Nosek, P. et al.: THE USE OF PRINCIPAL COMPONENT ANALYSIS FOR...

Kinesiology 55(2023)2:202-212

THE USE OF PRINCIPAL COMPONENT ANALYSIS

FOR REDUCTION OF TRAINING LOAD DATA
IN PROFESSIONAL SOCCER

Perry Nosek1,2, Matthew Andrew3, Mladen Sormaz4,

Barry Drust5, and Thomas E. Brownlee5
Leicester City Football Club, Leicester, UK
1

School of Sport and Exercise Science, Liverpool John Moores University, UK

2

3
Department of Sport and Exercise Sciences, Institute of Sport,
Manchester Metropolitan University, Manchester, UK
4
777 Partners Football Group, Miami, USA
5
School of Sport, Exercise and Rehabilitation Sciences, University of Birmingham, UK

Original scientific paper

DOI 10.26582/k.55.2.5

Abstract:
The aim of this study was to explore the use of principal component analysis (PCA) in understanding
multivariate relationships in soccer training load data. Training load data were collected from 20 professional
male soccer players during a 28-week in-season period. Twelve training load variables (total distance,
PlayerLoadTM, low-speed running distance, moderate-speed running distance, high-speed running distance,
sprint distance, moderate-speed running efforts, high-speed running efforts, sprint efforts, accelerations,
decelerations, and changes of direction) were collected during training sessions, with correlation analysis
revealing high intercorrelation between most variables (r = 0.04-0.98). Principal component analysis was
performed on datasets containing all players and on individual players. On the whole dataset, two principal
components were retained explaining a total of 81% of data variance. The first component comprised variables
associated with distances in speed zones and the second component changes of direction. Whilst some
individual variation existed among players, distances in speed zones were loaded on the first component
and inertial movement analysis variables, such as accelerations, decelerations, and changes of direction,
were loaded on the second component. These findings evidence the strong relationships between several
common training load variables and highlight the risk of data redundancy. By selecting variables from
each component, practitioners can reduce the number of variables reported whilst retaining as much of the
variation in data as possible.

Key words: training load, Global Positioning Units (GPS), multivariate analysis, football, feedback

Introduction sion-making of key stakeholders (e.g., coaches). For

example, coaches generally support the usefulness
To maximise soccer performance, professional and importance of training data collected by these
players participate in training sessions (i.e., prac- technologies (Nosek, Brownlee, Drust & Andrew,
tice) with the aim to improve technical, tactical, 2021; Weston, 2018), yet it is still unclear whether
physical, and psychological performance (Morgans, this information is impactful in aiding decision-
Orme, Anderson, & Drust, 2014; Williams & Reilly, making and may represent the ‘translational gap’
2000). From a physical perspective, the volume and often suggested within sport science (Eisenmann,
intensity of these sessions, known collectively as the 2017; Fullagar, McCall, Impellizzeri, Favero, &
training load (Impellizzeri, et al., 2019), are planned Coutts, 2019). Such gaps may be due to coaches
and manipulated to produce this desired response. perceiving a lack of a common goal with their sport
Utilisation of soccer training load monitoring tech- science departments as a barrier towards using
nologies, such as Global Positioning Systems (GPS), training data in coach decision-making (Nosek, et
is now considered common in soccer (Akenhead & al., 2021).
Nassis, 2015). The success of using such devices lies Feedback on performance related data is a key
in their ability to support and influence the deci- step in the coaching process (Franks & Goodman,

202
Nosek, P. et al.: THE USE OF PRINCIPAL COMPONENT ANALYSIS FOR... Kinesiology 55(2023)2:202-212

2008) and involves the collection, analysis, and the classification accuracy of PCs against the win/
delivery/communication of data to those respon- loss probabilities in rugby league and reported a
sible for decision-making (e.g., coaches; Buch- 90% accuracy. Similarly, Williams et al. (2017)
heit, 2017). Perhaps the most important part of this examined the relationship between variables in each
process may be the communication of this data to PC against injury risk in rugby union, reporting
coaches as if done poorly, it can create a barrier to that 4-week cumulative load, acute:chronic work-
data utilisation (Nosek, et al., 2021). Consequently, load ratio, and daily workload were the measures
the delivery format and complexity of chosen data describing the largest amount of variation in injury
must be considered carefully when designing feed- risk from the first three PCs, respectively. Taken
back to ensure efficient communication of key data together, these studies provide a framework to
to stakeholders and decision makers. Adding to reduce large datasets into groups of variables that
feedback complexity is potential disagreements can then be explored against outcome measures of
between coaches and sport scientists regarding the interest (Williams, et al., 2017).
information provided. For example, Weston (2018) Recently, PCA has been applied to training
reported differences between coaches and practi- data collected via GPS with the aim of reducing
tioners (sport scientists; fitness coaches; strength the number of variables used in training load moni-
and conditioning coaches) regarding the frequency, toring. For instance, Weaving et al. (2018) iden-
timing, and expertise requirements of training load tified 60-70% of training load variance in field-
reports. Furthermore, whilst practitioners identify based skills training could be provided by a PC
typically using 4-10 variables in reports (Akenhead containing total distance, sRPE or PlayerLoadTM,
& Nassis, 2015), coaches have suggested receiving supporting the use of these variables as moni-
too much information is a barrier towards using toring measures. Moreover, Weaving et al. (2019)
training data to inform decision-making (Nosek, et demonstrated how 12 training load variables could
al., 2021). These variables often show strong rela- be transformed into a 2D scatterplot, allowing for
tionships with each other, with intercorrelations heuristic decision-making such as modifying future
existing between various measures such as total training content after comparing present day data
distance, PlayerLoadTM (sum of accelerations in all to historical benchmarks. In soccer, Maughan et al.
planes collected via tri-axial accelerometer), high- (2021) illustrated that multiple measures of subject-
speed running, sprinting, session rating of perceived ive and external training load variables could be
exertion (sRPE), and heart rate (HR)-derived meas- reduced to two PCs that explained 83% of the vari-
ures during soccer training (Casamichana, Castel- ance within the data, one which contained all the
lano, Calleja-Gonzalez, San Román, & Castagna, variables to represent total training load, and one
2013; Maughan, MacFarlane, & Swinton, 2021; which contrasted subjective and external measures.
McLaren, et al., 2018). This high intercorrela- Furthermore, during the competitive season, PCA
tion may result in data ‘overload’ for coaches, and produced only one PC which housed all training
reporting many variables appears unnecessary and load variables excluding sprinting, which suggests
may increase complexity during decision-making all the variables used represent similar underlying
(Weaving, Beggs, Dalton-Barron, Jones, & Abt, information and could theoretically be used inter-
2019; Weaving, Marshall, Earle, Nevill, & Abt, changeably (Maughan, MacFarlane, & Swinton,
2014). 2022). Despite some interesting results here,
Reducing variable number (dimensionality research utilising PCA within professional soccer
reduction) can be complex, as simple variable is limited.
removal can lead to information loss. A popular Whilst most training load PCA studies analyse
dimension reduction technique is principal compo- whole squad data (Parmar, et al., 2018; Ryan,
nent analysis (PCA), which extracts important Kempton, & Coutts, 2021; S. Williams, et al.,
information from correlated variables and expresses 2017), this may result in individual player charac-
them as new, uncorrelated compound variables teristics being hidden in the analysis. One possible
named principal components (PCs) (Jolliffe, 1986; approach is to perform PCA on each player. Using
Jollife & Cadima, 2016). In a sporting context, this approach, Weaving et al. (2018) reported that
PCA has previously been used to examine tech- the same variables were loaded on each PC for all
nique analysis (Federolf, Reid, Gilgien, Haugen, rugby union players. Despite this, some individual
& Smith, 2014; Gløersen, Myklebust, Hallén, & variation in loadings were reported meaning that
Federolf, 2018), injury risk (Williams, Trewartha, utilising PC scores (i.e., standardized training load
Cross, Kemp, & Stokes, 2017), performance indi- data multiplied by PC loadings for each variable),
cators (Parmar, James, Hearne, & Jones, 2018), and as suggested by the authors, would only allow
training load (Weaving, et al., 2014, 2018, 2019; within-player comparisons and not between-player.
Weaving, Jones, Till, Abt, & Beggs, 2017). The PCs Whilst PCA appears a suitable method for dimen-
produced can then have hypotheses framed around sion reduction of training load data, its application
them. For example, Parmar et al. (2018) examined in soccer is not understood. Therefore, the main aim

