Running head: MULTIDIMENSIONAL ASSESSMENT OF TEAMWORK IN SPORT 1 2
The Development and Psychometric Properties of the Multidimensional Assessment of
3
Teamwork in Sport (MATS)
4 5
Authors: Desmond McEwan*, Bruno D. Zumbo, Mark A. Eys, & Mark R. Beauchamp
6 7
*Corresponding author (
[email protected])
8 9
Institutional Affiliations:
10
Desmond McEwan & Mark R. Beauchamp: School of Kinesiology, The University of British
11
Columbia, Vancouver, British Columbia, Canada
12
Bruno D. Zumbo: Department of Educational and Counseling Psychology, Faculty of Education,
13
University of British Columbia, Vancouver, British Columbia, Canada
14
Mark A. Eys: Departments of Kinesiology/Physical Education and Psychology, Wilfrid Laurier
15
University, Waterloo, Ontario, Canada
16 17
As accepted for publication in Journal of Sport & Exercise Psychology, ©Human Kinetics
18
Date of acceptance: March 15, 2018
19 20
MULTIDIMENSIONAL ASSESSMENT OF TEAMWORK IN SPORT 1 2
2
Abstract The purpose of this research was to develop a questionnaire to assess the multidimensional
3
construct of teamwork in sport and examine various aspects of validity related to that instrument. A
4
preliminary questionnaire was first created, and feedback on this instrument was then obtained from
5
a sample of team sport athletes (n = 30) and experts in sport psychology (n = 8). A modified version
6
of the questionnaire was then completed by 607 athletes from 48 teams, and five multilevel
7
confirmatory factor analyses were conducted to examine the structural properties of data derived
8
from this instrument. Evidence of adequate model-data fit along with measurement reliability was
9
obtained for each of the five models. Taken together, the results from this research provide support
10
for the content, substantive, and structural aspects of construct validity (cf. Messick, 1995) for data
11
derived from the 66-item Multidimensional Assessment of Teamwork in Sport.
12
Keywords: Team; Group; Processes; Effectiveness.
MULTIDIMENSIONAL ASSESSMENT OF TEAMWORK IN SPORT 1
The Development and Psychometric Properties of the Multidimensional Assessment of
2
Teamwork in Sport (MATS)
3
3
Teamwork is often noted as an important variable within the vernacular of sport. Coaches
4
frequently emphasize the importance of players working together, with athletes similarly attributing
5
team outcomes to the extent to which they work well with their teammates. Despite this seeming
6
importance of teamwork, there has been surprisingly limited research on this construct within sport
7
settings (Carron, Martin, & Loughead, 2012; McEwan & Beauchamp, 2014). Indeed, although
8
decades of research has accumulated within other team contexts such as health care, aviation,
9
business, and military settings, the evidence related to teamwork in sport has been sparse,
10 11
fragmented, and mostly anecdotal (McEwan & Beauchamp, 2014). In an effort to begin to fill this considerable hole in the field of sport psychology, McEwan
12
and Beauchamp (2014) conducted a theoretical and integrative review in order to provide a working
13
definition and conceptual framework of teamwork in sport. They defined teamwork as “a dynamic
14
process involving a collaborative effort by team members to effectively carry out the independent
15
and interdependent behaviors that are required to maximize a team’s likelihood of achieving its
16
purposes” (p. 233). Their conceptual framework—which includes five overarching aspects and 14
17
lower-order dimensions of teamwork—was informed by Rousseau, Aubé, and Savoie’s (2006)
18
teamwork framework. The framework by Rousseau et al. (2006) was itself based on a
19
comprehensive analysis of 29 frameworks used to study teamwork behaviors in a range of group
20
settings (e.g., health care, business, military) and, as such, can be considered a ‘meta-framework’ of
21
teamwork. Twelve of the dimensions within McEwan and Beauchamp’s (2014) framework focus on
22
behaviors associated with team task performance (i.e., regulation of team performance), while two
23
dimensions focus on behaviors associated with the management of team maintenance (i.e.,
24
interpersonal dynamics of a team).
