1. Top of page
  2. Abstract
  3. Method
  4. Results
  5. Discussion
  6. Conclusion
  7. Acknowledgments
  8. References


This study investigated the internal construct validity and dimensionality of the Melbourne Assessment of Unilateral Upper Limb Function (Melbourne Assessment), a widely-used measure of quality of upper limb movement, valid for children aged 2 years 6 months to 15 years with cerebral palsy.


Rasch analysis was used to assess of Melbourne Assessment raw scores for 163 children (94 males, 69 females; mean age 8y, SD 3y 5mo). Analysis was undertaken on the full scale comprising 37 scores and on groups of scores separated into four distinct movement subscales: range of movement, accuracy, dexterity, and fluency. Tests were conducted to evaluate overall model fit, item fit, suitability of the response options, unidimensionality, and differential item functioning (DIF) for sex, child age, and different raters.


The results did not support the unidimensionality of the 37-score scale. The four subscales showed adequate model fit after removal of some score items, and rescaling of others. The resulting subscales showed good internal consistency and no DIF for sex or child age.


This study provides empirical support for a revised version of the Melbourne Assessment which comprises 14 tasks and 30 movement scores grouped across four separate subscales. Further testing is required to assess the responsiveness of subscales to clinically important change.


Differential item functioning


Principal components analysis


Person Separation Index


Range of motion

The Melbourne Assessment of Unilateral Upper Limb Function (Melbourne Assessment) is a criterion referenced tool designed to measure the quality of upper limb movement in children aged 5 to 15 years with a neurological impairment.[1-3] Although originally designed for children as young as 5 years, recent studies have established the validity of using a modified version of the tool with children as young as 2 years 6 months of age.[4, 5] Since its publication in 1999, the Melbourne Assessment has been increasingly used as an outcome measure in clinical and research applications, both nationally and internationally.[6-8] Studies investigating the effectiveness of interventions such as intrathecal baclofen[6] and kinesiotaping[8] are using the Melbourne Assessment to evaluate change in children's upper limb movements. For the Melbourne Assessment to be a valid outcome measure, it is critical that it demonstrates strong psychometric properties. Studies undertaken at the time of developing the tool found the Melbourne Assessment to have high internal, test–retest, and intra- and interrater reliability for the overall test scores.[1]

The tool was developed using classical test theory approaches. Test developers undertook an extensive review of the literature and existing assessments, and sought input from expert clinicians to define the aspects or elements of movement that were typically observed when evaluating upper limb movement quality.[3] Four main elements of movement quality were identified: (1) amount of active range of movement at each upper limb joint (ROM); (2) accuracy of reach for, or placement of, an item; (3) dexterity of finger movements when grasping, releasing, and manipulating objects; and (4) fluency or smoothness of the movement. These four elements are observed during completion of 16 tasks, which were selected to mimic everyday activities. Two additional aspects of upper limb movement: ‘Bilateral co-ordination’ and ‘Speed’ were also included in the original scoring (Table 1). Scoring is completed by grading one or more elements of movement quality observed in the performance of each task on a 3-, 4-, or 5-point ordinal scale using specifically developed criteria. Across the 16 tasks there are a total of 37 movement scores, which are summed to provide a single total score of the child's upper limb movement quality.

Table 1. Components of movement evaluated by the original Melbourne Assessment of Unilateral Upper Limb Function
Upper limb task/actionItem no.Item titleElements of movement quality scored
Range of movementAccuracy of reach/placementFluencyDexterity of finger movementsBilateral coordinationSpeed
  1. X indicates element of movement is scored in item.

