You are on page 1of 21

Sports Biomechanics

ISSN: (Print) (Online) Journal homepage: https://www.tandfonline.com/loi/rspb20

The linear regression model provides the force-


velocity relationship parameters with the highest
reliability

Alejandro Pérez-Castilla, Pierre Samozino, Ivan Jukic, Eliseo Iglesias-Soler &


Amador García-Ramos

To cite this article: Alejandro Pérez-Castilla, Pierre Samozino, Ivan Jukic, Eliseo Iglesias-
Soler & Amador García-Ramos (2022): The linear regression model provides the force-
velocity relationship parameters with the highest reliability, Sports Biomechanics, DOI:
10.1080/14763141.2022.2058992

To link to this article: https://doi.org/10.1080/14763141.2022.2058992

Published online: 04 Apr 2022.

Submit your article to this journal

Article views: 213

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at


https://www.tandfonline.com/action/journalInformation?journalCode=rspb20
SPORTS BIOMECHANICS
https://doi.org/10.1080/14763141.2022.2058992

The linear regression model provides the force-velocity


relationship parameters with the highest reliability
Alejandro Pérez-Castilla a, Pierre Samozino b
, Ivan Jukicc, Eliseo Iglesias-Soler d

and Amador García-Ramos a,e


a
Department of Physical Education and Sport, Faculty of Sport Sciences, University of Granada, Granada,
Spain; bLaboratory of Human Movement Biology, University Savoie Mont BlancInter-university, Chambery,
France; cSchool of Engineering, Computer and Mathematical Sciences, Auckland University of Technology,
Auckland, New Zealand; dPerformance and Health Group, Department of Physical Education and Sport,
Faculty of Sports Sciences and Physical Education, University of a Coruna, A Coruña, Spain; eDepartment of
Sports Sciences and Physical Conditioning, Faculty of Education, Universidad Católica de la Santísima
Concepción, Concepción, Chile

ABSTRACT ARTICLE HISTORY


An a-posteriori multicentre reliability study was conducted to com­ Received 11 October 2021
pare the between-session reliability of the force-velocity relation­ Accepted 23 March 2022
ship parameters (force-intercept [F0], velocity-intercept [v0], and KEYWORDS
maximum power [Pmax]) between different regression models dur­ Training monitoring;
ing the bench press (BP) and bench press throw (BPT) exercises. multicentre reliability study;
Data from four and three studies were considered for the BP multiple-point method;
(n = 102) and BPT (n = 81) exercises, respectively. The force- muscle capacity; two-point
velocity relationships were determined using five regression mod­ method
els: linear multiple-point, linear two-point, quadratic polynomial,
hyperbolic, and exponential. All regression models provided F0 and
Pmax with acceptable reliability (cut-off CV ≤ 9.45%; cut-off ICC ≥
0.79) with the exceptions of F0 for the quadratic polynomial and
hyperbolic models (BPT) and Pmax for the exponential model (BP
and BPT). Only the linear multiple- and linear two-point models
provided v0 with acceptable absolute reliability (cut-off CV ≤ 7.72%).
Regardless of the exercise, the reliability of the three parameters
was generally higher for the linear multiple- and two-point models
compared to the other models (CVratio ≥ 1.22), and no significant
differences were observed between multiple- and linear two-point
models (CVratio ≤ 1.11). Linear regression models are recommended
to maximise the reliability of the force-velocity relationship para­
meters during the BP and BPT exercises.

Introduction
The force-velocity (F-v) relationship has been extensively used to assess the muscle
function during single- and multi-joint tasks (García-Ramos et al., 2017; Iglesias-Soler,
Fariñas et al., 2019). The most standard testing procedure consists of recording force and
velocity values under multiple loads (Alcazar et al., 2020; Giroux et al., 2016; Jiménez-
Reyes et al., 2017; Sreckovic et al., 2015). Given that the values of force and velocity
determined under different loads are inversely related (i.e., the increment of the load is

CONTACT Alejandro Pérez-Castilla alexperez@ugr.es


© 2022 International Society of Biomechanics in Sports
2 A. PÉREZ-CASTILLA ET AL.

associated with a higher force and a lower velocity), a mathematical function can be
applied to the force and velocity values determined against different loads allowing to
determine the maximal capacity of active muscles to produce force (F0), velocity (v0) and
power (Pmax; Alcazar et al., 2020; García-Ramos et al., 2017; García-Ramos, Torrejón,
Morales-Artacho et al., 2018; Iglesias-Soler, Fariñas et al., 2019). It has been argued that
the F-v relationship provides a deeper insight into muscle function than the standard
testing procedures conducted under a single mechanical condition (e.g., unloaded ver­
tical jump) because the force and velocity output obtained under a single mechanical
condition are inter-dependent (i.e., a higher relative net impulse that determines velocity
at take-off) and depend on the choice of the loading condition (Jaric, 2016). However, F0
and v0 are independent of each other (i.e., a change in F0 should not necessarily produce
a change in v0; Cuevas-Aburto et al., 2018). Furthermore, the F-v relationship can be used
for different purposes such as training prescription (Jiménez-Reyes et al., 2017), injury
management (Mendiguchia et al., 2016), fatigue monitoring (García-Ramos, Torrejón,
Feriche et al., 2018), or bilateral deficit assessment (Samozino et al., 2014). Therefore, the
assessment of the F-v relationship is not only of special relevance for a better under­
standing of the muscular function, but also for guiding sports training.
The shape of the F-v relationship has important implications in various aspects of
muscle physiology (Iglesias-Soler, Fariñas et al., 2019; Jaric, 2016). However, there is no
scientific consensus about the exact shape of the F-v relationship and the underlying
mechanisms accounting for it (Alcazar et al., 2020). Seminal studies published in the first
half of the 20th century revealed that the F-v relationship follows a non-linear, concave
upward function in isolated animal and individual human muscles and, therefore, it
could be expressed by exponential (‘exponential’) or hyperbolic (‘hyperbolic’) equations
(Fenn & Marsh, 1935; Hill, 1938). Of note is that in these seminal works authors were
discussing force production and fascicle shortening velocity in isolated muscles, and not
movement velocity and the approximation of external force production. However,
during multi-joint tasks the F-v relationship seems to be approximately linear (see,
Jaric (2015) for a review). Therefore, the F-v relationship has been determined by
means of linear regression models after collecting force and velocity data under multiple
loads (‘linear multiple-point’; Alcazar et al., 2020; Giroux et al., 2016; Jiménez-Reyes et al.,
2017; Sreckovic et al., 2015) or only two loads (‘linear two-point’; Garcia-Ramos et al.,
2018; Jaric, 2016). Finally, a quadratic polynomial regression model (‘quadratic poly­
nomial’) has been also used to fit the F-v relationship in several exercises such as the knee
extension, vertical jump, bench press or back-squat (Cuk et al., 2014; Iglesias-Soler,
Fariñas et al., 2019; Iglesias-Soler, Mayo et al., 2019). It is important to note that various
levels of organisation have been used to study the F-v relationship including molecular
and single-cell levels as well as single and multi-joint movements (Cormie et al., 2011).
Therefore, it should not be confused that while the early studies examined the muscle’s
force production and shortening velocity (Fenn & Marsh, 1935; Hill, 1938), recent
research evaluated the external force production and movement velocity (Giroux et al.,
2016; Jiménez-Reyes et al., 2017; Sreckovic et al., 2015).
Previous researchers have compared the goodness of fit between linear and various
non-linear regression models in an attempt to determine the most appropriate model to
fit the F-v relationship and calculate the associated parameters (Alcazar et al., 2020;
Iglesias-Soler, Fariñas et al., 2019; Iglesias-Soler, Mayo et al., 2019). However, it is
SPORTS BIOMECHANICS 3

important to note that the selection of the most suitable model should not only be based
on the goodness of fit of the F-v relationship, but also on the reliability and physiological
meaning of its parameters. To our knowledge, only Iglesias-Soler et al. (2019) have
compared the reliability of the F-v relationship parameters between a linear and a non-
linear model (i.e., quadratic polynomial model) during the bench press and back-squat
exercises, and found a higher reliability for the linear multiple-point model. Therefore, it
is evident that more studies are needed to comprehensively examine the most appro­
priate regression model for determining the maximal mechanical capacities of the
muscles to produce force (F0), velocity (v0) and power (Pmax).
To address this research question, force and velocity outputs were determined against
multiple loads in four studies for the bench press (BP; Cuevas-Aburto et al., 2018; Garcia-
Ramos et al., 2016; García-Ramos, Pestaña-Melet al., (2018); Iglesias-Soler, Mayo et al.,
2019) and three studies for the bench press throw (BPT) exercises (Cuevas-Aburto et al.,
2018; Garcia-Ramos et al., 2016; García-Ramos, Pestaña-Melero, et al., (2018)), and later
combined into an a-posteriori multicentre reliability study (Garcia-Ramos & Janicijevic,
2020). Specifically, the main objective of the present study was to compare the between-
session reliability of the F-v relationship parameters (F0,v0, and Pmax) between five
different regression models (linear multiple-point, linear two-point, quadratic polyno­
mial, hyperbolic, and exponential) during the BP and BPT exercises. We hypothesised
that the multiple- and two-point linear regression models would provide the
F-v relationship parameters with higher reliability than the quadratic polynomial
model (Iglesias-Soler, Mayo et al., 2019). Due to the paucity of similar studies, no specific
hypotheses were formulated regarding the reliability of the hyperbolic and exponential
regression models.