MULTIDIMENSIONAL ASSESSMENT OF TEAMWORK IN SPORT 1
4
With regard to regulating team performance (RTP), teamwork behaviors were
2
conceptualized by McEwan and Beauchamp (2014) to be enacted over four separate phases: (a)
3
preparation, (b) execution, (c) evaluation, and (d) adjustments. Three behavioral dimensions
4
comprise the preparation phase, which take place in advance of a team task (e.g., at the start of a
5
team’s season, before a game or competition). First, mission analysis involves defining a team’s
6
overall purpose or reasons for being together (e.g., to win a league championship, to have fun).
7
Second, goal specification involves setting team goals in order to achieve those team purposes (e.g.,
8
to obtain a certain race time in team kayaking, to score a specific number of runs per game in
9
baseball). Third, action planning involves establishing strategies for obtaining team goals (e.g.,
10
identifying defensive matchups in a basketball game, running drills in a practice).
11
Three dimensions comprise the execution phase of RTP, which occur during a team task
12
(i.e., while a team is competing). First, communication involves the information sharing that occurs
13
between teammates (e.g., field hockey players calling for a pass, a coxswain relaying stroke
14
commands in crew rowing). Second, coordination involves teammates being synchronized with
15
each other in terms of the sequencing and timing of their actions (e.g., quarterbacks passing the ball
16
once their receivers have run their route in football, a track team exchanging the baton at the ideal
17
time). Third, cooperation involves teammates working together in a unified manner (e.g., hockey
18
players blocking opponents’ shots for the benefit of their team, rugby players working together as
19
one strong entity in order to advance the ball in a scrum in a joint manner). Although the latter two
20
dimensions are similar in many regards, it is important to note that they are conceptually distinct.
21
With regard to coordination, the execution of a player’s task is dependent upon, and/or subsequently
22
links to, other players’ task executions. With regard to cooperation, team success is based on team
23
members helping one another and working collaboratively as a single, interconnected unit on a team
24
task.
MULTIDIMENSIONAL ASSESSMENT OF TEAMWORK IN SPORT 1
5
The third and fourth phases of RTP—evaluation and adjustments—occur after a team has
2
completed its task (e.g., during the halftime break in soccer, following the conclusion of a
3
tournament). The evaluation phase is comprised of two dimensions, which act as a means of self-
4
regulation. These dimensions include performance monitoring, whereby teams monitor how well
5
they have performed (e.g., whether the team won a recent game, how effectively the team worked
6
together) as well as systems monitoring, wherein teams monitor various conditions that may have
7
affected their performance (e.g., situations during a competition, changes in team personnel).
8
Following task execution, four teamwork behaviors comprise adjustments, the fourth phase
9
of RTP, whereby team members might diagnose issues in team functioning and devise solutions as
10
a result (i.e., team self-correction). First, problem solving involves identifying why a team has been
11
unsuccessful (e.g., poor coordination) and how they can perform better (e.g., carrying out certain
12
drills in practice focused on improving coordination). Second, innovation involves enacting novel
13
approaches in order to enhance team task execution (e.g., altering game strategies, trying players
14
out at new positions). Third, backing up involves teammates helping one another enhance their
15
performance (e.g., showing a teammate how to perform a certain manoeuvre, redistributing the task
16
responsibilities of players who are feeling overworked). The fourth dimension, intrateam coaching,
17
is similar to backing up behaviours but focuses on the verbal feedback that teammates provide to
18
each other following task performance (e.g., an experienced veteran player providing helpful advice
19
to a more inexperienced player, an injured player observing team competitions and making
20
suggestions to teammates on how they can be successful). The main distinction between these latter
21
two dimensions is that intrateam coaching involves verbal assistance between teammates whereas
22
backing up entails non-verbal assistance.
23 24
Rather than directly focused on team performance, the fifth aspect of teamwork—known as the management of team maintenance (MTM)—is focused on behaviors associated with keeping the
MULTIDIMENSIONAL ASSESSMENT OF TEAMWORK IN SPORT
6
1
team together, and ensuring that personal and/or interpersonal non-performance-related issues do
2
not preclude a team from functioning effectively. Two dimensions comprise this aspect of
3
teamwork. The first dimension, psychological support, involves teammates providing assistance to
4
one another in the form of emotional (e.g., sharing and listening to each other about events going on
5
in one’s life outside of sport), esteem (e.g., helping teammates feel confident about themselves),
6
informational (e.g., teammates providing advice to one another), and/or tangible (e.g., providing
7
concrete, instrumental assistance such as rides to and from practice) support. The second dimension,
8
integrative conflict management, involves managing interpersonal conflicts between team members
9
if/when they arise (e.g., resolving disagreements between teammates).
