constant and
variable trials, respectively (5). As a result, the authors
suggested that the cyclists self-selected pacing strategies
were suboptimal and affected by simulated variations in
environmental condition or wind direction. However, we are
unaware of any studies that have directly examined the
effects of altered environmental conditions (e.g., wind
direction) on the variations in power output during fieldbased research.
The aims of this study were therefore to (a) directly
examine the distribution of work during various stages of a
180-km field-based Ironman triathlon cycle phase, and (b)
determine the effect of wind direction on pacing strategy
and variations in power output. It was hypothesized that
through investigation of the patterns and possible variaPACING STRATEGIES DURING AN IRONMAN

tions in power output under these conditions, using

standard deviations (SD), discrete Fourier transformation,
and quadrant analysis, evidence for a dynamic complex
systems model would become apparent (32).

METHODS
Participants. Six well-trained male triathletes (x T SD:
age = 35 T 4 yr, mass = 78.8 T 6.5 kg, height = 179.4 T 5.5 cm,
O2peak=
sum of seven skinfolds = 51.4 T 12.8 mm, V
j1
j1
68 T 6 mLIkg Imin , maximal aerobic power = 405.7 T
21.9 W) were recruited 46 wk before the Ironman Western
Australia triathlon (Busselton, Western Australia) that took
place on November 28, 2004. Each of the participants had
previously completed an Ironman triathlon in less than
10 h 30 min. Before testing, all participants were informed
of the risks associated with the study and provided written
informed consent in accordance with the human research
ethical committees of The University of Western Australia
and Edith Cowan University.
Preliminary measurements. Two to three weeks
before the Ironman event, participants were measured for
height, mass, and the sum of seven skinfolds (triceps,
subscapular, biceps, supraspinale, abdominal, midthigh, and
calf) using calibrated Harpenden skinfold calipers (British
Indicators, England). Anthropometric measurements were
assessed by a Level 2 International Society for the Advancement of Kinanthropometry (ISAK) anthropometrist. Participants then performed a laboratory-based incremental
exercise test on a calibrated, wind-braked, cycle ergometer
(Evolution Pty. Ltd., Adelaide, Australia). The seat and
handlebar positions were adjusted to replicate the participants own habitual cycling position. The participants
used their own cycling shoes and pedal-cleat system. The
cycle ergometer had six gear ratios, and participants were
instructed to maintain their preferred cycling cadence
throughout the test. Power output was calculated using a
custom-made computer program (Cyclemax, School of
Human Movement and Exercise Science, The University of
Western Australia). Power output was determined at 0.2-s
increments and was displayed on a computer screen visible to
the participant. Throughout the duration of the test, oxygen
O2), ventilation, and respiratory exchange
consumption (V
ratio values were continuously recorded every 15 s using
Ametek gas analyzers (SOV S-3A and COV CD3A,
Pittsburgh, PA). The gas analyzers were calibrated immediately before and verified after each test, using three certified
gravimetric beta-grade gas mixtures (BOC Gases,
Chatswood, Australia). Throughout the tests, heart rate was
recorded every 5 s with a Polar Vantage heart rate monitor
(Polar Electro Oy, Kempele, Finland). Following a 10-min
warm-up (75 W), participants performed the incremental
exercise test. The test commenced at a workload of 100 W
for 4 min and increased 40 W every 4 min until volitional
O2peak was defined as the average of the
exhaustion. V
O2 values recorded during a 1-min
highest consecutive V
period, whereas maximal aerobic power (MAP) was defined
Medicine & Science in Sports & Exercised

Copyright @ 2006 by the American College of Sports Medicine. Unauthorized reproduction of this article is prohibited.