203
Nosek, P. et al.: THE USE OF PRINCIPAL COMPONENT ANALYSIS FOR... Kinesiology 55(2023)2:202-212

of the present study was to explore the use of PCA sive work in large areas of greater than 160 m2 per
to uncover multivariate relationships within soccer player over durations of 4-12 min, with increased
training load data. In line with previous research player numbers (8-22) to overload high-speed
(Casamichana, et al., 2013; Weaving, et al., 2018, running demands; and speed (MD-2; n = 29); work
2019) we hypothesized high intercorrelation would in moderately sized areas of 110-160 m2 focusing
exist within soccer training load data and that these on speed of play (Buchheit, Lacome, Cholley &
relationships would vary between players when Simpson, 2018), which involved various numbers
analysed with PCA. of players per activity. These themes typically
corresponded to MD-4, MD-3, and MD-2, respec-
Methods tively, though not in all cases. Training was typi-
cally conducted in the morning with games in the
Participants evening.
Twenty male professional soccer players (age Training content was designed to reflect the
20.50 ± 1.20 years, height 178.60 ± 6.60 cm [Seca fitness and physical status of individual players (i.e.,
213, Seca, Hamburg, Germany], weight, 80.10 ± number of days between games, acute and chronic
8.10 kg [Seca 876, Seca, Hamburg, Germany]) loadings, wellbeing responses to training) and
were recruited for this study. All players came therefore, whilst the areas, durations and players
from the same Professional Development Phase numbers outlined were typical of each theme, these
squad at an English Premier League club that was were not strict constraints for the sessions. Further-
competing in the Premier League 2-Divison 1 (i.e., more, some sessions included individual work and/
the highest tier) for the 2019/20 season. Goalkeepers or physical sessions designed to provide additional
were excluded from the analysis due to their vastly stimuli typically to substitutes known as ‘top-ups’,
different training and game physical requirements as required. Training data from rehabilitation, indi-
(Moreno-Pérez, et al., 2020). Players represented vidual, physical fitness and partially completed
the following primary positions: centre-back (n = (i.e., player injured during session) sessions were
4), full-back (n = 4), centre-midfield (n = 5), wide- excluded from the analyses.
midfield (n = 3) and forward (n = 4). The inclusion
criteria were that players must have completed a Equipment and procedures
minimum of 50 available sessions (mean 83 ± 12). During sessions, training load data were
Data were provided entirely as part of players’ normal recorded using 10 Hz GPS with an embedded 100
daily training routine; thus, no ethical approval Hz tri-axial accelerometer (Catapult Vector, Cata-
was required. The study did, however, conform pult Sports, Melbourne). Units were worn in a tight-
to the Declaration of Helsinki and Gatekeeper fitting vest placing the unit between the scapulae,
written consent was provided to allow data use. with players wearing the same unit for each session
to minimise inter-unit variability. All players had
Experimental design previous experience of wearing the vests and units
A longitudinal, observational design was used as it was part of routine working practice to collect
with training load data collected during 28 weeks of such data. Prior to training, units were turned on
the competitive, in-season period between August outside for 30 min to allow optimal connectivity
2019 and March 2020. Although the competitive with satellites. After each session, data were down-
season typically runs to May, the season concluded loaded into the manufacturer’s software (Open-
early due to the COVID-19 pandemic. Prior to the field v2.2) and inspected for artefacts (i.e., unre-
season, microcycle structure was designed with alistic spikes in velocity). Data were collected
input from coaching and sport science departments and analysed by a member of the sports science
to meet the tactical, technical, and physical require- department at the club who is a current professional
ments of the game. During weeks containing one doctorate student and lead author of the study. Data
match (MD; n = 18), this involved a recovery session were included if the number of connected satel-
the day after the match (i.e., match day plus one; lites was at least six and if the horizontal dilution
MD+1), followed by a day off. There would then of precision (HDOP) was <1.5 as per manufacturer
be three consecutive days of conditioned training guidance. Where these conditions were not met,
before a tactical themed activation session the day or full session data were not available due to other
before the next match (MD-1; n = 18). errors such as units running out of battery, these
The themes for the three conditioned training data were removed and replaced with positional
days were termed strength (MD-4; n = 25); inten- mean averages for that session (Jaspers, et al., 2018).
sive work in small areas of <110 m2 per player over This method was chosen to represent what had
duration bouts of 45-s to 4-min per activity, with occurred in the current session as opposed to using
small player numbers (2-12, to overload accelera- player previous data of sessions which may have
tion, deceleration, change of direction and meta- looked vastly different. This resulted in 11 out of
bolic demands; endurance (MD-3; n = 42); exten- 1466 (0.8%) individual player files being replaced.

