The theoretical framework underpinning the development of the GMCA is ecological dynamics. The theory suggests that movement responses are a consequence of the dynamic interactions between individual, task and environmental constraints (Newell et al., 1991) that are scaled based on the affordances (or opportunities to act) and action capabilities of the individual (Dicks et al., 2009; Button et al., 2020). Thus, a wide variety of movement experiences and exploration in movement contribute to greater movement competence (Seifert et al., 2013; Rudd et al., 2020a). As a child ages, the compounding effect of movement experiences accumulates and progressively, the child becomes more competent in movement (Clark & Metcalfe, 2002). They become more sensitive to the affordances and are thus better able to adapt movement solutions based on their action-capabilities. In other words, movement competence is age-related and dependent on engagement in varied movement experiences.

Movement competence is typically assessed with movement assessment batteries (MABs) evaluating performance across three constructs; stability, locomotion and object-control (Gallahue & Ozmun, 2006; Barnett et al., 2016). The stability construct includes movements that emphasize balance in static and dynamic situations which require coordination of the trunk and axial limb movements. Specific examples can include balancing on one foot statically or dynamic forms such as walking along a line. The locomotion construct includes movements that allow body transportation from one point to another, such as sliding, leaping and galloping. The object-control construct describes manipulative, interceptive and prehension movement types such as catching, throwing or striking with and without additional equipment such as gloves or rackets and tools such as hammers. It typically involves the transmission of force to manipulate, move or receive an object.

MABs would often measure movement competence by evaluating combinations of or all of the three constructs. The origin of MABs have a clinical origin and they were purposefully designed to identify children with poor competence who would potentially require clinical intervention. These batteries however remain commonly used in the general population as a test of general movement competence in research and as an indication of performance in general programmes such as physical education and sports (Cools et al., 2009; Giblin et al., 2014; French et al., 2018). Importantly, the original intent for MABs to identify deviations from typically developing individuals lead to the design of simplified and isolated assessment tasks. This undoubtedly creates a ceiling effect when used with general populations. A ceiling effect is observed when a large portion of the sample attains high scores in a test resulting in a skewness in statistical analyses. It indicates that a test is not challenging enough for the specific cohort. Hence, in the context of MABs, the presence of ceiling effects in some MABs indicate a lesser sensitivity at identifying individuals who are performing at the higher end of the movement spectrum (French et al., 2018).

These MABs tend to involve the observation of discrete tasks that are often sport-related (such as balancing, running, jumping, throwing and striking) in tightly regulated settings (Cools et al., 2009; Ng & Button, 2018). The tasks common to these MABs are performed in isolation and often stripped of context to promote reliability (e.g., performing a static dribble as compared to dribbling in an actual physical activity and/or sport-orientated setting) (Pinder et al., 2011). The decontextualized design of these activities results in assessment tasks that are over-simplistic and prone to bias from cultural differences between countries (Cools et al., 2009). This can result in unfair bias against children that may have not had the same opportunities to participate in sports and certain physical activities that involve the use of the assessed movement skill (Rudd et al., 2020b; Smits-Engelman et al., 2022). Essentially, these MABs are being used for purposes that they were not originally designed for.

Process and product-based assessments primarily differ in assessment forms. Process-based assessments utilise observational criteria to determine the quality of a particular movement (e.g., running form) that are criterion-referenced. Product-based assessments measure quantitative, outcome variables (e.g., running duration) and are often referenced with normative samples. Many MABs adopt criterion-based reference in evaluating movement competence by comparing the performances of individuals based on the ‘correctness’ of the technique employed in specific tasks (e.g., how to throw or catch a ball). In order to reliably reproduce the assessment settings, MABs are composed of static and isolated tasks that downplay or remove typical affordances that are present in a naturalistic setting (e.g., throwing a pass to an unmarked team-mate in a game situation). The technique-focused outcome typically requires a specific set of desired movement solutions (Kane, 2013). Hence, individuals who may be successful in achieving outcome goals through alternative, innovative forms of movement are penalised. Arguably, the assessment tasks do not allow children to demonstrate their ability to react and respond to changing constraints that are typical in an authentic setting (Newell et al., 1991; Cools et al., 2009).

