UEF//eRepository
DSpace https://erepo.uef.fi
Rinnakkaistallenteet Luonnontieteiden ja metsätieteiden tiedekunta
2020
Force-velocity profiling in ice hockey skating: reliability and validity of a simple, low-cost field method
Stenroth, Lauri
Informa UK Limited
Tieteelliset aikakauslehtiartikkelit
© 2020 Informa UK Limited, trading as Taylor & Francis Group All rights reserved
http://dx.doi.org/10.1080/14763141.2020.1770321
https://erepo.uef.fi/handle/123456789/8203
Downloaded from University of Eastern Finland's eRepository
1
Force-velocity profiling in ice hockey skating: reliability and validity of a simple, low-cost 1
field method 2
Lauri Stenroth, Paavo Vartiainen and Pasi A Karjalainen 3
Department of Applied Physics, University of Eastern Finland, Finland 4
5
ORCiDs:
6
Lauri Stenroth: https://orcid.org/0000-0002-7705-9188 7
Paavo Vartiainen: https://orcid.org/0000-0003-0974-0913 8
9
Corresponding author:
10
Lauri Stenroth 11
University of Eastern Finland 12
Department of Applied Physics 13
PO Box 1627 14
70211 Kuopio, Finland 15
tel. +358505649096, email: lauri.stenroth@uef.fi 16
2
Force-velocity profiling in ice hockey skating: reliability and validity of a simple, low-cost 17
field method 18
Abstract 19
In recent years, a simple method for force-velocity (F-v) profiling, based on split times, has 20
emerged as a potential tool to examine mechanical variables underlying running sprint perfor- 21
mance in field conditions. In this study, the reliability and concurrent validity of F-v profiling 22
based on split times were examined when used for ice hockey skating. It was also tested how a 23
modification of the method, in which the start instant of the sprint is estimated based on optimisa- 24
tion (time shift method), affects the reliability and validity of the method. Both intra- and inter- 25
rater reliability were markedly improved when using the time shift method (approximately 50%
26
decrease in the standard error of measurement). Moreover, the results calculated using the time 27
shift method highly correlated (r>0.83 for all variables) with the results calculated from a contin- 28
uously tracked movement of the athlete, which was considered here as the reference method. This 29
study shows that a modification to the previously published simple method for F-v profiling im- 30
proves intra- and inter-rater reliability of the method in ice hockey skating. The time shift method 31
tested here can be used as a reliable tool to test a player’s physical performance characteristic 32
underlying sprint performance in ice hockey skating.
33
Keywords: power, acceleration, speed, sprint, measurement error 34
35
3 Introduction
36
The ability to accelerate rapidly has the utmost importance for ice hockey players as the game 37
involves short intermittent sprints (Roczniok et al., 2016). Traditionally, players’ physical perfor- 38
mance regarding acceleration and maximal speed of movement has been assessed using maximal 39
effort sprint with the timing of either the whole sprint or a section of it. Depending on the distance 40
used and whether a standstill or flying start is used, different physical performance characteristics 41
are evaluated. However, the test only measures the average speed of the athlete over the given 42
distance and give limited information about the specific physical performance capabilities that 43
underlie the performance. Horizontal forces applied to the ground over the course of sprinting are 44
the main determinants of sprint performance and can give valuable information regarding both 45
physical and technical factors affecting the performance (Morin et al., 2011, 2012), but direct 46
measurement of these forces is only possible in laboratory conditions using an instrumented tread- 47
mill or floor-embedded force plates. Hence, a method allowing estimation of horizontal force gen- 48
eration in field conditions is highly valuable for practitioners.
49
An indirect method to estimate horizontal ground force generation of an athlete during maximal 50
sprinting performance, suitable for field tests, was recently validated against direct laboratory 51
measurements (Morin et al., 2019; Samozino et al., 2016). This macroscopic inverse dynamic 52
method takes advantage of the observation that an athlete's velocity as a function of time in maxi- 53
mal effort sprinting can be modelled accurately using an exponential function (Di Prampero et al., 54
2005). Therefore, it is possible to estimate instantaneous velocity based on discrete measurements 55
of an athlete's velocity as a function of time by fitting the exponential function to the observations.
56
Further, based on simple mechanics with the estimation of aerodynamic drag, horizontal force 57
applied to the body centre of mass can be estimated (Samozino et al., 2016). After horizontal force 58
4
generation has been estimated, a linear force-velocity profile and parabolic power-velocity profile 59
can be established. The process of obtaining these relationships can be called force-velocity (F-v) 60
profiling.
61
The force-velocity and power-velocity profiles describe sprint mechanics and give important in- 62
formation for strength and conditioning coach to assess physical performance characteristics un- 63
derlying and possibly limiting the athlete’s performance and the information can be used to indi- 64
vidualise and monitor training in a more detailed fashion compared to traditional timed sprints.
65
For example, Morin and Samozino presented a case of two athletes with similar 20-meter sprint 66
time but differences in their F-v profile (Morin & Samozino, 2016). In this example, the test results 67
may direct training either towards the early phases of the sprint acceleration or towards maximal 68
velocity training.
69
Until recently, it was not known if horizontal velocity in maximal ice hockey skating can be mod- 70
elled with the exponential model previously used for running (Morin et al., 2019; Samozino et al., 71
2016) but a recent study by Perez, Guilhem, & Brocherie (2019) showed that the model could also 72
be applies to skating. They also showed that acceptable inter-trial and test-retest reliability could 73
be obtained for F-v profiling in ice hockey players when using a radar to measure skating velocity.
74
Hence, F-v profiling is a promising method for assessing mechanical determinants of ice hockey 75
skating performance.
