Research in Developmental Disabilities 32 (2011) 1361–1369

Contents lists available at ScienceDirect

Research in Developmental Disabilities

Structural validity of the Movement ABC-2 test: Factor structure comparisons across three age groups Joerg Schulz a, Sheila E. Henderson b, David A. Sugden c, Anna L. Barnett d,* a

Department of Psychology, University of Hertfordshire, UK School of Psychology and Human Development, Institute of Education, University of London, UK c School of Education, University of Leeds, UK d Department of Psychology, Oxford Brookes University, UK b

A R T I C L E I N F O

A B S T R A C T

Article history: Received 15 April 2010 Received in revised form 6 January 2011 Accepted 15 January 2011 Available online 16 February 2011

Background: The Movement ABC test is one of the most widely used assessments in the field of Developmental Coordination Disorder (DCD). Improvements to the 2nd edition of the test (M-ABC-2) include an extension of the age range and reduction in the number of age bands as well as revision of tasks. The total test score provides a measure of motor performance, which can be used to help make a diagnosis of DCD. M-ABC-2 also provides 3 sub-scales for Manual Dexterity, Aiming and Catching and Balance but the validity of these conceptually derived sub-scales has not previously been reported. Aim: To examine the factor structure of the M-ABC-2 test across the three age bands (AB): AB1 (3–6-year olds), AB2 (7–10-year olds) and AB3 (11–16-year olds). Method: Data from the 2007 standardisation sample (N = 1172) were used in this study. Confirmatory factor analyses (CFA) and structural equation modelling (LISREL 8.8) were employed to explore the relationship between the tasks within each of the 3 age bands. A model trimming approach was used to arrive at a well fitting model. Results: In AB1 a complex factor structure emerged providing evidence for an independent general factor, as well as specific factors representing the 3 test components. In AB2 a final model emerged with four correlated factors, an additional distinction being drawn between static and dynamic balance. In addition, a 2nd order general factor explained a considerable amount of variance in each primary factor. In AB3 CFA supported the 3-factor structure of the M-ABC-2, with only modest correlations between each factor. Conclusions: The confirmatory factor analyses undertaken in this study further validate the structural validity of the M-ABC-2 as it has developed over time. Although its tasks are largely associated with the three sub-components within each age band, there was also clear evidence for a change in the factor structure towards differentiation in motor abilities with age. ß 2011 Elsevier Ltd. All rights reserved.

Keywords: Movement ABC Construct validity Factor analysis Developmental coordination disorder (DCD)

1. Introduction A wide range of movement assessments is used in the field of developmental coordination disorder (DCD), with different kinds of tests used for different purposes. In a recent review of available instruments and their functions, Barnett (2008)

* Corresponding author at: Department of Psychology, Oxford Brookes University, Headington Campus, Gipsy Lane, Oxford OX3 0BP, UK. Tel.: +44 01865 483680; fax: +44 01865 483887. E-mail addresses: [email protected] (J. Schulz), [email protected] (D.A. Sugden), [email protected] (A.L. Barnett). 0891-4222/$ – see front matter ß 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.ridd.2011.01.032