In recent times, some contemporary MABs have begun adopting a more purposeful approach in the design of assessment tasks as a consequence of growing concerns about their ability to distinguish between more or less competent individuals. These validated MABs include more dynamic tasks such as obstacle courses and game-based formats to evaluate movement in more representative contexts (e.g., Tylor et al., 2018; Flôres et al., 2021; Morley et al., 2021).

In our opinion, the game-based virtual environment of AVGs has considerable potential in offering a dynamic environment within controlled settings to potentially evaluate movement competence. Physical interaction with AVGs allows individuals to demonstrate their ability to accurately scale movement responses based on the changing affordances presented in modifiable tasks. Notably, the use of motion-sensing video game technology would ease some logistical constraints of space and equipment (e.g., typical requirements of MABs are large, unobstructed rooms such as school gyms) as well as the need for specialised training and assessors required by many MABs (Cools et al., 2009; Giblin et al., 2014). As gaming and e-sports continue to grow in popularity around the world (Franks et al., 2022), an active gaming platform may offer an inclusive means to develop general movement competency assessments that are less sensitive to cultural differences.

A new movement assessment instrument, the General Movement Competence Assessment (GMCA) was developed using the technology of the Microsoft Kinect with a series of AVGs created to assess various attributes of movement. In a previous study, Ng and others (2020) proposed the inclusion of a new movement construct (i.e., dexterity) in addition to the three commonly accepted constructs of movement competence (i.e., stability, locomotion and object-control) to provide a more encompassing description of movement competence suitable for the general population of healthy children. Before this work, dexterity had not been previously identified as an interdependent construct in the model of movement competence. Although the exploratory factor analysis showed that over 70% of the variance in performance on the GMCA games was explained by a four-construct model, a cautious approach was taken in generalising results with a limited sample size of 83 children (aged 8–10 years). This prompted further validation work with a focus on increasing sample size across wider age groups.

Robust assessments of children’s movement competency are of fundamental importance given the potential impact on health given declining levels of physical activity globally. Unfortunately, many existing movement assessment tools that were designed to classify children with very low competency over-emphasise technique instead of adaptation to constraints as well as creative movement alternatives to overcome movement problems. We propose that AVGs offer a promising platform to develop a new movement competency assessment tool that is suitable for use in the general population of children. The primary aim of the present study was to assess the internal structure of the GMCA in a sample of children 7–12 years old through confirmatory factor analysis of the proposed four-construct model consisting of dexterity, stability, locomotion and object-control (Ng et al., 2020).

The study sample consisted of 259 typically developing children, ranging from 7–12 years of age (M age = 9.97 years, SD = 1.61) from a variety of ethnicities. There were 138 boys (M age = 10.12 years, SD = 1.57) and 121 girls (M age = 9.80 years, SD = 1.64). All children were recruited through a convenience sampling method from a public primary school in Dunedin, New Zealand. As the GMCA had not been programmed to accommodate individual differences for children with disabilities, children with any physical impairment or disabilities (e.g., visual, hearing impaired, children with cerebral palsy, etc.) were excluded from the study. Written informed consent was attained from both parents and child participants and approval for the study (17/071) was obtained from the Human Ethics Committee of the participating institution.

Simulation work suggest that sample sizes for confirmatory factor analysis with maximum likelihood varies. Jak et al. (2014) classify sample sizes of n ≤ 250 as “small”. However, power estimates conducted for this study suggested sample sizes of between 200–300 for the test of closeness of fit (PCLOSE) for values of 0.769 and 0.928, respectively (MacCallum et al., 1996). Additionally, considering the degrees of freedom (df) for the extracted four-construct model of Ng et al. (2020) the sample size (n = 259) for the study was suitable based on the recommendations of MacCallum and others (1996) and fits the recommendations of other simulation work (Marsh et al., 1988; Kim, 2005; Wolf et al., 2013).