76
In that first report of F-v profiling in ice hockey skating, Perez et al. (2019) used a radar that costs 77
around $2000. The necessity to use costly devices (radar or laser) hinders the widespread use of 78
the method by coaches. A low-cost solution is to use split times measured using a high-speed video 79
camera capturing the movement in sagittal plane perspective and to fit an exponential model to the 80
split time data to estimate the position of the athlete as a function of time and, subsequently, the 81
5
F-v profile (Romero-Franco et al., 2017; Samozino et al., 2016). The method relies on accurate 82
detection of the beginning of the sprint and misidentification of the correct time instant may lead 83
to large errors in the results (Haugen et al., 2018). For running, a three-point start position, in 84
which both feet and one hand is touching the ground, has been used to facilitate accurate and 85
objective detection of the sprint start with the lift of the hand from the ground signifying the start 86
instant (Romero-Franco et al., 2017). Nevertheless, it has been observed that horizontal force gen- 87
eration begins before the hand lift. Therefore, a correction of the measured split times by adding 88
0.1 s to each split has been suggested to account for the time delay between the start of horizontal 89
force generation and the lift of the hand (Samozino, 2018).
90
There are two problems with using this methodology to assess ice hockey skating performance.
91
First, the three-point start position is not feasible for ice hockey skating, but when using a staggered 92
stance starting position, the first movement of the body may occur in different body parts in each 93
participant. This makes it difficult to set a clear definition for the beginning of the sprint start.
94
Secondly, if a constant correction of split times is used, individual differences between the first 95
observable movement and the start of the horizontal force generation are not considered. These 96
problems may negatively affect the reliability and accuracy of the low-cost, simple F-v profiling 97
method for ice hockey skating and therefore hinder the usability of the method. To overcome these 98
problems, the required correction of the split times could be obtained using optimisation, similarly 99
as has been done previously when using continuously measured velocity data to conduct the F-v 100
profiling (Morin et al., 2019; Samozino, 2018). In the proposed approach, athlete’s position as a 101
function of time is modelled using an exponential function. A parameter that shifts the modelled 102
position along the time axis is added to the equation and the value of this parameter is found in an 103
optimisation process that minimises the difference between the modelled and observed positions.
104
6
Thus, the optimisation finds a start instant which best matches with the assumption of exponen- 105
tially increasing position as a function of time.
106
The purpose of this study was to examine if an optimisation -based correction of split times can 107
improve inter-trial and inter- and intra-rater reliability of F-v profiling in ice hockey skating when 108
using the split time method. F-v profiling conducted using continuously tracked player movement 109
from a video recording was considered as the reference method since continuous tracking of the 110
movement gives us frequent data points facilitating accurate modelling of the instantaneous veloc- 111
ity during the sprint. Moreover, the data obtained using continuous tracking is comparable to data 112
obtained using laser device to measure instantaneous velocity, a method which was recently vali- 113
dated against direct measurement of ground reaction forces (Morin et al., 2019). Therefore, the 114
concurrent validity of the split time method, with and without optimisation, was examined by 115
comparing the results obtained with that method to the results obtained using continuously tracked 116
player movement. The hypothesis of the study was that an optimisation-based correction of split 117
times improves reliability and concurrent validity of the low-cost, simple F-v profiling method in 118
ice hockey skating.
119
Methods 120
Participants 121
Twelve male ice hockey players aged 18.4 years to 22.0 years old from elite Finnish junior league 122
or elite Finnish league volunteered as participants (height 1.82 ± 0.03 m, mass 83.9 ± 6.2 kg, and 123
body mass index 25.2 ± 1.8 kg/m2). The ethics committee of the Hospital District of Northern Savo 124
approved the study protocol, and all participants signed informed consent before participation in 125
the study.
126
7 Maximal 30-meter skating test
127
After a warm-up, participants performed a 30-meter maximal skating test. The test was performed 128
a minimum of five times by each participant. The recovery period between trials was 90 seconds.
129
Participants wore a helmet, skates, gloves, and a tracksuit and carried a hockey stick in one hand 130
during the test. Participants also had electro-goniometers attached to hip and knee joints, EMG- 131
electrodes on several lower limb muscles, and they carried a datalogger (mass 0.67 kg). Joint an- 132
gles and EMG-data were related to a separate study and not reported here. The time of the 30- 133
meter maximal skating test was measured using photocells (Chronojump Boscosystem®, Spain).
134
The first photocell was placed 25 cm ahead from the starting line to prevent accidental triggering 135
of the timing. The second photocell was placed 30 meters from the first photocell.
136
A high-speed camera (GoPro 3, GoPro Inc., USA, framerate 120 frames per second, resolution 137
1280 x 720 pixels) was used to capture video of the skating. The camera was equipped with a fish- 138
eye lens that allowed capturing the whole 30-meter trial using a fixed camera position. The line of 139
sight of the camera was positioned perpendicular to the skating direction. The distance of the cam- 140
era from the skating line was 16 meters, and the camera was positioned one meter above the ice 141
and 15 meters from the starting line in the direction of skating. The lens distortion in the horizontal 142
direction was considered by digitising the known locations of the split time marks (vertical sticks) 143
and fitting a third-order polynomial to the digitised points. The equation relating pixel values along 144
the horizontal axis of the image to physical distances was used to convert pixel coordinates to real- 145
world coordinates and was used for continuously tracked movement of the player. The video was 146
also used to measure split times of the skating. For this purpose, vertical sticks were placed on the 147
ice to mark the player’s position at 0, 2.5, 5, 7.5, 10, 15, 20, 25 and 30 meters from the start line 148
(to correct for the parallax error the actual stick positions were 0.94, 3.28, 5.63, 7.97, 10.31, 15.00, 149
8
19.69, 24.38 and 29.06 m from the start line). The split times were measured for every 2.5 meters 150
for the first 10 meters of the performance to increase available data for the portion of the perfor- 151
mance in which most acceleration occurs (Fig. 1). Skating line was parallel to the line of the sticks 152
with one-meter average separation between the lines.
