thinking about coaching? Get stronger by training with sms

Surprised by Chris Froome’s downhill victory? There may be thought (science) behind it!

-By: Tomas Swift-Metcalfe

Last modified: July 9, 2016

Pacing strategy and performance modelling in cycling… MIGHT explain that tactic.

I did my dissertation for my degree this winter on the topic of pacing strategy and performance modelling based on the Critical Power model. The subject was rather complex and as such I consider it a work in progress, since I barely scratched the surface of the subject. In fact, it’s been shelved for the time being and I wont get back to it as an academic, but possibly through some other realm (programming). I thought it interesting to see Chris Froome attacking on the descent and the tactic fit the logic underlying my hypothesis (in this case, ‘sprint’ to critical velocity, then get aero). Part of the story is in the literature review, second part in the actual write of the experiment I did.

Please ignore any editorial errors, I was working and training while all this was being done and isolated 2000 km away from the university, not to mention problems with the .docx closed source format (sometimes okay is good enough and being a non-conformist has it’s price!). Although I got a reasonable grade, I was horrified to see formatting error with formulas and such months past the date. For any future research projects I will use a version control system like GIT and publish everything so that others may build on it if they wish.

A lot of the logic and the maths used can be found in the initial literature review:

Optimisation of a variable pacing strategy in road cycling

In a nutshell, as air resistance builds exponentially, get aero on the down hill, as gravity is a constant, press on on the up hills. “Contextual Sprints” *might* provide a benefit based on the experiment. In experience as a pro cyclist for 6 year, I would say definitely, if a cyclist is skilful enough and strong enough and controls recovery periods adequately. I remember a few times with team mate Samuel Caldeira often ridding off the front of the peloton while controlling race on descents by adopting a similar stratagem. It follows logic from an earlier post that got pooped (with no sort of valid retort but an argument to authority) by a coach published in a peripheral academic journal, but that’s where the it started. As I mentioned this is a work in progress and I am not sure I will bother finishing it, but criticism is welcome. Anyway, the academic piece is bellow:

Development of a new model for individualized pacing strategy in non-drafting cycling events in hilly terrain.


Sprint efforts in specific contexts, “contextual sprint efforts” (CsEs), were investigated as potential modification to pacing strategy in hilly cycling time trials (TT). It was hypothesised that CsEs would improve average velocity. A 9189 m course with 149 m of ascent was chosen to test the effect of CsEs on average velocity, average power, power variability, perceived exertion and average heart rate. 15 participants, mean critical power (CP) (240 ∓ 55 W) performed two tests involving two 30 s efforts with 2 min recovery, two 2 min efforts with 4 min recovery and one 9189 m TT. The first (InitialTT) being a maximal self-paced effort, the second (InstructedTT) involving a written cue suggesting CsEs to quickly reach contextual critical velocity (CCV) given a change in topography, but otherwise self-paced. Between tests no statistical difference was found for: average velocity  (P = 0.2263), average power (P = 0.5214), rating of perceived exertion (RPE) (P = 0.08041) or average heart rate (P = 0.2192). CsE occurrence was significantly different (P =0.02139). External variables were controlled and not significantly different between conditions. These results indicate CsEs did not significantly affect finishing average velocity. A secondary objective tested certain factors against final average velocity for a linear relationship: recovery ratio between 30 s (30sRR) and 120 s (120sRR) efforts, mean 30 s power (30sW) and 120 s power (120sW) as well as critical power (CP). A strong association was found for CP (r = 0.7513) and 30sW (R2 = 0.7137) and a moderate association was found for 120sW (R2 = 0.6278). No linear association was found for the remaining factors. Multiple linear regression using mean 30 s power (30sW) and critical power (CP) had a strong association with outcome (adjusted R2  = 0.8994). This study indicates a test involving 30sW and CP together has a strong association with outcome in hilly TTs.


Eventual race performance and relative success can be thought of as a product of physical capacities and the application of these capacities in the environmental context. These factors can be quantified through performance modelling and pacing strategy. In this study, pacing strategy, is defined as how an athlete distributes power (W) during a TT. An optimal pacing strategy is defined as the pacing strategy that results in the highest average velocity for a cyclist on a given course.


Abbiss and Laursen (2008) reviewed the literature on pacing strategies in sport and note six different strategies: negative, all-out, positive, even, parabolic-shaped and variable pacing. Most of the strategies that have been examined in the literature do so in the context of stable or high controlled external variables, including topography. Atkinson et al. (2007) use simulated changes in topography in a laboratory, however it is not known whether these simulation reflect changes in resistance to motion given gradient or air resistance. Neither research observing variable pacing in race performances (Johnson et al., 2013) and laboratory based studies (Atkinson et al., 2007) consider aspects relating to contextual sprints to get up to speed and the contribution of the ATP-PCr metabolic pathway to effort, but rather focus on sustained efforts either side of steady state pace.


Atkinson et al. (2007) varied effort by 5% W either side of the mean value established in a 50-60 min TT. This variation in intensity alone led to an improvement in time to completion of a set amount of work (800 kJ) in 5 of their 7 participants. The researchers did report to which point in the effort the protocol was fully tolerated by the remaining 2 participants and over all results were not better for the group as a whole.