Reach1Reach forwardsXXX   
2Reach forwards to an elevated positionXXX   
3Reach sideways to an elevated positionXXX   
11Reach to brush from forehead to back of neckX X   
12Palm to bottomX X   
15Reach to opposite shoulderXXX   
16Hand to mouth and downXXX  X
Grasp4Grasp of crayon   X  
5Drawing grasp   X  
7Grasp of pellet   X  
Release6Release of crayonXX X  
8Release of pelletXX X  
Manipulation9Manipulation  XX  
Pointing (×4)10Pointing X X X X    
Bilateral transfer14Hand to hand transfer    X 
Total score items (37)  10118611

The recent emphasis on the importance of producing psychometrically sound, interval level measures to advance the field of health sciences[9-11] has suggested the need to further investigate several aspects of the Melbourne Assessment scale. The unidimensionality, one aspect of internal construct validity, of the Melbourne Assessment scale requires more targeted evaluation. The unidimensionality of a scale is one of the fundamental properties of any measurement. Unidimensionality indicates that all items within the one scale are measuring the same construct. For a summed score to be valid, all items contributing to that score should measure the same construct. In the Melbourne Assessment it is important that the unidimensionality of the total scale be examined, as currently the 37 movement scores are summed to provide one total score. It is possible that, given the same four elements of movement are scored repeatedly across the 16 tasks, groups of scores that measure the same element of movement may form four separate, unidimensional subscales.

Over the past two decades there has been increasing use of the Rasch measurement model[9, 12-15] in the development and evaluation of clinical tools in the health sciences. Rasch analysis has also been used to verify and refine the psychometric properties of outcome measures initially developed using traditional methods of test construction.[10, 11, 16] The aim of this study, therefore, was to use Rasch analysis to further investigate the internal construct validity of the Melbourne Assessment and to optimize the scaling of the measure to ensure clinicians and researchers could draw meaningful conclusions from scores obtained on the tool. Specifically, the focus of the study was to investigate whether scores on the Melbourne Assessment comprised a unidimensional scale, or instead, contained four psychometrically distinct subscales: (1) ROM; (2) accuracy; (3) dexterity; and (4) fluency.


  1. Top of page
  2. Abstract
  3. Method
  4. Results
  5. Discussion
  6. Conclusion
  7. Acknowledgments
  8. References

This study was an instrument validation study involving secondary analysis of a large set of de-identified Melbourne Assessment raw scores.


Copies of 121 sets of de-identified Melbourne Assessment raw scores were collected from six Australian researchers who had used the Melbourne Assessment as an outcome measure for children with cerebral palsy (CP) aged 5 to 15 years.[17-20] In addition, 20 sets of data used in the development of the original assessment[1] and 22 sets of scores collected during a study to extend the tool for use with children aged 2 to 4 years[5] were included. The following demographic data were available from every study: sex, age, type of movement disorder and distribution, and upper limb assessment. Gross Motor Function Classification System level and Manual Ability Classification System level were not available for all children.


For each contributing study the Melbourne Assessment was administered according to the guidelines in the test manual[3] by a clinician familiar with the tool. Each videotaped assessment was subsequently scored by an independent scorer. Across the included studies scoring was completed by five different scorers. Each scorer underwent separate training before commencing scoring the assessments specific to their study. The scorer's level of experience in working with children with CP ranged from 4 to 10 years or more. Data collected from intervention studies were raw scores obtained at baseline before the intervention.

Statistical analysis

Rasch analysis was conducted using the partial credit model of the Rasch Unidimensional Measurement Model (RUMM) 2030 software (RUMM Laboratory, Perth, WA, Australia) following the procedures recommended by Pallant and Tennant,[10] and Tennant and Conaghan.[11] Firstly, the dimensionality of the original 37-score scale of the Melbourne Assessment was assessed using principal components analysis (PCA) of the residuals. PCA was used to identify two subsets of item scores from the full 37-score scale that showed the most difference from one another. Rasch-derived values from these two sets of scores were compared using a series of Student's t-tests. The unidimensionality of the scale was supported if 5% or less of cases recorded significantly different scores on each subset of scores, or more specifically if the lower value of the 95% confidence interval was ≤0.05. In the event that support was not found for the unidimensionality of the full 37-score scale, Rasch methods were used to assess each of the four subscales for suitability of response format; overall fit to the Rasch model; individual item fit; local dependency; differential item functioning; dimensionality; and how well the item scores target the ability levels of children in the sample, that is, children with CP for whom the test had been developed.

