All children registered at two mainstream schools (total of 517 students) in the West Yorkshire region (United Kingdom) were invited to participate in the research. On the first testing session, we collected data from 495 children (235 male, 260 female, age range 3 years 2 months to 12 years 2 months, mean age = 7 years 2 months) on the manual control measure (22 children were either absent from school or opted out of testing). On the second session, we randomly selected classes across both schools within each year group and measured postural control in students from these classes. The final sample, with data recorded on postural stability in stance and manual control in sitting, comprised of 278 children (134 male, 144 female, age range 3 years 2 months to 11 years 10 months, mean age = 7 years 8 months). There were no children with severe disability within the schools, but it is probable that there were a number of children with neurodevelopmental problems (e.g. autism). We randomly sampled classes across both schools and did not exclude any child from these classes from participating, so our data are likely to be reasonably representative of a typical UK school population. Ethical approval for this study was obtained from the University of Leeds Ethics and Research committee.
Manual control measures
Participants completed a battery of fine motor tests called the Clinical Kinematic Assessment Tool (C-KAT). The C-KAT software, which is used to present the battery (Culmer et al. 2009), was created using a software development environment: LabVIEW (version 8.2.1, National Instruments TM) deployed on a Toshiba digitizing tablet portable computer (Toshiba Portégé, 14″ screen: 260 × 163 mm, 1,280 × 800 pixels, 32 bit colour, 60 Hz refresh time). The tablet’s screen provides a horizontal surface (in landscape orientation) similar to writing with a pen and paper using a stylus as an input device. Participants were seated at a desk with the tablet placed in front of them, 10 cm in from the table edge.
The C-KAT battery comprised of three visuo-spatial subtests completed in the following fixed order: tracking: participants placed their stylus, for 2 s, on a stationary green dot (10 mm in diameter) presented on the tablet screen, triggering it to begin moving around the screen by doing so. The dot’s movements followed a horizontal ‘figure-of-8’ spatial pattern (see Fig. 1a) continuously for 84 s (nine revolutions) and as it did so participants were asked to try and keep their stylus as close to the centre of it as possible. The motion was described by two oscillating sinusoidal waveforms in the axes of the screen.
These had a 2:1 ratio with respect to their frequencies and amplitudes, resulting in a ‘figure-8’ of 55 mm in height and 110 mm in width. This trial was repeated twice, first with no guide and then with a background ‘guideline’ provided to give the participant additional information on the spatial path the dot followed (see Fig. 1a). In each trial, the target dot’s speed immediately increased after a trio of revolutions. Starting ‘slow’ and increasing to ‘medium’ the ‘fast’ speeds (velocities of 42, 84 and 168 mm/s, respectively).
Aiming: in this subtest, participants were required to move from one target dot to another without lifting the stylus from the screen. The trial began when participants placed the stylus on the ‘start’ button for 2 s; this prompted the first target dot to appear. When the first target dot was reached, it disappeared and another appeared in a different location on the screen. Participants were instructed to move successively from one target dot to the next until they had made 75 aiming movements, after which a ‘finish’ block appeared. The trial terminated once they moved their stylus to it. Within these 75 movements, the participants moved successively to five fixed targeted locations (numbered in Fig. 1b) before being re-cued to the first location and repeating this sequence (i.e. participants resultant movements approximated drawing the star shape outlined in the fourth panel of Fig. 1b fifteen times in a row). Within the final 25 movements, pseudo-randomly distributed throughout them, six of the target dots would appear to ‘jump’ to the next target position before the participant had reached it; assessing participant’s capacity for making online corrections. ‘Baseline’ trials were denoted as the first 50 trials where no jump events had occurred. ‘Embedded’ trials were the normal trials in the last 25 movements with ‘jump’ trials being when the targets changed location mid-movement.
Tracing: this final subtest required participants to move their stylus along a static path between designated start and finish points, trying to trace as accurately as possible, whilst staying within the guidelines of the path (see Fig. 1c). The trial was initiated and the whole path appeared when the stylus was placed on the ‘start’ block for 1 s. A hollow box moved along the pattern to provide a guide for the ideal speed of movement. To try to avoid either very slow accurate tracing or rapid inaccurate tracing the participants were instructed to keep the stylus within the box throughout the trial (time for box to move from start to finish = 35 s). The trial was finished when the ‘finish’ block was reached. Participants completed three repetitions of two different tracing paths. The two paths were presented in alternate trials, totalling six tracing trials within the subtest.
Postural movement was calculated using a custom built motion-capture rig and force platform, specifically designed to be used in schools. The rig comprises a stereo-camera motion-capture system that measures the 3D position of an infra-red (IR) diode at 60 Hz. A battery-powered IR diode was placed on a light, inflexible plastic brace placed on the child’s head, which provided a measure of head movement (HM). In addition to the measure of movement at the head, a Nintendo Wii Fit board was used to simultaneously monitor the participant’s centre of pressure (COP) at 60 Hz.