10
As research on teamwork in sport is still in its infancy (cf. Carron et al., 2012; McEwan &
11
Beauchamp, 2014), it is perhaps unsurprising that there have not yet been any attempts to measure
12
this multidimensional construct. The availability of a psychometrically sound measure of teamwork
13
in sport is critical in order for this potentially important new line of research within sport
14
psychology to progress. Moreover, such an instrument has the potential to be useful to coaches and
15
sport psychology consultants. Specifically, coaches/consultants could have athletes complete the
16
teamwork measure, determine areas of strength and weakness with regard to teamwork, and target
17
those components that require improvement. With these potential uses in mind, the purpose of this
18
research was to develop a questionnaire designed to assess teamwork in sport and examine various
19
aspects of validity associated with measures derived from this instrument. This research was guided
20
by Messick’s (1995) unified view of validity, namely by the process of construct validation, which
21
comprises the collection of evidence and rationale that support the trustworthiness of data derived
22
from a focal measure (Messick, 1995). From this perspective, construct validation is considered an
23
ongoing process that occurs over the course of several studies, with six aspects of validity being
24
examined throughout (Hubley & Zumbo, 1996; Messick, 1995). These aspects include content (the
MULTIDIMENSIONAL ASSESSMENT OF TEAMWORK IN SPORT
7
1
relevance, representativeness, and technical quality of a questionnaire and its items), substantive
2
(respondents’ interpretations and judgments of a questionnaire and its items), structural (the extent
3
to which the model-data structure of measures from a questionnaire aligns with the conceptual
4
framework underpinning that construct), generalizability (the extent to which score properties and
5
interpretations generalize across populations, contexts, and tasks), external (convergence with
6
and/or discrimination between a focal construct and other variables), and consequential (the
7
implications that may result from using a test, such as providing a ‘basis for action’) validity
8
(Messick, 1995). Hence, validity needs to be considered at the outset of a research program as
9
opposed to being viewed as a ‘one-off’ procedure in a single study of test development (Anastasi,
10 11
1950; Hubley & Zumbo, 1996; Smith, 2005). The first phase of this construct validation process involves fully articulating a conceptual
12
framework and definition of the construct being studied (Clark & Watson, 1995; Cronbach &
13
Meehl, 1955; Smith, 2005). In the context of the current research, this was carried out via the
14
theoretical and integrative review by McEwan and Beauchamp (2014). While this framework
15
appears to provide a comprehensive and viable basis for examining teamwork, its utility within the
16
context of sport has yet to be tested. The next phase involves developing a psychometrically sound
17
questionnaire that accurately measures the focal variable, namely teamwork in sport. This phase
18
represented the overall focus of the research presented in this paper, which took place over two
19
studies. In Study 1, we sought to develop/collect a comprehensive pool of items, which was
20
informed by McEwan and Beauchamp’s (2014) framework and definition of teamwork. Once a
21
preliminary questionnaire was created, we sought to examine the content and substantive aspects of
22
validity by consulting with, and obtaining feedback from, the target population (i.e., team sport
23
coaches and athletes) as well as experts within the specific field of study (i.e., those with a PhD
24
focused on Sport Psychology; cf. Messick, 1995). The objective of Study 2 was to obtain evidence
MULTIDIMENSIONAL ASSESSMENT OF TEAMWORK IN SPORT
8
1
for the structural aspect of validity by examining data from a heterogenous sample of team sport
2
athletes who completed the questionnaire (cf. Clark & Watson, 1995; Flora & Flake, 2017;
3
Messick, 1995). Developing this questionnaire and obtaining support for the content, substantive,
4
and structural aspects of validity will enable researchers to examine the external, generalizability,
5
and consequential aspects of validity related to this instrument in future studies.