727

as the highest power output completed during the incremental test. If the participant finished partway through a
4-min stage, MAP was calculated in a pro rata manner. For
example, if a subject finished 2 min into the 300-W stage,
their MAP would be [(2 min/4 min]
40) + 300 = 320 W.
Ironman event. The 2004 Ironman Western Australia
Triathlon involved a 3.8-km swim, followed by a 180-km
cycle, and finally a 42.2-km marathon run. No
recommendations were given to participants as to quantities
and types of food and fluids that should be consumed during
their race, and fuel and fluid intake was ad libitum throughout
the event. However, all participants were experienced
triathletes, familiar with the requirements for adequate
ingestion of energy sources. Moreover, pre- and postrace
blood glucose concentrations (i-STAT Corporation, East
Windsor, NJ) were 7.7 T 1.7 and 6.4 T 1.4 mmolILj1
(P. Laursen, unpublished data, 2005), suggestive of the fact
that participants maintained euglycemia. Performance times
for the swim, cycle, and run phases for all triathletes were
retrieved from the race timing results that were posted
online following the event (http://www.ironmanwa.com).
Every 30 min, both ambient temperature and relative
humidity (rh) were recorded with a portable digital weather
tracker (Kestrel 4000, Nielsen-Kellerman, Australia). Wind
direction and speed were retrieved from the Bureau of
Meteorology (http://www.bmo.gov.au) following the event.
Both wind direction and speed were recorded from the
foreshore (Busselton Jetty) and taken as an average of the
10-min period before 9:00 a.m. and 3:00 p.m. Wind
direction was reported based on a 16-point compass.
Altitude was recorded every minute with the use of a Polar
heart rate monitor fitted to a subjects bike (s720i, Polar
Electro Oy, Kempele, Finland).
Before the event, each subjects own bicycle was fitted
with a calibrated SRM professional power meter (Schoberer Rad Mebtechnik, Germany). The SRM power meter
measured torque, cadence, and speed every 2 s (30 Hz)
throughout the cycle phase of the event. Heart rate was also
recorded every 2 s with the use of the SRM system and a
compatible chest electrode (Polar Electro Oy, Kempele,
Finland). One subject opted not to wear a chest electrode,
and thus heart rate data were available for only five of the
six participants. From the net torque per revolution and the
average velocity per revolution (cadence), the SRM system
calculates power output over an entire revolution. As a
result, power output is not an instantaneous measure, and
thus the power output reported does not contain artifact
caused by the instantaneous crank angle. From power
output and cadence reported by the SRM system, the net
torque over an entire revolution was calculated via the
following equation:
Net pedal torque power
60 = cadence
2
3 :1416

To perform quadrant analysis (A.R. Coggan, http://

home.earthlink.net/~acoggan/quadrantanalysis/) and to
be consistent with research into forcevelocity relationships,
the average effective pedal force over an entire revolution,
728

Official Journal of the American College of Sports Medicine

along with the circumferential pedal velocity, was calculated

based on the following equations:
Average effective pedal force power
60 =cadence
2
3:1416

crank length
Circumferential pedal velocity cadence
crank length
2
3:1416
 60

Each participants average effective pedal force and circumferential pedal velocity over the entire Ironman cycle
phase was used to separate force and velocity into quadrants
rather than forcevelocity at ventilation or lactate threshold.
This was done because subjects performed the cycle phase
of the event well below power output at lactate threshold
(71 T 5% LT).
The cycle phase of the Ironman triathlon consisted of
three flat (elevation change of 12.3 T 0.6 m per lap
recorded over three consecutive laps) 60-km laps, thus
allowing comparisons of pacing strategies used over
consecutive laps to be made. In addition, the specific
arrangement of the course and environmental conditions on
the day resulted in participants riding 10.9 km (from 11.45
to 22.35 km) of each lap with a near direct tailwind.
Following this, subjects performed a 180- turn and cycled
the same 10.9-km section of road (from 22.35 to 33.25 km)
into a headwind. This allowed for further comparisons to
be made on the influence of tailwinds and headwinds on
oscillatory pacing strategies during the cycle phase of the
event. Using customized software written in Labview
(V6.1; National Instruments Corporation, Austin, TX),
power output over the three consecutive laps and also
during the headwind and tailwind section of the road was
analyzed using discrete Fourier transformation to obtain a
power spectrum. The median frequency was then obtained
from the power spectrum.
Statistical analysis. The average and standard deviation (SD) of each dependent variable (power output,
cadence, torque, and speed) over each entire lap were
analyzed for significant effects (between laps) using a
one-way repeated-measures ANOVA. The SD of each
dependent variable (deviation) for each participant was
measured and offers an average of the oscillatory nature of
each dependent variable (power output, cadence, torque,
and speed). Median power output frequency was also
analyzed for significant effects (between laps) using a
one-way repeated-measure ANOVA. A two-way repeatedmeasure ANOVA, with two within-subject variables (lap
and wind direction) was used to determine significant
effects between lap and wind direction. Where a significant
effect was found between laps, the main effect was analyzed
using the least significant difference test for pairwise
comparisons. The least significant difference test was used
in the present study because of the overly conservative type
I error control of other post hoc comparisons. Least
significant difference does not correct for multiple comparisons resulting in an increased statistical power at the
risk of increased type I errors. All statistical tests were
conducted using SPSS version 10.0 (Chicago, IL), and data
http://www.acsm-msse.org

Copyright @ 2006 by the American College of Sports Medicine. Unauthorized reproduction of this article is prohibited.

TABLE 1. Average power output, cadence, net torque, speed, heart rate, performance
time, and median power frequency for successive laps.

Power (W)
Cadence (rpm)
Torque (NIm)
Speed (kgIhj1)
Heart rate (bpm)
Performance time (min)
Median power frequency
(Hz)

Lap 1

Lap 2

Lap 3

239 T 25
89 T 6
25.4 T 2.0
36.5 T 0.8
150 T 12
97.0 T 1.9
0.0606 T 0.0049

224 T 19
85 T 8
24.9 T 2.9
34.9 T 0.5
146 T 9
101.6 T 1.3
0.0624 T 0.0042

203 T 20*
82 T 8*
23.3 T 2.4
33.1 T 0.8
140 T 8
108.2 T 2.4
0.0605 T 0.004

* P G 0.05, vs lap 1; P G 0.01, vs lap 1.

are presented as means and standard deviations. For all

analyses, significance was accepted at P G 0.05.