153
Force-velocity profiling based on split times 154
Three fastest trials based on the time measured by the photocells were chosen for the measurement 155
of split times. Tracker 4.11.0 (http://physlets.org/tracker/) software was used to measure the split 156
times by manually selecting the frames in which the player passed the sticks. The body part that 157
was on the line of sight from the camera to the start line was used to detect the passing of each 158
stick. Two raters performed the analysis independently. In addition, the other rater (Rater 1) re- 159
peated the analysis for the fastest trial for intra-rater repeatability analysis. The spreadsheet devel- 160
oped by Morin and Samozino was used to calculate F-v profile and sprint mechanical variables 161
using the split times (Morin & Samozino, 2019). Position of the athlete was modelled as a function 162
of time with the equation 163
𝑥(𝑡) = 𝑣𝑚𝑎𝑥 ∙ (𝑡 + 𝜏𝑒−𝑡 𝜏⁄ ) − 𝑣𝑚𝑎𝑥 ∙ 𝜏 164
where vmax is the plateau of the velocity, i.e. maximum velocity of the model, and τ is acceleration 165
time constant. The constants (vmax and τ) were found using built-in solver function of Excel (Mi- 166
crosoft Corporation, Redmond, Washington, United States). The solver was set to minimise the 167
sum of squared differences between the modelled and actual positions of the athlete by altering 168
the constants. A nonlinear generalised reduced gradient algorithm was used as the solving method.
169
After estimating vmax and τ, the modelled instantaneous horizontal velocity was calculated using 170
the equation 171
9
𝑣(𝑡) = 𝑣𝑚𝑎𝑥 ∙ (1 − 𝑒−𝑡 𝜏⁄ ) 172
followed by calculation of modelled instantaneous horizontal acceleration as a derivative of the 173
velocity using the equation 174
𝑎(𝑡) =𝑣𝑚𝑎𝑥
𝜏 ∙ (𝑒−𝑡 𝜏⁄ ).
175
The horizontal force applied to the body centre of mass was estimated using the equation 176
𝐹(𝑡) = 𝑚𝑎(𝑡) + 𝐹𝑎𝑒𝑟𝑜(𝑡) 177
where m is athlete’s mass, and Faero is the aerodynamic drag force that is estimated from athlete's 178
body mass, height and instantaneous velocity and air temperature, and pressure (Arsac & Locatelli, 179
2002). The horizontal force was subsequently divided by body mass. Ratio of force was calculated 180
using the equation 181
𝑅𝐹 = 𝐹𝐻
√𝐹𝐻2+ 𝐹𝑉2 182
where FH is the estimated horizontal force applied to the body centre of mass, and FV is vertical 183
force applied to the body centre of mass, which is estimated to be equal to body weight over a full 184
step cycle due to constant vertical position of the centre of mass. Based on the calculations, the 185
sprint mechanical variables listed in the Table 1 were extracted and used in the subsequent anal- 186
yses. More details of the different variables and calculation methods are given in previous literature 187
(Morin & Samozino, 2016; Samozino et al., 2016).
188
To estimate the trial-specific correction of split times, the same spreadsheet and optimisation func- 189
tion that were used to estimate values for vmax and τ (Morin & Samozino, 2019) was used with an 190
addition of one more optimised parameter that named time shift. The time shift tries to remove 191
uncertainty in identifying the onset time of horizontal force generation by changing the duration 192
10
of the first time interval while maintaining the differences between other split times. The method 193
is later referred as the time shift method. The original and modified spreadsheets, including the 194
time shift, can be found from the supplementary material with a tool to calculate camera parallax 195
correction for sprint distance marks needed for measuring split times from video data.
196
For the time shift method, the equation used to model the athlete’s position was 197
𝑥(𝑡) = 𝑣𝑚𝑎𝑥∙ (𝑡 + 𝑐 + 𝜏𝑒(−𝑡+𝑐) 𝜏⁄ ) − 𝑣𝑚𝑎𝑥 ∙ 𝜏 198
where c is the time shift parameter used to correct the split times. It should be noted that the pa- 199
rameter c can have both positive and negative values. The subsequent analysis of the F-v profile 200
and sprint mechanical variables followed the same procedures as the F-v profiling without the time 201
shift.
202
Force-velocity profiling based on continuous tracking 203
The fastest trial was selected for continuous tracking of the player’s movement. Movement of the 204
player’s head was manually tracked using the Tracker software to acquire the player’s horizontal 205
position as a function of time. The amount of tracked frames was reduced by digitising every fifth 206
frame (0.042 s separation between digitised frames). The horizontal velocity of the player was 207
calculated from the position data dividing the frame-by-frame displacement by the time between 208
analysed frames in Matlab (Mathworks, MA, USA). The curve-fitting tool in Matlab was used to 209
fit a mono-exponential function to the velocity-time data (Cross et al., 2017; Morin et al., 2019).
210
The function used in the fitting was 211
𝑣(𝑡) = 𝑣𝑚𝑎𝑥 ∙ (1 − 𝑒−𝑡+𝑐 𝜏⁄ ).
212
As suggested by Samozino (2018), initial part of the movement was not used in the fitting process 213
due to the possible errors in the data in this phase (three first data points were exclude, i.e., the first 214
11
0.125 s). The equation used to model velocity passes through the origin. However, since time was 215
not set to zero at the beginning of the movement, parameter c was added to the equation (Samozino, 216
2018). The parameter c is analogous to the time shift used with the split time data. Finally, the 217
values for vmax and τ obtained from the fit were input to the same spreadsheet that was used to 218
calculate F-v profile based on split times to obtain F-v profile based on continuous tracking of the 219
players’ movement (Fig. 2). Therefore, the F-v profiling and calculation of sprint mechanical pa- 220
rameters followed identical procedures than were used for F-v profiling based on split times.