6
Study 1
7
The initial study of this research consisted of three parts. First, following procedures
8
outlined by Clark and Watson (1995) and DeVellis (2012), part A of the study involved creating an
9
initial pool of potential items that measured the 14 dimensions of teamwork. To do so, we referred
10
back to the conceptual framework and definition of teamwork (cf. McEwan & Beauchamp, 2014) in
11
order to create a pool of items that reflected those teamwork behaviors. In addition, we reviewed the
12
literature for previous attempts to measure teamwork in some way (in both sport settings and other
13
team contexts). A total of 566 items were identified as a result of this procedure (mean = 40 items
14
per dimension, range = 8 [systems monitoring] to 112 [conflict management]). Following
15
recommendations by Clark and Watson (1995), a rigorous and iterative process was then
16
undertaken to modify, amalgamate, or delete items where necessary. Specifically, the first and last
17
author examined items multiple times to ensure that: (a) the wording of the items was simple and
18
straightforward; (b) overly colloquial expressions were not used; (c) ‘double-barrelled’ items that
19
measure more than one component of a dimension (e.g., “we resolve conflict respectfully and
20
efficiently”) were avoided; and (d) items accurately reflected a teamwork behaviour (e.g., items
21
assessing members’ knowledge or affect with regard to their team were excluded or modified). This
22
process ultimately resulted in a preliminary 74-item measure of teamwork that we titled the
23
Multidimensional Assessment of Teamwork in Sport (MATS). A variety of response scales were then
24
considered including a dichotomous scale (i.e., where participants choose one of two response
MULTIDIMENSIONAL ASSESSMENT OF TEAMWORK IN SPORT
9
1
options), an ipsative scale (i.e., forced-choice scale that does not include a middle option), or a
2
Likert-type scale with various ranges (e.g., five-point, seven-point) and response formats (e.g., level
3
of agreement, frequency; Clark & Watson, 1995; DeVellis, 2012). Ultimately, a five-point, Likert-
4
type response scale (1 = strongly disagree; 2 = disagree; 3 = neither agree nor disagree; 4 = agree;
5
and 5 = strongly disagree) was deemed to be most appropriate in assessing the initial set of items
6
and was, therefore, selected in this preliminary measure.
7
A sample of 25 team sport athletes and five coaches (23 males, 7 females) then participated
8
in part B of the study. These participants competed in a variety of team sports, including volleyball
9
(n = 15), curling (n = 4), soccer (n = 3), rowing (n = 2), Canadian football (n = 1), baseball (n = 1),
10
ice-hockey (n =1), and multiple sports (n = 3). The participants competed at a range of levels,
11
including university (n = 18), recreational (n = 5), national (n = 4), and rep/select teams competing
12
against other teams within their district/region (n = 3). In this part of the study, participants
13
completed, and provided feedback on, the preliminary questionnaire in a focus group format
14
through use of a ‘retrospective think-aloud’ protocol (Willis, 2005). This was done in order to better
15
understand how members of the target population interpret and respond to items on the
16
questionnaire (i.e., examine response processes). Specifically, consulting with a sample of the target
17
population in this way helps to establish evidence of the content (i.e., the relevance,
18
representativeness, and technical quality of the items) and substantive (i.e., understanding how
19
respondents interpret and make sense of the items) aspects of validity (Messick, 1995).
20
After providing consent, participants were asked to complete (individually) a subset of items
21
from the preliminary MATS (approximately 18-19 items per session). The reason for having
22
participants only complete a subset of items (as opposed to the entire MATS) was to reduce
23
participant burden and ensure quality within their responses and the subsequent group discussion.
24
Participants were able to participate in multiple focus group sessions (i.e., provide feedback on
MULTIDIMENSIONAL ASSESSMENT OF TEAMWORK IN SPORT
10
1
separate sections of the questionnaire); on average, participants took part in two study sessions. A
2
total of 21 focus group sessions were ultimately carried out, each of which lasted approximately 45-
3
60 minutes and consisted of 2-5 participants. While going through the questionnaire individually,
4
participants were encouraged to make notes of any aspect of the instrument that they felt lacked
5
clarity or could otherwise be improved.