Each series of analyses commenced with inspection of the threshold map for the item scores in each subscale. Items with disordered score thresholds were identified and the category probability curves inspected to ascertain which thresholds did not discriminate between adjacent scores. Response categories for these items were corrected to achieve ordering of scores from easy to difficult. Overall fit to the model was assessed using χ2 statistics (with good fit indicated by a non-significant χ2 value after Bonferroni adjustment for the number of score items) and Fit Residual Standard Deviation values (<1.5). Individual item scores were assessed using χ2 statistics and individual fit residual values, with high positive fit residual values (+2.5) suggesting poor fit. Item dependency was identified by inspection of the residual correlations for values exceeding 0.30. Where residual correlations exceeded 0.3 and clinical judgement deemed one of the item scores provided no additional clinically important information, item scores were removed or combined and the analyses repeated.

The internal consistency of each subscale was assessed using the Person Separation Index (PSI), with values exceeding 0.70 considered acceptable, and those above 0.80 desirable.[21] Differential item functioning (DIF) for children's sex and age group, and for variation in scoring by different raters, was investigated using a Bonferroni adjusted p value. The person-item map for each refined subscale was also inspected to assess how well each subscale targeted the study population. Finally, the dimensionality of each subscale was evaluated using PCA and t-tests to determine if no more than 5% of cases achieved significantly different scores on two smaller subsets of item scores. Where subscales were found not to fit the Rasch measurement, model modifications were made (e.g. rescore or delete items) and the subscale retested. If support for a four subscale structure was found, PCA and t-tests were also used to assess dimensionality of the four subscale structure using sets of items across pairs of subscales.

Rasch-derived scores from the final revised version of each subscale were exported to spss Version 20 (IBM SPSS Statistics, IBM Corp., NY, USA). Pearson's correlation coefficients were calculated to assess the intercorrelation among the subscales to further understand if subscales were measuring discrete or related constructs.

There are no clear guidelines for the number of observations required to undertake Rasch analysis; however, Linacre[22] recommends that 150 sets of assessment raw scores would be required to undertake a full Rasch analysis on a tool. This number allows 99% confidence that the ‘true’ item difficulty was within (SD) half a logit of its reported estimate from the analysis.[22]

Approval for this study was received from the La Trobe University, Faculty of Health Sciences, Human Ethics Committee and The Royal Children's Hospital, Melbourne, Ethics into Human Research Committee.


  1. Top of page
  2. Abstract
  3. Method
  4. Results
  5. Discussion
  6. Conclusion
  7. Acknowledgments
  8. References

The final cohort comprised 163 sets of Melbourne Assessment raw scores (94 males, 69 females; age range 2–15y, mean age 8y, SD 3y 5mo). The type, topography, and severity of motor impairment of participants are described in Table 2. Separate analyses are reported for the full 37-score scale and then for each of the four identified movement subscales: (1) ROM; (2) accuracy; (3) dexterity; and (4) fluency.

Table 2. Type, topography and severity of motor impairment of the sample
Type of motor impairment and topographical distributionSeverityTotal, n (%)
Mild (n)Moderate (n)Severe(n)
Spasticity –Quadriplegia261642 (25.8)
Diplegia12719 (11.7)
Right hemiplegia721533 (20.3)
Left hemiplegia1027744 (27.0)
Triplegia26311 (6.7)
Athetosis426 (3.6)
Ataxia22 (1.2)
Hypotonia33 (1.8)
Dystonia – left upper limb123 (1.8)
Total329833163 (100)

Dimensionality of the original Melbourne Assessment

Results for the Rasch analyses undertaken on the full 37-score scale of the original tool are reported in Table 3. Disordered thresholds were detected for 10 of the 37 scores and the scale showed poor fit to the Rasch model (p<0.001). A PCA of the residuals for the rescaled scores from the original Melbourne Assessment did not support the unidimensionality of the full 37-score scale. Forty-three of 163 cases (26.38%, 95% CI 23.0–29.7%) were found to have statistically different scores on each subset of item scores. These results suggested that the four subscales should be assessed individually.