204
Nosek, P. et al.: THE USE OF PRINCIPAL COMPONENT ANALYSIS FOR... Kinesiology 55(2023)2:202-212

Data were split into individual drills, exported tional averages. Redundancy of the dataset was
into a custom Excel spreadsheet and assigned examined using repeated-measures correlation to
the corresponding session theme. The variables view the strength of relationships between the vari-
selected for analysis are described in Table 1. These ables (Bakdash & Marusich, 2017). The qualita-
were based primarily on research involving training tive descriptors for the magnitude of the correla-
and match load and associated fitness, fatigue and tions were: <0.1 trivial; 0.1 to 0.3 small; 0.3 to 0.5
injury responses, alongside some variables used as moderate; 0.5 to 0.7 large; 0.7 to 0.9 very large; 0.9
part of a club-wide monitoring philosophy (Barrett, to 1.0 almost perfect (Hopkins, 2010). Additionally,
Midgley, & Lovell, 2014; Bradley, et al., 2009; Lute- the Bartlett test of sphericity and the Kaiser-Meyer-
berget, Holme, & Spencer, 2018). Accelerations, Olkin (KMO) measure of sampling adequacy were
decelerations, and changes of direction were calcu- performed to assess the suitability of the data for
lated based on the manufacturer’s inertial move- PCA. The Bartlett test of sphericity was significant
ment analysis (IMA). This uses accelerometer and (p<.01) with a KMO of 0.7, with a KMO above 0.5
gyroscope data to count one-step efforts, the magni- suggesting data were suitable for PCA (Williams,
tude of which is expressed as delta velocity (m·s-1) Onsman, & Brown, 2010).
(Luteberget, et al., 2018). Although limited research Data were mean-centred and scaled to unit
has shown IMA counts in intensity bands to possess variance allowing equal weighting across vari-
moderate reliability, this is shown to improve when ables with differing measurement units such as
multiple bands, such as medium and high intensity, distances and counts. PCA was performed on the
are aggregated (Luteberget, et al., 2018) as in the whole dataset using the singular value decomposi-
present study. Further, it is hoped that improvements tion method where components with an eigenvalue
in effort detection algorithms through updates in the of >1 were retained for analysis and indicated the
manufacturer’s software have improved this. Dwell PC accounted for more variance than a single orig-
time was set at 0.5s for GPS variables. inal variable alone (Kaiser, 1960). Visual inspec-
Though HR-derived variables were collected, tion of the scree plot to identify the ‘elbow’ of the
due to a change in measurement devices mid- eigenvalues was also utilised to support this deci-
season and data recording issues that arose from sion (Nguyen & Holmes, 2019). Factor loadings,
this change, these variables were removed from the the strength of a variable’s relationship with the
analyses. PC, were considered meaningful if it exceeded 0.7
(Rojas-Valverde, Pino-Ortega, Gómez-Carmona,
Data reduction and analysis & Rico-González, 2020). Subsequently, the same
Prior to PCA, the dataset, which consisted analysis was performed on each individual players’
of 1466 rows of data, was explored for missing/ dataset using the same method. All analysis was
erroneous data, which may have occurred due to performed in R (Version 4.0.0) using the Facto-
units not being worn or poor satellite connectivity. MineR package for PCA (Lê, Josse, Rennes, &
These data were replaced with the session posi- Husson, 2008).

Table 1. Description of GPS training load variables

Variable (Abbreviation) Description

Total Distance (TD) Total distance covered during the session.
Accumulated accelerometer data across vertical, medio-lateral, and anterior-
Player LoadTM (PL)
posterior planes, divided by a scaling factor of 100.
Low-Speed Running Distance (LSRD) Distance covered between 0 and 4.5m·s-1.
Moderate-Speed Running Distance (MSRD) Distance covered between 4.5 and 5.5m·s-1.
High-Speed Running Distance (HSRD) Distance covered between 5.5 and 7m·s-1.
Sprint Distance (SD) Distance covered above 7m·s-1.
Moderate-Speed Running Efforts (MSRE) Number of efforts between 4.5 and 5.5m·s-1.
High-Speed Running Efforts (HSRE) Number of efforts between 5.5 and 7m·s-1.
Sprint Efforts (SE) Number of efforts above 7m·s-1.
Number of medium and high intensity accelerations above 2.5m·s-1 derived from
Accelerations (ACC)
inertial movement analysis (IMA) using accelerometer and gyroscope data.
Number of medium and high intensity decelerations above 2.5m·s-1 derived from
Decelerations (DEC)
IMA using accelerometer and gyroscope data.
Number of medium and high intensity changes of direction to the left above
Changes of Direction (COD)
2.5m·s-1 derived from IMA using accelerometer and gyroscope data.

205
Nosek, P. et al.: THE USE OF PRINCIPAL COMPONENT ANALYSIS FOR... Kinesiology 55(2023)2:202-212

Results Discussion and conclusions

The repeated-measures correlation matrix is The primary aim of the present study was to
shown in Table 2. All correlations were significant explore the use of PCA to uncover multivariate rela-
(p<.01) except for between sprint efforts (SE) and tionships within soccer training load data to reduce
changes of direction (COD) (p=.12). reporting redundant data to coaches. Correlation
Principal component analysis of the whole analysis revealed strong relationships between
dataset extracted two PCs as having an eigenvalue multiple training load variables with subsequent
of >1 with PC1 (65%) and PC2 (16%) accounting PCA identifying two PCs explaining a combined
accumulatively for 81% of the variance in training 81% of the variance in training load data. Although
load data (Table 3). The correlation between each similar within-player results were produced, some
training load variable and each PC is also shown variation in variable loadings on each PC existed
in Table 3. Whilst all variables were loaded some- suggesting that some individual player character-
what on PC1, those relating to overall volume and istics may be hidden by analysing the full dataset.
running in different speed zones were loaded above Correlation analysis revealed most variables had
the meaningful threshold (Figure 1). IMA variables a moderate to almost perfect relationship, although
were highest loaded on PC2, however, only COD some correlations between IMA-based variables
were above the meaningful threshold (Table 3). and running variables were trivial or small (Table 2).
When individual players were analysed, all This highlights the redundancy in the dataset in that
players produced two PCs with eigenvalues >1 (PC1 many of the variables will change at similar rates
eigenvalue range 6.32-9.13; PC2 eigenvalue range across the training period and supports the need to
1.3-2.65). PC1 accounted for 53-76% of the vari- perform data reduction techniques to explore these
ance, whilst PC2 accounted for a further 11-22%. relationships further. These results are similar to
Loadings between the variables and PCs for each others, which have reported very large to almost
player are shown in Table 4 and Table 5. perfect correlations between total distance (TD)