Method
Participants
This a-posteriori multicentre reliability study considered data from four different studies
for the BP exercise (Cuevas-Aburto et al., 2018; Garcia-Ramos et al., 2016; García-Ramos,
Pestaña-Melero, et al., (2018); Iglesias-Soler, Mayo et al., 2019; n = 102) and three studies
for the BPT exercise (Cuevas-Aburto et al., 2018; Garcia-Ramos et al., 2016; García-
Ramos, Pestaña-Melero, et al., (2018).) (n = 81; Table 1). All participants were physically
active and none of them reported any physical limitations, health problems or muscu­
loskeletal injuries that could compromise testing procedures. Participants were informed
of the study procedures and signed a written informed consent form prior to initiating
the study. All study protocols adhered to the tenets of the Declaration of Helsinki and
were approved by the Institutional Review Board.

Study design
An a-posteriori multicentre reliability study was conducted following the guidelines
proposed by Garcia-Ramos and Janicijevic (2020) to compare the reliability of the
F-v relationship parameters (F0, v0, and Pmax) between different regressions models
(linear multiple-point, linear two-point, quadratic polynomial, hyperbolic, and
4

Table 1. Characteristics of the study sample included in the multicentre reliability studies conducted with the bench press and bench press throw exercises.
A. PÉREZ-CASTILLA ET AL.

Exercise Study Sample size (n) Age (years) Body height (m) Body mass (kg) Bench press 1RM (kg)
Bench press Garcia-Ramos et al. (2016) 32 male sport science students 21.6 ± 2.9 1.77 ± 0.05 74.9 ± 7.4 76.6 ± 13.5
García-Ramos, Pestaña-Melero, et al., (2018). 30 male sport science students 21.2 ± 3.8 1.78 ± 0.07 72.3 ± 7.3 76.6 ± 12.8
Cuevas-Aburto et al. (2018) 8 male sport science students 24.0 ± 3.8 1.73 ± 0.07 80.0 ± 12.2 77.6 ± 21.0
6 male rugby players
5 male weightlifting competitors
Iglesias-Soler et al. (2019) 4 female rugby players 23.0 ± 2.5 1.69 ± 0.05 71.6 ± 9.5 74.7 ± 24.0
4 female judokas
5 male rugby players
8 male judokas
Multicentre study 102 22.5 ± 3.3 1.74 ± 0.06 74.7 ± 9.1 76.4 ± 17.8
Bench press throw Garcia-Ramos et al. (2016) 32 male sport science students 21.6 ± 2.9 1.77 ± 0.05 74.9 ± 7.4 76.6 ± 13.5
García-Ramos, Pestaña-Melero, et al., (2018) 30 male sport science students 21.2 ± 3.8 1.78 ± 0.07 72.3 ± 7.3 76.6 ± 12.8
Cuevas-Aburto et al. (2018) 8 male sport science students 24.0 ± 3.8 1.73 ± 0.07 80.0 ± 12.2 77.6 ± 21.0
6 male rugby players
5 male weightlifting competitors
Multicentre study 81 22.3 ± 3.5 1.76 ± 0.06 75.7 ± 9.0 76.9 ± 15.8
Data are means ± standard deviations. 1RM, one-repetition maximum.
SPORTS BIOMECHANICS 5

exponential) during the BP and BPT exercises. All participants were tested in two
sessions separated by 48–96 hours. Each session consisted of the assessment of force
and velocity values against multiple loads (see details below). Data of both sessions were
compared to determine the reliability of the F-v relationship parameters.

Procedures
Bench press testing procedure
The BP 1RM was determined in a preliminary session in two studies (Cuevas-Aburto
et al., 2018; Garcia-Ramos et al., 2016), while two familiarisation sessions were carried
out in the other two studies (García-Ramos, Pestaña-Melero, et al., (2018); Iglesias-Soler,
Mayo et al., 2019). The experimental sessions consisted of seven loads applied in an
incremental order (from 20% to 80% of 1RM in 10% increments; Garcia-Ramos et al.,
2016), an incremental loading test from 17 kg (9.5 ± 2.8 loads) (García-Ramos, Pestaña-
Melero, et al., (2018)) or 21.4 kg (8.0 ± 1.4 loads; Iglesias-Soler, Mayo et al., 2019) to the
1RM load, or six loads applied in a randomised order (load 1 = 1 kg; load 2 = 8 kg; load
3 = 29 kg; load 4 = 39.4 ± 5.5 kg; load 5 = 49.9 ± 10.0 kg; load 6 = 59.8 ± 15.0 kg) followed
by 1RM attempts (3.1 ± 1.0 incremental loads; Cuevas-Aburto et al., 2018). Data of the
80% 1RM load in the study of Garcia-Ramos et al. (2016) and the 1RM attempts in the
study of Cuevas-Aburto et al. (2018) were not considered for computing the
F-v relationship in the original studies. Participants performed 1–3 repetitions with
each load and 1–5 min of rest were implemented between different loads. All studies
evaluated the concentric-only BP exercise which was performed in a Smith machine
(Garcia-Ramos et al., 2016; García-Ramos, Pestaña-Melero, et al., (2018); Iglesias-Soler,
Mayo et al., 2019) or with free-weights (Cuevas-Aburto et al., 2018). It should be noted
that the BP exercise can be executed in a more controlled environment by restricting the
displacement of the barbell to the vertical direction and, consequently, it has been
suggested that machine-based exercises can displayed greater reproductivity than free-
weight exercises (Miller et al., 2020). Participants were instructed to perform the lifting
phase with maximal intent.

Bench press throw testing procedure


The BP 1RM was determined in a preliminary session in two studies (Cuevas-Aburto
et al., 2018; Garcia-Ramos et al., 2016), while two familiarisation sessions were carried
out in another study (García-Ramos, Pestaña-Melero, et al., (2018)). The experimental
sessions consisted of seven loads applied in an incremental order (from 20% to 80% of
1RM in 10% increments; Garcia-Ramos et al., 2016), an incremental loading test from
17 kg to the 1RM load (8.0 ± 1.3 loads) (García-Ramos, Pestaña-Melero, et al., (2018)), or
four loads applied in a randomised order (load 1 = 29 kg; load 2 = 39.4 ± 5.5 kg; load
3 = 49.9 ± 10.0 kg; load 4 = 59.8 ± 15.0 kg). Note that the 80% 1RM load was not
originally included in the study of Garcia-Ramos et al. (2016). Participants performed 1–
3 repetitions with each load and 4–5 min were implemented between different loads. The
three studies evaluated the concentric-only BPT exercise which was performed in a Smith
machine. Participants were instructed to throw the barbell as high as possible, but when
the load was so heavy that participants could not throw the barbell the test was completed
without throwing the barbell but performing the lifting phase with maximal intent.
6 A. PÉREZ-CASTILLA ET AL.