RESULTS
Throughout the cycle phase of the Ironman, ambient
temperature ranged from 21 to 25-C (23.7 T 1.7-C), and
relative humidity ranged from 52 to 84% rh (65 T 11%).
Wind direction (and speed) were reported to be northwest
(17 kmIhj1) at 9:00 a.m., and had shifted to west southwest
(30 kmIhj1) by 3:00 p.m. Thus, wind direction tended to be
from a westerly direction throughout the event. During the
race, participants were cycling between the times of 7:00 a.m.
and 12:30 p.m., and during postrace interview they reported
the noticeable head- or tailwind present on the cycle course,
depending on their cycle direction (traveling east vs west).
Overall triathlon performance time was 589.3 T 8.4 min
(within the top 10% of all finishers) with swim, cycle, and
run times equating to 56.9 T 5.5, 310.7 T 2.3, and 221.7 T
10.3 min, respectively. The participants completed the cycle
phase of the Ironman at a power output of 221 T 18 W (55 T
4% of MAP), a speed of 34.8 T 0.4 kmIhj1, a cadence of 85 T
7 rpm, and a heart rate of 146 T 10 bpm (83 T 5% HRmax).
Mean lap time significantly increased (P G 0.001), and
average power output, cadence, and speed all significantly
declined with increasing number of laps (P G 0.05; Table 1).
Average net torque (P = 0.17) and heart rate (P = 0.06)
tended to decline with increasing number of laps, but these
differences were not statistically significant (Table 1). The
median power frequency was not significantly different between
laps (P = 0.58; Table 1), however, and was significantly lower
during the headwind (0.0480 T 0.0083 Hz) compared with the
tailwind (0.0531 T 0.0101 Hz; P G 0.05) sections. Table 2
shows the SD in power output, cadence, torque, and speed.
None of these variables were found to change significantly
between laps (Table 2). In the tailwind and headwind sections
(10.9-km windows), average power output, cadence, and speed
all significantly declined as the number of laps increased (P G
0.05), whereas average net torque did not (P = 0.13; Fig. 1).
TABLE 2. Deviation (SD) in power output, cadence, net torque, speed, and heart rate
during each lap.
Lap 1
Power (W)
Cadence (rpm)
Torque (NIm)
Speed (kmIhj1)
Heart rate (bpm)

60 T 11
12 T 2
6.8 T 1.4
2.5 T 0.2
5T1

Lap 2
59
13
7.0
2.1
3

T7
T3
T 1.1
T 0.3
T1

PACING STRATEGIES DURING AN IRONMAN

Lap 3
55 T 7
14 T 3
6.7 T 1.2
2.2 T 0.2
4T1

FIGURE 1Average power output, cadence, torque, speed, and heart

rate during the tailwind and headwind sections of each lap.
Medicine & Science in Sports & Exercised

Copyright @ 2006 by the American College of Sports Medicine. Unauthorized reproduction of this article is prohibited.

729

TABLE 3. Time spent in each quadrant over progressive laps.

Lap 1
Quadrant
Quadrant
Quadrant
Quadrant

I
II
III
IV

1376 T 470
255 T 134
167 T 92
1110 T 501

Lap 2
893 T
560 T
365 T
1231 T

149
250
161
322

Lap 3
392
774
931
1149

T
T
T
T

233*
263*
429*
334

* P G 0.05, vs lap 1.

Moreover, average power output, cadence, torque, and speed

were not significantly different between the tailwind and
headwind sections (Fig. 1). The deviation in power output,
cadence, torque, and speed during the tailwind and headwind
sections were not significantly different between laps (Fig. 2).
The deviation in torque and speed, however, was significantly
greater in the headwind compared with the tailwind section
(both P G 0.05; Fig. 2), whereas the deviation in power
output and cadence were not (P = 0.65 and 0.27,
respectively; Fig. 2). Quadrant analysis (A.R. Coggan,
http://home.earthlink.net/~acoggan/ quadrantanalysis/)
revealed that, in comparison with the first lap, significantly
less time was spent in quadrant I, and more time was spent
in quadrants II and III during the third lap (Table 3, Fig. 3).
During both the tailwind and headwind sections, power
output and heart rate oscillated in a seemingly random
manner, both above and below the mean during the relevant
sections of the race (Fig. 4). Within both sections, power
output ranged from 0 to 363 W and heart rate ranged from
138 to 159 bpm. The mean power output and heart rate were
231 T 43 vs 228 T 47 W, and 152 T 2 vs 150 T 4 bpm over the
tailwind versus the headwind sections, respectively (Fig. 4).
The variation in power output was significantly greater than
the variation in heart rate during both headwind and tailwind
sections (both P G 0.001; Fig. 4).