Table 2. Correlation matrix (95% confidence intervals) for each training load variable during soccer training

TD PL LSRD MSRD HSRD SD MSRE HSRE SE ACC DEC COD

1
TD

0.98
PL (0.98-0.99) 1
AP
0.99 0.98
LSRD (0.99-1.00) (0.98-0.98) 1
AP AP
0.93 0.90 0.90
MSRD (0.92-0.94) (0.89-0.91) (0.88-0.91) 1
AP AP AP
0.88 0.85 0.83 0.89
HSRD (0.86-0.89) (0.83-0.86) (0.81-0.85) (0.87-0.90) 1
VL VL VL VL
0.67 0.65 0.63 0.60 0.75
SD (0.63-0.70) (0.61-0.69) (0.59-0.67) (0.56-0.64) (0.72-0.79) 1
L L L L VL
0.93 0.89 0.88 0.96 0.84 0.63
MSRE (0.92-0.93) (0.88-0.90) (0.86-0.89) (0.95-0.96) (0.83-0.86) (0.60-0.66) 1
AP VL VL AP VL L
0.85 0.80 0.78 0.89 0.94 0.75 0.92
HSRE (0.83-0.86) (0.78-0.82) (0.76-0.80) (0.88-0.90) (0.94-0.95) (0.72-0.77) (0.91-0.92) 1
VL VL VL VL AP VL AP
0.67 0.63 0.61 0.66 0.79 0.93 0.69 0.81
SE (0.65-0.70) (0.60-0.66) (0.58-0.64) (0.63-0.69) (0.77-0.80) (0.93-0.94) (0.66-0.72) (0.80-0.83) 1
L L L L VL AP L VL
0.37 0.45 0.40 0.24 0.24 0.26 0.24 0.15 0.13
ACC (0.32-0.43) (0.40-0.50) (0.35-0.45) (0.18-0.30) (0.18-0.30) (0.20-0.32) (0.19-0.29) (0.10-0.20) (0.08-0.18) 1
M M M S S S S S S
0.46 0.49 0.49 0.34 0.30 0.21 0.26 0.15 0.06 0.35
DEC (0.41-0.51) (0.44-0.54) (0.44-0.54) (0.28-0.39) (0.24-0.36) (0.15-0.27) (0.21-0.30) (0.09-0.20) (0.01-0.11) (0.30-0.41) 1
M M M M M S S S T M
0.48 0.55 0.51 0.36 0.31 0.22 0.28 0.13 0.04 0.58 0.65
COD (0.43-0.53) (0.50-0.59) (0.46-0.57) (0.30-0.41) (0.25-0.37) (0.16-0.28) (0.23-0.33) (0.08-0.19) (-0.01-0.09) (0.54-0.62) (0.62-0.69) 1
M L L M M S S S T L L

Note. AP – almost perfect, VL – very large, L – large, M – moderate, S – small, T – trivial; TD – Total Distance, PL – PlayerLoadTM,
MSRD – Moderate-Speed Running Distance, HSRD – High-Speed Running Distance, SD – Sprint Distance, MSRE – Moderate-
Speed Running Efforts, HSRE – High-Speed Running Efforts, SE – Sprint Efforts, ACC – Accelerations, DEC – Decelerations, COD
– Changes of Direction.

206
Nosek, P. et al.: THE USE OF PRINCIPAL COMPONENT ANALYSIS FOR... Kinesiology 55(2023)2:202-212

Table 3. Principal component analysis results for the training load data showing eigenvalues, percentage of variance explained,
cumulative variance explained, and component loadings for the first two PCs
PC1 PC2
Eigenvalue 7.78 1.96
% of total variance explained 64.85 16.35
Cumulative % of total variance explained 64.85 81.19
Variable loadings
TD 0.96 0.08
PL 0.95 0.15
LSRD 0.93 0.17
MSRD 0.91 -0.15
HSRD 0.88 -0.33
SD 0.74 -0.35
MSRE 0.95 -0.07
HSRE 0.93 -0.25
SE 0.79 -0.37
ACC 0.37 0.69
DEC 0.45 0.64
COD 0.47 0.75

Note. TD – Total Distance, PL – PlayerLoad , MSRD – Moderate-Speed Running Distance, HSRD – High-Speed Running Distance,
TM

SD – Sprint Distance, MSRE – Moderate-Speed Running Efforts, HSRE – High-Speed Running Efforts, SE – Sprint Efforts, ACC –
Accelerations, DEC – Decelerations, COD – Changes of Direction.
Loadings that met interpretation criteria (≥ 0.7) are highlighted in bold.