221
Statistical analysis 222
The normality of the data was tested using the Shapiro-Wilks test. Repeated measures analysis of 223
variance (ANOVA) was used to compare the mean values of the 30-meter time and the sprint 224
mechanical variables obtained from the three trials included in the analyses. The three trials were 225
also used for inter-trial reliability analysis. The fastest trial was used for inter- and intra-rater reli- 226
ability analyses. Repeated measures t-test was used to test systematic errors (mean differences) in 227
the intra- and inter-rater reliability analysis and to test differences in the mean values produced by 228
the different methods; split time method (ST), split time method with the time shift (ST-TS), and 229
continuous tracking (CT). The intraclass correlation coefficient (ICC) and standard error of meas- 230
urement (SEM) were calculated as measures of relative and absolute reliability, respectively, ac- 231
cording to Weir (Weir, 2005). ICC reflects the ability of the measure to differentiate individuals, 232
whereas SEM provides an estimate of the typical error of the measure. For the ICC calculations, 233
two-way random effects model for absolute agreement and single rater (ICC 2,1) was used. ICC 234
values were interpreted according to Koo and Li (2016) with the following cut points: <0.5 poor, 235
0.5-0.75 moderate, 0.75-0.9 good and >0.90 excellent reliability. SEM was calculated as the square 236
root of the mean square error from the repeated measures ANOVA. SEM values are presented as 237
12
a percentage of the mean. Pearson correlation coefficients were calculated to estimate the con- 238
sistency of the results calculated using simple, low-cost field methods (ST and ST-TS) with CT 239
(considered here as the reference method). The level of statistical significance was set at p<0.05.
240
All statistical analyses were conducted using IBM SPSS Statistics software (version 25, SPSS Inc., 241
IBM Company, Armonk, NY, USA).
242
Results 243
There was a significant difference in the 30-meter skating time between the three fastest trials 244
selected for analysis (4.14±0.10, 4.17±0.09 and 4.21±0.09 s, p<0.05 for all comparisons). How- 245
ever, none of the variables describing the F-v profile significantly differed from each other between 246
the trials (Table 3). Inter-trial reliability analyses yielded an ICC value of 0.811 and an SEM value 247
of 0.7% for the 30-m skating time. Without time shift, ICC values for the three fastest trials in the 248
different variables describing the profile ranged from poor to moderate in Rater 1 (ICC 0.351- 249
0.711) and from moderate to good in Rater 2 (ICC 0.515-0.859). SEM ranged from 2.6 to 11.3%
250
for Rater 1 and from 2.0 to 8.3% for Rater 2. The time shift had a marginal effect on the inter-trial 251
reliability.
252
There was a statistically significant mean difference between the values of Pmax (p=0.039) and 253
RFmax (p=0.032) between the analyses performed by the two raters when using the ST (Table 4).
254
In addition, the difference in all other parameters approached statistical significance for difference 255
(p<0.01). ICC values from inter-rater reliability analysis ranged from 0.495 to 0.759 (poor to good 256
reliability) and SEM values ranged from 1.2 to 9.1%. The time shift removed all the significant 257
differences between the raters and markedly improved reliability estimates. Good to excellent re- 258
liability (ICC ranging from 0.827 to 0.963) was obtained for all parameters and SEM values de- 259
creased on average by 48% when using the time shift. Based on SEM values the most reliable 260
13
variable was the estimate of skating speed at the 30-meter mark (Max speed) both with and without 261
the time shift (SEM 0.5% and 1.2%, respectively).
262
No significant mean differences were observed between the two repeated analyses of the same 263
trials by a single rater in any of the sprint mechanical parameters (Table 5). The differences be- 264
tween the analyses were further reduced when using the time shift. The F-v profiling showed mod- 265
erate to good relative intra-rater reliability without time shift (ICC ranging from 0.529 to 0.832) 266
and excellent reliability (ICC values ranging from 0.918 to 0.985) when using time shift. The time 267
shift reduced SEM values on average by 68%. The most reliable variable in the intra-rater relia- 268
bility analysis was the Max speed both without and with the time shift (SEM 1.1% and 0.5%, 269
respectively).
270
A comparison of the results obtained with the ST and the CT showed significant mean differences 271
in Max speed for Rater 1 (p=0.030) and in Pmax (p=0.015) and Max speed (p=0.024) for Rater 2 272
(Table 6). Correlations between the ST and CT methods were significant for Pmax (p=0.010, 273
r=0.706) and RFmax (p=0.040, r=0.598) in Rater 1 and for F0 (p=0.044, r=0.590), Pmax (p=0.008, 274
r=0.720), and RFmax (p=0.017, r=0.671) in Rater 2. The time shift introduced a systematic differ- 275
ence compared to the CT increasing the percentage differences between the methods. The values 276
obtained using the ST-TS and CT significantly differed in all variables (p<0.001). However, the 277
correlations between the methods improved markedly, ranging from 0.809 to 0.934 for Rater 1 and 278
from 0.873 to 0.922 for Rater 2 (p<0.01 for all correlations).
279
Discussion and implications 280
In this study, inter-trial and intra- and inter-rater reliability estimates for a simple, low-cost F-v 281
profiling method based on split times measured from video in ice hockey skating are presented.
282
14
Also the consistency of the method compared to the F-v profile estimated from velocity data de- 283
rived from continuous tracking of the player was assessed. The results showed that uncertainty in 284
the detection of the time instant of the beginning of horizontal force generation (start time) results 285
in significant measurement variability to the calculated sprint mechanical parameters. The results 286
also show that this uncertainty can be reduced by utilising an optimisation-based approach to esti- 287
mate the actual start time named here the time shift method. The concept of correcting start time 288
is not novel and has been previously used to account for the movement that occurs before trigger- 289
ing the timing when using timing gates to measure split times (Helland et al., 2019). However, 290
previously a constant correction has been used for each trial whereas, in our approach, the correc- 291
tion is unique to each trial and can be either positive or negative. Hence, the approach is similar to 292
the one that has been utilised for continuous velocity data obtained using laser or radar device 293
(Morin et al., 2019; Perez et al., 2019).