Table 3. Summary of Rasch analysis for the original 37-score scale and the four revised movement subscales on the Melbourne Assessment of Unilateral Upper Limb Function
SubscaleNo. of itemsItems with disordered response formatsOverall model fitItem fit residual mean (SD)Person fit residual mean (SD)PSILocal dependencyPCA % cases recording significantly differentscores on two subsets of items (95% CI)
Correlated items (r>0.3)Action
  1. aPearson's r approaching 0.3. PSI, Person Separation Index; PCA, Principal Components Analysis.

Original Melbourne Assessment374, 6.3, 7, 8.3, 9.1,10.2, 10.4, 11.1, 12.1, 16.2χ2=150.35 df=74 p=0.00000.21 (1.56)−0.19 (1.36)0.9747 pairs of correlations >0.326.38 (23.0 to 29.7)
Range9χ2=10.77 df=18 p=0.900.16 (0.91)−0.31 (1.00)0.89

1.1/2.1 (0.58)

1.1/1.3 (0.40)

2.1/3.1 (0.53)

6.1/8.1 (0.33)

Remove 2.1

Retain 6.1 and 8.1

8.0 (4.6 to 11.3)




0.01 (0.80)−0.17 (0.85)0.81

1.2/2.2 (0.40)

2.2/3.2 (0.46)

10.2/10.3 (0.39)

10.1/10.4 (0.27)

6.3/8.3 (0.53)

Remove 2.2

Combine 10.2 and 10.3

Combine 10.1 and 10.4

Retain 6.3 and 8.3





−0.38 (1.30)−0.35 (0.80)0.91NilNil4.29

χ2=6.07 df=14


−0.54 (1.22)−0.73 (1.53)0.92

1.3/2.3 (0.47)

2.3/3.3 (0.36)

1.3/3.3 (0.42)

Remove 2.3

Retain 1.3 and 3.3


Analyses conducted on the four subscales

Rasch analysis was conducted on each set of items representing the four movement subscales: ROM, accuracy, dexterity, and fluency. Two items from the original Melbourne Assessment (item 14 Hand to hand transfer, and item 16.4 Speed [of hand to mouth]) were not considered suitable for inclusion in any of the four identified subscales and were therefore not included in the subscale analyses.

Range of movement subscale (ROM)

Inspection of the score threshold map for the 10-item ROM subscale revealed items 11.1 (Reach forehead to back of neck) and 12.1 (Palm to bottom) to have disordered response formats. For each of these items, recoding of adjacent score categories resulted in ordered response formats. For example, in item 11.1 (Reach from forehead to back of neck) initial inspection of the category probability curves revealed the original score value of 2 was not discriminating between the adjacent score categories of 1 and 3. The original criterion for a score value of 3 was: ‘Compensatory and/or abnormal movement patterns involving one or two joints’; while a score value of 2 identified when three or more joints were involved. A score value of 1 was ‘Movement not achieved but can reach hand to head’. Further inspection of the item map revealed a reversal in levels of difficulty for the score values of 2 and 3 and so these score categories were collapsed. This resulted in rescaling the original 5-point scale (scored 01234) to a simpler 4-point scale (01223). The criterion for the rescored value of 2 is: ‘Compensatory and/or abnormal movement patterns involving any of the following joints – trunk, neck, shoulder, elbow, forearm, wrist, fingers’. Initial analyses of the rescaled subscale identified five items with high levels of item dependency (items: 1.1 [Reach forwards]/2.1 [Reach forwards elevated], 1.1 [Reach forwards]/3.1 [Reach sideways], 2.1 [Reach forwards elevated]/3.1 [Reach sideways] and 6.1 [Release of crayon]/8.1 [Release of pellet]). After considering the findings from the initial analyses of all subscales and application of clinical judgment, item 2.1 (Reach forwards elevated) was deemed to add no new clinical information and was removed from the scale. Items 6.1 (Release of crayon) and 8.1 (Release of pellet) were both retained as they provide unique clinical information.