DISCUSSION

FIGURE 2Deviation (SD) in power output, cadence, torque, speed,

and heart rate during the tailwind and headwind sections of each lap.

730

Official Journal of the American College of Sports Medicine

This study examined the variations in power output and

pacing strategies used during the 180-km cycling phase of
an Ironman triathlon event. The main findings of this study
were as follows: (a) most participants used a positive
pacing strategy during the cycle phase of the Ironman
triathlon; (b) an increase in lap performance time was not
associated with an increase in the deviation of power
output, cadence, torque, and speed; (c) a greater variation
was seen in torque and speed into a headwind compared
with a tailwind; and (d) power output constantly oscillates
during the cycle phase of the Ironman triathlon in a
seemingly random manner.
It has been suggested that under stable external conditions (i.e., environmental and geographic), a constant
pace is optimal during both prolonged track and road
cycling time trials (5,6,13,14,33). Evidence for this is
shown in a study by Padilla et al. (24), which examined the
speed of a cyclist during a successful 1-h track world
record attempt. From this study, it can be seen that the
cyclist was able to maintain a steady velocity throughout
the entire trial. Over the duration of the trial, the speed of
http://www.acsm-msse.org

Copyright @ 2006 by the American College of Sports Medicine. Unauthorized reproduction of this article is prohibited.

the cyclist per lap deviated very little from the target speed
(53.0 kmIhj1) or the actual mean of 53.040 kmIhj1 (24).
The ability to separate the 180-km cycle phase of the
present study into three 60-km laps has allowed us to
maintain relatively consistent external conditions. We have
found that lap performance times significantly increased
as the trial progressed (Table 1). This finding indicates
that these participants self-selected a positive pacing strategy during the cycle phase of the Ironman triathlon. Similarly, a number of researchers have shown progressive
reductions in running speed (18) and heart rate (19) during
self-paced ultraendurance events. Whether or not the
overall performance times of the participants in the present
study may have been improved had athletes adopted a
more consistent speed per lap is unclear and requires
further investigation.
It should be noted, however, that the progressive
reduction in lap speed found in the present study may
have been influenced as a result of changes in environmental conditions as wind speed increased by 13 kmIhj1
from 9:00 a.m. to 3:00 p.m. (5,33). The increase in lap
performance time, however, was associated with a decrease
in power output and cadence (Fig. 1) and a shift in the
forcevelocity relationship from quadrant I to quadrant II
and III (Table 2, Fig. 3), suggesting that environmental
conditions were not solely responsible for the significant
reduction in speed. The shift in the forcevelocity

FIGURE 4Oscillatory patterns in power output and heart rate of

an individual subject during the tailwind and headwind sections of
lap 2 (a case study).
PACING STRATEGIES DURING AN IRONMAN

FIGURE 3Quadrant analysis of an individual subject over progressive laps. Quadrants were separated by the mean effective pedal force
and circumferential pedal velocity of that subject, over the entire
Ironman event.

relationship from high force/high velocity (quadrant I) to

high force/low velocity (quadrant II) and low force or low
velocity (quadrant III) indicates that during the cycle phase
of the Ironman, these triathletes commenced the cycle
phase at a workload or power output that could not be
sustained for the entire 180 km. Further, this fact may
suggest a progressive shift in motor unit recruitment
strategies. In particular, it is believed that the gradual shift
from quadrant I to quadrant III indicates a general reduction
in the recruitment of fast-twitch muscle fibers (A.R. Coggan,
http://home.earthlink.net/acoggan/ quadrantanalysis/).
An initial high power output is a characteristic commonly found during self-paced simulated cycling time
trials of varying distance or with cyclists of differing
performance levels (3,5,14,22,35). To our knowledge,
however, the present study is the first to show such a
self-selected pacing strategy during the cycle phase of an
Ironman triathlon. The adoption of this initially high power
output suggests that the pacing strategy used during an
exercise event is perhaps not predetermined before beginning
the exercise task, but is perhaps more dynamic in nature.
Thus, participants likely began the cycle phase of the Ironman
triathlon at an intensity that they felt they could withstand for
the duration of the event. Following the adoption of this
initial strategy, however, it is likely that complex calculations
and resulting adjustments were made based on feedback from
numerous afferent sensory signals and the end point of the
exercise task (end of the cycle phase as well as the completion
of the entire triathlon) (32). As a consequence, exercise
intensity was reduced to ensure the athlete could complete
the entire Ironman triathlon without any physiological
system deviating too far from homeostasis.
How participants determine this initial power output
is of interest to exercise physiologists and coaches because it has been found that if cyclists are given feedback
on their power output and are instructed to reduce their
starting pace, overall performance times are usually improved (14,22). The initial high power output reported here
may be caused by (a) prior experiences in shorter time
trials or the lack of experience at the Ironman triathlon
Medicine & Science in Sports & Exercised

Copyright @ 2006 by the American College of Sports Medicine. Unauthorized reproduction of this article is prohibited.