294
Inter-trial repeatability 295
None of the variables of skating sprint mechanics significantly differed between the three fastest 296
trials, although in the 30-meter time, significant differences were observed. Some trials may have 297
had better initial acceleration, but worse performance in the later phases of the sprint, which pos- 298
sibly explains why systematic differences between the trials in the sprint mechanical variables 299
were not observed. Moreover, inter-trial reliability analysis revealed estimates for both relative 300
(ICC) and absolute (SEM) reliability that were systematically worse than estimates for intra-rater 301
reliability. This finding suggests that there were actual differences between the three trials and that 302
the observed variability is not just due to measurement error. Hence, selecting a trial with the 303
fastest average speed (30-meter time) for F-v profiling will not guarantee that all sprint mechanical 304
variables are maximised.
305
15
The effect of the time shift on the inter-trial reliability was inconsistent. On average, the time shift 306
improved reliability slightly for the Rater 1 but impaired the estimates slightly for the Rater 2. This 307
may be explained by actual differences between the trials. Therefore, even if the time shift method 308
would reduce analysis error, it would not improve inter-trial reliability. Interestingly, the time shift 309
method worsened inter-trial reliability for V0 in both raters, which may be explained by improved 310
sensitivity to detect small differences between trials in V0. 311
Inter- and intra-rater reliability 312
There were significant differences in Pmax and RFmax values between the analyses done by different 313
raters, which were removed by the time shift method. Hence, the differences in the values observed 314
between the raters were probably due to systematic differences in the detection of the movement 315
start instant. Therefore, it is preferable to use a single rater for the analyses. However, this is not 316
as critical when using the time shift as this method removes systematic differences between raters 317
and yields good to excellent relative inter-rater reliability (ICC). However, inter-rater reliability 318
remained still slightly worse than intra-rater reliability, even when using the time shift method.
319
Systemic differences between the measurements were not observed in the intra-rater reliability 320
analysis, but the analysis showed that a considerable amount of variability in the results is due to 321
random measurement error. Utilization of the time shift method markedly improved the intra-rater 322
reliability, which again suggests that the method reduced the amount of random measurement er- 323
ror. On average, ICC improved from 0.631 to 0.955 and SEM from 5.8% to 1.8%. FV slope and 324
Drf showed the largest errors with SEM of 3.4% for both variables. Hence, differences of less than 325
5% in sprint mechanical variables should be interpreted with caution. It should be noted that only 326
16
a single rater was included in the intra-rater reliability analysis and hence, the results are not gen- 327
eralizable. Therefore, intra-rater reliability may differ for different raters, but the observation that 328
the time shift method improves intra-rater reliability should not depend on the rater.
329
Consistency with continuous tracking of player’s movement 330
The F-v profile estimated from continuous tracking of player's movement was considered here as 331
the most accurate estimate of the true F-v profile since observations of player’s position were 332
obtained every 0.042 s throughout the sprint hence giving ample amount of data for modelling the 333
velocity as a function of time. Therefore, F-v profiles obtained using split times were compared 334
against F-v profiles obtained using continuous tracking.
335
The mean values obtained using the ST significantly differed from the values obtained using the 336
CT only in few variables, and these effects were not consistent between the raters. On the other 337
hand, mean values in all variables, and for both raters, were significantly different from the CT 338
when using the ST-TS. Compared to the CT, the ST-TS yielded smaller F0, Pmax, and RFmax and 339
larger V0, FV slope, Drf, and Max speed. The results may be explained by the fact that the time 340
shift was, in most cases, positive (average time shift +0.14 s) increasing the time from the start to 341
2.5 meters. Hence, the time shift lowered the average speed and acceleration at the beginning of 342
the sprint, which is the part of the sprint where F0, Pmax, and RFmax are observed. In addition, the 343
lower curvature of the velocity-time relationship causes the modelled speed to reach higher val- 344
ueswhich subsequently lead to the largest differences between the ST-TS and CT in FV slope and 345
Drf. The time shift had the smallest effect on Max speed, which occurs at an intermediate time 346
between initial acceleration and theoretical maximal speed and is thus a variable that may be least 347
affected by the curvature of the modelled velocity-time relationship.
348
17
Although time shift induced a significant bias to the sprint mechanical variables compared to the 349
valued obtained using the CT, the correlation between the results obtained based on the split times 350
and based on the CT markedly improved when using the ST-TS. The mean correlation coefficient 351
was 0.501 with the ST and only 5 out of 17 variables showed a statistically significant correlation.
352
In comparison, a strong relationship (mean correlation coefficient 0.899) was observed for all var- 353
iables with the ST-TS. Hence, it seems that the ST-TS reduces random measurement errors and 354
hence improves the concurrent validity of the simple, low-cost field method.
355
Practical implications and suggestions for future studies 356
Sprint mechanical parameters calculated using the macroscopic inverse dynamics approach uti- 357
lised in the current study are determined by the observations of athlete’s position at different time 358
instant during the sprint. Hence, the information provided by the methods is not different than 359
would be possible to detect directly from split times. However, the method provides a way to make 360
intuitive synthesis from the split times. Moreover, the sprint mechanical parameters are the key 361
determinant of sprint performance whereas split times are the results of the performance. There- 362
fore, the sprint mechanical parameters are more closely connected to the different physical and 363
technical capabilities of the athlete providing valuable information for coaches.
364
Improvements to the intra- and inter-rater reliability and concurrent validity by ST-TS supposedly 365
enhances the practical usability of the F-v profiling based on split times. Based on Perez et al.
366
(2019), averaging two trials will further improve test-retest reliability and hence this is suggested 367
practise for athlete monitoring. However, if the aim is to describe athlete’s peak performance, the 368
fastest trial out of several sprints can be selected for analysis since an exploration of the data of 369
the current study showed that all sprint mechanical variables, except V0, significantly correlate 370
with 30-meter time. Therefore, selecting the best trial based on the 30-meter time will ensure that 371
18
the estimated sprint mechanical variables closely match the athlete’s maximal capacity. For V0, a 372
relatively low standard error of measurement between trials was found when using the ST-TS.