6
Once all participants completed the questionnaire, the researcher facilitated the group
7
discussion about the MATS. Participants were encouraged to provide feedback on the questionnaire
8
at any point throughout the discussion, however minor or substantial (i.e., the “think-aloud”
9
component of the discussion). To begin the protocol, the researcher asked participants for their
10
feedback on the questionnaire’s five-point scale. Specifically, participants were asked if they felt
11
that they could respond to each item accurately with this scale, or if an amended scale—such as an
12
ipsative scale, dichotomous scale, or other Likert-type scale with fewer or additional response
13
choices—would be more useful. The researcher then went through each item on the questionnaire
14
with the participants and asked for their feedback with regard to an item's relevance to the context
15
of team sport, representativeness of the construct of teamwork, wording clarity, and potential item
16
redundancy. Specific probes for each item included the following: (a) “what, in your words, do you
17
feel is being asked in this item?”; (b) “did the answer choices include your response?”; (c) “did you
18
understand how to answer this question?”; and (d) “how would you modify this item?”. At the end
19
of each dimension of the questionnaire, participants were asked “was there anything in this section
20
that was left out that you feel is important with regard to [name of teamwork dimension]?”. Once all
21
sections were discussed, participants were given a final opportunity to provide any open-ended
22
feedback that they felt was relevant to the questionnaire. After going through each item within the
23
respective subsection of the questionnaire, participants were thanked for their participation and
24
provided with a $10 (CAD) honorarium. As an ongoing and iterative process, changes were made to
MULTIDIMENSIONAL ASSESSMENT OF TEAMWORK IN SPORT
11
1
the MATS throughout the course of this part of the instrument development (i.e., between focus
2
group sessions), until no further changes were identified (each subset of items was reviewed a
3
minimum of five times).
4
Finally, in part C of the study, a sample of eight experts from the field of sport psychology
5
(i.e., with a PhD focused on sport psychology or closely related subject) reviewed, and also
6
provided feedback on, the questionnaire. The purpose of this expert review was to further maximize
7
the content and substantive aspects of validity associated with measures derived from the
8
questionnaire (cf. Clark & Watson, 1995; DeVellis, 2012). After agreeing to take part in the study,
9
participants were sent a Word document version of the MATS. They were first given the definition
10
of teamwork and each of its dimensions (taken from McEwan & Beauchamp, 2014). Participants
11
then rated—on a seven-point scale from ‘-3: Not At All’ to ‘0: Uncertain’ to ‘3 Very Much’—each
12
item's relevance to and representativeness of the definition of teamwork, clarity of wording, and
13
item redundancy. There was also space for participants to provide comments on each item as well as
14
any general feedback about the questionnaire. Any items whereby the expert reviewers did not
15
provide a +3 rating for relevance, representativeness, and clarity, or -3 for redundancy were flagged
16
for potential revision or deletion. As with part B of this study, changes to the MATS were made
17
with each iteration of feedback provided by the experts. Hence, we continued to recruit experts who
18
would complete an updated version (i.e., based on feedback provided by previous participants) of
19
the MATS. This process concluded once no further substantive changes were identified.