1362

J. Schulz et al. / Research in Developmental Disabilities 32 (2011) 1361–1369

identified individually administered norm-referenced tests as one of the most commonly used in both clinical and research settings, the most popular being the Movement ABC (M-ABC, Henderson & Sugden, 1992; M-ABC-2, Henderson, Sugden, & Barnett, 2007, the Bruininks–Osteretsky Test of Motor Proficiency (BOT, Bruininks, 1978; BOT-2, Bruininks & Bruininks, 2005; McCarron Assessment of Neuromuscular Development (MAND), McCarron, 1982, 1997). Often these tests play an important role in making a diagnosis of DCD, as it is delineated by the American Psychiatric Association in DSM-IV (APA, 2000). Of the four criteria specified by the APA, the first, criterion A relates to the core motor difficulty and a diagnosis can only be considered when test scores indicate that the level of motor performance achieved is ‘‘substantially below’’ that expected for the child’s age. Given the importance of any test which might be used in this way, it is crucial to establish that it is a valid measure of the construct under consideration. Performance scores from the tests most frequently used in the field of DCD have been shown to be quite highly correlated (Larkin & Rose, 2005). For example, Kaplan, Wilson, Dewey, and Crawford (1998) found a correlation of 0.61 between the composite scores of the M-ABC and the BOT (short form, BOT-SF) in a sample of 379 children aged 8–17-year olds, and Croce, Horvat, and McCarthy (2001) reported correlations between these two tests ranging from 0.60 to 0.90 in a more restricted group between the age 5 and 12 years. In a group of 4–10-year-olds, Tan, Parker, and Larkin (2001) reported that rankings on the BOT-SF and the M-ABC had a correlation of 0.84 and M-ABC and MAND a correlation of 0.88. Not only do such substantial correlations between different measures of motor performance support their use as valid measures for clinical practice, they also suggest sufficient convergent validity for a common construct that we recognise as ‘motor ability’. However, just as with IQ and the concept of intelligence, the construct of ‘‘general motor ability’’ has been questioned. Almost 50 years ago for example, Henry (1958/61, 1968) held the view that motor abilities are specific to a particular task and that these are completely independent of one another. Schmidt and Lee (2005) outline a number of studies which support this view. In contrast, the factor analytic studies by Fleishman and Bartlett (1969) suggest a smaller number of motor abilities, with different tasks requiring some of the same underlying abilities for successful performance (particularly when the requirements of the two tasks are similar). However, this body of work is limited by its narrow focus on the performance of young adult males using seated manipulation tasks. Investigation of the structure of motor abilities in children is also limited. One of the most frequently quoted studies is that carried out over 30 years ago by Rarick, Dobbins, and Broadhead (1976) on 145 typically developing children aged 6–9 years. This work involved an examination of various aspects of motor control and coordination, similar to those included in current motor performance tests. The main factors to emerge from this study were labelled ‘Gross Limb-Eye Coordination’ (in which throwing tasks had high factor loadings), ‘Fine Visual Motor Coordination’ (including mainly fine manipulative tasks) and ‘Balance’. The study also included measures of fitness (strength, endurance, and flexibility) and body composition (body size and body fat), which loaded on other factors (see Keogh & Sugden, 1985 for a summary). More recently, in the context of test construction, Bruininks and Bruininks (2005) and Shih-Heng, Yi-Ching, Ching-Lin, Chien-Hui, and Sheng (2010) have identified similar factors. Given the limited data available on the structure of motor abilities in children of different ages, the construction of many motor performance tests has been largely driven by ‘‘common sense and clinical experience’’ (Henderson & Barnett, 1998, p. 455). Focussing on those most commonly used in the assessment of children with DCD we find that nearly all include a number of sub-components thought to represent the range of functionally relevant aspects of motor control and coordination underlying the activities of daily living that this group find most difficult. This sub-division of tasks allows for examination of a child’s profile across different performance areas (e.g., fine and gross motor items). Given that these performance profiles are often reported in the DCD literature and used to help plan intervention programmes, it is important to examine the relationships between specific tasks and the sub-components they are supposed to represent as well as the correlations among those sub-components. This investigation concerns the internal part of a construct validation process and has been referred to as the structural validity of a test (Messick, 1995; Wassermann & Bracken, 2003). Confirmatory factor analysis is regarded as the classic method for examining the structural validity of a test as specific hypotheses about the relationship between task performance (i.e. the manifest variables) and assumed underlying abilities (i.e. the latent variables) can be tested (Brown, 2006). To the extent that these hypotheses about the task design and the resulting test scores are empirically supported, strong evidence for the construct validity of the test in question is provided. The aim of the current work was to examine the structural validity of the M-ABC-2 test (Henderson et al., 2007) using the data collected during the standardisation of the test as our starting point. As with its predecessor, The M-ABC-2 test has three sub-components, now labelled ‘Manual Dexterity’ (MD), ‘Aiming and Catching’ (AC) and ‘Balance’ (BAL). The rationale for this division and the selection of tasks has been a continuous process beginning with the original publication of the Test in 1972, followed by the first revision in 1984 (Stott, Moyes, & Henderson, 1972, 1984) and the two editions of the renamed Movement ABC (Henderson & Sugden, 1992; Henderson et al., 2007). The three MD tasks focus on use of the hands to manipulate objects in various ways. This includes a measure of speed and accuracy in each hand separately, a timed bimanual task and an untimed drawing task. The AC component includes tasks requiring the projection of an object towards a target and the reception of an object in either one or two hands. The BAL component includes a static balance task and two dynamic balance tasks. One of the latter involves sustained, controlled movement (e.g. walking along a line) and one involves more explosive action (jumping or hopping). The age range of the second edition of the M-ABC-2 has been extended down to 3 years and up to 16 years and is organised across three age bands (AB, see Table 1). A similar set of eight tasks is included within each age band across the three sub-components MD, AC and BAL, with tasks increasing in difficulty with age.