The GMCA is a custom-written application (programmed in C++) that utilises the open-sourced Kinect for Windows Software Development Kit 2.0 to work with the Microsoft Kinect 2.0 system (Microsoft, Redmon, WA). The Kinect system consists of an infrared emitter, colour video and depth sensor. It tracks movements from the reflection of emitted infrared rays. The video and depth sensor captures three-dimensional movements and automatically locates and detects 25 joint centres of the human body.

The GMCA consists of five custom-programmed active video ‘games’. The five games are called Balance, Precision, Control, Swiftness, and Interception, with each game increasing in task difficulty as one meets the movement demands required by each stage (see Supplementary Table S1). The order of the various GMCA games were presented as follows ¹ : 1) Precision_unimanual, 2) Balance, 3) Precision_symmetrical, 4) Control, 5) Swiftness, 6) Precision_asymmetrical and 7) Interception. Once started the games would run automatically without the need for external administrators. Performance scores for each task were calculated based on how well children completed (or not) the progressively difficult levels of each game. Thus, if an individual did not do well in earlier stages of the game it was expected that they would not be as accomplished in the later stages.

The spatiotemporal demands of each task were pre-programmed into the software and these were modified to increase once a participant reached a specified threshold of achievement. Thus, the relative demands placed on the individual in terms of difficulty and complexity increased at each stage (e.g., the game Precision started with symmetrical pathways to asymmetrical pathways resulting in an increased difficulty and complexity of the task). Hence, individuals had to adapt their movement responses appropriately and it was expected that individuals at the higher end of the movement spectrum (Ng & Button, 2018) would have better performances even in the more complex stages of the games.

The GMCA assesses movement competence based on the measured variables from each game (for details see: Ng et al., 2020). For example, in the Balance game, a static balance pose is considered successful when the prescribed hands and feet positions were held for 3 s. The measured variable for Balance was the total number of successful poses held for each of the three stages. Importantly, the more competent individual would be better able to adapt the emergent postural control strategies and body configurations to suit the demands of the task. Recorded variables for the game, Precision, included the total time taken to move an object around set courses for three levels of difficulty. Measured variables from Control included the total number of balls used for the game and the total number of balloons popped. Control required individuals to first, juggle a virtual ball on the screen, then manoeuvre the ball to ‘pop’ a balloon that appeared at random locations on the screen. Hence, a proficient player would be able to control the juggle of a ball and use that same ball to pop the balloons that appeared throughout the game. On the other hand, a less able player would use more balls since a ball was ‘replaced’ when it was ‘lost’ (i.e., juggled out of control). Variables from the Swiftness game included the total amount of time taken to move between set places in the play area for each of the two levels of difficulty. Finally, the game, Interception, required participants to primarily, ‘save’ spaceships by hovering a hand over them as they appeared at randomised locations on the screen at the same time. A secondary task required participants to ‘intercept’ stray asteroids by touching them using the other free hand. The asteroids were programmed to take random flight paths and fly at random speeds throughout the game. The measured variables for Interception were the number of spaceships ‘saved’ and the number of spaceships ‘lost’ (or destroyed by the stray asteroids).

The standing height and weight of participants were measured with a portable stadiometer and a digital scale (UC-321, A&D Company Limited). The GMCA games were displayed on a Sony KDL-40EX400 40-inch 1,080 pixels HDTV. The TV was set upon a standing console 0.8 m from the ground. The Kinect Sensor was placed directly in front of the TV facing the game area. All trials were recorded to serve as a reference in the event of discrepancies.

In the present study, the game area measured 2.05 by 2.55 m. The distance between the Kinect sensor and the game area measured 2 m away from the front edge of the Kinect sensor to the front boundary of the play area. Play area boundaries were marked out with high-contrast coloured tape on the ground.