Table 4. Correlations between variables and PC1 for individual players

Player TD PL LSRD MSRD HSRD SD MSRE HSRE SE ACC DEC COD

1 0.98 0.98 0.97 0.96 0.96 0.88 0.97 0.97 0.92 0.28 0.67 0.63
2 0.98 0.98 0.97 0.96 0.96 0.88 0.97 0.97 0.92 0.37 0.40 0.30
3 0.96 0.94 0.93 0.81 0.84 0.65 0.95 0.91 0.75 0.28 0.57 0.55
4 0.95 0.94 0.90 0.84 0.54 0.60 0.96 0.88 0.56 0.69 0.59 0.76
5 0.98 0.96 0.96 0.95 0.95 0.86 0.97 0.96 0.90 0.47 0.50 0.39
6 0.98 0.98 0.97 0.95 0.91 0.82 0.96 0.94 0.85 0.35 0.42 0.64
7 0.96 0.95 0.94 0.93 0.86 0.74 0.95 0.94 0.81 0.57 0.50 0.57
8 0.96 0.94 0.92 0.90 0.87 0.78 0.95 0.91 0.80 0.27 0.43 0.25
9 0.93 0.87 0.85 0.79 0.70 0.77 0.91 0.88 0.78 0.17 0.16 0.18
10 0.95 0.94 0.91 0.80 0.71 0.80 0.89 0.85 0.77 0.41 0.51 0.61
11 0.98 0.98 0.96 0.96 0.92 0.85 0.96 0.95 0.90 0.64 0.60 0.51
12 0.97 0.98 0.96 0.94 0.94 0.66 0.96 0.95 0.80 0.60 0.52 0.71
13 0.97 0.97 0.94 0.93 0.89 0.85 0.96 0.93 0.90 0.55 0.49 0.60
14 0.96 0.91 0.90 0.84 0.84 0.80 0.94 0.91 0.79 0.40 0.39 0.34
15 0.98 0.98 0.97 0.92 0.95 0.52 0.95 0.95 0.64 0.34 0.68 0.79
16 0.97 0.95 0.92 0.93 0.85 0.54 0.96 0.93 0.66 0.11 0.48 0.51
17 0.98 0.97 0.96 0.93 0.95 0.71 0.96 0.96 0.78 0.58 0.38 0.50
18 0.98 0.97 0.97 0.98 0.96 0.85 0.97 0.98 0.85 0.41 0.18 0.12
19 0.97 0.94 0.93 0.89 0.82 0.82 0.94 0.91 0.88 0.43 0.53 0.49
20 0.97 0.93 0.93 0.84 0.76 0.75 0.94 0.91 0.81 0.39 0.54 0.52

Loadings that met interpretation criteria (≥ 0.7) are highlighted in bold.

and PlayerLoadTM (PL) in soccer (Casamichana, et lations of all other variables. In contrast, however,
al., 2013; Maughan, et al., 2021; Scantlebury, et al., whilst this study showed a very large relation-
2020), which is likely due to these measures being ship between TD and high-speed running distance
functions of the duration of the session and accumu- (HSRD), Scantlebury et al. (2020) reported only a

207
Nosek, P. et al.: THE USE OF PRINCIPAL COMPONENT ANALYSIS FOR... Kinesiology 55(2023)2:202-212

Table 5. Correlations between variables and PC2 for individual players

Player TD PL LSRD MSRD HSRD SD MSRE HSRE SE ACC DEC COD

1 -0.03 0.01 0.01 -0.16 -0.20 -0.11 -0.12 -0.15 -0.22 0.84 0.36 0.69
2 0.17 0.32 0.33 -0.28 -0.41 -0.42 -0.03 -0.29 -0.46 0.71 0.66 0.86
3 0.03 0.24 0.14 -0.34 -0.31 -0.28 -0.11 -0.24 -0.32 0.85 0.48 0.71
4 -0.23 -0.30 -0.37 0.29 0.70 0.64 0.07 0.41 0.68 -0.49 -0.43 -0.52
5 -0.02 0.11 0.04 -0.17 -0.22 -0.27 -0.09 -0.18 -0.25 0.64 0.68 0.82
6 0.03 0.11 0.12 -0.17 -0.29 -0.27 -0.09 -0.23 -0.32 0.72 0.69 0.68
7 0.06 0.13 0.12 -0.07 -0.33 -0.44 -0.02 -0.22 -0.39 0.35 0.69 0.68
8 0.13 0.26 0.23 -0.15 -0.40 -0.32 -0.02 -0.26 -0.43 0.81 0.76 0.84
9 0.20 0.37 0.33 -0.22 -0.47 -0.23 -0.05 -0.26 -0.28 0.59 0.82 0.87
10 0.10 0.26 0.23 -0.41 -0.53 -0.18 -0.28 -0.42 -0.19 0.75 0.66 0.68
11 0.04 0.09 0.10 -0.15 -0.20 -0.31 -0.12 -0.22 -0.28 0.38 0.56 0.72
12 -0.06 -0.02 -0.01 -0.23 -0.25 -0.09 -0.14 -0.21 -0.17 0.55 0.72 0.52
13 0.07 0.10 0.18 -0.25 -0.36 -0.26 -0.13 -0.28 -0.28 0.60 0.67 0.69
14 0.14 0.29 0.28 -0.28 -0.40 -0.34 -0.11 -0.28 -0.33 0.63 0.72 0.82
15 -0.10 -0.10 -0.13 -0.14 0.15 0.80 -0.11 0.11 0.72 0.11 -0.50 -0.34
16 0.15 0.26 0.30 -0.16 -0.41 -0.53 -0.01 -0.23 -0.56 0.50 0.60 0.72
17 0.02 0.07 0.09 -0.15 -0.21 -0.32 -0.09 -0.18 -0.33 0.40 0.75 0.77
18 0.01 0.10 0.04 -0.02 -0.12 -0.32 0.07 -0.01 -0.35 0.60 0.81 0.91
19 0.12 0.27 0.25 -0.29 -0.41 -0.34 -0.15 -0.32 -0.30 0.70 0.56 0.76
20 0.09 0.29 0.23 -0.30 -0.54 -0.27 -0.13 -0.34 -0.35 0.75 0.62 0.75

Loadings that met interpretation criteria (≥ 0.7) are highlighted in bold.

ship in the present study. These differences may be

due to the method used to define HSRD (distance
covered above 61% of a player’s maximum velocity
compared with an arbitrary threshold of 5.5m·s-1 in
the present study) or the large differences in training
output and methodology used by semi-professional
players used by Scantlebury et al (2020). Whilst
correlations between running-based variables were
strong, those between IMA variables and running-
based variables were weaker. This could be due to
the different physical qualities being captured by
IMA variables since the one-step actions of accel-
erations (ACC), decelerations (DEC) and changes
of direction (COD) are highly independent of any
running effort in different speed zones. For example,
to produce an effort in the sprint speed zone, a player
must travel through all previous zones, thus accu-
mulating TD, low-speed running distance (LSRD),
Note. TD – Total Distance, PL – PlayerLoadTM, MSRD –
moderate-speed running distance (MSRD), HSRD
Moderate-Speed Running Distance, HSRD – High-Speed and sprint distance (SD), which can be achieved
Running Distance, SD – Sprint Distance, MSRE – Moderate- without registering any ACC, DEC or COD. Taken
Speed Running Efforts, HSRE – High-Speed Running Efforts, together, the reported high intercorrelation suggests
SE – Sprint Efforts, ACC – Accelerations, DEC – Decelerations,
COD – Changes of Direction.
that all these measures may lead to data redundancy
and unnecessary data overload for coaches.
Figure 1. PCA loading plot for the two extracted principal Performing PCA on the whole dataset uncov-
components. ered two PCs identified as having eigenvalues >1
indicating these new composite variables account
moderate relationship. Furthermore, Scantlebury et for more variance in training load data than a
al. (2020) reported only a trivial correlation between single original variable (Kaiser, 1960). The first
PL and HSRD compared with a very large relation- PC accounted for 65% of the total variance in the

