Adolescent Awkwardness: Alterations in Temporal Control Characteristics of Posture with Maturation and the Relation to Movement Exploration.

A phenomenon called adolescent awkwardness is believed to alter motor control, but underlying mechanisms remain largely unclear. Since adolescents undergo neurological and anthropometrical changes during this developmental phase, we hypothesized that adolescents control their movements less tightly and use a different coordinative structure compared to adults. Moreover, we tested if emerging differences were driven by body height alterations between age groups. Using 39 reflective markers, postural movements during tandem stance with eyes open and eyes closed of 12 adolescents (height 168.1 ± 8.8 cm) and 14 adults were measured, in which 9 adults were smaller or equal than 180 cm (177.9 ± 3.0 cm) and 5 taller or equal than 190 cm (192.0 ± 2.5 cm). A principal component analysis (PCA) was used to extract the first nine principal movement components (PMk). The contribution of each PMk to the overall balancing movement was determined according to their relative variance share (rVARk) and tightness of motor control was examined using the number of times that the acceleration of each PMk changed direction (Nk). Results in rVARk did not show significant differences in coordinative structure between adolescents and adults, but Nk revealed that adolescents seem to control their movements less tightly in higher-order PMk, arguably due to slower processing times and missing automatization of postural control or potential increases in exploration. Body height was found to not cause motor control differences between age groups.


Introduction
During maturation, adolescents are suggested to experience a period of altered motor control called 'adolescent awkwardness' [1]. During this phase of development, it is assumed that adolescents experience neurological, multisensory and information processing changes resulting in less efficient postural control compared to adults [2]. While postural control deficits in 10-14-year-old individuals are clearly evident from reduced performance in standing balance and dynamic functional tests [3,4], underlying differences in the behavior of the control system remain largely unclear. Investigating movement strategies and differences in the temporal structure of how movement strategies are controlled during balance tasks, offers a novel approach for assessing differences in sensorimotor control between adolescents and adults.
Postural control can be studied using an inverted pendulum model, characterized by its sway amplitude and sway frequency [5,6]. Efficient postural control develops during maturation from strategies during standing balance tasks with a specific focus on the number of interventions that the postural control system deploys in each movement dimension to maintain balance. We counted as an intervention when the movement acceleration changed direction, since acceleration changes indicate changes in muscle activation. Variance in movement strategies between the use of dominant and non-dominant leg during one leg stance were found [36] and might be correlated with a higher injury risk in downhill skiers [37]. Alterations between adults and elderly were found in the number of interventions of the control system, whereas the elderly were suggested to control less tight in some, but not in every movement dimension during tandem stance [38].
The purpose of this study was to investigate differences in stable postural control between adolescents and adults during tandem stance through application of a PCA on 3D kinematic data quantifying full-body balancing movements. The primary goal was to compare (1) the coordinative structure of postural movements, specifically the contribution of different movement dimensions used by adolescents and adults during stance; and (2) the control of individual movement strategies. It was expected that adolescents show more variable movement strategies to maintain balance (hypothesis 1), specifically an increase in anterior-posterior movement is expected to account for exploration in a dimension with low task relevance following the minimal intervention principle. Furthermore, if differences exist between control strategies in adolescents and adults, we expect that adolescents show fewer postural interventions (less tight control) compared to adults as an effect of the increased exploration function (hypothesis 2). However, it should be noted that any significant differences found between adolescents and adults could potentially just be the effect of body height on coordination. An exploratory analysis was done to investigate this potentially confounding effect.