Abiss et al. (2006) looked at pacing strategy in well-trained triathletes taking part in an Ironman event. The triathletes adopted a dynamic pacing strategy given context, which in this case was dominated by wind with athletes increasing effort into the wind, hence to some degree experienced athletes may vary effort in certain context to greater effect, suggesting a variable and dynamic approach may be prefered by cyclists naturally.


An even distribution of effort is commonly thought to be optimal (Thomas et al., 2012), however in road cycling, rarely are conditions found that remain constant and as such an even pacing strategy would be difficult to apply. Thomas et al. (2012) report increases physiological perturbation and perception of effort given a variable effort with work rates set at 72% and 142% of the average power of a self paced trial in a 1:1.5 ratio. For this study it was hypothesised that physiological and psychological perturbations could be minimised by only suggesting brief sprints at key points on a course with the rest of the effort being self-paced. It was hypothesised that sprint and short anaerobic glycolytic efforts held greater weight of importance than typically afforded by current performance models and pacing strategies contemplate where short sprints are not accounted and efforts above CP tend to be sustained high intensity efforts over relatively long periods of time, in the case of Thomas et al. (2012) ~60 s and longer in other studies utilizing less variation (Atkinson et al., 2007) which had cyclist perform 5% above self-paced average power till producing 100 kJ, or ~5-10 min depending the efficiency and sustainable power of the participant.    


Key to cycling performance is power development given topographical context and that high intensity efforts may be of a relevance not previously regarded. To maximise performance, CCV or sustainable velocity for a given context, must be regarded for each context and sprint effort may be useful in reaching CCV more quickly and for less physiological perturbation. Brickley et al. (2007) state that between variable efforts and steady effort if net energy cost remains similar, physiological perturbation remains similar supporting the noting rapid accelerations and changes to output given a change in context are achievable.


Both mathematical models (Swain, 1997) and laboratory based research (Atkinson et al., 2007) demonstrate an improvement for adopting variable pacing for specific individuals. However other studies, both mathematical (Padilla et al., 2000) and experimental (Thomas et al., 2012) do show a benefit for even pacing, in static or highly controlled environments such as on laboratory ergometers or in velodromes. Hence may not apply to exterior TTs. The variable power output strategy postulated by Swain (1997) theoretically resulted in a quicker time to completion of a TT. Similar results were obtained independently using the equations presented by the model of road cycling power by Martin et al. (1998) using the concept of CsE. For the power variation suggested by Swain (1997), however it seems unlikely many cyclists could sustain the suggested variation of 10% above sustainable VO2 given later findings by Atkinson et al (2007) where certain participants among their sample were not able to sustain a variation of 5% yet alone the 10% variability suggested by Swain (1997). In terms of variation of effort this study investigates only a suggested increase in CsEs, which appear tolerable even under extreme stress given the finding by Marcora and Staiano (2010) and observations by Menaspà et al. (2015) where participants were able to develop high intensity efforts even after prolonged high intensity exercise.


To know to what degree a cyclist can sustain a variable pacing strategy, quantification of sustainable power and anaerobic power is necessary. Laboratory measures such as lactate threshold testing, VO2max or CP can detect events analogous with sustainable power and measure anaerobic work capacity (AWC). The CP concept, originally conceived by Monod and Scherrer (1965), tested in the field can produce reliable and valid results (Karsten et al., 2014) in one day. One of the typical limitations of CP is that it is usually tested over several days. Karsten et al. (2014) however use an extensive recovery period between bouts of exercise (30 min) which may not be practical in terms of both time and managing the participant. Other research claims the reliability and validity of a single 3 min bout to test CP (Vanhatalo et al., 2007), with the final 45 s equivalent to CP implying AWC is exhausted in ~02:15 min. However these measures of AWC and yet alone a single maximal 3 min bout of exercise still do not distinguish sprint efforts relying on the ATP-PCr system from effort relying on anaerobic glycolysis. Other commonly used performance measures such as Functional Threshold Power (FTP) (Allen and Coggan, 2006) do not consider AWC at all aside a 5% correction on maximum average 20 min W. Miller et al. (2014) showed FTP was able to predict race performance in cross-country mountain biking (R2 = 0.736), while not exactly the same a hilly TT among, this was the only example of the predictive capacity of FTP in a variable cycling event. Likewise CP predicted performance with greater validity (R2 = 0.943) in cross-country mountain biking (Miller and MacDermid, 2015) e another study. This experiment set out to examine whether 30 s effort and 2 min, corresponding to effort which are 73% and 37% anaerobic (∓10%) (Gastin, 2001) could further build on CP to predict performance in variable TTs and whether CsEs in pacing on variable courses were of any significant relevance to outcome.


All descriptive results are presented as mean (∓SD).


15 participants (female n = 2, male n = 13) took part in the experiment. Participants had an age of 33.1 (∓ 4.8) years and a mean weight of 67.53 (∓ 6.43) kg. Participants were recruited from triathlon and cycling clubs. This number of participants was chosen based on observation of sample populations used in similarly designed experiments reviewed previously and subsequent to a power analysis for the paired t-test (power = 80%, n = 9.9).


All participants were between the ages of 26 and 40, amateur and federated in national federations for either cycling or triathlon. Participants were informed in their native language about what the study would entail, what information was collected and how it would be used. All participants had been subject to a sports medical exam in the past year and all participants provided signed informed consent. The study was approved by the Ethics Committee of Manchester Metropolitan University on 16th October 2015.