J. Schulz et al. / Research in Developmental Disabilities 32 (2011) 1361–1369

1363

Table 1 Individual tasks and scores obtained for each age band in the M-ABC-2 test. Tasks

Age band 1, 3–6 years

Age band 2, 7–10 years

Age band 3, 11–16 years

Raw score

Placing pegs Preferred handa Non-preferred hand Threading lace

MD3

Drawing trail 1

Drawing trail 2

Turning pegs Preferred handa Non-preferred hand Triangle with nuts and bolts Drawing trail 3

Completion time in seconds

MD2

Posting coins Preferred handa Non-preferred hand Threading beads

Manual Dexterity (MD) MD1

Aiming and Catching (AC) Catch Catching beanbag

Throw

Catching with one hand Best hand Other hand Throwing at wall target

Number of errors

Number of correct catches out of 10

Throwing beanbag onto mat

Throwing beanbag onto mat

One-leg balance

One-board balance

Two-board balance

Number of seconds balanced (to maximum of 30)

Dynamic BAL1

Best leg Other leg Walking heels raised

Dynamic BAL2

Jumping on Mats

Best leg Other leg Walking heel-to-toe forwards Hopping on mats

Walking toe-to-heel backwards Zig-zag hopping

Number of correct steps (to maximum of 15) Number of correct jumps/ hops out of 5

Best leg Other leg

Best leg Other leg

Balance (BAL) Static BAL

a

Catching with two hands

Completion time in seconds

Number of correct throws out of 10

Preferred hand is taken as the hand used to write/draw with.

The aim of the current study was to examine the structural validity of the M-ABC-2 in each age band separately using confirmatory factor analyses. We conducted these psychometric analyses in two stages beginning with a strictly confirmatory model testing approach followed by model trimming (Brown, 2006; Kline, 2005). Hence, we first tested the factor pattern matrix for a simple structure involving three different underlying motor abilities ‘Manual Dexterity’, Aiming and Catching’ and ‘Balance’ as latent variables and their relationships with the tasks as specified by the design of the M-ABC2 (see Table 1). Because a simple structure hypothesis of this type represents a very stringent psychometric model likely to be rejected, the second stage of our analyses involved improvement of each model via model trimming in order to find a well fitting model for each age. Finally, we compared the factor structures between the age bands for similarities and differences. 2. Method 2.1. The normative sample A stratified sample of 1172 children (566 boys and 606 girls), aged 3–16 years, from the UK, comprised the normative sample. There were 431 in AB1, 333 in AB2 and 408 in AB3. For the project as a whole, a stratified sampling plan was developed to ensure that representative proportions of children from each demographic group in the UK would be included. Data gathered from the 2001 Census provided the basis for this stratification. The sampling plan involved a cell structure that identified appropriate numbers of children for each cell, defined in terms of age, gender, geographic region, race/ethnic group, and parental education (see M-ABC-2 test manual for full details). Goodness of fit tests involving the stratification variables confirmed the representativeness of this sample in relation to the 2001 Census data. On only one variable, geographic region, did the sample proportions differ significantly from the proportions specified by the census data (x2 = 52.3, p < 0.001). For example, some areas in the North of England were slightly over represented and some in the South under represented. 2.2. The M-ABC-2 tasks Within each age band of the test, eight tasks are grouped under the three headings: Manual Dexterity, Aiming and Catching and Balance. The tasks for each age band are shown in Table 1 and their raw scores include ‘time in seconds’, ‘number of errors’ and ‘number of correct attempts’. For some tasks, a higher score indicates better performance (e.g. number of seconds balanced and number of correct jumps) whereas for others a lower score indicates better performance (e.g. completion time and number of errors).