731

distance, (b) lower levels of fatigue-related afferent sensory feedback at the beginning of the trial (i.e., pain,
perceived exertion, muscle glycogen content), and (c) an
initial high anxiety level or excitement when competing in
the event.
Interestingly, the reductions in power output, cadence,
and speed over the three laps in this study did not follow
significant increases in the deviation in any measured
variables (e.g., torque, speed, power). Based on classic
cause-and-effect models, it was hypothesized that as the
trial progressed, afferent sensory signals related to fatigue
(i.e., reduced muscle glycogen content, perceived pain, and
changes in neuromuscular recruitment) (1) would have
induced a greater influence on the pacing strategies used,
even in these well-trained triathletes. Consequently, it was
proposed that as the trial progressed, participants would
find it more and more difficult to maintain a steady or
optimal pace, resulting in a progressively greater variation
in power output, cadence, torque, and speed; however, this
was not found (Table 2). Although power output significantly declined (Table 1), no significant increase occurred
in the deviation of these variables with increased cycling
distance (Table 2). Further, as the event progressed (laps),
no significant change was noted in the median power
frequency (Table 1), suggesting that no considerable shift
occurred in the distribution (i.e., amplitude and frequency)
of power output. This may suggest that the reductions in
power output were not directly caused by the inability to
maintain physiological homeostasis. Instead, the decrease
in power, without a significant increase in the variation,
may indicate that the reduction in power output was
controlled in a more dynamic, nonlinear manner.
Although the above information may suggest that a
constant power output applied to the pedals would provide
the optimal pacing strategy during the entire cycle phase of
the Ironman, the determination of the best possible pacing
strategy is complicated by the effects of external conditions
such as wind. As mentioned, evidence for this comes from
a study performed by Atkinson and Brunskill (5), which
examined the effects of a simulated 8.05-kmIhj1 headwind
and tailwind over a 16.1-km simulated time trial. It was
found that the optimal pacing strategy involved cyclists
increasing power output into the headwind (5% above
mean), followed by a reduction in power output with a
tailwind (5% below mean). This pacing strategy induces
faster performance times primarily because of the greater
time that is spent in the headwind compared with the
tailwind section. Thus, in comparison with a constant
power output, a lower power output during a tailwind
section but with a greater power output during a headwind
section results in an improved overall performance time
(5,7,33). Field-based support for this comes from a study
by Atkinson and Edwards (7), which showed that during
the 1997 British womens time trial championship, cyclists
who road the first, wind-assisted half of a 16-km time trial
at a relatively slower pace, but then increased pace into a
headwind, produced the fastest overall performance times.
However, a limitation of the aforementioned study is that
732

Official Journal of the American College of Sports Medicine

power output was not measured, and thus variations in

speed may have been the result of altered aerodynamic
positioning and not direct variations in power output or
pacing strategy. In addition, laboratory-based research into
the effects of wind on performance has been limited by the
inaccurate replication of the wind (no use of fans), which
may induce a psychological influence on exercise performance. Although it has been found that wind can affect
exercise performance in the heat (4), we are unaware of
any studies that have looked specifically at the effect of
convective cooling on pacing strategies; however, increased convective cooling has recently been shown to
have a profound effect on exercise performance (30).
In the present study, it can be seen that, although average
power output, cadence, and speed significantly declined as
the cycle phase progressed (Fig. 1), no significant
difference was found in these variables between the
headwind and tailwind sections. This suggests that participants self-selected a suboptimal pacing strategy and may
have benefited from a tactic whereby power output was
increased into the wind but reduced during the tailwind
section. The fact that participants maintained a constant
power output over both headwind and tailwind sections,
however, lends support for a recent dynamic pacing
hypothesis (32) in that any variation in power output
(greater power into the headwind) would have likely
resulted in greater physiological and psychological strain
and a consequent reduction in power to maintain a
comfortable exercise intensity or an intensity that
allows homeostasis to be maintained. In the present study,
it was hypothesized that as participants rode into a
headwind there would be a significant decrease in cadence
accompanied by the reduction in speed, as found during
high mountain ascents. This was not the case. It is likely that
these differences in cadence between uphill cycling and
riding into a headwind result from different riding technique (i.e., standing) while riding uphill compared with a
more consistent time trial position during flat individual time
trials (12).
Interestingly, although the variation in power and
cadence was not significantly different between headwind
and tailwind sections, the deviation in speed was significantly greater into the headwind compared with the
tailwind sections (Fig. 2). As the deviation in power output
remained constant, it is likely that the greater deviation in
speed into a headwind was caused by greater air resistance
or aerodynamic drag (6). It should be noted, however, that
because neither average power output nor average speed
was significantly different between headwind and tailwind
sections, the possibility that participants adopted a more
aerodynamic position into a headwind to counteract the
increased resistive force (wind resistance) cannot be ruled
out. Interestingly, the deviation in torque was also
significantly greater in the headwind compared with the
tailwind section. It is hypothesized that this variation in
torque was the result of the participants altered riding style
as participants attempted to maintain both a constant
cadence and speed, despite the deviation in speed caused
http://www.acsm-msse.org