Controlled variables were recorded before each of the two tests. Both tests followed the procedure detailed below:


  • Two 30 s maximal effort with 2 min active recovery,
  • Two 2 min maximal effort, 4 min active recovery,
  • One 9189 m TT on a hilly course.


Both 30 s and 2 min efforts took place on a slightly rising (1.3%) road with an even gradient. Recovery involved cycling slowly back to the start. The large lap involved a hilly circuit with and 149 m of ascent and 146 m of descent with start and finish occurring at slightly different places the course was measured at 9189 m. The course consisted of 3.6 km of climbing, 2.3 km flat and 3.3 km downhill, the course had a minimum altitude of 85 m and a maximum of 149 m.


Throughout both tests the following variables were recorded: heart rate (b.p.m.) and GPS location (1 hz sampling, error ∓ 8.7 m) were recorded using a Garmin 910XT (Garmin International Inc., Olathe, Kansas, USA). Velocity (m.s-1) and altitude was inferred by the Garmin 910XT from position and time data. GPS derived altitude data presents with an error 5-10% (Menaspà et al., 2014) hence was corrected using altitude correction application Strava (Strava Inc, San Francisco, CA, USA) which uses a resolution of 30 m against a database of survey values providing consistent measurements of altitude. W was directly recorded using a PowerTap (Saris Cycling Group, Inc., Fitchburg, Wisconsin, USA). The PowerTap was found by Bertucci et al. (2005) to offer a degree of validity and reliability to render it suitable for submaximal laboratory tests presenting to a coefficient of variation of 0.9 to 2.9 %. For sprint efforts PowerTap could report as much as 8% less power if a low cadence was used (Bertucci et al., 2005). Gardner et al. (2004) also found the PowerTap to be valid and reliable with a mean error of -2.7 ∓0.1% over an 11 month period. Gardner et al. (2004) also mention the PowerTap can be significantly affected by large temperature differences (8-21 ºC). The PowerTap was calibrated prior to each test lap. Cadence (r.p.m.) was indirectly inferred in real time by the PowerTap from variations in power delivery. Participants used their own bicycles for the test with the PowerTap fitted.


After each test participants rated their effort on the 6-20 Borg scale (Borg, 1998), between 1 and 2 min after completing the TT.


Before each test, participants were asked to abstain from heavy exercise in the three days prior and replicate their diet and training from the first test, for the second test. Tests took place within the same hour for each test and the InstructedTT took place between 3 and 7 days after the InitialTT.


Before InstructedTT, participants were given the following instructions in writing in their native language to follow and explained further verbal when necessary:


“When you reach near the end of a climb and about to begin a descent, don’t sit up, rather try and accelerate into the down hill to gain momentum and bring you up to speed quickly. For the climbs and remaining parts of the course, pace them as you would normally.”

Controlled variables

Before each test participants were asked to keep a diet and exercise log for the three days before and replicate these behaviour in the second test.


Weather data (temperature, wind, wind direction, air pressure, humidity, precipitation) was obtained online from a meteorological station at the site (Autódromo Internacional do Algarve) and recorded.


Body weight (including clothing, excluding shoes and helmet) was recorded with a BC-1500 Ironman® scale (Tanita Corporation of America, Inc., Clearbrook, Illinois, USA) Equipment weight (bicycle, shoes, bottles and helmet) was measured by standing on the scales with the full equipment and subtracting this from body weight measured previously.


While this experiment does not look at specifically psychological aspects of pacing, mood state was controlled using a Portuguese adaptation of the POMS (Viana et al., 2001). Given findings by Marcora et al. (2009) which detail changes in performance for mentally tired participants it was considered necessary to control psychological aspects. After each test, participants were also asked to report Borg RPE (6-20).


After each test participants were asked to inform of any mechanical or traffic incidents.

Data Analysis

Raw data was converted to comma separated values for analysis. Mean (∓ SD) values for W, HR, speed, RPE were calculated and presented for both TTs. From the W data collected, recovery ratios, variability using weighted power (WP), CsEs and CP were calculated and presented as mean (∓ SD) for both conditions. For paired samples, once all other assumptions of normality were met, normality was tested using the Shapiro-Wilks test and visual inspection using the boxplot method. Subsequent to the data being assumed normal, paired t-tests were conducted on each of the aforementioned variables between conditions, where data was assumed not normal the Wilcoxon signed rank test was used. Statistical significance was set at P < 0.05.


Correlations between performance (average speed) and potential predictors of performance (energy expenditure, efficiency, 30sW, 120sW, 30sRR, 120sRR and CP) were tested using linear regression and multiple linear regression. Multiple linear regression was used on 30sW and CP and the data was not further fitted using polynomial functions.


Controlled variables were analysed using paired t-tests followed by a Bonferroni correction.


In order to discern whether a participant used a variable pacing strategy, variability was calculated as a function of weighted power over average power. Weighted power uses the following algorithm developed by Allen and Coggan (2006):



Contextual efforts

Due to the fact that variability as defined above does not discern context, further analysis was conducted to detect CsEs. Two factors, assumed to be of key importance were analyzed: effort intensity and context. Effort intensity simply refers to power over time and context refers to the topographical situation that effort was produced in.