Results of the analysis undertaken on the refined nine-item ROM subscale are reported in Table 3. The overall model fit statistics were non-significant, indicating no serious misfit to the model (p=0.90). The final PSI was high (0.89). There was no evidence of DIF for sex or age; however, for scorer variability, three items (items 11.1 [Reach from forehead to back of neck], 15.1 [Reach to opposite shoulder] and 16.1[Hand to mouth]) recorded p values exceeding the adjusted criterion, indicating individual scorers were likely to score children with equal levels of ROM ability differently on these items. As these items had not shown misfit to the model they were retained in the analyses. Inspection of the person-item threshold map (see Fig. 1a) showed that, although item score thresholds are spread across the scale of person abilities, more score items target the lower and mid-range ability levels of children in this sample. PCA supported the unidimensionality of the nine-item scale.


Figure 1. Person-item threshold maps for the revised subscales: (a) range of movement; (b) accuracy; (c) dexterity (d) fluency.

Download figure to PowerPoint

Accuracy of reach and pointing

Three items in the 11-item accuracy subscale (items 6.3 [Release of crayon], 8.3 [Release of pellet] and 16.2 [Hand to mouth]) were identified as having disordered score thresholds. Recoding of adjacent score categories established ordered response formats for items 6.3 (Release of crayon) and 16.2 (Hand to mouth); however, minor disordering remained for item 8.3 (Release of pellet). The scoring criteria for items 8.3 (Release of pellet) and 6.3 (Release of crayon) were designed to be consistent; as this consistency was maintained with the recoding, the rescaled response format for item 8.3 (Release of pellet) was retained with minor disordering. Substantial correlations between the residuals were identified for seven items (1.2 [Reach forwards]/2.2 [Reach forwards elevated], 2.2 [Reach forwards elevated]/3.2 [Reach sideways], 10.2 [Pointing green square]/10.3 [Pointing yellow square] and 6.3 [Release crayon]/8.3 [Release pellet]). Item 2.2 (Reach forwards elevated) was subsequently removed from the scale and items 10.2 (Pointing green square) and 10.3 (Pointing yellow square) were combined into one pointing item (10.1 Far pointing). The other two score items for item 10 (item 10.1 [Pointing red square] and item 10.4 [Pointing blue square]) were also combined together into one item (10.2 Near pointing) to create one far and one near pointing item. Items 6.3 (Release crayon) and 8.3 (Release pellet) both provided clinically important information so both were retained.

Rasch analysis of the refined eight-item accuracy scale showed good fit to the model (p=0.09) and a high PSI statistic of 0.81 (see Table 3). No DIF was found for sex or age; however, individual scorers appeared to score three of the accuracy items differently (items 6.3 [Release crayon], 10.1 [Pointing red square], and 15.2 [Reach opposite shoulder]). These items were retained as they did not show any evidence of misfit. Figure 1b shows that item score thresholds are also focused on the lower and mid-range ability levels of children in this sample. The unidimensionality of the revised accuracy subscale was supported (see Table 3).

Dexterity of reach, grasp, and manipulation

Two items in the six-item dexterity subscale (items 4 [Grasp of crayon] and 9.1 [Manipulation]) required recoding of adjacent score categories to create ordered score thresholds. Analyses of the rescaled subscale revealed no significant correlations between item residuals. A non-significant item-trait χ2 statistic (p=0.49) indicated adequate fit to the model (see Table 3). All items showed fit to the model and the PSI was high at 0.91. No DIF was present for sex or age; however, for differences between individual scorers’ DIF was found for items 5 (Drawing grasp) and 8.2 (Release of pellet). Inspection of the person-item threshold in Figure 1c revealed that item score thresholds are spread across the range of children's abilities, although the spread of thresholds is sparse. Results of the PCA reported in Table 3 supported the unidimensionality of the six-item scale.