Copyright @ 2006 by the American College of Sports Medicine. Unauthorized reproduction of this article is prohibited.

by the increased aerodynamic drag (deceleration). In fact,

Sanderson and Black (29) found that both the peak propulsive
torque and the peak retarding torque (from three cycle
revolutions) significantly increased following a cycling test
to exhaustion (at 80% peak power output). No significant
change was noted, however, in net angular torque (29).
A further finding in the present study was that, despite
similar average power outputs in the headwind and tailwind
sections, the median power frequency was significantly
greater during the tailwind section. This shift in the power
spectrum suggests that the participants altered the distribution
of power output (i.e., the amplitude, frequency of power, or
both) during headwind and tailwind sections. Whether this
shift in the power spectra is evidence of altered centrally
controlled pacing strategies (34) or, possibly, the result of
participants altering their riding style in an attempt to
maintain a constant speed, despite differences in aerodynamic drag, requires further investigation. In particular, the
discrete Fourier transformation and determination of median
frequency, as used in the present study, only indicates a
change in the distribution of power output and is not
sensitive to variations in amplitude, frequency, and time. It
is therefore evident that future research into pacing and the
distribution of power during cycling should incorporate
more in-depth analytical methods such as amplitude
probability distribution function analysis (16) or exposure
variation analysis, as used in electromyography analysis
(21). Such analytical techniques would allow us to determine
whether possible shifts in the distribution of power as seen in
the present study are the result of variations in the amplitude,
frequency, or time spent in varying power output bands.
Historically, when examining power output and the
optimal pacing strategies used during cycling, researchers
have assessed the mean value of certain variables over a
relatively long distance or large time unit (i.e., average
power output per lap) (31). With the development of the
SRM power meter, performance can be monitored at much
shorter time increments (average of every 1 s). This then
allows for a more sensitive determination of the fluctuations in power production that occur over an entire event.
From this, additional support for the involvement of a
nonlinear dynamic process occurring during the cycle
phase of the Ironman can be seen (Fig. 4). To our
knowledge, the present study is the first to directly show
the oscillatory patterns of power output that occur during a
prolonged cycling event and their involvement with
dynamic pacing strategies. As mentioned, the optimal
pacing strategy during prolonged exercise is thought to be
one in which a constant or steady power output is applied
throughout the duration of the event (5,6,13,14,33). From
Figure 4, however, it can be seen that throughout both
headwind and tailwind sections, power output increases
and decreases in a seemingly random manner. In fact,
throughout these two sections, the profiles of power output
in this participant ranged from 0 to 363 W and can be seen
to contain multiple peaks and troughs (Fig. 4). Because the
cycle course was relatively flat (elevation change of 12.3 T
0.6 m per lap) and it is unlikely that wind direction
PACING STRATEGIES DURING AN IRONMAN

deviated significantly throughout these sections, the variation in power output would not solely be related to
environmental and geographical factors. Further, the cycle
phase of the Ironman triathlon is essentially an individual
time trial event, free from drafting behind competitors or
team members wheels. Thus, the stochastic nature of
power output and heart rate seen in the present study are
unrelated to drafting or group dynamics. We therefore
believe that the oscillatory variations in power output
shown in the present study indicate that multiple inputs are
responsible for the control of power output during the cycle
phase of the Ironman event, and thus provide evidence for
the interrelationship of the numerous linear physiological
fatigue systems reacting in a complex nonlinear pattern
(32,34). Within this complex systems model, it is believed
that multiple peripheral physiological systems provide
feedback to the brain, resulting in a dynamic, nonlinear
control of power to maintain homeostatic control of each
physiological system (17). The continual oscillatory variations in power output, therefore, may indicate a delay
between afferent signaling and efferent control of power
output (17). Interestingly, the fluctuations in power output
were significantly more variable than the oscillations in heart
rate shown over the same time period (P G 0.05; Fig. 4). This
then supports the premise that, although the measurement of
heart rate may indicate the physiological stress, current
portable heart rate monitors are not adequately sensitive to
show minor fluctuations in power output and, thus, pacing
(5). As a result, future studies should address the cyclic oscillatory patterns of power output during cycling and their influence
on pacing during field cycling performance. In addition, research
should examine the effects of more or less oscillation (variation)
on pacing and overall exercise performance.
In conclusion, the present study has shown that, although
these well-trained participants finished within the top 10%
of the Ironman Western Australia event, power output and
speed declined with increasing distance, suggesting that
pacing strategies during the cycle phase of the event may
have been suboptimal. As a result, participants may have
benefited from the use of a more constant speed over the
duration of the cycle phase of the event. Further, this study
has shown that average power output, cadence, net torque,
and speed were not significantly different between headwind and tailwind sections. The deviation (variation) in
speed and torque, however, were significantly greater into
a headwind, which was most likely caused by an increase
in aerodynamic drag. Finally, this study has provided some
evidence to support the hypothesis that pacing during
cycling is controlled through continual dynamic regulation strategies that concomitantly attempt to maintain
homeostasis while completing known amounts of work as
fast as possible.