Participants
A convenience data sample of 26 participants (Table 1), originally recorded for another project [39] was analyzed for the purpose of this study (Supplementary Materials File S1). Exclusion criteria were diagnosed injuries, concussion, or other neurological disorders within the last six months, as well as self-reported problems concerning joint, tendons or muscles. Further, in the current study only data sets were included in which participants were able to stand without any kind of visible compensatory movement that led to a change in position. The current study specifically aimed at comparing motor control between adults and adolescents in stable trials. Visible instabilities, which can be expected to trigger additional postural control mechanisms, were avoided. Therefore, making a small step, lifting the toe or taking away the hand off the hip led to exclusion. The reason for including only such trials was to ensure that only data of the tandem stance during "successful" motor control without the need of compensation were analyzed. The included participants were divided into two groups for maturation status (12 adolescents and 14 adults). To test for the effect of body height, adults were further separated into two height groups; n = 9 smaller or equal than 180 cm, and n = 5 taller or equal than 190 cm (incidentally none of the recruited participants were between 181 cm and 189 cm, which was the main motivation behind this split). Volunteers were healthy and physically active persons. All three groups differed significantly (p < 0.05) in height and weight, only in age no differences occurred between the adult groups. In the Supplementary Materials, a file with more detailed subject characteristics can be found.

Measurement Procedure
Participants were instructed to stand as still as possible on a marked area on the ground in tandem stance (one foot in front of the other, the toe of the back foot slightly touching the heel of the foot in front). Tandem stance was selected, because it was considered a moderately challenging balance exercise for both participant groups. Participants were free to decide which foot they want to place in front and were advised to choose the one on which they felt most confident while keeping balance. Two trials were conducted, always with the same foot in front. For the first trial, they were asked to focus their gaze on a target cross placed 5.5 m in front of them at 1.75 m height and to stand as still as possible for 60 s. Visible compensatory movements were avoided. Hence, making a step, lifting the toe or foot, or moving the hand off the hip led to the trial being excluded and the participant was asked to re-do the trial. For the second trial, they were asked to again stand as motionless as possible, but with closed eyes. The eyes-closed trials were conducted for only 30 s, since pilot testing had revealed problems (especially among adolescents) with maintaining balance for 60 s with eyes closed in tandem stance. Between the trials, the volunteers were given rest periods of at least 90 s, in which they were free to move around in the room. The order of the two trials was not randomized, since the eyes open trial was used to evaluate if participants are stable enough for the eyes closed trial as well. Therefore, we dispensed randomization to prevent falls in the eyes closed trial.

Instrumentation
Kinematic movement data of the volunteers were recorded at 250 Hz. An 8-camera Vicon motion tracking system (Vicon Motion Systems Ltd., Oxford, UK) collected the 3D coordinates of 39 retro-reflective markers that were attached on the participant's skin or tight sport-clothing using double-sided tape. The markers were positioned on anatomical landmarks in accordance with the "Full-Body Plug-In Gait" provided by Vicon. Modified sweatbands were used to attach the markers on the head and on the wrist. The Vicon software (Vicon Nexus, Version 2.2.3; Vicon Motion Systems Ltd., Oxford, UK) was used for reconstructing the marker trajectories.