1364

J. Schulz et al. / Research in Developmental Disabilities 32 (2011) 1361–1369

2.3. Procedure Ethical approval for the study was obtained from Oxford Brookes University. Details of testing are described in detail in the test manual (Henderson et al., 2007). Children were individually tested in their regular educational setting on the set of age-appropriate tasks from the M-ABC-2 test. All testers were appropriately qualified and extensive training on the administration and scoring of the test was provided to ensure consistency. 3. Data preparation and analysis 3.1. Distribution of the task scores The distributions of raw scores on each task were first explored using indices of descriptive statistics. Due to the nature of the measurements of the tasks (i.e. some measure time taken and others counts) the shape of the distributions varied considerably from slightly skewed to heavily skewed with some tasks also having extremely peaked distributions. The hopping tasks, for example, had a very narrow range of scores and to reduce the amount of skewness they were summed to give a total ‘hopping’ score. The multivariate kurtosis also suggested some non-normality (Mardia’s kappa = 1.45) in the distributions of the scores. These raw data were then used to generate covariance matrices for the subsequent analyses, outlined below. 3.2. Confirmatory factor analysis of the Movement ABC-2 tasks and model fitting strategy Structural equation modelling (SEM) was employed using LISREL8.8 to reveal the factor structure underlying the correlations between tasks within each of the three age bands. Following current recommendations regarding the model fitting strategy for SEM analyses (Asparouhov & Muthen, 2009; Brown, 2006; Thompson, 2004), we first specified and tested the psychometric model that had the strongest theoretical justification; if accepted, any such model would provide the strongest evidence for structural validity of the test under investigation. The first model therefore represented the threecomponent design of the M-ABC-2 test (Table 1) as a confirmatory factor analysis. Thus, the loading matrix was specified as having a simple structure strictly in accordance with the test design where the three factors MD, AC and BAL impacted only on their respective component tasks; no double loadings were allowed and no correlated measurement errors, but the factors were expected to correlate. Since pure simple structure solutions are difficult to achieve, we continued the SEM analyses moving from a strict confirmatory phase of testing to a more exploratory phase. For this we applied, a model trimming approach, where the models were improved step by step by adding further and/or dropping insignificant (p > 0.05) parameters to arrive at a good fitting model for each age band. This post hoc modification of the models was informed by checking the t-values of the parameters as well as the standardised residuals with large residuals (>3.00) indicating some misspecification in the model. Where appropriate, the correlation matrix of first-order factors representing the components of the test was itself investigated for a general factor by testing the fit of a 2nd-order factor model. Because of the non-normality of the data all analyses were conducted on covariance matrices using robust ML estimation in LISREL, with the advantage of producing corrected standard errors for the parameter estimates. Furthermore, the Satorra– Bentler global goodness of fit test was used which re-scales the Chi-square to take non-normality of distributions into consideration. Several recommended fit indices were also inspected when assessing the fit of the models: the root mean square error of approximation (RMSEA), the non-normed fit index (NNFI), the adjusted goodness of fit (AGFI) and the standardised root mean residual (SRMR) (Bentler, 2007; Kline, 2005). The correlation matrices of the tasks for each age band are displayed in Table 2. 4. Results 4.1. CFA for age band 1 (3–6 years) The first model specifying 3 correlated factors with a simple structure of the loading matrix and no error correlations, was clearly rejected, Satorra–Bentler x2 (df = 32) = 410.65, p < 0.001, and all fit indices confirmed that its fit was rather poor, RMSEA = 0.17, NNFI = 0.76, AGFI = 0.70, SRMR = 0.19 and there were a number of large standardised residuals >5.00. Post hoc modifications of this model allowing double loadings and error correlations did not produce a model with an acceptable fit and an interpretable factor structure. In particular, there remained many substantial correlations between the measurement errors of the tasks, clearly pointing towards another systematic source of variation. We therefore extended the modelling and adopted a type of CFA which specified a general factor in addition to the task specific factors. Importantly though, the new general factor was specified to be independent of the three specific motor ability factors. This model therefore states that each task is influenced simultaneously by a general factor underlying all items as well as by a specific motor ability factor influencing a group of items only. The result of the goodness of fit test for this model suggested that it should be accepted, Satorra–Bentler x2 (df = 24) = 33.44, p < 0.095, with all fit indices suggesting that its fit was excellent, RMSEA = 0.03, NNFI = 0.99, AGFI = 0.96, SRMR = 0.023. The p-value for the test of close fit was very high, p = 0.92; the largest standardised residual was just 2.25; all its parameters were significant (t-value > 1.96).

J. Schulz et al. / Research in Developmental Disabilities 32 (2011) 1361–1369

1365

Table 2 Correlation matrices of the M-ABC-2 for each age band. Age band 1, N = 431 MD1 npref hand MD2 MD3 Catch Throw Static BAL best leg Static BAL other leg Dynamic BAL1 Dynamic BAL2 Age band 2, N = 333 MD1 npref hand MD2 MD3 Catch Throw Static BAL best leg Static BAL other leg Dynamic BAL1 Dynamic BAL2 Age band 3, N = 408 MD1 npref hand MD2 MD3 Catch best hand Catch other hand Throw Static BAL Dynamic BAL1 Dynamic BAL2