208
Nosek, P. et al.: THE USE OF PRINCIPAL COMPONENT ANALYSIS FOR... Kinesiology 55(2023)2:202-212

dataset with running- and volume-based variables in that it will reduce 12 training load variables down
showing meaningful relationships with the compo- to just two yet still retaining 81% of the variance,
nent (Table 3). The second PC accounted for 16% the new variables, reported in arbitrary units, may
of the total variance where COD was the only vari- increase the complexity of the report and misun-
able with a meaningful relationship with the compo- derstanding of coaches. For example, understanding
nent, although the other IMA variables loaded just what an increase of 1AU for PC1 and making infer-
short of the threshold (Table 3) yet within other ences around this is very difficult. Applying this
reported acceptable thresholds (Rojas-Valverde, et method to individually analysed datasets would
al., 2020). This demonstrates the independent infor- further reduce interpretability if different vari-
mation that running-based metrics and explosive ables for different players were used to create the PC
actions provide which is unsurprising given their scores which would make between-player compari-
differing physical requirements. These results are sons impossible. Furthermore, the repeated-meas-
dissimilar to Maughan et al. (2021) who reported ures nature of training load monitoring would
that accelerations and decelerations measured by mean the standardized data is updated with each
GPS were loaded on the first component with other training session, thus adjusting the PCA model and
running-based metrics such as TD, PL, HSRD and rendering comparisons between sessions using PC
subjective measures of session rating of perceived scores as unworthwhile.
exertion (sRPE). The differences here may highlight An alternative approach to using the PCA infor-
the benefit of using IMA-derived accelerations and mation is to simply select variables from each PC
decelerations over those produced by GPS, as these that are highly loaded. In its simplest form, this may
provide additional information to running-based be selecting a single variable from each PC, such as
GPS variables. The results here show some simi- those that have the highest validity and reliability
larities to those by Scantlebury et al. (2020) who or practicality (Ryan, et al., 2021). Although many
also showed that TD and PL were heavily weighted variables are available to choose from PC1, TD has
on PC1; however, in contrast to this study, they been shown to be a valid and reliable measure that is
reported high loadings for HSRD on PC2. This is easily understood by coaches (Johnston, Watsford,
likely due to differences in variables collected as no Kelly, Pine, & Spurrs, 2014). Yet this measure does
IMA data were reported (Scantlebury, et al., 2020). not account for any distances in higher speed zones
This shows how the results of PCA are dependent which are an important aspect of soccer match play
on the variables used, meaning practitioners should (Barnes, Archer, Hogg, Bush, & Bradley, 2014).
perform PCA on their own dataset and not use Selecting a single variable from PC2 may be
results from research studies. more difficult as the IMA variables show similar
When PCA was performed on individual levels of validity and reliability as well as inter-
players, similar results were found; however, some pretability (Luteberget, et al., 2018). In this regard,
individual differences were observed. For example, practitioners may aggregate variables that are both
COD was highly associated with PC1 for three loaded above the threshold on the PC and share
players (Table 4) and on PC2 there were various the same measurement units. For example, HSRD
IMA variables (Table 5) that were meaningfully and SD could be grouped from the first PC, whilst
loaded for each player. These results contrast the ACC, DEC and COD could be grouped from the
main findings of Weaving et al. (2018) who showed second PC. Using these two new aggregate vari-
that all rugby union players had sRPE, TD and PL ables would give a coach an idea of the volume
meaningfully loaded on PC1 and HSRD loaded on of high-speed running and the number of intense
PC2. These differences could be attributed to the actions in the session, with both giving unique
type of dataset analysed, with only skills training information and expressed in interpretable units
used in Weaving et al. (2018), which will likely yield as opposed to the arbitrary units of PC scores.
different results due to the differences in relation- It could be suggested that sport scientists utilise
ships between training load variables in different PCA results, their domain-specific expertise, and
modes of training (Lovell, Sirotic, Impellizzeri, input from other key stakeholders such as coaches
& Coutts, 2013; Weaving et al., 2014). Whilst the to co-create an impactful feedback tool (Richter,
individual variation highlights the unique training O’Reilly, & Delahunt, 2021).
outputs between players, selecting different vari- Despite the potential application of PCA seen
ables for different players would seem nonsensical here, several limitations exist. No internal training
and would make the evaluation of training sessions load variables such as HR-based measures and
difficult for coaches. sRPE were used in this study, which may provide
To make use of PCA results, practitioners have additional information due to the individual internal
several options. By multiplying the standardized response to external training load (Impellizzeri,
training data by the loadings on each PC, PC scores et al., 2019) and the differences in internal load
can be produced giving a single score for each PC between different training formats (Owen, Wong,
(Weaving, et al., 2019). Whilst this may seem ideal Mckenna, & Dellal, 2011). Furthermore, only one

209
Nosek, P. et al.: THE USE OF PRINCIPAL COMPONENT ANALYSIS FOR... Kinesiology 55(2023)2:202-212

soccer team was studied, meaning the results likely information is fed back to them regarding GPS-
depend on the periodisation model and training derived training load data (Nosek, et al., 2021). Data
strategy of the team, and thus may not be gener- reduction was undertaken using PCA which identi-
alisable to other soccer teams or team sports. The fied two PC’s, suggesting a multivariate approach
team studied showed a distinct variation in load- is needed when utilising training data. Results
ings throughout the training week whereby different from both the whole dataset and individual anal-
physical qualities were targeted on each day, which ysis demonstrated how PCA can be used to uncover
may in part explain the correlations between vari- multivariate relationships between twelve training
ables and PCA loadings. It is recommended that load variables, with variables relating to volume
sport scientists perform PCA on their own data set and running distances in speed zones associated
to provide insights relating to their team’s specific with the first PC and IMA-derived intensive effort
periodisation and loading strategy. Finally, speed variables mostly associated with PC2. Practitioners
zones analysed were arbitrary, which fails to account can therefore be confident that by reporting vari-
for differences in fitness and athleticism across the ables from each PC they capture unique informa-
squad (Hunter, et al., 2014). Future research may tion compared to using multiple variables from a
look to perform PCA using individualised speed single PC. The impact of these results, however,
and acceleration zones as well as comparing vari- relies on the collaboration between sport science
able relationships between positions and competi- practitioners and coaches to select variables that
tive level. help answer coach questions, such as those perti-
In sum, the present study aimed to address nent to planning and evaluating training.
concerns from soccer coaching staff that too much