Data Processing
MatLab R2019b (The Mathworks Inc., Natick, MA, USA) and the "PManalyzer" [40] were used to process the kinematic data. In all trials, the first 5 s were omitted to avoid settle-in effects, nine asymmetrical markers were removed, and gaps in marker-trajectories were reconstructed using a PCA-based procedure [41,42]. Then, animated stick figures were created for each trial in original motion. These video-representations were again screened for movements listed in the exclusion criteria, e.g., making a small movement with one foot or slightly shifting a hand off the hips. If minor, yet visible, compensatory movements were recognized, data were excluded in this step to analyze 'successful' motor control only. For each trial, a 16 s period was extracted, in which participants were stable without any disallowed movements.
Then, data of participants with the left foot in front (n = 6) were mirrored and relabeled, such that for the analysis all participants appeared to stand with their right foot in front [38]. The normalization procedure before performing the PCA contained three steps; (1) the subject mean was first subtracted to center the data [34]; (2) data were normalized to the mean Euclidean distance [38,43]; and (3) the data were weighted according to the relative weight contribution represented by each marker [44]. The data sets from all participants were concatenated to create one 208,000 × 90 input matrix (lines: 26 participants, two trials of 16 s recorded at 250 Hz; columns: three-dimensional (3D) coordinates of 30 markers). The rows of the matrix contained the available posture information about the participant's position at a certain time point and were interpreted as 90-dimensional posture vectors [33,[45][46][47]. After performing the normalization and concatenation of the data from all volunteers and all trials, one PCA could be conducted for the whole dataset, allowing a direct comparison of the resultant movement components between all included trials and all participants. The PCA was calculated as eigenvector decomposition of the covariance matrix of the data. The resultant eigenvectors PC k (where k specifies the order of the eigenvector) thus form a new basis in the vector space of the posture vectors. As kinematic movement data were analyzed, each PC k describes one linear pattern of correlated marker movements. Two other output variables of the PCA are eigenvalues EV k and scores PP k (t). The scores represent the posture vectors in the new PC k -basis and are obtained through a basis transformation of the original data onto the new basis. Therefore, the scores can be interpreted as "principal positions" PP k (t), since they are a representation of positions in posture space [36,38,42,48,49]. The principal position PP k (t) together with the eigenvector PC k define one (the k-th) component of the whole postural movement. We call these movement components "principal movements" (PM k ) as they describe linear representations of the involved movement dimensions. How much contribution each PM k provided to the overall postural variance across all participants is indicated by the EV k [33,[45][46][47]49].
To perform further analyses with an analogue and subject-specific variable, we calculated the relative variances rVAR k from PP k (t). The rVAR k quantify how much each PM k contributed to the whole postural variance for each trial [34,48]. The variable rVAR k was evaluated to identify differences in movement strategy and movement dimensions between the trials and participants (Hypothesis 1).
The idea of the PP k representing the positions of the body can be expanded by (double) differentiation to calculate the "principal accelerations" PA k (t) of each PM k . By counting the "number of zero crossings" (N k ) in the PA k time-series, one obtains a variable to investigate the temporal characteristics of how the movement components are controlled (Hypothesis 2) [38]. Each PA k can be interpreted as the result of the interplay between agonistic and antagonistic muscle action. If more N k are present, a tighter movement control can be suggested as the sensorimotor system makes more frequent corrections. Reductions in N k may be related to longer processing times for postural adjustments or could indicate that a movement component is not controlled as tightly, for example, because the control of PM k is deprioritized [36][37][38].

Statistics
The two dependent variables in the current study were rVAR k to detect differences in movement strategy and N k to investigate alterations in the 'tightness' of motor control. Shapiro-Wilk tests were used to check for normal distribution of the data for each group, task, and movement component.
The primary goal of the statistical analysis was to investigate effects of the between-subject factor "group" (adolescent/adult) and the within-subject factor "eye condition" (eyes closed/eyes open) and potential interaction effects on rVAR k and N k . Since rVAR k were not normally distributed, separate Mann-Whitney U tests were performed to compare rVAR k in each movement component k between adolescents and adults and between eye conditions. "Group", "eye condition" and interaction effects of the normally distributed N k were tested using a repeated-measures MANOVA to implement all variables in one statistical model. In the case of significant main effects, subsequent t-tests were computed. The effect sizes were calculated using Rosenthals r for non-parametric tests [50] and partial eta square η p 2 for parametric tests.
Since prerequisites for an ANCOVA calculation were not fulfilled, the secondary goal of determining potential effects of body height was investigated by splitting the adult group into two subgroups: (1) Adults smaller or equal than 180 cm (Adults ≤ 180) and (2) adults taller or equal than 190 cm (Adults ≥ 190). Effects of the new three-leveled "group" variable and of the within-subject factor "eye condition" on rVAR k and N k were analyzed similarly to the first approach: since rVAR k were not normally distributed, separate Kruskal-Wallis tests were performed. Effects on the normally distributed N k were tested using a repeated-measures MANOVA.
The p-values corresponding to all post-hoc or pairwise comparisons for individual movement components were adjusted according to the Holm-Bonferroni correction to lower the risk of type-I error. The newly calculated thresholds were p < 0.0055, p < 0.0063, p < 0.0071, and p < 0.0083 in the first, second, third, and fourth rank, respectively. All statistical tests were performed using SPSS (IBM SPSS Statistics, Version 24, SPSS Inc., Chicago, IL, USA).