MD1 pref hand

MD1 npref hand

0.80 0.46 0.17 0.12 0.11 0.20 0.18 0.11 0.07 MD1 pref hand

0.48 0.10 0.13 0.09 0.14 0.12 0.01 0.14 MD1 npref hand

0.50 0.44 0.18 0.19 0.24 0.27 0.28 0.26 0.20 MD1 pref hand

0.38 0.22 0.17 0.27 0.28 0.24 0.32 0.22 MD1 npref hand

0.66 0.43 0.36 0.15 0.17 0.20 0.08 0.02 0.09

0.43 0.34 0.27 0.26 0.22 0.17 0.01 0.13

MD2

0.28 0.32 0.15 0.26 0.23 0.25 0.23 MD2

0.20 0.14 0.13 0.29 0.29 0.27 0.22 MD2

0.19 0.30 0.30 0.23 0.09 0.09 0.01

MD3

0.41 0.38 0.53 0.47 0.43 0.34 MD3

0.20 0.12 0.32 0.22 0.34 0.24 MD3

0.20 0.17 0.11 0.17 0.03 0.11

Catch

Throw

0.39 0.46 0.44 0.36 0.31

0.43 0.38 0.37 0.25

Catch

Throw

0.27 0.27 0.15 0.19 0.18

0.21 0.20 0.16 0.16

Static BAL best leg

Static BAL other leg

Dynamic BAL1

0.87 0.56 0.29

0.48 0.26

0.38

Static BAL best leg

Static BAL other leg

Dynamic BAL1

0.73 0.30 0.35

0.21 0.24

0.51

Catch best hand

Catch other hand

0.86 0.40 0.06 0.11 0.05

0.44 0.07 0.18 0.08

Throw

0.01 0.13 0.07

Static BAL

Dynamic BAL1

0.27 0.25

0.16

The final model for age band 1 showing a general factor independent of the three motor specific ability factors is displayed in Fig. 1. With the exception of the two ‘posting coins’ tasks all other items have a substantial loading (>0.50) on the general factor, GF, and were higher than the corresponding values on the specific motor ability factors. The ‘walking heels raised’ task has no loading on the balance factor, but is solely governed by the GF factor. All three correlations among the motor ability factors are weak (r < 0.30). 4.2. CFA for age band 2 (7–10 years) As before in age band 1, a 3 correlated factors model with a simple structure of the pattern matrix was tested first and rejected by the global goodness of fit test, Satorra–Bentler x2 (df = 32) = 124.6, p < 0.001. Inspection of several fit indices confirmed that its fit was poor, RMSEA = 0.094, NNFI = 0.83, AGFI = 0.85, SRMR = 0.089, the largest standardised residual was 5.72. Post hoc modifications for this model first involved adding double loadings and correlated measurement errors, but this improved the model fit only partially. Consequently, the factor structure had to be altered and was refined by allowing two balance factors to take account of the difference in ability when performing the static as opposed to the dynamic balance tasks. The result of the goodness of fit test for this final model with two double loadings and no measurement error correlation was acceptable, Satorra–Bentler x2 (df = 27) = 37.70, p = 0.08; all fit indices also suggested an excellent model fit, RMSEA = 0.035, NNFI = 0.98, AGFI = 0.95, SRMR = 0.038. The p-value for the test of close fit was high, p = 0.84. The largest standardised residual was only 3.16. All parameters of the model were significant (t-values > 1.96). The factor pattern matrix showed that each task except the ‘drawing trail’ had a substantial loading (>0.50) on its respective factor. The correlations among the four factors ranged from modest (r = 0.36) to substantial (r = 0.58) suggesting sufficient common variance to assume an underlying higher-order factor representing general not domain specific motor abilities. Thus, a 2nd-order CFA model was specified (Fig. 2) producing an excellent model fit, Satorra–Bentler x2 (df = 29) = 39.11, p = 0.10; all fit indices were excellent, RMSEA = 0.032, NNFI = 0.98, AGFI = 0.95, SRMR = 0.039. The p-value for the test of close fit was high, p = 0.87; the largest standardised residual was just 2.82. The 2nd order general factor explained a considerable amount of variance in each primary factor; manual dexterity 61%, aiming and catching 54%, static balance 34% and dynamic balance 39%.

[()TD$FIG]

1366

J. Schulz et al. / Research in Developmental Disabilities 32 (2011) 1361–1369

Fig. 1. General and specific factor model of the Movement ABC-2 tasks for AB1 (3–6 years) Satorra–Bentler x2 (df = 24) = 33.4, p = 0.10, RMSEA = 0.030; MD, Manual Dexterity; AC, Aiming and Catching; BAL, Balance; GF, general factor. Note: All parameters are standardised and significant at p < 0.01.