Measurement equipment and data analysis


The values of force and velocity within the propulsive phase (i.e., the time interval from
the initiation of the vertical movement of the barbell until the acceleration of the barbell
drops below −9.81 m·s−2) were computed with a linear position transducer (Real Power
Pro Globus, Codogne, Italy; Cuevas-Aburto et al., 2018; Garcia-Ramos et al., 2016) or
a linear velocity transducer (T-Force System; Ergotech, Murcia, Spain) (García-Ramos,
Pestaña-Melero, et al., (2018); Iglesias-Soler, Mayo et al., 2019). Displacement-time or
velocity-time data were always recorded at a frequency of 1,000 Hz. The mass of the arms
(10% of body mass) and the friction force of the Smith machine were only considered for
force computations as in the study of Cuevas-Aburto et al. (2018). The F-v relationship
was modelled using five different regression models (see, Figure 1 for illustration):
(I and II) a least-square linear regression model (V[F] = v0 + aF) applied to all loading
conditions (‘linear multiple-point’) or only to the two most distant loads (‘linear two-
point’). v0 represents the velocity-intercept and a is the slope of the F-v relationship and
they enabled the calculation of F0 (force-intercept; F0 = v0/a) and Pmax (Pmax = v0·F0/4).
(III) a quadratic polynomial regression model (V[F] = aF2 + bF + c) applied to all
loading conditions (‘quadratic polynomial’). The intercept c is equivalent to v0 in the
linear regressions models and a indicates the concavity of the curve (the higher a, the
higher the concavity). F0 was obtained by calculating the root of the polynomial at which
the first derivative is negative. Pmax was estimated by finding zero of the derivative of the
expression of power output (i.e., velocity times force; Iglesias-Soler, Fariñas et al., 2019;
Iglesias-Soler, Mayo et al., 2019).
(IV) a hyperbolic regression model ([F + a][V + b] = [F0]b]) applied to all loading
conditions (‘hyperbolic’). F0is the force-intercept estimated by Hill’s equation, and a and
b are constants. v0 was calculated as the velocity-intercept (v0 = F0b/a) and Pmax was
identified as the apex of the power-velocity relationship.
(V) an exponential regression model (V = Ae−F/k) applied to all loading conditions
(‘exponential’). A represents the v0 and K is the exponential force constant (i.e., the force
produced at around 37% of v0). Pmax was estimated by finding zero of the derivative of the
expression of power (i.e., velocity times force; Iglesias-Soler, Fariñas et al., 2019; Iglesias-
Soler, Mayo et al., 2019).
All individual regressions were obtained considering velocity as a dependent variable
and force as the predictor variable. The order of the variables was only exchanged to
obtain F0 in the exponential model. In several cases, the quadratic polynomial and
hyperbolic models failed to cut the velocity-intercept making impossible to determine
v0. Participants were excluded from the reliability analyses for the models in which v0
could not be obtained. Each individual regression model was obtained by using
a customised 2019 Microsoft Excel® spreadsheet (version 16.32, Microsoft
Corporations, Redmond, Washington, USA). The distance from the experimental points
to the interception axes is presented in Table 2.

Statistical analyses
The goodness of fit of the F-v relationships was assessed by the r2 coefficient, which is
presented through the median values and range. Reliability was assessed through the
standard error of the measurement expressed in absolute (SEM) and relative values
SPORTS BIOMECHANICS 7

Figure 1. Force-velocity relationships modelled by the linear multiple-point, linear two-point, quad­
ratic polynomial, hyperbolic, and exponential models of a representative participant during the bench
press (upper panel) and bench press throw (lower panel) exercises. Note: All loads were considered to
fit the different regression models, with the exception of the linear two-point model that only
considered the two most distant loads. Regressions equations and coefficients of determination (r2)
are depicted.

(CV = SEM/participants’ mean score × 100) and the intraclass correlation coefficient
(ICC; model 3,1) with their corresponding 95% confidence intervals (CI). The reliability
was reported for each independent study as well as for the multicentre reliability studies
conducted with the BP and BPT exercises (García-Ramos & Jaric, 2019). Acceptability
cut offs (i.e., upper [CV] and lower [ICC] bounds of 95% CI) for the CV and ICC were set
at ≤ 10% and ≥ 0.75, respectively (James et al., 2017; Koo & Li, 2016). The ratio between
two CVs was used to compare the reliability between the different regression models. The
8 A. PÉREZ-CASTILLA ET AL.

Table 2. The distance from experimental points to the intercept axes.


Exercise Study F-v model F0 (%) v0 (%)
Bench press Garcia-Ramos et al. (2016) Linear multiple-point 16 ± 4 41 ± 5
Linear two-point 18 ± 3 41 ± 4
Quadratic polynomial 20 ± 6 37 ± 23
Hyperbolic 24 ± 11 71 ± 7
Exponential 24 ± 6 58 ± 19
García-Ramos, Pestaña-Melero et al. (2018) Linear multiple-point 6±2 37 ± 4
Linear two-point 8±2 38 ± 4
Quadratic polynomial 10 ± 3 44 ± 7
Hyperbolic 11 ± 6 80 ± 5
Exponential 13 ± 4 50 ± 9
Cuevas-Aburto et al. (2018) Linear multiple-point 6±5 3±5
Linear two-point 8±4 5±1
Quadratic polynomial 10 ± 6 6±1
Hyperbolic 9±4 35 ± 8
Exponential 35 ± 8 7±2
Iglesias-Soler et al. (2019) Linear multiple-point 8±5 39 ± 11
Linear two-point 9±4 41 ± 9
Quadratic polynomial 10 ± 5 59 ± 187
Hyperbolic 11 ± 4 49 ± 14
Exponential 13 ± 6 80 ± 9
Bench press throw Garcia-Ramos et al. (2016) Linear multiple-point 16 ± 5 41 ± 3
Linear two-point 18 ± 4 42 ± 3
Quadratic polynomial 22 ± 6 43 ± 17
Hyperbolic 25 ± 8 61 ± 16
Exponential 24 ± 6 72 ± 5
García-Ramos, Pestaña-Melero, et al. (2018) Linear multiple-point 5±4 30 ± 6
Linear two-point 7±5 35 ± 5
Quadratic polynomial 10 ± 9 42 ± 8
Hyperbolic 14 ± 5 60 ± 14
Exponential 11 ± 4 74 ± 4
Cuevas-Aburto et al. (2018) Linear multiple-point 24 ± 6 51 ± 9
Linear two-point 24 ± 5 52 ± 9
Quadratic polynomial 22 ± 9 109 ± 254
Hyperbolic 26 ± 6 52 ± 9
Exponential 31 ± 7 77 ± 9
Data are means ± standard deviations. F0, maximum force; v0, maximum velocity

smallest important ratio between two CVs was considered to be higher than 1.15 (Fulton
et al., 2009; Garcia-Ramos & Janicijevic, 2020). Reliability analyses were performed by
means of a custom Excel spreadsheet (Hopkins, 2000). Alpha was set at 0.05.