Data was normalized using Z-scores for both W and percent gradient data for the entire lap duration for each participant. Z-scores were chosen to normalized the data used in the CsE analysis as this method was more tolerant of errors in the data, such as outlying data points. Z-scores effectively polarized the data, for power data this had the effect of granting a score of 1 to values about 10% higher than an individual’s critical power, thus distinguishing efforts with a significant anaerobic contribution for those less significant. For gradient data, this effectively returned zero for descents and 1 for uphill sections.


The z-scores were then summed and averaged for each data point:




The effect of this was to attribute a score of < 0.5 where power was not significant and >0.5 where an effort was significant in amplitude and context.


In order to detect transition phase between ascent and descent, the following formula was used, where i = power for each point in the data:



This checked the average CsE 10 s before a and ahead of particular point for a significant effort (near 1) and checking ahead for the change in effort and context (near 0). If it detects an effort, this value will be greater than 0, with high values (>0.5) being deliberate sprints in context. A 10 s window either side was chosen because it relates to the duration of a short sprint.


CsE cannot distinguish efforts longer than 10 s where there is a significant grade and significant power produced, from an effort of not significant power on a not significant grade as under both circumstances zero is returned.

Example usage of the CsE score

This example shows the CsE score distinguishing between between two laps by the same participant completed at the same average velocity (10.31 m.s-1) coincidentally, but different average power (337 W and 323 W), under both InitialTT and InstructedTT conditions. Elevation data and power data were normalized between 0-1 for illustration purposes only. Where the CsE line crosses 0.5 on the y-axis are efforts that would be flagged as deliberate efforts over the crest of a climb:


Figure 1.) Average velocity 10.31 m.s-1, power 337 W, CsE score 24.22.


Figure 2.) Average velocity 10.31 m.s-1, power 323 W, CsE score 34.41.


From the two figures it can be seen peaks in power coinciding with peaks on the course occur with greater magnitude in the InstructedTT and these efforts are flagged by CsE (orange line).


Of 15 participants, 10 yielded full datasets, 5 more yielded partial data. Of the 15 sets of data, two participants performed under outlying differences in wind speed and their data was removed from calculations of CsE. Four participants failed to complete the 9189 m lap, hence only 7210 km of the lap could be used for analysis. Lap average velocity was used in place of lap time due to correction made to lap length for participants who failed to complete the full lap, these data was used used for within participant analysis only. The shortened lap was still sufficiently long for all participant for other purposes such as testing CP (Vanhatalo et al., 2011). Five participants had heart rate (HR) data proved invalid containing either null values, or values lying well beyond those those considered normal (>220 b.p.m.).

Measured dependant performance variables

For none of the measured performance variables was a statistically significant difference found. A smaller sample (n = 9) was used to compare HR data due to data corruption for several participant files. Except for heart rate, sample size was 13. Data from the last 2 participants to be tested was removed due to excessive differences in wind. RPE did not meet all the assumptions of normality when tested with the Shapiro-Wilk test (W = 0.7903, p-value = 0.002765) and as such was tested with the Wilcoxon signed rank test (V = 37, p-value = 0.08041) indicating the perceived effort was not significantly different. All other tests on the dependent variables listed below were performed with paired t-tests.


Dependant variable IntialTT InstructedTT t( degrees of freedom), p.
Power output (W) 254 (∓ 57) 251 (∓ 59) t(12) = 0.8037, p = 0.4372
Lap speed (m.s-1) 8.72 (∓ 0.97) 8.76 (∓ 1.04) t(12) = -0.9034, p = 0.3841
Mean HR (beat.min-1) 173 (∓ 9.4) 171 (∓ 8.8) t(8) = 1.3494, p = 0.2192
General RPE 16.9 (∓ 0.64) 16.4 (∓ 1.0) V = 37, p-value = 0.08041*

Table 1.) Mean (∓SD) for dependant variables for both tests. *RPE was tested with the Wilcoxon signed rank test.

Inferred variables

From measured dependant variables CP and weighted power were calculated. Specifically relating to analysis of pacing strategy application CsE and variability were calculated:

Inferred variable IntialTT InstructedTT t( degrees of freedom), p.
CP (W) 246 (∓ 55) 244 (∓ 58) t(12) = 0.3375, p = 0.7415
Weighted Power (W) 290 (∓ 65) 289 (∓ 70) t(12) = 0.1853, p = 0.8561
CsE 33.23 (∓ 9.49) 37.82 (∓ 7.91) t(12) = -2.6445, p = 0.02139
Variability 1.15 (∓ 0.05) 1.16 (∓ 0.06) t(12) = -0.7372, p = 0.4752)

Table 2.) Mean (∓SD) for inferred variables for both tests.


Of the inferred variables calculated, only CsE detected a significant difference between conditions signifying participants responded to the treatment. In terms of 4 participants presented with a drop in variability. Among the 4 participants that presented with a drop in variability are the participants that presented with the greatest magnitude change in average lap speed (4.55% and -5.91% respectively).


For both conditions combined mean CsE score was 34.30 (∓ 8.78). CsE was further analysed using boxplots to understand the true extent of change in this variable.


Figure 3.) Change in CsE score between conditions.