Fluency of movement

All items in the eight-item fluency subscale showed ordered thresholds. Further analysis revealed substantial correlations among the residuals for three items (item 1.3 [Reach forwards]/2.3 [Reach forwards elevated], 2.3 [Reach forwards elevated]/3.3 [Reach sideways], and 1.3 [Reach forwards] /3.3[Reach sideways]). Item 2.3 (Reach forwards elevated) was consequently removed and the analyses repeated on the seven-item scale. A non-significant χ2 item-trait interaction statistic (p=0.96) indicated adequate fit to the model (see Table 3). The PSI statistic was 0.92. No DIF was found for sex or age but one item showed DIF for differences in individual scorers, item 9.2 (Manipulation). The person-item threshold map (see Fig. 1d) showed that score thresholds are spread across the full range of ability levels for children in this sample, although again the spread of items is sparse. PCA results supported the unidimensionality of the seven-item fluency subscale.

Dimensionality of the four subscale structure

Evidence of multidimensionality across the four subscales was found as more than 5% of cases achieved significantly different scores on sets of items from the fluency subscale and all other subscales, and also for sets of items from the ROM and dexterity subscales.

Relationship between subscale scores

Moderate to high, statistically significant (p<0.001), correlations were recorded between scores on ROM, dexterity and accuracy subscales: ROM/accuracy 0.74, ROM/dexterity 0.80, and accuracy/dexterity 0.81. The fluency subscale, however, showed lower correlations with the other subscales: fluency/ROM=0.52, fluency/accuracy 0.43, and fluency/dexterity 0.45.


  1. Top of page
  2. Abstract
  3. Method
  4. Results
  5. Discussion
  6. Conclusion
  7. Acknowledgments
  8. References

Rasch analysis, used in this study to further investigate the psychometric properties of the Melbourne Assessment, identified aspects of the tool that required modification. The scale was found to be multidimensional, with support found for four discrete unidimensional subscales, each measuring a distinct attribute of movement quality. This differs from the original assumption made at the time of developing the tool that all scores could be summed to provide one total score to measure upper limb movement quality. The revised version of the scale (Melbourne Assessment 2) resulting from this study, provides a measure of a child's quality of unilateral upper limb movement across four separate subscale scores: (1) range; (2) accuracy; (3) dexterity; and (4) fluency.

Rasch analysis identified a number of additional modifications needed to the Melbourne Assessment. Seven of the original 37 scores exhibited disordered response formats requiring rescoring, and 11 pairs of scores revealed high residual intercorrelations, suggesting item redundancy. Based on these results, and clinical judgement, the Melbourne Assessment was reduced in size, and restructured, resulting in a 9-score active range of movement subscale, an 8-score accuracy subscale, a 7-score fluency subscale, and a 6-score dexterity subscale. The final modified version of each subscale demonstrated good internal consistency, with high PSI values ranging from 0.81 and 0.92. In addition, the absence of evidence of differential item functioning for children's sex or age provides support that Melbourne Assessment 2 scores are not influenced by sex or age, thus clinicians can be confident that any differences in scores observed over time or between children of differing ages is a result of variations in their level of movement quality. Subscales showed overlap, but provide measurement of four clinically discrete elements of movement quality.

The refinements identified from this study have now been implemented in the production of a revised version of the tool, the Melbourne Assessment 2.[23] The Melbourne Assessment 2 is available for purchase from The Royal Children's Hospital, Melbourne via the website ( The Melbourne Assessment 2 comprises 14 tasks and 30 scores. The scaling refinements incorporated in the revised tool improve the measurement potential of the scale and enhance clinician interpretation of test scores. The restructure of the scale into four separate, unidimensional subscales enables measurement of the specific qualitative elements of movement range, accuracy, dexterity, and fluency. Summing of scores into four subscale total scores also allows clinicians to identify the particular element(s) of movement quality a child might be experiencing greatest difficulty with, or where an intervention may have the greatest impact.