Results
Bench press
The F-v relationships were strong for the linear multiple-point (r2 = 0.976 [0.637, 1.000]),
quadratic polynomial (r2 = 0.990 [0.852, 1.000]), hyperbolic (r2 = 0.984 [0.846, 1.000])
and exponential (r2 = 0.937 [0.702, 0.996]) models. However, v0 could not be obtained in
61 out of 204 relationships using the quadratic polynomial model and in 11 out of 204
relationships using the hyperbolic model.
The reliability of the F-v relationship parameters obtained using the different regres­
sion models are presented in Table 3. All regression models revealed an acceptable
reliability for F0 and Pmax (cut-off CV values ≤ 9.45%; cut-off ICC values ≥ 0.79) in the
multicentre study with the only exception of Pmax for the exponential model (cut-off CV
Table 3. Reliability of the force-velocity (F-v) relationship parameters obtained for the independent studies as well as for the multicentre study using the linear
multiple-point, linear two-point, quadratic polynomial, hyperbolic, and exponential models during the bench press exercise.
Study F-v model Sample size (n) Parameter Session 1 (mean ± SD) Session 2 (mean ± SD) SEM CV (%) (95% CI) ICC (95% CI)
Garcia-Ramos et al. (2016) Linear multiple-point 32 F0 (N) 701 ± 123 700 ± 111 26 3.67 (2.94, 4.87) 0.96 (0.91, 0.98)
v0 (m·s−1) 2.32 ± 0.31 2.30 ± 0.38 0.17 7.36 (5.90, 9.79) 0.77 (0.58, 0.88)
Pmax (W) 404 ± 77 399 ± 77 32 8.02 (6.43, 10.7) 0.83 (0.69, 0.92)
Linear two-point 32 F0 (N) 723 ± 135 716 ± 124 22 3.11 (2.49, 4.14) 0.97 (0.94, 0.99)
v0 (m·s−1) 2.32 ± 0.32 2.41 ± 0.39 0.19 8.01 (6.43, 10.7) 0.73 (0.51, 0.86)
Pmax (W) 415 ± 83 428 ± 89 23 5.52 (4.42, 7.33) 0.93 (0.86, 0.97)
Quadratic 8 F0 (N) 728 ± 157 659 ± 85 62 8.99 (5.94, 18.3) 0.82 (0.34, 0.96)
polynomial v0 (m·s−1) 2.34 ± 0.75 1.71 ± 0.70 0.61 30.2 (20.0, 61.5) 0.35 (−0.42, 0.82)
Pmax (W) 415 ± 66 456 ± 80 18 4.06 (2.69, 8.27) 0.96 (0.83, 0.99)
Hyperbolic 24 F0 (N) 790 ± 301 776 ± 206 104 13.3 (10.3, 18.6) 0.85 (0.68, 0.93)
−1
v0 (m·s ) 6.65 ± 9.67 4.03 ± 2.51 7.27 136 (105, 191) −0.06 (−0.45, 0.34)
Pmax (W) 410 ± 83 410 ± 91 41 10.1 (7.84, 14.2) 0.79 (0.57, 0.90)
Exponential 32 F0 (N) 776 ± 146 753 ± 113 66 8.57 (6.87, 11.4) 0.76 (0.56, 0.88)
v0 (m·s−1) 4.97 ± 1.35 4.89 ± 1.53 0.87 17.6 (14.2, 23.5) 0.65 (0.39, 0.81)
Pmax (W) 436 ± 95 429 ± 102 47 10.9 (8.72, 14.5) 0.78 (0.60, 0.89)
García-Ramos, Pestaña-Melero, et al. (2018) Linear multiple-point 30 F0 (N) 798 ± 126 797 ± 123 19 2.43 (1.93, 3.26) 0.98 (0.95, 0.99)
v0 (m·s−1) 2.14 ± 0.20 2.13 ± 0.20 0.12 5.83 (4.64, 7.83) 0.62 (0.33, 0.80)
Pmax (W) 427 ± 79 424 ± 74 22 5.06 (4.03, 6.80) 0.93 (0.85, 0.96)
Linear two-point 30 F0 (N) 812 ± 136 814 ± 134 16 1.93 (1.54, 2.60) 0.99 (0.97, 0.99)
v0 (m·s−1) 2.19 ± 0.22 2.22 ± 0.20 0.12 5.60 (4.46, 7.52) 0.67 (0.42, 0.83)
Pmax (W) 444 ± 87 451 ± 82 25 5.63 (4.48, 7.57) 0.92 (0.83, 0.96)
Quadratic 18 F0 (N) 807 ± 124 833 ± 145 37 4.50 (3.38, 6.75) 0.93 (0.83, 0.97)
−1
polynomial v0 (m·s ) 2.40 ± 0.49 2.58 ± 0.37 0.39 15.8 (11.8, 23.6) 0.20 (−0.28, 0.60)
Pmax (W) 426 ± 76 432 ± 73 24 5.55 (4.16, 8.32) 0.91 (0.78, 0.97)
Hyperbolic 29 F0 (N) 852 ± 129 868 ± 142 56 6.52 (5.17, 8.82) 0.84 (0.69, 0.92)
v0 (m·s−1) 2.84 ± 0.72 12.01 ± 46.53 34 444 (353, 601) 0.00 (−0.37, 0.36)
Pmax (W) 438 ± 71 439 ± 68 30 6.79 (5.39, 9.18) 0.83 (0.66, 0.91)
Exponential 30 F0 (N) 860 ± 143 855 ± 139 27 3.17 (2.53, 4.27) 0.97 (0.93, 0.98)
v0 (m·s−1) 7.45 ± 3.03 6.92 ± 2.07 1.51 21.0 (16.8, 28.3) 0.67 (0.42, 0.83)
Pmax (W) 562 ± 132 538 ± 95 66 12.0 (9.52, 16.1) 0.69 (0.44, 0.84)
SPORTS BIOMECHANICS

(Continued)
9
10

Table 3. (Continued).
Study F-v model Sample size (n) Parameter Session 1 (mean ± SD) Session 2 (mean ± SD) SEM CV (%) (95% CI) ICC (95% CI)
Cuevas-Aburto et al. (2018) Linear multiple-point 19 F0 (N) 798 ± 193 801 ± 206 25 3.12 (2.36, 4.61) 0.99 (0.96, 0.99)
v0 (m·s−1) 2.33 ± 0.20 2.34 ± 0.15 0.12 4.94 (3.73, 7.30) 0.60 (0.22, 0.83)
Pmax (W) 570 ± 134 470 ± 130 26 5.55 (4.20, 8.21) 0.97 (0.91, 0.99)
Linear two-point 19 F0 (N) 820 ± 212 820 ± 221 32 3.92 (2.97, 5.80) 0.98 (0.95, 0.99)
v0 (m·s−1) 2.38 ± 0.26 2.37 ± 0.16 0.12 5.16 (3.90, 7.63) 0.70 (0.37, 0.87)
Pmax (W) 495 ± 159 490 ± 148 23 4.71 (3.56, 6.96) 0.98 (0.95, 0.99)
A. PÉREZ-CASTILLA ET AL.

Quadratic 18 F0 (N) 840 ± 244 831 ± 228 34 4.04 (3.03, 6.05) 0.98 (0.95, 0.99)
polynomial v0 (m·s−1) 2.41 ± 0.28 2.41 ± 0.17 0.13 5.44 (4.08, 8.15) 0.70 (0.36, 0.88)
Pmax (W) 452 ± 122 448 ± 123 29 6.47 (4.86, 9.70) 0.95 (0.87, 0.98)
Hyperbolic 19 F0 (N) 830 ± 212 828 ± 217 17 2.09 (1.58, 3.10) 0.99 (0.99, 1.00)
v0 (m·s−1) 2.44 ± 0.27 2.43 ± 0.16 0.13 5.31 (4.02, 7.86) 0.67 (0.33, 0.86)
Pmax (W) 450 ± 122 452 ± 124 28 6.30 (4.76, 9.32) 0.95 (0.88, 0.98)
Exponential 19 F0 (N) 1172 ± 279 1159 ± 307 104 8.95 (6.76, 13.2) 0.89 (0.73, 0.95)
v0 (m·s−1) 3.56 ± 0.67 3.67 ± 0.52 0.48 13.3 (10.1, 19.7) 0.38 (−0.08, 0.70)
Pmax (W) 381 ± 107 382 ± 102 24 6.41 (4.84, 9.47) 0.95 (0.88, 0.98)
Iglesias-Soler et al. (2019) Linear multiple-point 21 F0 (N) 786 ± 223 796 ± 238 23 2.94 (2.25, 4.25) 0.99 (0.98, 1.00)
v0 (m·s−1) 2.00 ± 0.29 1.98 ± 0.29 0.09 4.65 (3.55, 6.71) 0.91 (0.79, 0.96)
Pmax (W) 394 ± 129 394 ± 131 17 4.28 (3.28, 6.19) 0.99 (0.96, 0.99)
Linear two-point 21 F0 (N) 795 ± 238 806 ± 252 22 2.69 (2.06, 3.89) 0.99 (0.98, 1.00)
v0 (m·s−1) 2.05 ± 0.26 2.01 ± 0.29 0.13 6.21 (4.75, 8.97) 0.81 (0.58, 0.92)
Pmax (W) 414 ± 150 408 ± 147 21 5.14 (3.93, 7.42) 0.98 (0.96, 0.99)
Quadratic 14 F0 (N) 728 ± 253 722 ± 228 40 5.52 (4.00, 8.89) 0.98 (0.93, 0.99)
polynomial v0 (m·s−1) 1.57 ± 0.86 1.74 ± 0.39 0.54 32.7 (23.7, 52.7) 0.38 (−0.17, 0.75)
Pmax (W) 370 ± 129 364 ± 129 13 3.50 (2.54, 5.65) 0.99 (0.97, 1.00)
Hyperbolic 19 F0 (N) 790 ± 247 781 ± 249 49 6.20 (4.69, 9.18) 0.97 (0.91, 0.99)
v0 (m·s−1) 2.65 ± 1.52 2.08 ± 0.46 1.08 45.5 (34.4, 67.3) 0.08 (−0.38, 0.51)
Pmax (W) 389 ± 135 383 ± 133 19 4.90 (3.70, 7.24) 0.98 (0.96, 0.99)
Exponential 21 F0 (N) 837 ± 246 846 ± 257 37 4.41 (3.37, 6.37) 0.98 (0.95, 0.99)
v0 (m·s−1) 7.17 ± 2.87 8.07 ± 4.43 1.88 24.7 (18.9, 35.7) 0.76 (0.50, 0.90)
Pmax (W) 535 ± 200 563 ± 214 74 13.5 (10.3, 19.5) 0.88 (0.73, 0.95)
(Continued)
Table 3. (Continued).
Study F-v model Sample size (n) Parameter Session 1 (mean ± SD) Session 2 (mean ± SD) SEM CV (%) (95% CI) ICC (95% CI)
Multicentre study Linear multiple-point 102 F0 (N) 765 ± 166 767 ± 170 23 3.03 (2.66, 3.51) 0.98 (0.97, 0.99)
v0 (m·s−1) 2.20 ± 0.29 2.19 ± 0.31 0.13 6.01 (5.28, 6.97) 0.80 (0.72, 0.86)
Pmax (W) 421 ± 104 419 ± 103 25 5.98 (5.26, 6.94) 0.94 (0.92, 0.96)
Linear two-point 102 F0 (N) 782 ± 178 783 ± 182 23 2.91 (2.56, 3.38) 0.98 (0.98, 0.99)
v0 (m·s−1) 2.24 ± 0.29 2.26 ± 0.32 0.15 6.65 (5.85, 7.72) 0.76 (0.67, 0.83)
Pmax (W) 438 ± 119 442 ± 115 24 5.40 (4.75, 6.26) 0.96 (0.94, 0.97)
Quadratic 58 F0 (N) 787 ± 206 781 ± 198 45 5.74 (4.85, 7.03) 0.95 (0.92, 0.97)
polynomial v0 (m·s−1) 2.19 ± 0.68 2.20 ± 0.54 0.45 20.3 (17.2, 24.8) 0.48 (0.25, 0.65)
Pmax (W) 419 ± 107 424 ± 109 25 5.88 (4.97, 7.19) 0.95 (0.92, 0.97)
Hyperbolic 91 F0 (N) 818 ± 224 817 ± 201 66 8.07 (7.05, 9.45) 0.91 (0.86, 0.94)
v0 (m·s−1) 3.72 ± 5.26 5.83 ± 26.34 19 399 (349, 468) −0.01 (−0.21, 0.19)
Pmax (W) 423 ± 102 422 ± 104 31 7.30 (6.37, 8.54) 0.91 (0.87, 0.94)
Exponential 102 F0 (N) 887 ± 242 878 ± 244 62 7.00 (6.15, 8.11) 0.94 (0.91, 0.96)
v0 (m·s−1) 5.89 ± 2.67 5.94 ± 2.88 1.31 22.1 (19.5, 25.7) 0.78 (0.69, 0.85)
Pmax (W) 483 ± 151 480 ± 148 57 11.9 (10.5, 13.8) 0.85 (0.79, 0.90)
F0, maximum force; v0, maximum velocity; Pmax, maximum power; SD, standard deviation; SEM, standard error of measurement; CV, coefficient of variation; ICC, intraclass correlation coefficient;
95% CI, 95% confidence interval. Bold numbers indicate an unacceptable reliability (cut-off CV values > 10%; cut-off ICC values < 0.75).
SPORTS BIOMECHANICS
11
12 A. PÉREZ-CASTILLA ET AL.