Inferred variables and predictors of outcome

Mean energy expenditure (kJ) was calculated for both conditions and corrected to a value of kJ per km (efficiency). For InitialTT efficiency was 28.01 (∓ 3.84) while for the InstructedTT it was 27.59 (∓ 4.16) this difference was not statistically significant (P = 0.154) and essentially reflects a reduction in time and power for the InstructedTT. For all but three subjects economy of effort was lower in the InstructedTT.  


Linear Associations

Linear associations were tested using 26 data pairs. A weak association was found between efficiency and average velocity F(1,24 = 29.78, p = 0.00001278 (R2 = 0.4417) and between global energy expenditure for the full 9189 m lap and average velocity a moderate association was found F(1,16) = 27.97, p= 0.000007343, R2 = 0.6134.


Linear regression was further used to examine the association between 30sW and outcome F(1, 18) = 44.87, p = 0.0000002785, R2 = 0.7137. For the relationship between average 120sW effort and outcome F(1,18) = 30.36, p = 0.000003114, R2 = 0.6279. CP was tested against average velocity yielding F(1,18) = 70.02, p= 0.0000000027, R2 = 0.75 presenting the best association without come of any of the inferred or measured performance variables. Due to missing data 30sW and 120sW data, only 20 tests could be used for comparing 30sW, 120sW and CP data with outcome.


30sRR (0.99 ∓ 0.1337469209) and 120sRR (0.98 ∓ 0.06552246817) were both separately plotted against average velocity and no linear relationship was evident, further further analysis using linear regression confirmed this (R2 = 0.191) for 30sRR and (R2 = 0.014) for 120sRR. No linear relationship was evident in plots of speed and CsEs. CsE was tested against outcome for all data in both conditions using linear regression (R2 = 0.177) no relationship was found.


CP and 30sW were analysed against speed using multiple linear regression resulting F(2,18) = 90.43, p = 0.000000004, adjusted R2 = 0.8994. From this result, the following equation was derived to predict final average velocity (m.s-1):

=4.8986423 +CP* 0.0121975 + 30sW * 0.0018766  

Controlled variables

Two tailed paired samples t-tests were conducted on the following controlled variables between both conditions: wind speed, air pressure, air temperature, mood state and humidity.

Controlled variable IntialTT InstructedTT t( degrees of freedom), p.
wind speed (m.s-1) 3.95 (∓ 6.27) 3.97 (∓ 3.09) t(12) = -0.0433, p = 0.9662
air pressure (mbar) 1026 (∓ 2.81) 1029 (∓ 5.75) t(12) = -2.8108, p = 0.01573
air temperature (º C) 17.84 (∓ 2.54) 16.46 (∓ 3.18) t(12) = 1.2136, p = 0.2483
humidity (%) 73.69 (∓ 11.32) 65.38 (∓11.92) t(12) = 1.8224, p = 0.08091
mood state (POMS) 33.13 (∓4.79) 34.30 (∓ 9.05) t(12) = -0.1973, p = 0.8469

Table 3.) Mean (∓SD) for controlled variables for both tests.


To further investigate if difference in air pressure was actually significant, alveolar partial pressure was calculated for mean pressure under both condition: pO2 109 and 110 which was considered insignificant. Resistance to motion was tested using the equations of Martin et al. (1998) using mean values for all the relevant variables and a frontal area of 0.5 m, rolling resistance of 0.4 and no rise, yielded 11.29 m.s-1 and 11.31 m.s-1 for both average pressures, or a difference of about 1.4 s over 9.189 km. Air pressure was not deemed a significant factor. Two outliers were identified for POMS in the InstructedTT with scores of 47 and 60.


The data from controlled variables were subject to a Bonferroni correction yielding the following p-values: wind, 1.00000; air pressure, 0.07865; air temperature, 1.00000; humidity, 0.40455; mood state, 1.00000. Supporting the previous assumption that air pressure differences were not significant.


The experiment was designed to test the effect of dynamic pacing -specifically sprinting in certain contexts on outcome measured as average velocity and to test for potential predictors of performance: CP, 30sRR, 120sRR, 30sW and 120sW. Controlled variables  remained stable enough not to severally challenge the reliability of the experiment.


For testing CP, Vanhatalo et al. (2011) recommend efforts of between 2 and 15 min that have a difference of at least 5 min in magnitude between efforts, the test used in this study was designed considering these limits. While CP has been shown to be testable with a single maximal 3 min bout of exercise (Vanhatalo et al., 2007), it was chosen to test CP with several bouts of exercise: Two repeated efforts of the same duration were used for 30sW and 120sW, rather than efforts of different durations to provide a value for recovery ratio. No association was found between recovery ratio and outcome. While each component of a normal CP test is done on separate days, it was chosen for the test to follow this format in order to see actual difference between bouts for the given recovery recovery period. The 30 s intervals were considered less problematic than the 2 min intervals: Hazell et al. (2010) found no significant difference between 2 and 4 min rest intervals for 30 s maximal sprints. Bogdanis et al. (1995) place the half life of phosphocreatine resynthesis at 56.6 s (∓ 7.3) suggesting participants would be mostly recovered after 2 min. Shorter sprint efforts may be more useful in testing the capacity of the ATP-PCr system since in 30 s this system only relate to ~28% of power produced (Smith and Hill, 1991). Any future protocol based on this study should consider using 10 s sprints in place of the 30 s sprints.