PCA Results
The first nine analyzed movement components explained 98.37% of the overall movement variance during the balancing trials. A video sequence, which can be found in the Supplementary Materials (Videos S1-S3), contains visualizations of these movement components. The extreme positions of each component along with the respective eigenvalues EV k and qualitative descriptions are presented in in Figure 1.

PCA Results
The first nine analyzed movement components explained 98.37% of the overall movement variance during the balancing trials. A video sequence, which can be found in the Supplementary Material (Video S1-S3), contains visualizations of these movement components. The extreme positions of each component along with the respective eigenvalues EVk and qualitative descriptions are presented in in Figure 1.   (Table 2), results were inconclusive. In three movement components (PM 4 , PM 5 , PM 9 ), medium effect sizes and small p-values were observed, however, these p-values did not meet the threshold of significance after the Holm-Bonferroni correction. The results for the eyes-closed trials (Table 3) did not suggest a difference in the coordinative structure between adolescents and adults. The comparison of rVAR k between the eyes open and eyes closed condition (without Table) revealed p-values smaller than 0.05 in rVAR 2 (z = −2.222, p = 0.026, n = 26) and in rVAR 6 (z = −2.197, p = 0.028, n = 26), however, significance was not reached when applying the Holm-Bonferroni correction.

Temporal Control Characteristics-N k
When analyzing the number of zero crossings (N k ), the group effect between adults and adolescents was significant (F(9,16) = 5.587, p = 0.001, η p 2 = 0.759). Significant differences in N k were found in some but not in all PM k as listed in Table 4. Adults always presented a higher N k than adolescents except for N 1 . There was no significant group × eye condition interaction effect (F(9,16) = 0.381, p = 0.927, η p 2 = 0.177), nor an eye condition effect (F(9,16) = 1.710, p = 0.167, η p 2 = 0.490).

Relative Variance-rVAR k
Preliminary differences between adolescents and small and tall adults (according to p < 0.05) could be observed in rVAR 3 , rVAR 5 and rVAR 9 but did not meet the threshold of significance after the Holm-Bonferroni correction. Except for the lower-order components, rVAR 1 and rVAR 2 , adolescents appeared on average to contribute more rVAR k compared to smaller adults while there was little difference compared to taller adults (Table 5). In the eyes-closed trial (without Table), no significant differences were found. The group effect between small adults, tall adults and adolescents was statistically significant (F(18,32) = 2.943, p = 0.004, η p 2 = 0.623). Individual comparisons are presented in Table 6 and visualized in Figure 2. There was no significant group × eye condition interaction effect (F(18,32) = 0.892, p = 0.591, η p 2 = 0.334), nor an eye condition effect (F(9,15) = 1.823, p = 0.146, η p 2 = 0.522).  2. Visualization of the differences in "number of zero crossings" between adolescents (dark blue), adults smaller or equal than 180 cm (blue) and adults taller or equal than 190 cm (light blue) in the different PMk. Bold PMk with hashtag (#) indicate statistically significant main effects, asterisks (*) between the bars indicate significant post-hoc differences between groups.

Main Results-Motor Control Differences between Adolescents vs. Adults
The current study investigated alterations in the coordinative structure of postural movements and in the control of individual movement strategies between adolescence compared to adulthood. Figure 2. Visualization of the differences in "number of zero crossings" between adolescents (dark blue), adults smaller or equal than 180 cm (blue) and adults taller or equal than 190 cm (light blue) in the different PM k . Bold PM k with hashtag (#) indicate statistically significant main effects, asterisks (*) between the bars indicate significant post-hoc differences between groups.