References
Akenhead, R., & Nassis, G.P. (2015). Training load and player monitoring in high-level football: Current practice and
perceptions. International Journal of Sports Physiology and Performance, 11(5), 587-593. https://doi.org/10.1123/
ijspp.2015-0331
Bakdash, J.Z., & Marusich, L.R. (2017). Repeated measures correlation. Frontiers in Psychology, 8, 1-13. https://doi.
org/10.3389/fpsyg.2017.00456
Barnes, C., Archer, D., Hogg, B., Bush, M., & Bradley, P. (2014). The evolution of physical and technical performance
parameters in the English Premier League. International Journal of Sports Medicine, 35(13), 1095-1100. https://
doi.org/10.1055/s-0034-1375695
Barrett, S., Midgley, A., & Lovell, R. (2014). PlayerLoadTM: Reliability, convergent validity, and influence of unit
position during treadmill running. International Journal of Sports Physiology and Performance, 9(6), 945-952.
https://doi.org/10.1123/IJSPP.2013-0418
Bradley, P.S., Sheldon, W., Wooster, B., Olsen, P., Boanas, P., & Krustrup, P. (2009). High-intensity running in
English FA Premier League soccer matches. Journal of Sports Sciences, 27(2), 159-168. https://doi.
org/10.1080/02640410802512775
Buchheit, M. (2017). Want to see my report, coach. Aspetar Sports Medicine Journal, 6, 36-43.
Buchheit, M., Lacome, M., Cholley, Y., & Simpson, B.M. (2018). Neuromuscular responses to conditioned soccer
sessions assessed via GPS-Embedded accelerometers: Insights into tactical periodization. International Journal
of Sports Physiology and Performance, 13(5), 577-583. https://doi.org/10.1123/ijspp.2017-0045
Casamichana, D., Castellano, J., Calleja-Gonzalez, J., San Román, J., & Castagna, C. (2013). Relationship between
indicators of training load in soccer players. Journal of Strength and Conditioning Research, 27(2), 369-374.
https://doi.org/10.1519/JSC.0b013e3182548af1
Eisenmann, J. (2017). Translational gap between laboratory and playing field: New era to solve old problems in sports
science. Translational Journal of the American College of Sports Medicine, 2(8), 37-43.
Federolf, P., Reid, R., Gilgien, M., Haugen, P., & Smith, G. (2014). The application of principal component analysis to
quantify technique in sports. Scandinavian Journal of Medicine and Science in Sports, 24(3), 491-499. https://
doi.org/10.1111/j.1600-0838.2012.01455.x
Franks, I.M., & Goodman, D. (2008). A systematic approach to analysing sports performance. Journal of Sports
Sciences, 4(1), 49-59. https://doi.org/10.1080/02640418608732098
Gløersen, Ø., Myklebust, H., Hallén, J., & Federolf, P. (2018). Technique analysis in elite athletes using principal
component analysis. Journal of Sports Sciences, 36(2), 229-237. https://doi.org/10.1080/02640414.2017.1298826