Participants’ anthropometric measures were recorded 1 week before data collection. Age was calculated by subtracting the date of the testing date from the birth date of each child. The start and end of the GMCA application were controlled by an experimenter operating the computer at each testing station. The GMCA does not require any specialised training, nor does it require the presence of a tester with specialised knowledge of motor performance for the GMCA was programmed to run automatically. All participants engaged in the GMCA trial once immediately after their familiarisation trial. Including familiarisation, the entire GMCA test duration ranged from approximately 10–18 min per individual. More competent individuals completed GMCA trials faster ² .

A proposed four-construct model of the GMCA was validated for the study sample using confirmatory factor analysis. To determine the suitability of the data for factorial analysis, data were screened using Kaiser-Meyer-Olkin (KMO) and Bartlett’s test of sphericity. Data is suitable for factor analysis when the KMO value is more than 0.6 (Kaiser & Rice, 1974) and when Bartlett’s test is significant (p < 0.05). When assumptions were met, confirmatory factor analysis was conducted using the maximum likelihood method of estimation in AMOS 24.

The fit of the tested model was interpreted from various fit indices. On top of the chi-square (χ2) statistic and df, other goodness-of-fit indices were used to determine model fit. These were the χ2 divided by the df (χ2/df), Comparative Fit Index (CFI; Bentler, 1990), Lewis-Tucker Index (TLI; Tucker & Lewis, 1973), root mean square of approximation (RMSEA; Steiger & Lind, 1980; Browne & Cudeck, 1993) with confidence intervals (CI) and probability of the test of close fit (PCLOSE; Hu & Bentler, 1999).

The χ ² statistic measures the overall fit of the model with a higher probability (p > 0.05) indicating a closer fit between the tested model and the perfect fit (Bollen, 1989). Instances of good fitting models being rejected with the test of exact fit due to the large χ ² statistic relative to the df have been highlighted in the literature (Jöreskog & Sörbom, 1993). Thus, other alternative indices of fit were used to address the limitations associated with the χ ² statistic.

Alternate fit indices (i.e., χ ²/df, CFI, TLI and RMSEA) are typically used as adjuncts to the χ ² statistic. χ²/df provides an indicator of fit with values of less than 2 being considered an adequate fit (Wheaton et al., 1977). The CFI is a revision of the Normed Fit Index (NFI; Bentler & Bonnet, 1980) that takes sample size into account since a limitation of NFI is the underestimation of fit in small samples (Byrne, 2013). The TLI yields values from 0.0 to 1.0. Values closer to 1.0 are indicative of a good fit. CFI and TLI values more than 0.9 were interpreted as “acceptable”, while values more than 0.95 were “good” (Hu & Bentler, 1999).

The next fit statistic, RMSEA, is postulated to be one of the most informative fit indexes as it considers the error of approximation through the provision of CIs (Browne & Cudeck, 1993). RMSEA values of less than 0.05 are indicative of a “good” fit; 0.05 to 0.08, “fair” and 0.08 to 0.10, “mediocre” (Browne & Cudeck, 1993). Non-etheless, Hu and Bentler (1999) propose that RMSEA values of up to 0.06 can still be considered a good fit. CI substantiates the RMSEA value by providing additional information regarding the precision of estimates (MacCallum et al., 1996). For example, if the lower bound of the RMSEA’s CI is above 0 and less than 0.05, then the probability of the χ ² statistic being less than 0.05 is expected (MacCallum et al., 1996). Additionally, if the upper bound of the CI is above 0.05, it would be an indication of a plausible good-fitting model.

PCLOSE is a test for the closeness of fit. Specifically, it tests the hypothesis that the RMSEA value is “good” for the sample population. The probability for the PCLOSE test should be p > 0.50 (Browne & Cudeck, 1993).