value = 13.8%). On the contrary, the v0 did not meet the criterion of acceptable reliability
for any of the regression models (cut-off CV values ≥ 24.8%; cut-off ICC values ≤ 0.72)
with the only exception of absolute reliability for the linear multiple- and two-point
models (cut-off CV values = 6.97% and 7.72%, respectively). In general, the comparison
of the CV values in the multicentre reliability study revealed a higher reliability of F0 and
v0 for the linear multiple-point and linear two-point models compared to the quadratic
polynomial (CVratio = 1.89 and 1.97 for F0; 3.38 and 3.05 for v0, respectively), hyperbolic
(CVratio = 2.66 and 2.77 for F0; 66.39 and 60.00 for v0, respectively) and exponential
models (CVratio = 2.31 and 2.41 for F0; 3.68 and 3.32 for v0, respectively). Pmax was
obtained with a higher reliability using the linear multiple-point, linear two-point and
quadratic polynomial compared to the hyperbolic (CVratio = 1.22, 1.35 and 1.24, respec­
tively) and exponential models (CVratio = 1.99, 2.20 and 2.02, respectively; Figure 2).

Bench press throw


The F-v relationships were strong for the linear multiple-point (r2 = 0.974 [0.762, 1.000]),
quadratic polynomial (r2 = 0.992 [0.769, 1.000]), hyperbolic (r2 = 0.990 [0.762, 1.000])
and exponential (r2 = 0.972 [0.757, 0.998]) models. However, v0 could not be obtained in
82 out of 162 relationships using the quadratic polynomial model and in 11 out of 162
relationships using the hyperbolic model.
The reliability of the F-v relationship parameters obtained using the different regression
models are presented in Table 4. All regression models generally revealed an acceptable
reliability for F0 and Pmax (cut-off CV values ≤ 7.88%; cut-off ICC values ≥ 0.85) in the
multicentre study with the exceptions of F0 for the quadratic polynomial (cut-off CV
value = 12.9%; cut-off ICC value = 0.68) and hyperbolic (cut-off CV value = 11.4%; cut-off
ICC value = 0.74) models and Pmax for exponential model (cut-off CV value = 12.4%; cut-
off ICC value = 0.73). On the contrary, the v0 did not meet the criterion of acceptable
reliability for any of the regression models (cut-off CV values ≥ 36.2%; cut-off ICC values
≤ 0.71) with the only exception of absolute reliability for the linear multiple- and two-
point models (cut-off CV values = 7.06% and 7.33%, respectively). In general, the
comparison of the CV values in the multicentre reliability study revealed a higher relia­
bility of F0 and v0 for the linear multiple-point and linear two-point models compared to
the quadratic polynomial (CVratio = 2.42 and 2.39 for F0; 5.04 and 4.86 for v0, respectively),
hyperbolic (CVratio = 2.50 and 2.47 for F0; 27.5 and 26.49 for v0, respectively) and
exponential models (CVratio = 1.58 and 1.56 for F0; 5.13 and 4.94 for v0, respectively).
Pmax was obtained with a higher reliability using the linear multiple-point, linear two-point
and quadratic polynomial compared to the hyperbolic (CVratio = 1.42, 1.34 and 1.47,
respectively) and exponential models (CVratio = 2.27, 2.13 and 2.34, respectively).

Discussion and implications


This a-posteriori multicentre reliability study combined the data of four studies for the BP
exercise (Cuevas-Aburto et al., 2018; Garcia-Ramos et al., 2016; García-Ramos, Pestaña-
Melero, et al., 2018; Iglesias-Soler, Mayo et al., 2019) and three studies for the BPT
exercise (Cuevas-Aburto et al., 2018; Garcia-Ramos et al., 2016; García-Ramos, Pestaña-
Melero, et al., 2018) to elucidate which regression model provides the F-v relationship
SPORTS BIOMECHANICS 13

Figure 2. Comparison of the reliability of the force-velocity relationship parameters between the linear
multiple-point, linear two-point, quadratic polynomial, hyperbolic, and exponential models during the
bench press (upper panel) and bench press throw (lower panel) exercises. F0, maximum force; v0,
maximum velocity; Pmax, maximum power. a, significantly more reliable than the quadratic polynomial
model; b, significantly more reliable than the hyperbolic model; c, significantly more reliable than the
exponential model (CVratio > 1.15).

parameters with the highest reliability from data obtained commonly obtained in prac­
tice. The main finding of this study was that all regression models, with some exceptions
for the quadratic polynomial and hyperbolic models for F0 and the exponential model for
Pmax, provided F0 and Pmax with an acceptable absolute and relative reliability. However,
only the linear multiple- and two-point models provided v0 with acceptable absolute
reliability, while the relative reliability of v0 was unacceptable for all regression models.
Regardless of the exercise, the reliability of F0 and v0 was higher using the linear multiple-
14

Table 4. Reliability of the force-velocity (F-v) relationship parameters obtained for the independent studies as well as for the multicentre study using the linear
multiple-point, linear two-point, quadratic polynomial, hyperbolic, and exponential models during the bench press throw exercise.
Study F-v model Sample size (n) Parameter Session 1 (mean ± SD) Session 2 (mean ± SD) SEM CV (%) (95% CI) ICC (95% CI)
Garcia-Ramos et al. (2016) Linear multiple-point 32 F0 (N) 694 ± 118 690 ± 110 19 2.68 (2.15, 3.56) 0.98 (0.95, 0.99)
v0 (m·s−1) 2.70 ± 0.23 2.68 ± 0.29 0.15 5.44 (4.36, 7.23) 0.71 (0.48, 0.85)
Pmax (W) 469 ± 95 464 ± 96 21 4.47 (3.58, 5.94) 0.96 (0.91, 0.98)
Linear two-point 32 F0 (N) 709 ± 121 707 ± 116 16 2.28 (1.83, 3.03) 0.98 (0.96, 0.99)
v0 (m·s−1) 2.74 ± 0.23 2.69 ± 0.30 0.16 5.94 (4.76, 7.89) 0.65 (0.39, 0.81)
Pmax (W) 487 ± 98 478 ± 106 24 4.99 (4.00, 6.63) 0.95 (0.90, 0.97)
A. PÉREZ-CASTILLA ET AL.