373
Hence, although V0 may not be maximised if choosing the best trial based on 30-meter time for 374
the analysis, the value will still probably well describe the athlete’s current performance level.
375
Future studies should establish minimal detectable change estimates using a test-retest setting and 376
investigate if the time shift improves sensitivity to detect changes in sprint mechanics due to train- 377
ing, injury, or fatigue, providing further support for its practical usability. Further investigations 378
of validity against direct measurements of horizontal force generation or against laser or radar 379
measurements are warranted. Moreover, future studies could establish regression analysis -based 380
correction to the ST-TS method if good absolute agreement between the results obtained using 381
continuous position data (video, laser or radar) and simple low-cost method is required.
382
Conclusion 383
The current study shows that the previously established simple, low-cost field method for F-v 384
profiling provides, on average, similar results compared to the results obtained using continuous 385
tracking of movement in ice hockey skating. However, both intra- and inter-rater reliability and 386
concurrent validity were only moderate. These findings suggest that random measurement errors 387
in the simple, low-cost field methods hinder practical use of the method in ice hockey skating. A 388
methodological variation of the previously established simple, low-cost field methods, in which 389
the instant of sprint start is found by optimisation (time shift method), was shown to improve intra- 390
and inter-rater reliability and concurrent validity. The method is also likely to improve intra- and 391
inter-rater reliability of split time -based F-v profiling in sprint running, providing improvement 392
to the approach proposed by Samozino et al. (2016), but this should be investigated in future stud- 393
ies.
394
19
In conclusion, the result of the current study showed that a simple, low-cost field method based on 395
split times analysed from a high-speed video can be used as a reliable method to test important 396
mechanical determinants of skating performance in ice hockey players, especially when utilising 397
the optimisation-based approach to reduce error associated with identification of sprint start.
398
Acknowledgments 399
The authors wish to thank Mr. Sami Kaartinen for his help with the measurements and participant 400
recruitment. This work was supported by the European Regional Developments Fund and the Uni- 401
versity of Eastern Finland under the project: Human measurement and analysis - research and 402
innovation laboratories (HUMEA, project identifiers: A73200 and A73241). No potential conflict 403
of interest was reported by the authors.
404
References 405
Arsac, L. M., & Locatelli, E. (2002). Modeling the energetics of 100-m running by using speed 406
curves of world champions. Journal of Applied Physiology, 92, 1781–1788. doi:
407
10.1152/japplphysiol.00754.2001 408
Cross, M. R., Brughelli, M., Samozino, P., & Morin, J. B. (2017). Methods of Power-Force- 409
Velocity Profiling During Sprint Running: A Narrative Review. Sports Medicine, 47, 1255–
410
1269. doi: 10.1007/s40279-016-0653-3 411
Di Prampero, P. E., Fusi, S., Sepulcri, L., Morin, J. B., Belli, A., & Antonutto, G. (2005). Sprint 412
running: A new energetic approach. Journal of Experimental Biology, 208, 2809–2816. doi:
413
10.1242/jeb.01700 414
Haugen, T. A., Breitschädel, F., & Samozino, P. (2018). Power-Force-Velocity Profiling of 415
Sprinting Athletes. Journal of Strength and Conditioning Research, (in press). doi:
416
20 10.1519/JSC.0000000000002890
417
Helland, C., Haugen, T., Rakovic, E., Eriksrud, O., Seynnes, O., Mero, A. A., & Paulsen, G.
418
(2019). Force–velocity profiling of sprinting athletes: single-run vs. multiple-run methods.
419
European Journal of Applied Physiology, 119, 465–473. doi: 10.1007/s00421-018-4045-2 420
Morin, J.-B., Bourdin, M., Edouard, P., Peyrot, N., Samozino, P., & Lacour, J. R. (2012).
421
Mechanical determinants of 100-m sprint running performance. European Journal of 422
Applied Physiology, 112, 3921–3930. doi: 10.1007/s00421-012-2379-8 423
Morin, J.-B., Edouard, P., & Samozino, P. (2011). Technical ability of force application as a 424
determinant factor of sprint performance. Medicine and Science in Sports and Exercise, 43, 425
1680–1688. doi: 10.1249/mss.0b013e318216ea37 426
Morin, J.-B., & Samozino, P. (2016). Interpreting power-force-velocity profiles for 427
individualized and specific training. International Journal of Sports Physiology and 428
Performance, 11, 267–272. doi: 10.1123/ijspp.2015-0638 429
Morin, J.-B., & Samozino, P. (2019). Data from: Spreadsheet for sprint acceleration force- 430
velocity-power profiling [Dataset]. ResearchGate. Retrieved from https://www.re- 431
searchgate.net/publication/321767606_Spreadsheet_for_Sprint_acceleration_force-velocity- 432
power_profiling 433
Morin, J.-B., Samozino, P., Murata, M., Cross, M. R., & Nagahara, R. (2019). A simple method 434
for computing sprint acceleration kinetics from running velocity data: Replication study 435
with improved design. Journal of Biomechanics, 94, 82–87. doi:
436
10.1016/j.jbiomech.2019.07.020 437
Perez, J., Guilhem, G., & Brocherie, F. (2019). Reliability of the force-velocity-power variables 438
21
during ice hockey sprint acceleration. Sports Biomechanics, (in press). doi:
439
10.1080/14763141.2019.1648541 440
Roczniok, R., Stanula, A., Maszczyk, A., Mostowik, A., Kowalczyk, M., Fidos-Czuba, O., &
441
Zając, A. (2016). Physiological, physical and on-ice performance criteria for selection of 442
elite ice hockey teams. Biology of Sport, 33, 43–48. doi: 10.5604/20831862.1180175 443
Romero-Franco, N., Jiménez-Reyes, P., Castaño-Zambudio, A., Capelo-Ramírez, F., Rodríguez- 444
Juan, J. J., González-Hernández, J., … Balsalobre-Fernández, C. (2017). Sprint 445
performance and mechanical outputs computed with an iPhone app: Comparison with 446
existing reference methods. European Journal of Sport Science, 17, 386–392. doi:
447
10.1080/17461391.2016.1249031 448
Samozino, P. (2018). A Simple Method for Measuring Force, Velocity and Power Capabilities 449
and Mechanical Effectiveness During Sprint Running [Adobe Digital Editions version]. In 450
J.-B. Morin & P. Samozino (Ed.), Biomechanics of training and testing: innovative 451
concepts and simple field methods (pp. 237–267). Cham: Springer International Publishing.