Main Results-Motor Control Differences between Adolescents vs. Adults
The current study investigated alterations in the coordinative structure of postural movements and in the control of individual movement strategies between adolescence compared to adulthood. Considering our first hypothesis, the expectation of differences in movement strategies between adolescents and adults, which would mostly manifest in anterior-posterior direction, can only be partially supported. Some significant differences (p < 0.05) were identified between conditions for a number of relative variance variables. However, after controlling for alpha error, the statistical significance disappeared. However, results are in accordance with our second hypothesis; the expectation that less tight control would be shown by adolescents. This was supported by the differences in the N k between adolescents and adults for movement components PM 2 , PM 5 and PM 6 . It should be noted that these PM k describe movement components predominantly representing motion in anterior-posterior direction, as it was hypothesized. This finding is strengthened by the fact that it remained significant even with a relatively conservative correction for multiple comparisons.
Showing differences in selected PM k but not in all analyzed PM k agrees well with the minimum intervention principle [23,24]. It is stated that motor control intervenes if task-relevance is at risk but allowing movement variability as long as the primary goal is not affected negatively. Considering that the significances appeared in PM k representing movements in anterior-posterior direction, which is suggested to be more stable in tandem stance than the medio-lateral direction and therefore is less task-relevant to control, results appear to be conclusive. Moreover, they agree with previous studies on effects in postural control, again revealing differences in some but not all PM k , e.g., due to aging [38]. Contrary to results in the current study, older individuals presented differences especially in task-relevant movement components like medio-lateral ankle sway [38], which can be interpreted as a sign of changes in sensorimotor control due to e.g., degeneration of structures [51]. On the other hand, the differences in our results might be driven by the still developing automatisms in adolescents [13]. Adolescents may control task-relevant movement components in a rather reliable way; however, redundant components show alterations leading to the observed variability. Results are further in line with previous studies investigating leg dominance [36,37,52], sensory perturbation [48], or dual tasking [39] and its effect on postural control, presenting some but not all PM k to be affected. Although adolescents are suggested to heavily rely on visual information [13], no differences could be observed between the eyes-open and eyes-closed trial. Since the trials were not randomized, a learning or training effect leading to habituation effects might have occurred which would be explainable by the concept of exploration (a concept well established in perception-action theory [53,54]). Then, however, the absence of any "eye condition" effect could be interpreted as a result of exploration, which in this case seems to be sufficient to maintain balance. Since our study design is not well-suited to answer this question, this assumption remains speculative.
The altered activity in anterior-posterior movement components in adolescents' tandem stance confirms the expectation that exploration would be increased in adolescents. To the knowledge of the authors, this study was the first to hypothesize that exploration patterns would follow a pattern predictable by optimal feedback control theory. This raises the question whether these theories are of compatible nature. It is a general limitation of optimal control theory that control laws are only defined during movement; perhaps perception-action theory could be used as a supplemental theory to answer the question of how we decide to engage in these movements [53,55,56]. In contrast, the control laws defined by optimal control theory can provide insight into the weighting of alternative movement plans for ongoing action, a process similar to what has been hypothesized in the affordance competition hypothesis [57]. More work is required to determine the feasibility of combining these conceptual frameworks, which could be in the field of both hypothesis testing and more philosophical work.