No research was found relating specifically the rate of regeneration of AWC from a 2 min all out effort, however in the experiment itself recovery ratio was 0.98 suggesting that the 4 min active recovery suggested in the protocol was sufficient. Recovery ratio in the 30 s interval 0.99 was greater than those observed by Hazell et al. (2010) which saw a recovery ratio of 0.89 who used twice the recovery period (4 min). While the 2 min maximal efforts used in this test protocol may not be enough to completely exhaust AWC (Chidnok et al., 2013) while findings by Vanhatalo et al. (2007) would indicate a 2:15 period for the exhaustion of AWC. Dekerle et al. (2015) found that rates of AWC are not constant and as such the use of 3 min max effort may better ensure exhaustion of AWC across a broader range of individuals. Future protocols may consider using 3 min max effort, sufficient to completely exhaust AWC (Chidnok et al., 2013) as well as providing a more accurate measure of the rate of replenishment of AWC once this has been completely exhausted. AWC is most often treated as a finite reserve and not a capacity with regenerative ability, yet it seems probable that some degree of regeneration occurs every time power drops below critical power. Both 30sRR and 120sRR were examined to see whether they had a relationship with performance, suggesting that for this study regeneration of AWC was not a factor, either due to participants maintaining some reserve, or this reserve adequately regenerated between extreme efforts. 3 min efforts may provide the best compromise for testing CP and AWC, however if any future tests are run concurrently such as this one, a different recovery period may be required.


CP and 30sW did present with strong an association with outcome while 120sW a less strong outcome. These findings were counterintuitive as the metabolic regime employed in 30sW was distinctly different to that in the TT. From this analysis It is not possible to know whether a higher 30sW signifies more frequent or efficacious use of sprint efforts or whether some other related mechanism allows for better performance in the TT. It’s not know whether the CP and 30sW combination would predict performance in conditions that are not variable. It is assumed that since CP approximates the average power over maximal efforts up to about 30 min and has been shown to predict 16.1 km TT performance (Black et al., 2015), that it alone would provide the most accurate predictor of performance over the combination of CP and 30sW found in this study, in situations with less topographic variation such as long climbs and flat time trials. Future investigation could analyse to what degree a course/athlete combination would benefit from either a variable or steady pacing strategy.


No association was detected between CsE and outcome, perhaps because CsE does not describe the magnitude or length of an effort. Future models need to look at a broader range of contexts the magnitude and duration of effort developed in these contexts and give them a weighting according to relevance of context.


To examine CsE scores better above 32 was chosen to define a deliberate use of CsE sprints as part of pacing strategy. Of 13 valid samples, 8 participants had a CsE score of above 32 in InitialTT (mean 39.12 ∓ 4.5). In the InstructedTT this number increase 10 (mean 41.37 ∓ 4.6). Three participants showed no deliberate adoption of the strategy as per the experimental treatment in the InstructedTT.


Participants that naturally adopted CsE sprints (n=8) saw a small increase in CsE score (mean 3.31) between the two TTs. Of those that did not naturally adopt CsE sprints (n=5), 1 negatively adopted the strategy (-6.4), 2 presented with little difference between the conditions (-0.4), 2 saw a large increase in CsE (mean = 11.75).


Figure 4.) Paired results for the application of CsEs.


Neither CsEs, variability, CP or were able to account for improvements in performance and drops in average power for 2 participants, hence further analysis was done on the distribution of effort. This was done by counting 1 s interval at CP as measured in the InitialTT and subsequently, CP + 40%, CP + 60%, CP + 80% and CP + 100%. Certain participants (N= 2) improved performance while reducing average power and with little apparent change in CsEs, they still presented with a greater number of efforts in the extreme domain, the sum of efforts above CP 40% in both conditions was 38 and 91 for one participant, 295 and 360 for the other suggesting a deliberate increase in extreme efforts, although in a different context to that detectable by the CsE method in its current form. CsE only partly accomplished the task it was designed for.


Partially asserting the idea that AWC is not a single quantity is the concept of functional reserve capacity (FRC). Marcora and Staiano (2010) found among a group of 10 rugby players who cycled to exhaustion that a residual capacity roughly three times as high as the intensity level required to induce this exhaustion existed. FRC would indicate capacity to sprint contextually in situations such as sprinting into a descent at the end of an “all out” effort on a climb. This conclusion is also supported by Menaspà et al. (2015) who found that constant high intensity effort for 10 min did not affect subsequent sprint performance in elite cyclists. Hence it was assumed that a sprint at the end of an all out climb was viable for athletes and may provide an effective pacing strategy while reducing homeostatic disturbance. In this experiment however no benefits were gained sprinting. The benefits that may exist are probably later negated by reductions in effort in other areas of the course.


Vanhatalo et al. (2011) refers to individuals that have high CP having relatively small AWC supporting the observations by Atkinson et al. (2007) that cyclists with a ventilatory threshold closer to  their VO2max were less tolerant to the imposed variations in intensity. In the InitialTT, 8 participants naturally adopted the CsE strategy electively without instruction, so of which despite having a relatively small AWC (less than double CP in the 2 min test), it is though that CsEs generally are too short to elicit a large degree of physiological perturbation, and as such better tolerated and are employed even by cyclists with a small AWC.