The enthusiastic participation of each subject in this study is

greatly appreciated. The authors would also like to thank the
organizers and sponsors of the 2004 Ironman Western Australia
event. This study was supported by an Edith Cowan University,
Faculty of Computing and Health Science Small Grant (#2147620).
Medicine & Science in Sports & Exercised

Copyright @ 2006 by the American College of Sports Medicine. Unauthorized reproduction of this article is prohibited.

733

REFERENCES
1. ABBISS, C. R., and P. B. LAURSEN. Models to explain fatigue during
prolonged endurance cycling. Sports Med. 35:865898, 2005.
2. ALBERTUS, Y., R. TUCKER, A. S. C. GIBSON, E. V. LAMBERT, D. B.
HAMPSON, and T. D. NOAKES. Effect of distance feedback on pacing
strategy and perceived exertion during cycling. Med. Sci. Sports
Exerc. 37:461468, 2005.
3. ANSLEY, L., E. SCHABORT, A. ST. CLAIR GIBSON, M. LAMBERT, and
T. D. NOAKES. Regulation of pacing strategies during successive
4-km time trials. Med. Sci. Sports Exerc. 36:18191825, 2004.
4. ARMADA-DA-SILVA, P. A., J. WOODS, and D. A. JONES. The effect of
passive heating and face cooling on perceived exertion during
exercise in the heat. Eur. J. Appl. Physiol. 91:563571, 2004.
5. ATKINSON, G., and A. BRUNSKILL. Pacing strategies during a cycling
time trial with simulated headwinds and tailwinds. Ergonomics
43:14491460, 2000.
6. ATKINSON, G., R. DAVISON, A. JEUKENDRUP,and L. PASSFIELD. Science
and cycling: current knowledge and future directions for research.
J. Sports Sci. 21:767787, 2003.
7. ATKINSON, G.,and B. EDWARDS. Pacing strategy and cycling
performance: field data from the 1997 British 16 km time-trial
championship. In: Proceedings of the Third Annual Congress of
the European College of Sports Science, Liverpool: Centre for
Health Care Development, 1998, p. 211.
8. BANGSBO, J., K. MADSEN, B. KIENS, and E. A. RICHTER. Effect of
muscle acidity on muscle metabolism and fatigue during intense
exercise in man. J. Physiol. 495:587596, 1996.
9. CALLOW, M., A. MORTON, and M. GUPPY. Marathon fatigue: the role
of plasma fatty acids, muscle glycogen and blood glucose. Eur. J.
Appl. Physiol. Occup. Physiol. 55:654661, 1986.
10. DE KONING, J. J., M. F. BOBBERT, and C. FOSTER. Determination of
optimal pacing strategy in track cycling with an energy flow
model. J. Sci. Med. Sport 2:266277, 1999.
11. FABIATO, A., and F. FABIATO. Effects of pH on the myofilaments
and the sarcoplasmic reticulum of skinned cells from cardiac and
skeletal muscles. J. Physiol. 276:233255, 1978.
12. FARIA, E. W., D. L. PARKER, and I. E. FARIA. The science of
cycling: factors affecting performancepart 2. Sports Med.
35:313337, 2005.
13. FOSTER, C., J. J. DEKONING, F. HETTINGA, et al. Effect of competitive
distance on energy expenditure during simulated competition. Int.
J. Sports Med. 25:198204, 2004.
14. FOSTER, C., A. C. SNYDER, N. N. THOMPSON, M. A. GREEN, M.
FOLEY, and M. SCHRAGER. Effect of pacing strategy on cycle
time trial performance. Med. Sci. Sports Exerc. 25:383388,
1993.
15. GONZALEZ-ALONSO, J., C. TELLER, S. L. ANDERSEN, F. B. JENSEN,
T. HYLDIG, and B. NIELSEN. Influence of body temperature on the
development of fatigue during prolonged exercise in the heat.
J. Appl. Physiol. 86:10321039, 1999.
16. JONSSON, B. Quantitative electromyographic evaluation of muscular
load during work. Scand. J. Rehabil. Med. 6:6974, 1978.
17. LAMBERT, M., A. ST CLAIR GIBSON, and T. D. NOAKES. Complex
systems model of fatigue: integrative homoestatic control of
peripheral physiological systems during exercise in humans. Br.
J. Sports Med. 39:5262, 2005.
18. LAMBERT, M. I., J. P. DUGAS, M. C. KIRKMAN, G. G. MOKONE, and
M. R. WALDECK. Changes in running speeds in a 100 km ultramarathon race. J. Sports Sci. Med. 3:167173, 2004.
19. LAURSEN, P. B., E. C. RHODES, R. H. LANGILL, D. C. MCKENZIE, and