Quadratic 10 F0 (N) 723 ± 143 699 ± 141 50 6.97 (4.80, 12.7) 0.91 (0.67, 0.98)
−1
polynomial v0 (m·s ) 2.87 ± 0.82 2.73 ± 0.73 0.74 26.4 (18.1, 48.2) 0.10 (−0.53, 0.66)
Pmax (W) 472 ± 135 474 ± 136 18 3.80 (2.62, 6.95) 0.99 (0.95, 1.00)
Hyperbolic 29 F0 (N) 783 ± 140 765 ± 183 77 9.96 (7.90, 13.5) 0.79 (0.60, 0.89)
v0 (m·s−1) 7.74 ± 12.9 4.92 ± 4.61 9.75 154 (122, 208) −0.02 (−0.38,
0.34)
Pmax (W) 484 ± 96 471 ± 104 35 7.32 (5.81, 9.89) 0.89 (0.77, 0.94)
Exponential 32 F0 (N) 775 ± 143 763 ± 120 37 4.77 (3.83, 6.34) 0.93 (0.86, 0.96)
v0 (m·s−1) 5.90 ± 1.42 6.18 ± 1.85 0.92 15.3 (12.2, 20.3) 0.70 (0.47, 0.84)
Pmax (W) 511 ± 112 517 ± 132 40 7.74 (6.21, 10.3) 0.90 (0.81, 0.95)
García-Ramos, Pestaña-Melero, et al. Linear multiple-point 30 F0 (N) 783 ± 127 790 ± 125 23 2.91 (2.32, 3.92) 0.97 (0.94, 0.99)
(2018) v0 (m·s−1) 2.21 ± 0.24 2.19 ± 0.18 0.12 5.54 (4.41, 7.44) 0.67 (0.42, 0.83)
Pmax (W) 431 ± 77 432 ± 73 20 4.58 (3.65, 6.16) 0.94 (0.87, 0.97)
Linear two-point 30 F0 (N) 813 ± 190 835 ± 201 46 5.58 (4.21, 8.25) 0.95 (0.88, 0.98)
v0 (m·s−1) 2.35 ± 0.24 2.34 ± 0.19 0.13 5.36 (4.27, 7.21) 0.68 (0.43, 0.83)
Pmax (W) 472 ± 96 474 ± 89 22 4.72 (3.76, 6.35) 0.95 (0.89, 0.97)
Quadratic 9 F0 (N) 740 ± 121 786 ± 127 47 6.20 (4.19, 11.9) 0.89 (0.60, 0.97)
polynomial v0 (m·s−1) 2.43 ± 0.38 2.55 ± 0.18 0.29 11.6 (7.82, 22.2) 0.09 (0.58, 0.68)
Pmax (W) 396 ± 77 397 ± 77 20 4.94 (3.34, 9.47) 0.95 (0.81, 0.99)
Hyperbolic 29 F0 (N) 888 ± 160 893 ± 153 47 5.26 (4.18, 7.12) 0.92 (0.83, 0.96)
v0 (m·s−1) 6.66 ± 13.7 4.92 ± 3.79 10.19 176 (140, 238) −0.03 (−0.39,
0.33)
Pmax (W) 445 ± 76 438 ± 66 26 6.00 (4.76, 8.11) 0.87 (0.74, 0.94)
Exponential 30 F0 (N) 844 ± 143 852 ± 141 30 3.48 (2.77, 4.68) 0.96 (0.92, 0.98)
v0 (m·s−1) 6.31 ± 1.96 6.17 ± 1.48 0.88 14.1 (11.3, 19.0) 0.75 (0.55, 0.88)
Pmax (W) 500 ± 86 499 ± 74 44 8.75 (6.97, 11.8) 0.72 (0.49, 0.86)
(Continued)
Table 4. (Continued).
Study F-v model Sample size (n) Parameter Session 1 (mean ± SD) Session 2 (mean ± SD) SEM CV (%) (95% CI) ICC (95% CI)
Cuevas-Aburto et al. (2018) Linear multiple-point 19 F0 (N) 816 ± 189 840 ± 196 45 5.42 (4.09, 8.01) 0.95 (0.88, 0.98)
v0 (m·s−1) 2.45 ± 0.24 2.37 ± 0.30 0.18 7.45 (5.63, 11.0) 0.58 (0.18, 0.81)
Pmax (W) 498 ± 122 495 ± 123 25 5.05 (3.82, 7.47) 0.96 (0.91, 0.99)
Linear two-point 19 F0 (N) 813 ± 190 835 ± 201 46 5.58 (4.21, 8.25) 0.95 (0.88, 0.98)
v0 (m·s−1) 2.49 ± 0.22 2.42 ± 0.28 0.19 7.74 (5.85, 11.5) 0.45 (0.01, 0.74)
Pmax (W) 505 ± 120 501 ± 123 26 5.11 (3.86, 7.55) 0.96 (0.90, 0.98)
Quadratic 5 F0 (N) 824 ± 285 852 ± 281 121 14.5 (8.66, 41.6) 0.95 (0.62, 0.99)
polynomial v0 (m·s−1) 0.25 ± 3.98 0.92 ± 2.90 0.97 166 (99, 476) 0.99 (0.92, 1.00)
Pmax (W) 538 ± 181 525 ± 152 28 5.32 (3.20, 15.3) 1.00 (0.99, 1.00)
Hyperbolic 14 F0 (N) 847 ± 202 961 ± 302 106 11.7 (8.50, 18.9) 0.85 (0.61, 0.95)
v0 (m·s−1) 2.46 ± 0.27 2.98 ± 1.30 0.97 35.6 (25.8, 57.3) −0.06 (−0.56,
0.47)
Pmax (W) 515 ± 130 515 ± 133 32 6.25 (4.53, 10.1) 0.95 (0.85, 0.98)
Exponential 19 F0 (N) 911 ± 249 956 ± 266 80 8.55 (6.46, 12.6) 0.91 (0.79, 0.97)
v0 (m·s−1) 7.52 ± 9.74 5.96 ± 5.12 3.54 52.5 (39.7, 77.7) 0.81 (0.57, 0.92)
Pmax (W) 529 ± 137 521 ± 119 55 10.5 (9.08, 12.4) 0.82 (0.73, 0.88)
Multicentre study Linear multiple-point 81 F0 (N) 755 ± 148 762 ± 151 29 3.80 (3.29, 4.50) 0.96 (0.94, 0.98)
v0 (m·s−1) 2.46 ± 0.32 2.43 ± 0.34 0.15 5.97 (5.17, 7.06) 0.81 (0.71, 0.87)
Pmax (W) 462 ± 99 459 ± 97 21 4.63 (4.01, 5.48) 0.95 (0.93, 0.97)
Linear two-point 81 F0 (N) 768 ± 154 775 ± 156 30 3.84 (3.32, 4.54) 0.96 (0.94, 0.98)
v0 (m·s−1) 2.54 ± 0.29 2.50 ± 0.30 0.16 6.19 (5.37, 7.33) 0.72 (0.60, 0.81)
Pmax (W) 485 ± 102 482 ± 103 24 4.92 (4.26, 5.82) 0.95 (0.92, 0.97)
Quadratic 24 F0 (N) 751 ± 169 764 ± 176 70 9.19 (7.15, 12.9) 0.85 (0.68, 0.93)
polynomial v0 (m·s−1) 2.18 ± 1.99 2.29 ± 1.49 0.67 30.1 (23.4, 43.3) 0.86 (0.71, 0.94)
Pmax (W) 457 ± 134 456 ± 126 20 4.49 (3.49, 6.30) 0.98 (0.95, 0.99)
Hyperbolic 72 F0 (N) 838 ± 166 855 ± 213 80 9.50 (8.16, 11.4) 0.83 (0.74, 0.89)
v0 (m·s−1) 6.28 ± 11.95 4.54 ± 3.87 8.91 164 (141, 197) −0.01 (−0.24,
0.22)
Pmax (W) 474 ± 99 466 ± 100 31 6.58 (5.66, 7.88) 0.91 (0.85, 0.94)
Exponential 81 F0 (N) 832 ± 179 841 ± 185 50 6.00 (5.20, 7.10) 0.93 (0.89, 0.95)
v0 (m·s−1) 6.43 ± 4.89 6.13 ± 2.83 1.92 30.6 (26.5, 36.2) 0.77 (0.67, 0.85)
Pmax (W) 529 ± 137 521 ± 119 55 10.5 (9.08, 12.4) 0.82 (0.73, 0.88)
SPORTS BIOMECHANICS