4.3. CFA for age band 3 (11–16 years) As before, we first tested a 3 correlated factors model with a simple structure loading matrix which was rejected by the global goodness of fit test Satorra–Bentler x2 (df = 32) = 71.05, p < 0.001. However, several fit indices, RMSEA = 0.055, NNFI = 0.93, AGFI = 0.93, and SRMR = 0.056, clearly indicated that this was not actually a very poor fitting model and could probably be improved by some post hoc specification; the largest standardised residual was 3.52. Step by step post hoc modification added 3 double loadings and one measurement error correlation to the model (Fig. 3). The goodness of fit test for the final model was acceptable, Satorra–Bentler x2 (df = 28) = 38.41, p = 0.09, and all fit indices suggested an excellent model fit, RMSEA = 0.030, NNFI = 0.98, AGFI = 0.96, and SRMR = 0.036. The p-value for the test of close fit was high, p = 0.93. The largest standardised residual was only 2.53. All parameters of the model were significant (tvalue > 1.96). The factor pattern matrix provided evidence that a separation of the tasks into three distinct content areas appears justified. Only three tasks (i.e. turning pegs with preferred hand, triangle with nuts and bolts, and throwing at wall target) have small double loadings on MD or AC. All the other tasks only have loadings on the factors they are supposed to tap. Secondly, the factor correlations (Fig. 3) are weak to modest only. There is a modest correlation between MD and AC (r = 0.37) whereas the correlations of these two factors with the BAL factor are very weak. As was done for age band 1, the common variance shared by the three domain factors can be represented by a 2nd order factor (Fig. 3); this factor explained 54% of the variance in Manual Dexterity and 26% in Aiming and Catching but just 9% in Balance. Since this 2nd-order factor model is equivalent to a 3-correlated factors model its goodness of fit is identical to the one reported for the latter above. 5. Discussion The eight M-ABC-2 tasks included within the three components of the test were originally selected to represent aspects of motor control and coordination that seem to underlie many activities of daily living for children. It is also clear that these are areas in which children with motor difficulties often struggle. In clinical settings, it is common to find children whose profile across the three components of the M-ABC-2 varies dramatically. To the many users of the test, therefore these components have content validity and seem functionally relevant. Furthermore, the three components of the test reflect quite well different strands of research in the area of motor development. For example Haywood and Getchell (2005) devote entirely separate chapters of their book to explain the development of ‘manipulative skills’, ‘ballistic skills’ and ‘locomotion’ (including dynamic balance). In the present study, we have subjected the 3-component structure of the M-ABC-2 test to an extensive statistical analysis designed to explore its structural validity using the standardisation sample which consisted mainly of typically developing children. Although the test structure remains identical across age in terms of its three domains of testing, age appropriate tasks had to be developed for each age band separately. Consequently, we investigate the structural validity of the test in each age group separately.

[()TD$FIG]

J. Schulz et al. / Research in Developmental Disabilities 32 (2011) 1361–1369

1367

Fig. 2. 2nd-order factor model of the Movement ABC-2 tasks for AB2 (7–10 years). Satorra–Bentler x2 (df = 29) = 39.11, p = 0.10; RMSEA = 0.032; MD, Manual Dexterity; AC, Aiming and Catching; BAL stat, Static balance; BAL dyn, Dynamic balance; GF mot, 2nd order general factor. Note: All parameters are standardised and significant at p < 0.01.

As expected, the relationships between the tasks of the M-ABC-2 and its three domains of testing could not be represented by a 3-factors model with a perfect simple structure of the factor loading matrix. However, some post hoc modifications established that in each age band a well fitting factor model corroborated the structural validity of the test, particularly so in age bands 1 and 2. In these two age bands, most of the tasks had a substantial loading on the domain specific factor they are supposed to measure and there were few double loadings. This pattern of the factor loadings suggests both that the tasks are reliable measures of the three ability factors MD, AC and BAL and that there is evidence for convergent validity of the tasks towards the three core constructs of the M-ABC-2 underlying movement skills. Finally, the modest to moderate correlations between the ability factors provide evidence for their discriminant validity. For the youngest children (AB1), model trimming revealed a rather complex factor structure underlying the correlations among the tasks. This ‘hierarchical factor model’ (Yung, Thissen, & McLeod, 1999), suggests that their performance depends partly on a general factor independent of motor abilities, with loadings higher than those on the corresponding domain factors. Two interpretations as to the nature of this g-factor could be considered. It may be a motivational factor as children in this age band were rather young (3–6 years) and so their willingness to fully engage with the tasks might crucially depend on their current mood. A more likely interpretation, however, is that it represents the children’s current general developmental status, in particular individual differences in biological maturity relating to their growth and physical strength (Gallahue & Ozmun, 2006), as well as their specific environmental experiences. It is interesting to note that only the task ‘posting coins’ had a high loading on the specific factor MD, but a very small loading on the general factor. There is not an obvious explanation for these differences, although one could speculate that the timed ‘posting coins’ task may be considered a more ‘pure’ assessment of fine motor abilities while the untimed ‘drawing trail’ taps a broader aspect of development and handeye control influenced by experience. In AB2 the independent general factor has disappeared and four well correlated specific motor ability factors emerged largely in line with the division of the tasks into the M-ABC-2 sub-components. At this age, a distinction emerged between the tasks measuring static and dynamic balance. Although this is consistent with some research (Drowatsky & Zuccato, 1967; Tsiglis, Zachopoulou, & Mavridis, 2001), it may be that these two balance factors are also partly capturing method variance due to the different nature of the static (i.e. one-board balance on each leg) as opposed to the dynamic tasks (i.e. walking and hopping). For 11–16-year olds, however, the ‘two board balance’ item is performed with both feet on the board at the same time and therefore yields only a single score ruling out the possibility of distinguishing between method and ability variance. The substantial correlations between the four specific motor ability factors could be explained by a second-order factor representing general motor abilities. This is in stark contrast to the finding of a strong independent general factor in AB1 representing differences in developmental level associated with physical maturation. The existence of such a second-order general factor in the model provides evidence that children in this age band tended to have similar performance levels across the three motor domains suggesting considerable continuity in the developmental processes governing their movement skills. For the oldest age group (AB3), there was even clearer support for a separation of the tasks into three components as the factor pattern matrix almost had a simple structure and the factor correlations were weak to modest only. Compared to the model for AB2, the weakening of the correlations between the three factors suggests that there is an ongoing developmental process towards differentiation and specialisation of movement abilities in children influencing a number of distinct movement skills. As the three domain factors became more independent, accordingly, the strength of the higher-order factor