The finding that 10 score items across the four subscales showed DIF for different raters was not unexpected. Assessment of inter-rater agreement undertaken on the 37 scores of the original Melbourne Assessment had also identified 11 scores for which inter-rater agreement was moderate or poor (weighted kappa values ranging between 0.39 and 0.60).[1] The presence of variation in raters’ scorings of these items creates a case for providing raters with consistent training in reliable scoring of the tool. Web-based resources which provide demonstration in the administration of assessment tasks and training in the scoring of individual movements are now accessible to purchasers of the tool. Guidelines are also provided for establishing reliable administration and scoring of the tool. If raters achieve the recommended level of scoring reliability, scores from different raters can be used with confidence in clinical and research applications.

The construction of the Melbourne Assessment 2 has arisen from the findings of the Rasch Analysis implemented in this study. It is now important that a study be undertaken with data collected using the Melbourne Assessment 2 to confirm these findings. Linacre[22] also states that the sample size for a definitive analysis should be 20 times the number of items in the scale (i.e. 20×30 items in the refined scale), suggesting that future testing using larger samples is needed to support the findings of this study. In addition, other aspects of the scale also require further investigation. Assessment is needed of the ability of the four subscales to discriminate between differing levels of limitation in children's manual abilities. If subscale scores are found to correlate strongly with the level of upper limb function, it may be that the quantitative scores of the Melbourne Assessment 2 could be used to support applications for educational funding or therapy services for individual children. Subscale scores may also provide clinicians and researchers with an objective means of grouping children of like ability together in studies of treatment effectiveness or long-term follow-up.

Further research is also required to determine the ability of the scale of the Melbourne Assessment 2 to detect small, but clinically important, changes in a child's unilateral upper limb movement quality. A study to establish the standard error of measurement and smallest detectable difference, as well as the amount of change needed for each subscale to show a ‘clinically important change’ is currently being undertaken.


  1. Top of page
  2. Abstract
  3. Method
  4. Results
  5. Discussion
  6. Conclusion
  7. Acknowledgments
  8. References

The application of Rasch analysis to further assess the internal construct validity of the Melbourne Assessment identified a number of psychometric weaknesses in the original scale. Modifications to the structure and scaling of item scores resulted in an improved four subscale tool (the Melbourne Assessment 2) with each subscale showing fit to the Rasch model, high internal consistency and evidence of unidimensionality. Further testing, on larger samples and confirmation of the sensitivity of the Melbourne Assessment 2 to detect clinically important changes in a child's upper limb movement quality is now needed.


  1. Top of page
  2. Abstract
  3. Method
  4. Results
  5. Discussion
  6. Conclusion
  7. Acknowledgments
  8. References

This study was supported by a La Trobe University Australian Postgraduate Award and Murdoch Childrens Research Institute Scholarship. The authors also sincerely thank their valued colleagues Josie Duncan, Margaret Wallen, Helen Bourke-Taylor, Catherine Elliott, and Siobhan Reid for their gracious sharing of Melbourne Assessment data.