F0, maximum force; v0, maximum velocity; Pmax, maximum power; SD, standard deviation; SEM, standard error of measurement; CV, coefficient of variation; ICC, intraclass correlation coefficient;
95% CI, 95% confidence interval. Bold numbers indicate an unacceptable reliability (cut-off CV values > 10%; cut-off ICC values < 0.75).
15
16 A. PÉREZ-CASTILLA ET AL.

and two-point models compared to the quadratic polynomial, hyperbolic and exponen­
tial models, while the reliability of Pmax was higher using the linear multiple-point, linear
two-point and quadratic polynomial compared to the hyperbolic and exponential mod­
els. These results suggest that linear regression models during BP and BPT exercises
provide the F-v relationship parameters with the highest reliability.
The goodness of fit of the F-v relationships was strong for all regression models in the
BP and BPT exercises. In line with previous studies, a slightly higher r2 was observed for
the quadratic polynomial model compared to the linear and exponential models
(Iglesias-Soler, Fariñas et al., 2019; Sreckovic et al., 2015). By contrast, some studies did
not find differences in r2 between the linear and quadratic polynomial models during the
BP, back-squat and vertical jump exercises (Cuk et al., 2014; Iglesias-Soler, Mayo et al.,
2019). It has been suggested that the apparent linearity of the F-v relationship observed in
this (see, Table 2) and previous studies could be a consequence of the narrow range of
loads tested (Alcazar et al., 2019, 2020). Moreover, it has recently been suggested that the
F-v relationship could be better fitted by a double-hyperbolic function when the experi­
mental data range from 35% to 92% or 18% to 97% of F0 during the unilateral leg press
and BP exercises, respectively (Alcazar et al., 2020). However, in that study (Alcazar et al.,
2020), both exercises were evaluated using partial ranges of motion to obtain experi­
mental data close to the high-force/low-velocity region of the F-v relationship. In the
present study, the Edman’s double-hyperbolic model could not be applied because we did
not obtain experimental points close to F0 following a double-hyperbolic shape despite
that the 1RM load was included in several studies. Therefore, although it is possible that
the F-v relationship could be better modelled by a double-hyperbolic function, it is
impractical because this model cannot be applied to the data typically determined during
common monitoring procedures in a real-world setting (e.g., different load ranges and
exercises tested using a full range of motion).
In addition to the goodness of fit, the reliability and physiological meaning of the
F-v relationship parameters should be taken into consideration when selecting the most
appropriate model. Moreover, the reliability has to be tested from data commonly
obtained during testing in practice, notably in terms of F-v spectrum range. Our findings
show that different regression models included in the current study generally provided F0
and Pmax with an acceptable absolute and relative reliability (except the quadratic
polynomial and hyperbolic models for F0 and the exponential model for Pmax). Only
the linear regression models (i.e., multiple- and two-point models) provided v0 with
acceptable absolute reliability, while the relative reliability of v0 was unacceptable for all
regression models. Practitioners should therefore question the use of v0, which is also
used to compute Pmax, to differentiate performance between athletes and also to identify
real changes in individual performance and make training decisions. The lower reliability
observed for v0 could be explained by the greater extrapolation from the experimental
points to v0 in exercises performed against gravity (see, Table 2; García-Ramos et al.,
2021). These results are partially in consensus with previous studies that also reported an
acceptable reliability for the three F-v relationship parameters (F0,v0, and Pmax) using
linear models and for F0 and Pmax using a quadratic polynomial model during the
different variants of the BP exercise (Garcia-Ramos et al., 2016; Iglesias-Soler, Mayo
et al., 2019; Pérez-Castilla et al., 2018; Sreckovic et al., 2015). In addition, supporting our
hypothesis, the multiple- and two-point linear regression models provided the
SPORTS BIOMECHANICS 17

parameters of the F-v relationship (except for Pmax) with a higher reliability than the
quadratic polynomial model. This is in line with the findings of Iglesias-Soler et al. (2019)
who also observed more reliable parameters using the multiple-point linear model than
the quadratic polynomial model during the BP and back-squat exercises. These results
are probably caused by an inconsistent curvature of the F-v relationship fitted by the
quadratic polynomial model. Specifically, 32 out of 143 relationships for the BP and 3 out
of 80 relationships for the BPT displayed downward concavity profiles, limiting the
physiological meaning of the F-v relationship by yielding lower, illogical values of F0 and
v0. Furthermore, this is the first study to show that linear models enable determination of
the F-v relationship parameters with a greater reliability than both hyperbolic and
exponential models. The worst reliability of both non-linear models can also be attrib­
uted to higher, illogical values of the F-v relationship parameters due to a mismatch of the
function with respect to the intercept axes and, more specifically, to the velocity-intercept
(see, Figure 1). Even more importantly, the function did not cut the velocity-intercept on
numerous occasions for the quadratic polynomial (30–50%) and hyperbolic models (5–
7%), whereas a double adjustment by exchanging the force and velocity variables is
necessary with the exponential model to report all F-v relationship parameters (Iglesias-
Soler, Fariñas et al., 2019; Iglesias-Soler, Mayo et al., 2019). Taken together, our results
suggest an advantage of the linear models considering their higher reliability, simplicity
and physiological meaning of its parameters.
Although this a-posteriori multicentre reliability study provides novel findings regard­
ing the most suitable model for determining the F-v relationship parameters, some
limitations need to be addressed. First, while linear models can better fit the
F-v relationships considering a relatively narrow range of loads, non-linear models
could be more beneficial when experimental data points closers to the F0 and v0 are
considered (Alcazar et al., 2019, 2020). However, it should be noted that, in this study, the
F-v relationship was assessed with a full range of motion with a frequent inclusion of the
most extreme loading conditions that can be applied in practical settings (from 1 kg to
1RM). Second, the use of the force and velocity values within the propulsive phase could
have affected the shape of the F-v relationship. Namely, while the propulsive phase
represents approximately the same range of motion for all loads evaluated in the BPT,
the range of motion that corresponds to the propulsive phase can increase with the
increment of the load during the BP. Nevertheless, both the goodness of fit and the
reliability of the F-v relationship parameters did not differ between the BP and BPT
regardless of the regression model considered.

Conclusions
The absolute and relative reliability of F0 and Pmax for all regression models was
acceptable (with some exceptions for the quadratic polynomial and hyperbolic models
for F0 and exponential model for Pmax). Instead, the absolute reliability of v0 was only
acceptable for the linear multiple- and two-point models, while the relative reliability
of v0 was inacceptable for all regression models. Therefore, v0 would not only be an
inappropriate parameter to differentiate performance between athletes, but also to
identify real changes in performance specially if obtained from non-linear models.
The linear multiple- and two-point models provided the F-v relationship parameters
18 A. PÉREZ-CASTILLA ET AL.

with a higher reliability than the quadratic polynomial, hyperbolic and exponential
models, while the reliability did not differ between the two linear models.
Importantly, the quadratic polynomial and hyperbolic models failed to reveal the v0
in several tests (30–50% and 5–7%, respectively) and, together with the exponential
model, they usually yielded illogical F-v relationship parameters, limiting its physio­
logical meaning. Taken together, the linear regression model provides the
F-v relationship parameters with the highest reliability, simplicity and apparent
physiological meaning.

Disclosure statement
No potential conflict of interest was reported by the author(s).

Funding
The author(s) reported there is no funding associated with the work featured in this article.