452
doi: 10.1007/978-3-319-05633-3_11 453
Samozino, P., Rabita, G., Dorel, S., Slawinski, J., Peyrot, N., Saez de Villarreal, E., & Morin, J.
454
B. (2016). A simple method for measuring power, force, velocity properties, and 455
mechanical effectiveness in sprint running. Scandinavian Journal of Medicine and Science 456
in Sports, 26, 648–658. doi: 10.1111/sms.12490 457
Weir, J. P. (2005). Quantifying test-retest reliability using the intraclass correlation coefficient 458
and the SEM. Journal of Strength and Conditioning Research, 19, 231–240. doi:
459
10.1519/15184.1 460
461
22
Table 1. Description and relevance of the different sprint mechanical variables determined with the force-velocity profiling.
Variable name Unit Description Meaning or relevance
F0 N/kg Theoretical maximal force Determines the maximal horizontal acceleration capacity when velocity is zero.
V0 m/s Theoretical maximal velocity Gives indication of athlete’s velocity capacity.
Pmax W/kg Maximal power Peak power output, i.e. maximum rate of change in kinetic energy.
FV slope Ns/kgm Slope of the force-velocity relationship Higher value (smaller slope) indicates smaller decrements in acceleration capacity with increase in velocity.
Drf %s/m Slope of the ratio of force-velocity relationship Higher value (smaller slope) indicates smaller decrements in the ability to direct ground reaction force horizontally with increase in velocity.
Max speed m/s Speed at 30-meters Maximal velocity obtained during the 30-meter sprint.
RFmax % Maximal value of the ratio of force Gives indication of the ability to direct ground force horizontally at the beginning of the sprint and is related to both technical and physical capabilities.
23
Table 2. Descriptive values of the sprint mechanical variables calculated using the split time, split time with time shift and continuous tracking methods.
Split time Split time with time shift Continuous tracking
Trial 1 Trial 2 Trial 3 Trial 1 Trial 2 Trial 3 Trial 1
F0 (N/kg) 6.48±0.87 6.73±0.54 6.51±0.75 5.39±0.63 5.49±0.51 5.37±0.67 6.43±1.04
V0 (m/s) 9.30±0.47 9.15±0.26 9.21±0.46 10.01±0.46 9.89±0.37 9.96±0.52 9.08±0.39
Pmax (W/kg) 15.00±1.54 15.39±1.19 14.95±1.47 13.43±1.20 13.55±1.06 13.32±1.27 14.52±1.95
FV slope (Ns/kgm) -0.70±0.12 -0.74±0.07 -0.71±0.10 -0.54±0.08 -0.56±0.07 -0.54±0.09 -0.71±0.14
Drf (%s/m) -6.58±1.12 -6.90±0.61 -6.67±0.09 -5.10±0.76 -5.24±0.60 -5.11±0.86 -6.69±1.26
Max speed (m/s) 8.57±0.24 8.53±0.19 8.52±0.26 8.81±0.19 8.79±0.19 8.78±0.26 8.41±0.19
RFmax (%) 41.63±2.59 42.48±1.67 41.70±2.31 38.31±2.48 38.72±1.94 38.16±2.43 41.06±3.13
Values are from the analyses performed by Rater 1 and are presented as mean±SD. Different trials represent the three fastest trials selected for analysis. Continuous tracking was performed only for the fastest trial. F0, theoretical maximal force; V0, theoretical maximal velocity; Pmax, maximal power; FV slope, slope of the force-velocity relationship; Drf, slope of the ratio of force-velocity relationship; Max speed, speed at 30-meters; RFmax, maximal value of the ratio of force.
24
Table 3. Inter-trial reliability of the different sprint mechanical variables calculated using the split time and split time with time shift methods.
Split time Split time with time shift
p-value ICC SEM% p-value ICC SEM%
Rater 1
F0 (N/kg) 0.506 0.411 8.6 0.677 0.660 6.6
V0 (m/s) 0.446 0.500 3.2 0.743 0.343 3.7
Pmax (W/kg) 0.454 0.580 6.1 0.490 0.841 3.5
FV slope (Ns/kgm) 0.549 0.352 11.3 0.758 0.538 10.0
Drf (%s/m) 0.551 0.351 11.0 0.761 0.527 10.1
Max speed (m/s) 0.581 0.711 1.5 0.825 0.596 1.6
RFmax (%) 0.370 0.486 3.8 0.495 0.737 3.1
Rater 2
F0 (N/kg) 0.843 0.731 5.9 0.594 0.710 6.2
V0 (m/s) 0.923 0.515 2.5 0.822 0.198 3.8
Pmax (W/kg) 0.794 0.859 3.8 0.393 0.887 3.1
FV slope (Ns/kgm) 0.848 0.614 8.3 0.715 0.557 9.8
Drf (%s/m) 0.847 0.601 8.2 0.724 0.541 9.7
Max speed (m/s) 0.868 0.646 1.5 0.989 0.497 1.7
RFmax (%) 0.745 0.845 2.0 0.432 0.797 2.8
P-value is for the difference in mean values between trials (repeated measured ANOVA). ICC, intraclass correlation coefficient; SEM%, standard error of measurement presented as a percentage of the mean; F0, theoretical maximal force; V0, theoretical maximal velocity; Pmax, maximal power; FV slope, slope of the force-velocity relationship; Drf, slope of the ratio of force-velocity relationship; Max speed, speed at 30-meters; RFmax, maximal value of the ratio of force.