None of the athletes tested were specifically trained in sprinting and three had relatively little experience of the sport. None of the participants adopted other techniques aimed at increasing CCV or reducing effort, such as using an aerodynamic tuck on descents. CsE score worked in this specific scenario to detect specific efforts that related to the application of the particular pacing strategy being applied. CsE did not explain power distribution efficacy completely, with some participants producing better performances for similar CsE values. Further analysis did reveal an increase in time and magnitude of efforts for those participants that did not show an improvement, suggesting there are  other contexts in which extreme efforts can produce a benefit.


In terms of predictors of performance, 30sW, counter intuitively had a stronger linear relationship with outcome than any of the other variables analyzed except CP. These factors when combined were seemingly the strongest predictor of performance and merit further research using independant tests.

Assumptions and experimental flaws

Wind direction was accounted for, but no analysis was conducted on it on the assumption that as the course was on a roughly circular circuit the effect of wind direction would negate itself. Precipitation data was not analyzed since 14 participants performed each test in the same conditions (dry) and 1 participant performed both tests with wet road conditions.


It was assumed that participants once provided with a meal diary for the 24 h before the experiment would repeat their stated food intake and that participants would arrive at the test in a similar state of fatigue following the same training logged in the InitialTT.


Despite participants being familiarised with the course before hand and 3-7 days separating both tests, a learning effect exist. Although large efforts were made to make this ecologically valid test reliable, effect size would need to be large to be significant given the potentially compounding effect of so many controlled variables and even the reliability of the test apparatus. In terms of the validity of using a hilly TT to test CP, this is a potential flaw despite Stevens and Dascomb (2015) finding TTs are valid to test performance, including somewhat variable course, the variability of the courses analysed is unlikely to be as high as in this experiment.


Pacing strategies reviewed in Abbiss and Laursen (2008) may too be simplistic in attempting to describe cycling pacing on variable terrain. This study found intra-individual difference in power distribution can vary significantly, for similar results in average velocity, perceived exertion and average heart rate.


Atkinson (2007) and Thomas et al. (2013) have demonstrated a benefit in terms of work done of using a variable pacing strategy in a lab test, but not utilising deliberate sprint efforts. This study addressed this sprint in an ecologically valid setting, it did not find significant benefit in terms of time saved on a course, nor significant differences in other variables except in the amount of time performing CsEs.


CP was able to a large degree explain performances, when CP and 30sW were taken together a stronger relationship was observed suggestion that a protocol such as the one detailed may be useful in determining performance. This would require further validation.


Future research should involve a test and TT one in a controlled environment (lab) and one in the field so that any predictions made in the test are completely independant of the TT result.

Word count: 6349


Abbiss, C. R., Quod, M., Martin, D., Netto, K., Nosaka, K., Lee, H., Surriano, R., Bishop, D. and Laursen, P. (2006) ‘Dynamic pacing strategies during the cycle phase of an Ironman triathlon.’ Medicine and Science in Sports and Exercise, 4(38) pp.726-734.


Abbiss, C. R. and Laursen, P. B. (2008) ‘Describing and understanding pacing strategies during athletic competition.’ Sports Medicine, 38(3) pp. 239–252.


Allen, H. and Coggan, A. (2006) Training and racing with a power meter, 2nd edition,  Boulder, CO., VeloPress.


Atkinson, G., Peacock, O. and Law, M. (2006) ‘Acceptability of power variation during a simulated hilly time trial.’ International Journal of Sports Medicine, 2(28) pp. 157-163.


Bertucci, W., Duc, S., Villerius, V., Pernin, J. and Grappe, F. (2005) ‘Validity and reliability of the PowerTap mobile cycling powermeter when compared with the SRM device.’ International Journal of Sports Medicine, 26(10) pp. 868–873.


Brickley, G., Green, S., Jenkins, D., McEinery, M., Wishart, C., Doust, J. and Williams, C. (2006) ‘Muscle metabolism during constant- and alternating-intensity exercise around critical power.’ International Journal of Sports Medicine, 28(4) pp. 300-305.


Black, M., Bailey, S., Jones, A., Morgan, P. and Vanhatalo, A. (2015) ‘Conventional critical power test predicts 16.1 km road time trial performance.’ In The Physiological Society, Proceedings of The Physiological Society. Cardiff, UK (2015) [no page number].


Borg, G. (1998) Borg’s Perceived Exertion and Pain Scales, Champaign, Il., Human Kinetics.


Bogdanis, G. C., Nevill, M. E., Boobis, L. H., Lakomy, H. K. and Nevill, A. M. (1995) ‘Recovery of power output and muscle metabolites following 30 s of maximal sprint cycling in man,’ The Journal of Physiology, 482(2) pp. 467-480.


Chidnok, W., Dimenna, F., Bailey, S., Wilkerson, D., Vanhatalo, A. and Jones, A. (2013) ‘Effects of pacing strategy on work done above critical power during high-intensity exercise.’ Medicine and Science in Sports and Exercise, 45(7) pp. 1377–85.


Dekerle, J., de Souza, K. M., de Lucas, R. D., Guglielmo, L. G. A., Greco, C. C. and Denadai, B. S. (2015) ‘Exercise tolerance can be enhanced through a change in work rate within the severe intensity domain: Work above critical power is not constant.’ Public Library of Science (PLoS), 10(9) e.0138428.