20
Several modifications to the questionnaire were made as a result of parts B and C of this
21
study. Some of these changes were minor (e.g., improving the look/presentation of the
22
questionnaire), while others were more substantive. For one, the response scale of the questionnaire
23
was changed to a 7-point Likert-type scale (1 = completely disagree, 2 = mostly disagree, 3 =
24
slightly disagree, 4 = neither agree nor disagree, 5 = slightly agree, 6 = mostly agree, 7 = completely
MULTIDIMENSIONAL ASSESSMENT OF TEAMWORK IN SPORT
12
1
agree), as the majority of participants indicated that this expanded scale allowed them to provide
2
responses that more accurately reflected their thinking and appraisals compared to the original 5-
3
point scale. In particular, several participants in the early focus groups of part B of the study
4
remarked that having more than one response option between the neutral and completely agree
5
options would allow them to distinguish between items that were just slightly above the ‘neither
6
agree nor disagree’ mark with those that were more highly above this response but not quite
7
reaching the ‘completely agree’ mark. After adjusting this scale, all participants in subsequent focus
8
groups felt that this scaling was appropriate—no participants in either part B or C of the study
9
indicated that an alternate scale (including the original five-point scale) would be more useful
10 11
compared to this seven-point scale. Second, preambles that described the dimension at the beginning of each section of the
12
questionnaire were expanded and included examples to help foster understanding of each
13
dimension. Third, two new items were created and 32 existing items were deleted, modified, or
14
combined in some way; these amendments were made in order to avoid item redundancy and
15
eliminate items that were (a) unsuitable/inappropriate to sports or (b) included the use of potentially
16
confusing or colloquial terms. For example, an item “Teammates’ actions are executed in harmony
17
with each other” from the original version of the MATS was deleted as several participants
18
remarked that (a) this item was subsumed within the other items of the coordination section, and (b)
19
the phrase ‘in harmony’ was overly colloquial. These modifications ultimately resulted in a revised
20
70-item version of the MATS (consisting of 14 teamwork dimensions and four to seven items per
21
dimension), with support for the content and substantive aspects of validity related to this
22
instrument. The MATS displays a Grade 7-8 (aged 13+ years) reading ability score (Flesch-Kincaid
23
Grade Level = 7.7, Flesch Reading Ease score = 55.1; cf. Flesch, 1948; Kincaid, Fishburne, Rogers,
24
& Chissom, 1975).
MULTIDIMENSIONAL ASSESSMENT OF TEAMWORK IN SPORT 1 2
13
Study 2 The next stage in the construct validation process involved examining the structural aspect
3
of validity of data derived from the MATS. Specifically, we sought to examine the extent to which
4
the items from the MATS aligned with/loaded onto their respective teamwork dimension, as well as
5
whether the factor structure of those dimensions coincided with the conceptual framework of
6
teamwork in sport (cf. McEwan & Beauchamp, 2014). We also sought to examine the reliability of
7
each of the 14 subscales of the MATS. In order to accomplish these objectives, a heterogeneous
8
sample (in terms of sport type, age, sex, and competitive level) of teams and athletes completed the
9
MATS. Five multilevel, multidimensional confirmatory factor analytic models were then carried out
10
in order to examine the structural properties of data derived from the MATS. The multilevel
11
component of the analyses stems from the fact that the athletes were clustered/nested within the
12
teams from which they were sampled.
13
Methods
14
Following institutional ethics approval, a heterogeneous sample (cf. Clark & Watson, 1995)
15
of 607 athletes (65.3% males; mean age = 17.7 years, range = 13-73) from 48 teams were recruited
16
to participate in the study. These teams competed in a range of sports (both interdependent and
17
coactive) including ice-hockey (n = 8), basketball (n = 7), soccer (n = 6); baseball (n = 5), curling (n
18
= 5), water polo (n = 4), rugby (n = 3), volleyball (n = 2), track and field (n = 2), lacrosse (n = 1),
19
cycling (n = 1), field hockey (n = 1), crew rowing (n = 1), swimming (n = 1), and synchronized
20
swimming (n = 1). The majority (n = 36) of these teams were rep/select teams competing against
21
other teams in their region/district; the competitive levels of the remaining teams included seven
22
university teams, two national teams, two provincial teams, and one professional team. The
23
researcher met with the teams before or after practices at the teams’ practice facilities (between
MULTIDIMENSIONAL ASSESSMENT OF TEAMWORK IN SPORT 1
January 2016 – June 2017) where participants (i.e., athletes within those teams) were asked to
2
complete a demographic form as well as the updated (i.e., based on study 1) 70-item MATS.