ORCID
Alejandro Pérez-Castilla http://orcid.org/0000-0001-5535-2087
Pierre Samozino http://orcid.org/0000-0002-1665-870X
Eliseo Iglesias-Soler http://orcid.org/0000-0002-1212-4188
Amador García-Ramos http://orcid.org/0000-0003-0608-8755

References
Alcazar, J., Csapo, R., Ara, I., & Alegre, L. M. (2019). On the shape of the force-velocity relationship
in skeletal muscles: The linear, the hyperbolic, and the double-hyperbolic. Frontiers in
Physiology, 10(10), 769. https://doi.org/10.3389/fphys.2019.00769
Alcazar, J., Pareja-Blanco, F., Rodriguez-Lopez, C., Navarro-Cruz, R., Cornejo-Daza, P. J., Ara, I.,
& Alegre, L. M. (2020). Comparison of linear, hyperbolic and double-hyperbolic models to
assess the force–velocity relationship in multi-joint exercises. European Journal of Sport Science,
21(3), 359–369. https://doi.org/10.1080/17461391.2020.1753816
Cormie, P., McGuigan, M. R., & Newton, R. U. (2011). Developing maximal neuromuscular
power: Part 1–biological basis of maximal power production. Sports Medicine, 41(1), 17–38.
https://doi.org/10.2165/11537690-000000000-00000
Cuevas-Aburto, J., Ulloa-Díaz, D., Barboza-González, P., Chirosa-Ríos, L. J., & García-Ramos, A.
(2018). The addition of very light loads into the routine testing of the bench press increases the
reliability of the force–velocity relationship. PeerJ, 6, e5835. https://doi.org/10.7717/peerj.5835
Cuk, I., Markovic, M., Nedeljkovic, A., Ugarkovic, D., Kukolj, M., & Jaric, S. (2014). Force-velocity
relationship of leg extensors obtained from loaded and unloaded vertical jumps. European
Journal of Applied Physiology, 114(8), 1703–1714. https://doi.org/10.1007/s00421-014-2901-2
Fenn, W. O., & Marsh, B. S. (1935). Muscular force at different speeds of shortening. The Journal of
Physiology, 85(3), 277–297. https://doi.org/10.1113/jphysiol.1935.sp003318
Fulton, S. K., Pyne, D., Hopkins, W., & Burkett, B. (2009). Variability and progression in
competitive performance of Paralympic swimmers. Journal of Sports Sciences, 27(5), 535–539.
https://doi.org/10.1080/02640410802641418
SPORTS BIOMECHANICS 19

García-Ramos, A., Feriche, B., Pérez-Castilla, A., Padial, P., & Jaric, S. (2017). Assessment of leg
muscles mechanical capacities: Which jump, loading, and variable type provide the most
reliable outcomes? European Journal of Sport Science, 17(6), 690–698. https://doi.org/10.1080/
17461391.2017.1304999
Garcia-Ramos, A., & Janicijevic, D. (2020). Potential benefits of multicenter reliability studies in
sports science: A practical guide for its implementation. Isokinetics and Exercise Science, 28(2),
199–204. https://doi.org/10.3233/IES-192242
Garcia-Ramos, A., Jaric, S. (2018). Two-point method: A quick and fatigue-free procedure for
assessment of muscle mechanical capacities and the one-repetition maximum. Strength and
Conditioning Journal, 40(2), 54–66. https://doi.org/10.1519/SSC.0000000000000359
García-Ramos, A., & Jaric, S. (2019). Optimization of the force–velocity relationship obtained from
the bench-press-throw exercise: An a posteriori multicenter reliability study. International Journal
of Sports Physiology and Performance, 14(3), 317–322. https://doi.org/10.1123/ijspp.2018-0457
Garcia-Ramos, A., Jaric, S., Padial, P., & Feriche, B. (2016). Force-velocity relationship of upper
body muscles: Traditional versus ballistic bench press. Journal of Applied Biomechanics, 32(2),
178–185. https://doi.org/10.1123/jab.2015-0162
García-Ramos, A., Pérez-Castilla, A., & Jaric, S. (2021). Optimisation of applied loads when using
the two-point method for assessing the force-velocity relationship during vertical jumps. Sports
Biomechanics, 20(3), 274–289. https://doi.org/10.1080/14763141.2018.1545044
García-Ramos, A., Pestaña-Melero, F. L., Pérez-Castilla, A., Rojas, F. J., & Haff, G. G. (2018).
Differences in the load-velocity profile between 4 bench press variants. International Journal of
Sport Physiology and Performance, 13(3), 326–331. https://doi.org/10.1123/ijspp.2017-0158
García-Ramos, A., Torrejón, A., Feriche, B., Morales-Artacho, A. J., Pérez-Castilla, A., Padial, P., &
Jaric, S. (2018). Selective effects of different fatigue protocols on the function of upper body
muscles assessed through the force–velocity relationship. European Journal of Applied
Physiology, 118(2), 439–447. https://doi.org/10.1007/s00421-017-3786-7
García-Ramos, A., Torrejón, A., Morales-Artacho, A. J., Pérez-Castilla, A., & Jaric, S. (2018). Optimal
resistive forces for maximizing the reliability of leg muscles’ capacities tested on a cycle ergometer.
Journal of Applied Biomechanics, 34(1), 47–52. https://doi.org/10.1123/jab.2017-0056
Giroux, C., Rabita, G., Chollet, D., & Guilhem, G. (2016). Optimal balance between force and
velocity differs among world-class athletes. Journal of Applied Biomechanics, 32(1), 59–68.
https://doi.org/10.1123/jab.2015-0070
Hill, A. (1938). The heat of shortening and the dynamic constants of muscle. Proceedings of the
Royal Society of London. Series B - Biological Sciences, 126(843), 136–195. https://doi.org/10.
1098/rspb.1938.0050
Hopkins, W. (2000). Calculations for reliability (Excel spreadsheet). A New View of Statistics.
Retrieved January 06, 2021, from http://www.sportsci.org/resource/stats/relycalc.html#excel
Iglesias-Soler, E., Fariñas, J., Mayo, X., Santos, L., & Jaric, S. (2019). Comparison of different
regression models to fit the force–velocity relationship of a knee extension exercise. Sports
Biomechanics, 18(2), 174–189. https://doi.org/10.1080/14763141.2018.1442873
Iglesias-Soler, E., Mayo, X., Rial-Vázquez, J., Morín-Jiménez, A., Aracama, A., Guerrero-
Moreno, J. M., & Jaric, S. (2019). Reliability of force-velocity parameters obtained from
linear and curvilinear regressions for the bench press and squat exercises. Journal of Sports
Sciences, 37(22), 2596–2603. https://doi.org/10.1080/02640414.2019.1648993
James, L. P., Roberts, L. A., Haff, G. G., Kelly, V. G., & Beckman, E. M. (2017). Validity and
reliability of a portable isometric mid-thigh clean pull. Journal of Strength and Conditioning
Research, 31(5), 1378–1386. https://doi.org/10.1519/JSC.0000000000001201
Jaric, S. (2015). Force-velocity relationship of muscles performing multi-joint maximum perfor­
mance tasks. International Journal of Sports Medicine, 36(9), 699–704. https://doi.org/10.1055/
s-0035-1547283
Jaric, S. (2016). Two-load method for distinguishing between muscle force, velocity, and
power-producing capacities. Sports Medicine, 46(11), 1585–1589. https://doi.org/10.1007/
s40279-016-0531-z
20 A. PÉREZ-CASTILLA ET AL.

Jiménez-Reyes, P., Samozino, P., Brughelli, M., & Morin, J. B. (2017). Effectiveness of an indivi­
dualized training based on force-velocity profiling during jumping. Frontiers in Physiology, 7,
677. https://doi.org/10.3389/fphys.2016.00677
Koo, T. K., & Li, M. Y. (2016). A guideline of selecting and reporting intraclass correlation
coefficients for reliability research. Journal of Chiropractic Medicine, 15(2), 155–163. https://
doi.org/10.1016/j.jcm.2016.02.012
Mendiguchia, J., Edouard, P., Samozino, P., Brughelli, M., Cross, M., Ross, A., Gill, N., &
Morin, J. B. (2016). Field monitoring of sprinting power–force–velocity profile before, during
and after hamstring injury: Two case reports. Journal of Sports Sciences, 34(6), 535–541. https://
doi.org/10.1080/02640414.2015.1122207
Miller, R. M., Freitas, E. D. S., Heishman, A. D., Koziol, K. J., Galletti, B. A. R., Kaur, J., &
Bemben, M. G. (2020). Test-retest reliability between free weight and machine-based movement
velocities. Journal of Strength and Conditioning Research, 34(2), 440–444. https://doi.org/10.
1519/JSC.0000000000002817
Pérez-Castilla, A., Jaric, S., Feriche, B., Padial, P., & García-Ramos, A. (2018). Evaluation of muscle
mechanical capacities through the two-load method: Optimization of the load selection. Journal
of Strength and Conditioning Research, 32(5), 1245–1253. https://doi.org/10.1519/JSC.
0000000000001969
Samozino, P., Rejc, E., Di Prampero, P. E., Belli, A., & Morin, J.-B. (2014). Force–velocity
properties’ contribution to bilateral deficit during ballistic push-off. Medicine and Science in
Sports and Exercise, 46(1), 107–114. https://doi.org/10.1249/MSS.0b013e3182a124fb
Sreckovic, S., Cuk, I., Djuric, S., Nedeljkovic, A., Mirkov, D., & Jaric, S. (2015). Evaluation of
force-velocity and power-velocity relationship of arm muscles. European Journal of Applied
Physiology, 115(8), 1779–1787. https://doi.org/10.1007/s00421-015-3165-1

You might also like