Gardner, A. S., Stephens, S., Martin, D. T., Lawton, E., Lee, H. and Jenkins, D. (2004) ‘Accuracy of SRM and power tap power monitoring systems for bicycling’ Medicine and Science in Sports & Exercise, 36(7) pp. 1252-1258.


Gastin, P. B. (2001) ‘Energy System Interaction and Relative Contribution During Maximal Exercise’, Sports Medicine, 31(10) pp. 725-741.


Hazell, T., Macpherson, R., Gravelle, B. and Lemon, P. (2010) ‘10 or 30-s sprint interval training bouts enhance both aerobic and anaerobic performance.’ European Journal of Applied Physiology, 110(1) pp. 153-160.


Johnson, E. C., Pryor, J. L., Casa, J. D., Belval, L. N., Vance, J. S., DeMartini, J. K., Maresh, C. M. and Armstrong, L. E. (2013) ‘Bike and run pacing on downhill segments predict Ironman triathlon relative success’ Journal of Science and Medicine in Sport, 18(1) pp. 82-87.


Karsten, B., Beedie, C., Jobson, S. A., Hopker, J. G. and Stevens, L. (2014) ‘Validity and reliability of critical power field testing.’  European Journal of Applied Physiology, 115(1) pp. 197-204.


Marcora, S. M., Staiano, W. and Manning, V. (2009) ‘Mental fatigue impairs physical performance in humans’ Journal of Applied Physiology, 106(3) pp. 857-864.


Marcora, S. M. and Staiano, W. (2010) ‘The limit to exercise tolerance in humans: mind over muscle?’ European Journal of Applied Physiology, 109(4) pp. 763-770.


Martin, J. C., Milliken, D. L., Cobb, J. E., McFadden, K. L. and Coggan A. R. (1998) ‘Validation of a Mathematical Model for Road Cycling Power’ Journal of Applied Biomechanics, 14(3) pp. 276-291.


Menaspà, P., Impellizzeri, F., Haakonssen, E., Martin, D. and Abbiss, C. (2014) ‘Consistency of commercial devices for measuring elevation gain.’ International Journal of Sports Physiology and Performance, 9(5) pp. 884–6.


Menaspà, P., Martin, D., Victor, J. and Abbiss, C. (2015) ‘Maximal sprint power in road cyclists after variable and Nonvariable high-intensity exercise.’ Journal of Strength and Conditioning, 29(11) pp. 3156-61.


Miller, M. C., Moir, G. L. and Stannard, S. R. (2014) ‘Validity of using functional threshold power and intermittent power to predict cross-country mountain bike race outcome.’ Journal of Science and Cycling, 3(1) pp. 16-20.


Miller, M. C. and Macdermid, P. W. (2015) ‘Predictive validity of critical power, the onset of blood lactate and anaerobic capacity for cross-country mountain bike race performance,’ Sports and Exercise Medicine Open Journal, 4(1) pp. 105–110.


Monod, H. and Scherrer, J. (1965) ‘The work capacity of a synergic muscular group.’ Ergonomics, 8(3) pp. 329–338.


Padilla, S., Mujika, I., Angulo, F. and Goiriena, J. (2000) ‘Scientific approach to the 1-h cycling world record: A case study.’ Journal of Applied Physiology, 89(4) pp. 1522-1527.


Smith, J. C. and Hill, D. W. (1991) ‘Contribution of energy systems during a Wingate power test.’ British Journal of Sports Medicine, 25(4) pp. 196–199.


Stevens, C. J. and Dascombe, B. J. (2015) ‘The reliability and validity of protocols for the assessment of endurance sports performance: An updated review.’ Measurement in Physical Education and Exercise Science, 19(4) pp. 177–185.


Swain, D. P. (1997) ‘A model for optimizing cycling performance by varying power on hills and in wind.’ Medicine & Science in Sports & Exercise, 29(8) pp. 1104–1108.


Thomas, K., Stone, M. R., Thompson, K. G., St. Clair Gibson, A. and Ansley, L. (2012) ‘The effect of self- even- and variable-pacing strategies on the physiological and perceptual response to cycling.’ European Journal of Applied Physiology, 8(112) pp. 3069-3078.


Thomas, K., Stone, M., St Clair Gibson, A., Thompson, K. and Ansley, L. (2013) ‘The effect of an even-pacing strategy on exercise tolerance in well-trained cyclists.’ European Journal of Applied Physiology, 113(12) pp. 3001–3010.


Vanhatalo, A., Doust, J. and Burnley, M. (2007) ‘Determination of critical power using a 3-min all-out cycling test.’ Medicine and Science in Sports and Exercise, 39(3) pp. 548–55.


Vanhatalo, A., Jones, A. M. and Burnley, M. (2011) ‘Application of Critical Power in Sport’ International Journal of Sports Physiology and Performance, 6(1), pp. 128-136.


Viana, M. F., de Almeida P. L. and Santos, R. C. (2001) ‘Adaptação portuguesa da versão reduzida do Perfil de Estados de Humor – POMS’ Análise Psicológica, 1(19) pp. 77-92.

Did you enjoy this post?

Sign up to my newsletter (sent about once a month) to keep up to date with all the latest training tips and info from the Algarve.
* = required field

Leave a Reply