[()TD$FIG]

1368

J. Schulz et al. / Research in Developmental Disabilities 32 (2011) 1361–1369

Fig. 3. 2nd-order factor model of the Movement ABC-2 tasks for AB3 (11–16 years) Satorra–Bentler x2 (df = 28) = 38.4, p = 0.09; RMSEA = 0.030, MD, Manual Dexterity; AC, Aiming and Catching; BAL, Balance; GF mot, 2nd order general factor. Note: All parameters are standardised and significant at p < 0.01.

representing a general motor ability dimension has weakened, particularly in relation to the balance factor and thus locomotion as one of the essential movement functions. Altogether therefore, this model offers strong support for the structural validity of the M-ABC-2 test and its assessment of three distinct dimensions of motor performance, at least in the age group 11–16. As far as interpretation of the test results are concerned, therefore, confidence in the validity of the three components of the M-ABC-2 test becomes stronger as a child grows older. These developmental changes are also supported by data from the BOT-2, whereby associations between the sub-components are reported to decrease slightly with age (Bruininks & Bruininks, 2005). It certainly seems logical that early on children start to develop what are described by Burton and Miller (1998) as ‘fundamental movement skills’. These are ‘‘the locomotor and object-control skills. . .used by persons in all cultures of the world’’ (p. 58). Once these emerge there will be opportunities for the child to practise, refine and combine them to develop more complex skills specific to particular tasks and contexts. These developmental aspects are emphasized in models of motor development and are widely reflected in the literature with separate descriptions of ‘rudimentary movement abilities’ in infancy, ‘fundamental movement abilities’ in childhood and subsequent refinement of skills in the context of sports with the development of ‘specialized movement abilities’ in adolescence (for example see Clark & Metcalfe, 2002; Gallahue & Ozmun, 2006). To our knowledge there is no relevant empirical research to draw on to support our findings. It is clear that further work in this area is needed in studies designed specifically for the purpose of examining movement factors at different ages, as data from assessment tests is not ideally suited to address this issue in a comprehensive manner. The possible theoretical implications of our results across the three age bands have been outlined above in terms of support for the validity of the test and also, more generally, in our understanding of the structure of movement abilities in children. Future work should examine these factors in greater detail and, in particular, compare the factor structure for children with movement difficulties. Practical implications of the work are perhaps limited until such further information is gathered. In the meantime, we can make some tentative suggestions in relation to the use of test scores from typically developing children. In age band 1 the focus should be primarily on the total test score. In age bands 2 and 3 the profile of scores across the three sub-components becomes more meaningful and can be used with greater confidence. 6. Conclusion Overall our findings lend support for the structural validity of the M-ABC-2 test. Future work should focus on a similar examination of the structural validity of the test in children with movement difficulties, with whom the M-ABC-2 is usually employed. Acknowledgements We are grateful to Pearson Assessment, Action Medical Research and the Freemason’s Grand Charity for funding this work and to Professor John Rust, Professor Susan Golombok and Dr. Emma Lycett for their expert assistance in sample selection. We thank all the children who participated in this study and our many testers who administered the Movement ABC-2.