The four-construct model of the GMCA was validated in this population of children aged 7–12 years old. Modification indices generated by AMOS 24 were only considered if proposed modifications were theoretically grounded, else, modifications made would reflect minute changes in the model according to the nuances of the sample (Byrne, 2013). In addition, a second-order confirmatory factor analysis was conducted to examine if the four constructs loaded onto the higher-order variable of movement competence.

Preliminary examination confirmed there was no significant difference in age within the sample for boys and girls (p = 0.11). Table 1 shows the results of the internal consistency reliability analysis for each of the four constructs measured by the GMCA. For all constructs, omega coefficients were above the recommended 0.7 value.

Confirmatory factor analysis with maximum likelihood estimation was conducted to test the internal structure of the GMCA as extracted from a previous study (Ng et al., 2020). Assumptions testing indicated that the sample data was suitable for factorial analysis. The KMO value was 0.86 which indicated excellent suitability and Bartlett’s test was significant (p < 0.001).

The path diagrams (i.e., Figures 1–3) of confirmatory factor analysis comprise all 11 measured variables included in the analysis as well as the four specified constructs of the GMCA. Each construct consists of the measured GMCA game variables (also known as observed variables) and is influenced by a random measurement error, indicated by the associated error term (e.g., e1, e2, e3, etc.). Each observed variable regresses onto its respective construct. Finally, the constructs co-vary via the corresponding covariate arrows in path diagrams from the specified model.

The specified model (Model 1; see Figure 1) was based on the four-construct model extracted from the exploratory factor analysis of Ng and others (2020). The four constructs measured were balance, locomotion, object-control and dexterity. The variable, Balance Stage 1, was specified to double load onto the locomotion construct as dynamic balances are proposed to influence the performance of locomotion tasks (Bril & Brenière, 1993).

The initial confirmatory factor analysis for the specified four-construct model (Model 1; see Figure 4.2) found an adequate fit (χ ²(37) = 60.006; p = 0.010; χ ²/df = 1.622; CFI = 0.988; TLI = 0.983; RMSEA = 0.049, CI 0.024–0.071; PCLOSE = 0.500).

In model 1, the standardized regression weight of Balance Stage 1 loading onto the locomotion construct was −0.072. Notably, factor loadings (i.e., regression weight) of less than 0.4 are not considered valuable to the overall model fit (Sireci, 2007). Thus, the initial model (Model 1; see Figure 1) was re-specified with the removal of the double loading between Balance Stage 1 and the locomotion construct and confirmatory factor analysis was conducted again.

The second model (Model 2; see Figure 2) supports the data characteristics well based on fit indices (χ ²(38) = 60.588; p = 0.011; χ ²/df = 1.594; CFI = 0.989; TLI = 0.983; RMSEA = 0.048, CI 0.023–0.070; PCLOSE = 0.533) which accounted for 74.85% of the variance. Modification indices suggested several ways to enhance the model fit, however, the suggested changes were not theoretically justifiable.

With the first-order confirmatory factor analysis establishing a good fit of the specified four-construct model of the GMCA, a second-order confirmatory factor analysis was undertaken that specified for each construct to load onto the higher-order variable of movement competence. The four second-order latent variables of stability, dexterity, locomotion and object-control were specified to load directly into movement competence (see Figure 3).

Results indicated that an adequate fit was achieved (χ ²(40) = 67.376; p = 0.004; χ ²/df = 1.684; CFI = 0.986; TLI = 0.981; RMSEA = 0.052, CI 0.029–0.072; PCLOSE = 0.429) and accounted for 95.44% of the variance (approximately 20% more variance than the first order model accounted for). Table 2 provides a tabled comparison of the various fit indices for the three tested models.