25
Table 4. Inter-rater reliability of the different sprint mechanical variables calculated using the split time and split time with time shift methods.
Split time Split time with time shift
p-value ICC SEM% p-value ICC SEM%
F0 (N/kg) 0.050 0.625 6.9 0.755 0.926 3.2
V0 (m/s) 0.070 0.495 2.7 0.679 0.827 1.8
Pmax (W/kg) 0.039 0.759 4.5 0.999 0.963 1.8
FV slope (Ns/kgm) 0.058 0.565 9.1 0.650 0.901 4.7
Drf (%s/m) 0.058 0.562 8.9 0.643 0.898 4.8
Max speed (m/s) 0.141 0.726 1.2 0.338 0.930 0.5
RFmax (%) 0.032 0.702 2.8 0.881 0.925 1.8
P-value is for the difference in mean values between analyses performed by Rater 1 and Rater 2. ICC, intraclass correlation coefficient; SEM%, standard error of measurement presented as a percentage of the mean; F0, theoretical maximal force; V0, theoretical maximal velocity; Pmax, maximal power; FV slope, slope of the force-velocity relationship; Drf, slope of the ratio of force-velocity relationship; Max speed, speed at 30-meters; RFmax, maximal value of the ratio of force.
26
Table 5. Intra-rater reliability of the different sprint mechanical variables calculated using the split time and split time with time shift methods.
Split time Split time with time shift
p-value ICC SEM% p-value ICC SEM%
F0 (N/kg) 0.392 0.548 7.8 0.378 0.965 2.1
V0 (m/s) 0.839 0.670 2.6 0.956 0.920 1.3
Pmax (W/kg) 0.296 0.665 5.2 0.090 0.985 1.0
FV slope (Ns/kgm) 0.435 0.529 10.4 0.542 0.951 3.4
Drf (%s/m) 0.443 0.533 10.2 0.560 0.949 3.4
Max speed (m/s) 0.830 0.832 1.1 0.670 0.944 0.5
RFmax (%) 0.451 0.642 3.3 0.323 0.971 1.1
P-value is for the difference in mean values between repeated analyses performed by Rater 1. ICC, intraclass correlation coefficient; SEM%, standard error of measurement presented as a percentage of the mean; F0, theoretical maximal force; V0, theoretical maximal velocity; Pmax, maximal power; FV slope, slope of the force-velocity relationship; Drf, slope of the ratio of force-velocity relationship; Max speed, speed at 30-meters; RFmax, maximal value of the ratio of force.
27
Table 6. Concurrent validity of the different sprint mechanical variables calculated using the split time and split time with time shift methods compared to the values calculated using the continuous tracking method.
Split time Split time with time shift
% difference p-value r p-value % difference p-value r p-value
Rater 1
F0 (N/kg) 0.85 0.846 0.520 0.083 -16.14 < 0.001 0.927 < 0.001
V0 (m/s) 2.41 0.192 0.211 0.510 10.16 < 0.001 0.834 0.001
Pmax (W/kg) 3.28 0.260 0.706 0.010 -7.51 0.003 0.904 < 0.001
FV slope (Ns/kgm) -1.46 0.805 0.405 0.192 -23.94 < 0.001 0.917 < 0.001
Drf (%s/m) -1.69 0.772 0.392 0.208 -23.83 < 0.001 0.915 < 0.001
Max speed (m/s) 2.00 0.030 0.423 0.171 4.83 < 0.001 0.805 0.002
RFmax (%) 1.39 0.465 0.598 0.040 -6.71 < 0.001 0.934 < 0.001
Rater 2
F0 (N/kg) 7.26 0.086 0.590 0.044 -16.49 < 0.001 0.908 < 0.001
V0 (m/s) 0.16 0.899 0.317 0.315 10.50 < 0.001 0.909 < 0.001
Pmax (W/kg) 7.83 0.015 0.720 0.008 -7.51 0.004 0.873 < 0.001
FV slope (Ns/kgm) 6.59 0.216 0.485 0.110 -24.62 < 0.001 0.921 < 0.001
Drf (%s/m) 6.15 0.237 0.472 0.121 -24.52 < 0.001 0.922 < 0.001
Max speed (m/s) 1.20 0.058 0.568 0.054 5.06 < 0.001 0.921 < 0.001
RFmax (%) 4.28 0.024 0.671 0.017 -6.82 < 0.001 0.894 < 0.001
Differences were calculated as the split time method – continuous tracking and presented as a percentage of continuous tracking value (% difference). r, Pearson correlation coeffi- cient; F0, theoretical maximal force; V0, theoretical maximal velocity; Pmax, maximal power; FV slope, slope of the force-velocity relationship; Drf, slope of the ratio of force-velocity relationship; Max speed, speed at 30-meters; RFmax, maximal value of the ratio of force.
28 Figure captions
Figure 1. Screenshot of a video record of a skating trial. The vertical sticks used for F-v profiling are highlighted.
Inset shows an enlarged view of the player and sticks at the starting position of the trial.
29
Figure 2. Example data and force-velocity and power-velocity profiles for one participant. An example of velocity-time data obtained from continuous tracking of a player’s movement is shown in panel A. Panel B shows a comparison of the three methods used in the study. The dotted line represents the position-time relationship modelled based on values obtained from continuous tracking of the player’s position. Black dots and the line is for split time measurements without time shift, and gray dots and line are for split time measurement with the time shift. In this example, the time shift applied to the data was +0.133 s. Panel C shows the resulting force- velocity (continuous line) and power-velocity (dotted line) profiles for the three methods. Black is for split time method without time shift, light gray for split time method with the time shift, and dark gray for continuous tracking of player’s movement.
30