J. Schulz et al. / Research in Developmental Disabilities 32 (2011) 1361–1369

1369

References APA. (2000). DSM-IV-TR. Diagnostic and statistical manual of mental disorders (4th ed.). Washington, DC: American Psychiatric Association. Asparouhov, T., & Muthen, B. (2009). Exploratory structural equation modeling. Structural Equation Modeling, 16, 397–438. Barnett, A. L. (2008). Motor assessment in developmental coordination disorder: From identification to intervention. International Journal of Disability, Development and Education, 55, 113–129. Bentler, P. (2007). On tests and indices for evaluation structural models. Personality and Individual differences, 42, 825–829. Brown, T. A. (2006). Confirmatory factor analysis for applied research. New York: The Guilford Press. Bruininks, R. H. (1978). Bruininks-Oseretsky Test of Motor Proficiency. Circle Pines, MN: American Guidance Service. Bruininks, R. H., & Bruininks, B. D. (2005). Bruininks–Oseretsky test of motor proficiency (2nd ed.). Windsor: NFER-Nelson. Burton, A. W., & Miller, D. E. (1998). Movement skill assessment. Champaign, IL: Human Kinetics. Clark, J. E., & Metcalfe, J. M. (2002). The mountain of motor development: a methaphor. In J. E. Clark & J. H. Humphrey (Eds.), Motor development: Research and reviews (pp. 163–190). Reston, VA: NASPE Publications. Croce, R. V., Horvat, M., & McCarthy, E. (2001). Reliability and concurrent validity of the movement assessment battery for children. Perceptual and Motor Skills, 93, 275–280. Drowatsky, J. N., & Zuccato, F. C. (1967). Interrelationships between selected measures of static and dynamic balance. Research Quarterly, 38, 509–551. Fleishman, E. A., & Bartlett, C. J. (1969). Human abilities. Annual Review of Psychology, 20, 349–380. Gallahue, D. L., & Ozmun, J. C. (2006). Understanding motor development. Infants, children, adolescents, adults (6th ed.). Madison, WC: Brown & Benchmark. Haywood, K. M., & Getchell, N. (2005). Life span motor development (4th ed.). Champaign, IL: Human Kinetics. Henderson, S. E., & Barnett, A. L. (1998). The classification of specific motor coordination disorders in children: Some problems to be solved. Human Movement Science, 17, 449–469. Henderson, S. E., & Sugden, D. A. (1992). The movement assessment battery for children. London: The Psychological Corporation. Henderson, S. E., Sugden, D. A., & Barnett, A. L. (2007). Movement assessment battery for children [examiner’s manual] (2nd ed.). London: Pearson Assessment. Henry, F. M. (1961). Reaction time–movement time correlations. Perceptual and Motor Skills, 12, 63–66. Henry, F. M. (1968). Specificity vs. generality in learning motor skill. In R. C. Brown & G. S. Kenyon (Eds.), Classical studies on physical activity. Englewood Cliffs, NJ: Prentice-Hall. (original work published 1958). Kaplan, B. J., Wilson, B. N., Dewey, D., & Crawford, S. G. (1998). DCD may not be a discrete disorder. Human Movement Science, 17, 471–490. Keogh, J., & Sugden, D. A. (1985). Movement skill development. New York: Macmillan. Kline, R. (2005). Principals and practice of structural equation modelling (2nd ed.). New York: The Guildford Press. Larkin, D., & Rose, E. (2005). Assessment of developmental coordination disorder. In D. Sugden & M. Chambers (Eds.), Children with developmental coordination disorder (pp. 135–154). London: Whurr. McCarron, L. T. (1982). McCarron assessment of neuromuscular development. Dallas, TX: Common Market Press. McCarron, L. T. (1997). McCarron assessment of neuromuscular development: Fine and gross motor abilities. Texas: McCarron-Dial Systems Inc. Messick, S. (1995). Validity of psychological assessment. American Psychologist, 50, 741–749. Rarick, G. L., Dobbins, D. A., & Broadhead, G. D. (1976). The motor domain and its correlates in educationally handicapped children. Engelwood Cliffs, NJ: Prentice-Hall. Schmidt, R. A., & Lee, T. D. (2005). Motor control and learning. A behavioral emphasis (4th ed.). Champaign, IL: Human Kinetics. Shih-Heng, S., Yi-Ching, Z., Ching-Lin, S., Chien-Hui, L., & Sheng, K. W. E. (2010). Development and initial validation of the Preschooler Gross Motor Quality Scale. Research in Developmental Disabilities, 31, 1187–1196. Stott, D. H., Moyes, F. A., & Henderson, S. E. (1972). The test of motor impairment. San Antonio, TX: The Psychological Corporation. Stott, D. H., Moyes, F. A., & Henderson, S. E. (1984). The test of motor impairment—Henderson revision. San Antonio, TX: The Psychological Corporation. Tan, S. K., Parker, H. E., & Larkin, D. (2001). Concurrent validity of motor tests used to identify children with motor impairment. Adapted Physical Activity Quarterly, 18, 168–182. Tsiglis, N., Zachopoulou, E., & Mavridis, T. (2001). Evaluation of the specificity of selected dynamic balance tests. Perceptual and Motor Skills, 90, 827–833. Thompson, B. (2004). Exploratory and confirmatory factor analysis: Understanding concepts and applications. Washington, DC: American Psychological Association. Wassermann, J. D., & Bracken, B. (2003). Psychometric characteristics of assessment procedures. In I. B. Weiner, D. K. Freedheim, J. A. Graham, & J. A. Naglieri (Eds.), Handbook of psychology. Vol. 10. Assessment psychology. New Jersey: John Wiley. Yung, Y., Thissen, D., & McLeod, L. D. (1999). On the relationship between the higher-order and the hierarchical factor model. Psychometrica, 64, 113–128.