To measure movement competence, process-based assessment approaches have been recommended (Ulrich, 2000; Stodden et al., 2009). Indeed, the GMCA utilises a product-based approach towards the measurement of movement competence. Process and product-based assessments primarily differ in assessment forms. Although the validity of both forms of assessments to measure movement skills have been raised before (Stodden et al., 2008; Logan et al., 2012), previous studies have suggested associations between the two (Roberton & Konczak, 2001; Miller et al., 2007; Mally et al., 2011), thus, highlighting that both forms have their merits and that results from both assessment forms are valid for the purposes that they were designed for.

From an ecological dynamics perspective, the implications on practice are that learning or assessment tasks in the movement context should in design and execution strive to ensure that the link between perception and action remains and is not left decoupled by design. When assessment tasks are decoupled or decontextualized, it limits the opportunity to provide an accurate description of movement competence. A simulated assessment environment should have elements of the performance environment to ensure representativeness (Chow et al., 2007; Dicks et al., 2009; Pinder et al., 2011). This allows individuals to demonstrate their capacity to adapt efficient movement forms that are self-organised based on inherent individual differences (Schöllhorn et al., 2002) in addition to the demands presented by dynamic movement situations which are found in activities of daily living physical activity and sport at all levels of participation. The effectiveness, efficiency and quality of movement can then be judged based on a contextualised movement problem that keeps the perception-action coupling intact.

Understanding the process of movement or assessing its quality is suggested to be an important feature in determining the efficiency of movement (Ulrich, 2000) and process-based MABs often assess children’s movement skills based upon a mature, expert-like form (Stodden et al., 2008). Notably, that approach fails to consider the influence of individual differences on movement responses (Vella et al., 2023). Importantly, variability of and within observed movements is inherently present due to the unique individual differences of every individual (Chow et al., 2007). In addition, there are no universally ideal or expert-like patterns of movement (Davids et al., 2013; Seifert et al., 2013). Hence, the GMCA was developed as a product-based assessment that is concerned with the movement outcome since the process of executing movement would be unique to each individual, based upon their action capabilities and interaction with task and environmental constraints.

One of the merits of this study is its utilisation of relatively low-cost, portable video game technology that can be operated without specialised training to help ease some of the constraints of current MABs such as the need for trained assessors (Cools et al., 2009). As the GMCA was written with an open-source application, there is potential to further programme customised games that could suit various population samples. There is potential for the GMCA to also be used as a supplementary teaching aid for the general population. There is also considerable promise for the GMCA with its AVG modality to be used in conjunction with intervention programmes for clinical populations (Camara Machado et al., 2017; Page et al., 2017). The large sample of children utilised and naturalistic settings (i.e., in school classes) were also strengths of the study. Notably, our results suggest stability competence has a critical influence on the other three movement constructs which signals future work to establish if the influence is variable across the age groups.

There are some limitations in this study. First, the model fit was specific to the study sample and some caution has to be heeded in generalising to other populations. Future studies should consider its validity in other populations. Assessing validity is an ongoing process of evaluating data that is first derived from a specific population. Hence, more than one source of evidence is necessary (Messick, 1989). Furthermore, as evident in past validation studies of MABs, a critical limitation has been raised concerning the incongruent results found when particular MABs were used in different populations from normative samples (Chow et al., 2006; van Waelvelde et al., 2008; Bardid et al., 2016). Hence, future studies should consider validating the GMCA in other populations to further supplement the validity evidence (Zumbo & Chan, 2014). Secondly, the efficacy of any assessment will be in its discriminative validity to detect changes over time however this was not yet established in this study. Therefore, to further validate the GMCA, future research should determine the GMCA’s sensitivity in tracking developmental changes. Lastly, it is also recommended that the relationship between dexterity and overall movement competence be explored further to inform the design of future movement assessments that are suitable for use amongst the general population. Future studies should explore the role of dexterity in distinguishing children across the spectrum of movement competence. This could be achieved through concurrent validation studies between the dexterous tasks of the GMCA and the other validated assessment batteries such as the Brunicks-Oseretsky Test of Motor Proficiency (Bruininks, 2005) and Movement ABC-2 (Henderson et al., 2007).