  1. Top of page
  2. Abstract
  3. Method
  4. Results
  5. Discussion
  6. Conclusion
  7. Acknowledgments
  8. References
  • 1
    Randall M, Carlin J, Chondros P, Reddihough D. Reliability of the Melbourne Assessment of Unilateral Upper Limb Function. Dev Med Child Neurol 2001; 43: 7617.
  • 2
    Randall M. Development of ‘The Melbourne Assessment of Unilateral Upper Limb Function’: a quantitative test of quality of movement in children with cerebral palsy. In: Fong HD, editor. Focus on Cerebral Palsy Research. New York: Nova Science Publishers, 2005: 93122.
  • 3
    Randall M, Johnson L, Reddihough D. The Melbourne Assessment of Unilateral Upper Limb Function: Test Administration Manual. Melbourne: Royal Children's Hospital, 1999.
  • 4
    Randall M, Imms C, Carey L. Establishing validity of a modified Melbourne Assessment for children aged 2–4 years. Am J Occup Ther 2008; 62: 37383.
  • 5
    Randall M, Imms C, Carey L. Further evidence of validity of the modified Melbourne Assessment for neurologically impaired children aged 2 to 4 years. Dev Med Child Neurol 2012; 54: 4248.
  • 6
    Motta F, Stignani C, Antonello C. Upper limb function after intrathecal baclofen treatment in children with cerebral palsy. J Pediatr Orthop 2008; 28: 916.
  • 7
    Ozer K, Chesher SP, Scheker LR. Neuromuscular electrical stimulation and dynamic bracing for the management of upper-extremity spasticity in children with cerebral palsy. Dev Med Child Neurol 2006; 48: 5593.
  • 8
    Yasukawa A, Patel P, Sisung C. Pilot study: investigating the effects of Kinesio Taping in an acute pediatric rehabilitation setting. Am J Occup Ther 2006; 60: 10410.
  • 9
    Krumlinde-Sundholm L, Eliasson A-C. Development of the Assisting Hand Assessment: a Rasch-built measure intended for children with unilateral upper limb impairments. Scand J Occup Ther 2003; 10: 1626.
  • 10
    Pallant J, Tennant A. An introduction to the Rasch measurment model: an example using the Hospital Anxiety and Depression Scale (HADS). Br J Clin Psychol 2007; 46: 118.
  • 11
    Tennant A, Conaghan PG. The Rasch measurement model in rheumatology: what is it and why use it? When should it be applied, and what should one look for in a Rasch paper? Arthritis Rheum 2007; 57: 135862.
  • 12
    Haley SM, Coster WJ, Ludlow LH, Haltiwanger JT, Andrellos PJ. Pediatric Evaluation of Disability Inventory (PEDI). Boston: New England Medical Center Hospitals, 1992.
  • 13
    Palisano RJ, Rosenbaum P, Bartlett D, Livingston MH. Content validity of the expanded and revised Gross Motor Function Classification System. Dev Med Child Neurol 2008; 50: 74450.
  • 14
    Young N, Williams J, Yoshida K, Wright J. Measurement properties of the activities scale for kids. J Clin Epidemiol 2000; 53: 12537.
  • 15
    Penta M, Thonnard J-L, Tesio L. ABILHAND: a Rasch-built measure of manual ability. Arch Phys Med Rehabil 1998; 79: 103842.
  • 16
    Conaghan P, Emerton M, Tennant A. Internal construct validity of the Oxford Knee Scale: evidence from Rasch measurement. Arthritis Rheum 2007; 57: 13637.
  • 17
    Wallen M, O'Flaherty SJ, Waugh MC. Functional outcomes of intramuscular botulinum toxin type a and occupational therapy in the upper limbs of children with cerebral palsy: a randomized controlled trial. Arch Phys Med Rehabil 2007; 88: 110.
  • 18
    Elliott C, Reid S, Alderson J, Lloyd D, Elliott B. Quantitative analysis of upper limb function in children with and without cerebral palsy: comparison of postural strategies and upper limb functional movements in children aged 7–13 with and without cerebral palsy (CP) using 3D motion analysis. The Australasian Academy of Cerebral Palsy and Developmental Medicine, 2004 19–20 March; Melbourne.
  • 19
    Bourke-Taylor H. Melbourne Assessment of Unilateral Upper Limb Function: construct validity and correlation with the Pediatric Evaluation of Disability Inventory. Dev Med Child Neurol 2003; 45: 926.
  • 20
    Reid S. Effects of an upper limb strength training program for children with cerebral palsy. (Thesis). University of Western Australia, 2005.
  • 21
    Nunnally JC. Psychometric Theory, 2nd edn. New York: McGraw-Hill, 1978.
  • 22
    Linacre J. Sample size and item calibration stability. Rasch Measur Transct 1994; 7: 328.
  • 23
    Randall M, Johnson L, Reddihough D. The Melbourne Assessment 2: A Test of Unilateral Upper Limb Function [Online assessment tool with instructional component]. Melbourne: The Royal Children's Hospital, 2011.