3
Data Analysis
14
4
Drawing directly from the a priori conceptual framework by McEwan and Beauchamp
5
(2014), five separate multilevel confirmatory factor analyses (MLCFA) of the data from the 70-item
6
MATS were conducted corresponding to each of the five aspects of teamwork—preparation,
7
execution, evaluation, adjustments, and management of team maintenance (see below for model
8
specification and rationale). Factors and indicators were standardized separately at the athlete and
9
team levels. Within each model, intraclass correlations (ICCs) were calculated to estimate how
10
much variance in each item was observed at the group level. If item-level ICCs are greater than .20,
11
then the clustering of the data should not be ignored (Muthén & Satorra, 1995). We also computed
12
both within-team (i.e., level 1) and between-team (i.e., level 2) ordinal composite reliability (CR)
13
values to examine the internal consistency of each of the 14 subscales of the MATS (Bollen, 1989;
14
Geldhof, Preacher, & Zyphur, 2014; Zumbo, Gadermann, & Zeisser, 2007). Computing these within
15
and between ordinal CR coefficients accounts for the clustering of the data (i.e., athletes within
16
teams) and is, therefore, a more accurate estimator of reliability compared to other estimators, such
17
as Cronbach’s alpha which assumes a single-level factor structure (Geldhof et al., 2014).1 An
18
acceptable indication of reliability is evident if CR values exceed .70 (DeVellis, 2012; Hair, Black,
19
Babin, Anderson, & Tatham, 2006). Finally, we computed Average Variance Extracted (AVE)
20
scores, which measure the convergence among the items in each model. It is recommended that the
21
AVE values for each factor not exceed the squared correlations between that factor and any other
22
variable (Fornell & Larcker, 1981).
23
In accordance with guidelines for conducting MLCFA (Brown, 2006; Muthén, & Muthén,
24
2017), the individual items (i.e., observed variables or indicators) were first specified to load onto
MULTIDIMENSIONAL ASSESSMENT OF TEAMWORK IN SPORT
15
1
their theoretically-aligned teamwork dimension (i.e., latent factors). Within the models examining
2
the preparation, execution, and adjustments phases of teamwork, those lower-order latent factors
3
were then specified to load onto a second-order latent factor designed to reflect an omnibus/global
4
representation of each of those phases. Specifically, preparation was modelled (see Figure 1) to be
5
comprised of three lower-order latent variables measuring mission analysis (5 items), goal
6
specification (6 items), and planning (7 items). Execution was also modelled (see Figure 2) to
7
consist of three lower-order latent variables measuring coordination (4 items), cooperation (5
8
items), and communication (5 items). Adjustments was modelled to be comprised of problem
9
solving (5 items), innovation (4 items), intrateam coaching (4 items), and backing up (5 items).
10
Since it is not mathematically possible to test a higher-order structure from two lower-order factors
11
(a minimum of three lower-order factors is required, otherwise the model would be under-
12
identified; Brown, 2006), we specified a first-order measurement model for the evaluation phase
13
(see Figure 3) that included two correlated latent factors related to performance monitoring (6
14
items) and systems monitoring (4 items). Similarly, we specified a first-order measurement model
15
for the management of team maintenance (see Figure 5) that included two correlated latent factors
16
related to conflict management (5 items) and psychological support (5 items).2
17
The factor analyses were carried out in Mplus Version 8 (Muthén & Muthén, 2017). A
18
variety of estimation methods were considered in conducting these analyses, including Weighted
19
Least Squares Means and Variance adjusted (WLSMV), Maximum Likelihood (ML), and Bayesian
20
estimation. While each method has strengths and limitations, we ultimately chose to use the
21
WLSMV method of estimation, as this method performs best when conducting CFA with ordinal
22
data and is also suitable when dealing with a large number of participants (e.g., over 200) and
23
multilevel data structures (Brown, 2006; Muthén, Muthén, & Asparouhov, 2015). Following
24
recommendations provided by Brown (2006) and Jackson and Gillaspy (2009), global model fit was
MULTIDIMENSIONAL ASSESSMENT OF TEAMWORK IN SPORT
16
1
evaluated with a variety of fit indices. First, we considered the chi-square test to examine absolute
2
fit. However, several authors have suggested that obtaining a non-significant 𝜒2 value is highly
3
unrealistic (Barrett, 2007; Hu & Bentler, 1995) and, although this value is commonly reported,
4
supplemental indices should be relied on more heavily to examine fit (Brown, 2006). Thus, we
5
computed two indices of absolute fit—Root Mean Square Error of Approximation (RMSEA) and
6
Standardized Root Mean Square Residual (SRMR)—and two indices of comparative fit—
7
Comparative Fit Index (CFI) and Tucker-Lewis Index (TLI). Adequate fit between the hypothesized
8
model and the data occurs when CFI and TLI values are close to or greater than .90 (ideally >.95),
9
SRMR values are close to or below .08, and RMSEA values are less than .08 (ideally