Secondary Results-Differences between Adolescents, Smaller and Taller Adults
It was a secondary purpose of this study to analyze whether effects that had been attributed to differences between adolescence and adulthood could also be explained by body height. Thereto, the adult groups were split into two subgroups based on height. Independently of the sensory situation (eyes open or eyes closed), we found that all three significant effects that we had observed between adolescents and the group of all adults (in N 2 , N 5 and N 6 ) could be explained by differences between adolescents and the smaller adults. In fact, we found an additional effect between groups in N 7 . Further, when comparing the number of zero crossings between the three groups ( Figure 2), we found that in all movement components, with the exception of PM 1 , that the N k of both the adolescents and the tall adults were smaller than those of the smaller adults. This suggests that body height might play a role in the control of PM 1 , but developmental stage is more important for the observed differences in postural control in the other PM k . We would reason that very tall individuals sometimes also demonstrate forms of 'motor awkwardness'; in this sense, it is fitting that we find a similar pattern in tall adults and adolescents in our data, although this suggestion is speculative since the differences between tall and less-tall adults were not significant in the current study.
In PM 1 , which is the largest movement component representing 60% of the postural variance, both variables relating to relative variance (rVAR 1 ) and tightness of control (N 1 ) showed changes that would be consistent with the differences in body height (an increase in rVAR 1 and a decrease in N 1 with increasing height), but these differences were not significant. Thus, if body height is assumed to affect postural control during tandem stance, then rVAR 1 and N 1 would be the suitable variables for further investigations.
Finally, we assessed differences in postural control between adolescents and adults when standing with eyes open vs. eyes closed. Previous research had suggested that adolescents rely more on visual sensory information due to a lack of automatized control strategies [13], therefore, different results were expected between the eyes-open and eyes-closed trials. However, the prediction was not confirmed by our results.

Limitations
The relatively small sample size and subsequent division into three groups is a limitation of the current study. Still, significant differences could be observed and effect sizes varied from medium to very large (Tables 2-6) in several of the movement components. This limitation affects particularly the analysis related to body height, as we only had five adults in the 'tall group'. However, the observation that p-values did not increase despite the now smaller sample size is, in our opinion, good evidence for the assertion that the age differences are not driven by differences in body height. Nevertheless, further research with larger sample size is required to examine whether there are effects of body height.
A general limitation to the current study comes from unclarity over definitions of 'awkwardness' available within the research field. We cannot be sure that all adolescent participants were influenced by the phenomenon of adolescent awkwardness, since there is no definition or test for adolescent awkwardness. A longitudinal design would be recommended for a future follow-up study to monitor the maturation process and the associated development of postural control characteristics.
We did not randomize the order of the eyes-open and eyes-closed trials to avoid falls of the participants and the data recording varied in time for the trials. As a consequence, habituation and motor acquisition effects cannot be ruled out, which is a limitation when comparing these conditions.
With the methodological approach used in the current study, over 98% of the postural variance was captured within the nine analyzed principal movements. This indicates the representativeness of the approach to the actual behavior. It could be reasoned that more principal movements would result in an even better fit, however, the authors decided against analyzing more components as this would reduce the interpretability of the results (and all further components explain less than 0.25% of the variance per individual component). However, this implies that the results are an approximation of the actual postural movements, as the whole set of PM k would be necessary to fully reconstruct the original motion.

Conclusions
The current study could find differences in the temporal structure of how postural movement components are controlled between adolescents and adults. These differences were in line with the hypothesis of slower processing of motor commands in adolescents as a phenomenon of "adolescent awkwardness". Our results further suggest that the observed differences in motor control are not a result of the smaller body height of adolescents; on the contrary, tall adults showed in some movement components similar control characteristics to the adolescent group.
Author Contributions: F.W. and P.F. initially conceived the experiment. F.W. designed and conducted the experiment together with F.T., who was involved in the measurement process and gathered parts of the data; F.W. produced and analyzed the results and wrote the first draft of the paper. F.W. and P.F. contributed equally interpreting the results, S.v.A. and M.M. gave valuable input how to write and structure the manuscript. F.W., M.M., S.v.A. and P.F. revised and finalized the manuscript. All authors have read and agreed to the published version of the manuscript.

Funding:
The corresponding author received a scholarship from the University of Innsbruck (funding-code: 2017/PSY/SPORT-23).