210
Nosek, P. et al.: THE USE OF PRINCIPAL COMPONENT ANALYSIS FOR... Kinesiology 55(2023)2:202-212

Hopkins, W.G. (2010). Linear models and effect magnitudes for research, clinical and practical applications. Sportscience,
14, 49-56. https://doi.org/sportsci.org/2010/wghlinmod.htm
Hunter, F., Bray, J., Towlson, C., Smith, M., Barrett, S., Madden, J., Abt, G., & Lovell, R. (2014). Individualisation of
time-motion analysis: A method comparison and case report series. International Journal of Sports Medicine,
36(01), 41-48. https://doi.org/10.1055/s-0034-1384547
Impellizzeri, F.M., Marcora, S.M., & Coutts, A.J. (2019). Internal and external training load: 15 years on. International
Journal of Sports Physiology and Performance, 14(2), 270-273. https://doi.org/10.1123/ijspp.2018-0935
Jaspers, A., Kuyvenhoven, J.P., Staes, F., Frencken, W.G.P., Helsen, W.F., & Brink, M.S. (2018). Examination of the
external and internal load indicators’ association with overuse injuries in professional soccer players. Journal
of Science and Medicine in Sport, 21(6), 579-585. https://doi.org/10.1016/J.JSAMS.2017.10.005
Johnston, R.J., Watsford, M.L., Kelly, S.J., Pine, M.J., & Spurrs, R.W. (2014). Validity and interunit reliability of 10 Hz
and 15 Hz GPS units for assessing athlete movement demands. Journal of Strength and Conditioning Research,
28(6), 1649-1655. https://doi.org/10.1519/JSC.0000000000000323
Jolliffe, I.T. (1986). Principal components in regression analysis. In Principal Component Analysis (pp. 129-155). https://
doi.org/10.1007/978-1-4757-1904-8_8
Jollife, I.T., & Cadima, J. (2016). Principal component analysis: A review and recent developments. In Philosophical
Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences (Vol. 374, Issue 2065).
Royal Society of London. https://doi.org/10.1098/rsta.2015.0202
Kaiser, H.F. (1960). The application of electronic computers to factor analysis. Educational and Psychological
Measurement, 20(1), 141-151. https://doi.org/10.1177/001316446002000116
Lê, S., Josse, J., Rennes, A., & Husson, F. (2008). FactoMineR: An R package for multivariate analysis. Journal of
Statistical Software, 25(1), 1-18. doi: 10.18637/jss.v025.i01
Lovell, T.W.J., Sirotic, A.C., Impellizzeri, F.M., & Coutts, A.J. (2013). Factors affecting perception of effort (session
rating of perceived exertion) during rugby league training. International Journal of Sports Physiology and
Performance, 8(1), 62-69. https://doi.org/10.1123/IJSPP.8.1.62
Luteberget, L.S., Holme, B.R., & Spencer, M. (2018). Reliability of wearable inertial measurement units to measure
physical activity in team handball. International Journal of Sports Physiology and Performance, 13(4), 467-473.
https://doi.org/10.1123/ijspp.2017-0036
Maughan, P.C., MacFarlane, N.G., & Swinton, P.A. (2021). Relationship between subjective and external training load
variables in youth soccer players. International Journal of Sports Physiology and Performance, 16(8), 1127-
1133. https://doi.org/10.1123/IJSPP.2019-0956
Maughan, P.C., MacFarlane, N.G., & Swinton, P.A. (2022). The influence of season phase on multivariate load
relationships in professional youth soccer. Journal of Sports Sciences, 40(3), 345-350. https://doi.org/10.1080/
02640414.2021.1993642
McLaren, S.J., Macpherson, T.W., Coutts, A.J., Hurst, C., Spears, I.R., & Weston, M. (2018). The relationships between
internal and external measures of training load and intensity in team sports: A meta-analysis. Sports Medicine,
48(3), 641-658. https://doi.org/10.1007/s40279-017-0830-z
Moreno-Pérez, V., Malone, S., Sala-Pérez, L., Lapuente-Sagarra, M., Campos-Vazquez, M. A., & Del Coso, J. (2020).
Activity monitoring in professional soccer goalkeepers during training and match play. International Journal
of Performance Analysis in Sport, 20(1), 19-30. https://doi.org/10.1080/24748668.2019.1699386
Morgans, R., Orme, P., Anderson, L., & Drust, B. (2014). Principles and practices of training for soccer. Journal of
Sport and Health Science, 3(4), 251-257. https://doi.org/10.1016/j.jshs.2014.07.002
Nguyen, L.H., & Holmes, S. (2019). Ten quick tips for effective dimensionality reduction. PLOS Computational Biology,
15(6), e1006907. https://doi.org/10.1371/journal.pcbi.1006907
Nosek, P., Brownlee, T.E., Drust, B., & Andrew, M. (2021). Feedback of GPS training data within professional English
soccer: A comparison of decision making and perceptions between coaches, players and performance staff.
Science and Medicine in Football, 5(1), 35-47. https://doi.org/10.1080/24733938.2020.1770320
Owen, A.L., Wong, D.P., Mckenna, M., & Dellal, A. (2011). Heart rate responses and technical comparison between
small- vs. large-sided games in elite professional soccer. Journal of Strength and Conditioning Research, 25(8),
2104-2110. https://doi.org/10.1519/JSC.0b013e3181f0a8a3
Parmar, N., James, N., Hearne, G., & Jones, B. (2018). Using principal component analysis to develop performance
indicators in professional rugby league. International Journal of Performance Analysis in Sport, 18(6), 938-949.
https://doi.org/10.1080/24748668.2018.1528525
Richter, C., O’Reilly, M., & Delahunt, E. (2021). Machine learning in sports science: Challenges and opportunities.
[Editorial] Sports Biomechanics, published ahead of print. https://doi.org/10.1080/14763141.2021.1910334
Rojas-Valverde, D., Pino-Ortega, J., Gómez-Carmona, C.D., & Rico-González, M. (2020). A systematic review of
methods and criteria standard proposal for the use of principal component analysis in team’s sports science.
International Journal of Environmental Research and Public Health, 17(23), 8712. https://doi.org/10.3390/
IJERPH17238712
Ryan, S., Kempton, T., & Coutts, A.J. (2021). Data reduction approaches to athlete monitoring in professional Australian
football. International Journal of Sports Physiology and Performance, 16(1), 59-65. https://doi.org/10.1123/
IJSPP.2020-0083

211
Nosek, P. et al.: THE USE OF PRINCIPAL COMPONENT ANALYSIS FOR... Kinesiology 55(2023)2:202-212

Scantlebury, S., Till, K., Beggs, C., Dalton-Barron, N., Weaving, D., Sawczuk, T., & Jones, B. (2020). Achieving a desired
training intensity through the prescription of external training load variables in youth sport: More pieces to the
puzzle required. Journal of Sports Sciences, 38(10), 1124-1131. https://doi.org/10.1080/02640414.2020.1743047
Weaving, D., Beggs, C., Dalton-Barron, N., Jones, B., & Abt, G. (2019). Visualizing the complexity of the athlete-
monitoring cycle through principal-component analysis. International Journal of Sports Physiology and
Performance, 14(9), 1304-1310. https://doi.org/10.1123/ijspp.2019-0045
Weaving, D., Dalton, N.E., Black, C., Darrall-Jones, J., Phibbs, P.J., Gray, M., Jones, B., & Roe, G.A.B. (2018). The
same story or a unique novel? Within-participant principal-component analysis of measures of training load in
professional rugby union skills training. International Journal of Sports Physiology and Performance, 13(9),
1175-1181. https://doi.org/10.1123/ijspp.2017-0565
Weaving, D., Jones, B., Till, K., Abt, G., & Beggs, C. (2017). The case for adopting a multivariate approach to optimize
training load quantification in team sports. Frontiers in Physiology, 8. https://doi.org/10.3389/fphys.2017.01024
Weaving, D., Marshall, P., Earle, K., Nevill, A., & Abt, G. (2014). Combining internal- and external-training-load
measures in professional rugby league. International Journal of Sports Physiology and Performance, 9(6),
905-912. https://doi.org/10.1123/ijspp.2013-0444
Weston, M. (2018). Training load monitoring in elite English soccer: A comparison of practices and perceptions
between coaches and practitioners. Science and Medicine in Football, 2(3), 216-224. https://doi.org/10.1080/2
4733938.2018.1427883
Williams, B., Onsman, A., Brown, T., Andrys Onsman, P., & Ted Brown, P. (2010). Exploratory factor analysis: A
five-step guide for novices. Journal of Emergency Primary Health Care (JEPHC), 8(3), article 990399. https://
doi.org/10.33151/ajp.8.3.93
Williams, S., Trewartha, G., Cross, M.J., Kemp, S.P.T., & Stokes, K.A. (2017). Monitoring what matters: A systematic
process for selecting training-load measures. International Journal of Sports Physiology and Performance,
12(Suppl 2), S2-101-S2-106. https://doi.org/10.1123/ijspp.2016-0337

Submitted: January 31, 2022

Accepted: May 29, 2023
Published Online First: November 8, 2023

Correspondence to:
Tom Brownlee, Ph.D.
School of Sport, Exercise and Rehabilitation Sciences
University of Birmingham
Birmingham, B15 2TT UK
Phone: +44 7775 333932
E-mail: [email protected]

212