On another testing session (separated by at least 2 days from the manual control assessment) participants were asked (i) to stand with their feet shoulder width apart with their eyes closed for 30 s, then (ii) to stand with their feet shoulder width apart with their eyes fixed on a target placed 1 m away at eye level. During both conditions (hereafter referred to as ‘eyes closed’ and ‘eyes open’, respectively), the participants were constantly observed to ensure compliance. HM data were filtered using a 10 Hz dual-pass Butterworth filter, and the COP data were filtered using a wavelet filter described in Flatters et al. (2014a). After filtering, the 3D and 2D path lengths subtended by the IR diode and COP, respectively, were calculated (in mm) for each 30 s trial. Allowing time for measurement equipment set-up and rest-breaks, this session lasted approximately 3 min.
Defining outcomes measures
Postural measures: we wished to analyse both head movement (HM) and centre of pressure (COP) variables separately and also to create a composite measure of these two variables, to provide an index of overall postural stability. Shapiro–Wilks tests indicated normality assumptions were met for HM and COP measures (p’s > 0.05). Thus, z-score transformations could be used to convert participants’ HM and COP scores to a unified scale. This then allowed for a mean of these two scores to be calculated, giving a ‘composite posture’ score. In order to control for the well-established age differences in motor control, we experimented with three different approaches to standardisation. First, each participant’s scores on HM and COP were standardised in relation to their means and standard deviations within the respective school years within the sample—(Years 1, 2, 3, 4, 5 and 6). Second, participants were grouped based on the year of birth and standardised in relation to each group’s means and SDs. For this, the following groups were used: 2009–2008; 2007–2006; 2005–2004; 2003–2002; and 2001–2000. Finally, scores were standardised relative to the entire sample and age included as a covariate in subsequent statistical analysis. Irrespective of the approach taken the same pattern of results was observed during analysis, demonstrating the robustness of the results. For conciseness, from here on, we only report results in which z-scores were calculated based on the first approach described (standardisation by school year). Note that age standardisation, when z-scoring, was also used with respect to all the following fine motor control measures.
Manual control measures: for every trial within every subtest of the C-KAT battery, the KAT software recorded the position of the stylus at a rate of 120 Hz and at the end of each testing session, these raw positional data were filtered using a 10 Hz dual-pass Butterworth filter. From these filtered time-series, a wide variety of spatial, temporal and frequency-based kinematic metrics could be calculated (see Culmer et al. 2009). To avoid data mining, only a specific a priori determined subset of all the potential variables was analysed. These were selected on the following criteria: (i) variables had to be normally distributed or responsive to transforms that enforced this (e.g. reciprocal, natural log). This legitimated z-score transformations of such variables, in turn allowing multiple outcomes relating to given subtests to be averaged to give composite scores for each subtest. (ii) Variables had to be at least moderately correlated with age (r > 0.3), implying they were a meaningful measures of some characteristic of the development of fine motor control. (iii) The variables needed to relate to a measure that directly indicated performance on the task (as per explicit instructions to the children and implicit within the task design). Application of these criteria meant that the following kinematic variables were selected as outcomes for the respective C-KAT subtests.
For the tracking subtest, the spatio-temporal accuracy of the participant at each sampled time point was measured as the two-dimensional distance from the stylus to the dot centre (i.e. root-mean-square error [RMSE]). Across data points, a mean value for RMSE was calculated for each of the six experimental conditions (i.e. one per speed [slow, medium, fast] for both background conditions [without guideline, with guideline]). To capture the spatial accuracy of the shape subtended during pursuit, a second metric (path accuracy [PA]) was calculated, as the mean of the minimum distances from input to the ideal path across all data points (within each condition). These twelve measures of RMSE and PA (i.e. two metrics, three speeds and two background conditions) had reciprocal transforms applied to normalise the distributions and were then converted to standardised z-scores. A composite score for tracking was calculated as the arithmetic average of these twelve z-scores.
For the aiming subtest, median values for both the reciprocal movement time (MT) and the log normalised jerk (NJ) of the aiming movements made within each of the three experimental conditions were calculated separately (i.e. baseline, embedded and jump conditions). These six values were then z-score standardised and a composite score for the aiming subtest calculated by averaging these six standardised scores.
For tracing, the minimum 2D distance between the idealised reference path and the stylus was calculated for each sampled time point within a trial. For each of the six trials, the arithmetic mean of these values was taken as a measure of shape reproduction accuracy, termed path accuracy (PA). Despite continuous monitoring of the participants by the experimenter, a number of participants were unable to adhere to the instructions to stay within the moving on-screen box with their stylus, whilst tracing. Thus, interpretation of participants’ accuracy during these trails was potentially confounded by a lack of standardisation for their speed. Consequently, in order to control for variation in time, a ‘penalised path accuracy’ (PPA) metric was calculated that adjusted PA score with respect to movement time (MT). The ideal trial time, including the 1 s delay at the onset of the trial, was 36 s. To normalise path accuracy for task time, PA was inflated by the percentage that participants’ actual MT deviated from the ideal 36 s value. For each of the six trials, the PPA value a reciprocal transformation was applied to normalise its distribution before being z-score transformed and a composite performance score for this subtest calculated as the mean of these six values.
Finally, an overall battery score for the C-KAT was calculated as the arithmetic mean of the respective tracking, aiming and tracing composite scores.