734

Official Journal of the American College of Sports Medicine

20.

21.

22.

23.

24.

25.

26.

27.

28.

29.
30.

31.

32.

33.

34.

35.

36.

J. E. TAUNTON. Relationship of exercise test variables to cycling

performance in an Ironman triathlon. Eur. J. Appl. Physiol.
87:433440, 2002.
LIEDL, M. A., D. P. SWAIN, and J. D. BRANCH. Physiological effects
of constant versus variable power during endurance cycling. Med.
Sci. Sports Exerc. 31:14721477, 1999.
MATHIASSEN, S. E., and J. WINKEL. Quantifying variation in
physical load using exposure-vs-time data. Ergonomics 34:
14551468, 1991.
MATTERN, C. O., R. W. KENEFICK, R. KERTZER, and T. J. QUINN.
Impact of starting strategy on cycling performance. Int. J. Sports
Med. 22:350355, 2001.
NIKOLOPOULOS, V., M. J. ARKINSTALL, and J. A. HAWLEY. Pacing
strategy in simulated cycle time-trials is based on perceived rather
than actual distance. J. Sci. Med. Sport 4:212219, 2001.
PADILLA, S., I. MUJIKA, F. ANGULO,and J. J. GOIRIENA. Scientific
approach to the 1h cycling world record: a case study. J. Appl.
Physiol. 89:15221527, 2000.
PADILLA, S., I. MUJIKA, J. ORBANANOS, and F. ANGULO. Exercise
intensity during competition time trials in professional road
cycling. Med. Sci. Sports Exerc. 32:850856, 2000.
PALMER, G. S., J. A. HAWLEY, S. C. DENNIS, and T. D. NOAKES.
Heart rate responses during a 4-d cycle stage race. Med. Sci. Sports
Exerc. 26:12781283, 1994.
PATERSON, S., and F. E. MARINO. Effects of deception of distance
on prolonged cycling performance. Percept. Motor Skill 98:
10171026, 2004.
PICHON, A. P., C. DE BISSCHOP, M. ROULAUD, A. DENJEAN, and Y.
PAPELIER. Spectral analysis of heart rate variability during
exercise in trained subjects. Med. Sci. Sports Exerc. 36:1702
1708, 2004.
SANDERSON, D. J., and A. BLACK. The effect of prolonged cycling
on pedal forces. J. Sports Sci. 21:191199, 2003.
SAUNDERS, A. G., J. P. DUGAS, R. TUCKER, M. I. LAMBERT, and T. D.
NOAKES. The effects of different air velocities on heat storage and
body temperature in humans cycling in a hot, humid environment.
Acta Physiol. Scand. 183:241255, 2005.
SMITH, M. F., R. C. DAVISON, J. BALMER, and S. R. BIRD.
Reliability of mean power recorded during indoor and outdoor
self-paced 40 km cycling time-trials. Int. J. Sports Med.
22:270274, 2001.
ST. CLAIR GIBSON, A., and T. D. NOAKES. Evidence for complex
system integration and dynamic neural regulation of skeletal
muscle recruitment during exercise in humans. Br. J. Sports Med.
38:797806, 2004.
SWAIN, D. P. A model for optimizing cycling performance by
varying power on hills and in wind. Med. Sci. Sports Exerc.
29:11041108, 1997.
TUCKER, R., C. BOLGER, A. BESTER, C. VAUGHAN, T. D. NOAKES, and
A. ST. CLAIR GIBSON. A novel mathematical model of pacing
strategy during self-paced exercise. Med. Sci. Sports Exerc.
37:S100S101, 2005.
TUCKER, R., L. RAUCH, Y. X. R. HARLEY, and T. D. NOAKES.
Impaired exercise performance in the heat is associated with an
anticipatory reduction in skeletal muscle recruitment. Pflugers
Arch. 448:422430, 2004.
WALTERS, T. J., K. L. RYAN, L. M. TATE, and P. A. MASON. Exercise
in the heat is limited by a critical internal temperature. J. Appl.
Physiol. 89:799806, 2000.

http://www.acsm-msse.org

Copyright @ 2006 by the American College of Sports Medicine. Unauthorized reproduction of this article is prohibited.