Arm dominance is an emergent effect of practice executing complex trajectory shapes required by tools and objects

Significance

Limb dominance is often taken as evidence that the dominant hemisphere is intrinsically better at motor control. We tested an alternative: dominance reflects asymmetric practice with tools and objects requiring precise control of complex trajectory shapes. Using markerless 3D reaching plus geometric trajectory shape analysis, we found little or no dominant advantage during ordinary reaching or when arm dynamics were challenged with wrist weight. In contrast, a forearm-mounted stick revealed a strong dominant advantage, driven by poor nondominant control of the stick-tip trajectory. Crucially, dominance disappeared when writing with a novel effector, the elbow. Practice with either elbow produced symmetric improvements, surpassing nondominant hand performance. These findings reframe skill dominance as learned, tool-specific trajectory control rather than general dynamic superiority.

Abstract

Limb dominance is a human behavioral characteristic with many cultural, practical, scientific, and clinical implications. Yet why the dominant limb performs better across a range of motor skill-requiring tasks remains unanswered. Is it because of an intrinsic hemispheric advantage or instead is it the result of life-long practice with the dominant side? We tested these alternatives using two tasks either cross sectionally or after training. The first was 3D reaching with either an inertial challenge or the need to use a stick-like tool. The second required participants to write with their dominant and nondominant elbows. We applied a geometric analysis to quantify movement-trajectory shape. We show that 1) tool-use unmasks markedly inferior control in the nondominant arm, and this is because tools impose the need to generate unfamiliarly shaped movement trajectories; and 2) there is no general dominant limb motor control advantage, only task-specific experience or practice riding on top of an initial preference. These results reframe dominance as predominantly about learned control of tool kinematics rather than baseline asymmetry in control of limb dynamics.

Most people are right-handed (1–5). What does this mean? Confusingly, it has come to mean two related things (6). The first is the preference to use the right arm and/or hand for most tasks. The second is that the dominant right arm is more skilled than the nondominant left arm at most tasks, e.g., handwriting and throwing (7, 8). It is important that these two aspects of dominance not be conflated. Evidence suggests that preference is present even in utero and predicts which arm will become dominant after birth (9, 10). Thus, we take it as an established empirical starting point that hand preference has biological roots, likely involving genetic and neurodevelopmental factors, and we do not seek to address it any further here. Our question is about the origin of dominance: the superior skill of the preferred hand. Is it also the result of an innate advantage intrinsic to the dominant hemisphere, or is it instead acquired via greater accumulated practice with the preferred limb? These two questions are separable—one can accept a biological origin for preference while remaining agnostic about whether the skill asymmetry that develops on top of it is itself either innate or practice-dependent. Our paper addresses only this second question. A straightforward consequence of preference is that it leads to more practice with the preferred arm, with the result that it becomes more skilled. In this scenario, there is no need to posit an additional dominant hemisphere advantage in motor control; skill is never a general capacity but just a practice effect that manifests in a task-specific manner. Any apparent generality could just be because there is some degree of overlap between tasks in their control requirements, e.g., writing, chopstick use, throwing, and conducting an orchestra all require the dominant limb to take the primary trajectory-shaping role over an object or tool.

A counterhypothesis to “it’s just practice” would be that the preference for the left hemisphere in right-handers is precisely because it has lateralized specializations that make it better for skilled motor control. This is indeed the position put forward by the Dynamic Dominance Hypothesis (DDH) (11), which proposes that the key factor that results in the better performance on the dominant side is control of limb dynamics (11, 12), for which the dominant hemisphere is specialized. A key prediction of this account is therefore that any task imposing a greater intersegmental dynamic challenge should reveal or amplify the dominant advantage—independent of whether that specific task has been practiced asymmetrically. The hypothesis is primarily based on planar reaching tasks, for which marked differences in trajectory shape for the two arms are reported (12, 13). Two additional points need to be made about the DDH. First, it was originally formulated only in terms of what the dominant limb was better at (11, 12, 14), additional considerations about complementary capacities of the nondominant limb came later (15, 16) and are not relevant to the truth of the original hypothesis. Second, all the experiments addressing the DDH involved the arm and not the hand—the apparatus used did not even require hand grip (11, 12). This is important because there tends to be a hand-centric view of both dominance and of tool-use that is misleading. For many tools—such as an axe, a lathe, a violin bow, a golf-club, or a tennis racket—proximal dexterous control by the arm is at least as important as for the hand (17, 18). When throwing a ball accurately—the hand is only there largely to time release. In this paper, we use tool in the more general sense of an object that requires complex trajectory control by the arm and/or hand for effective use. We will also use the term object in this paper to emphasize this broader definition.

Regarding the idea of practice, the proponents of the DDH, to the best of our knowledge, have never tested to see if the advantage in the control of dynamics on the dominant side can be matched by the nondominant arm through training. This is important because if such training is possible then the dynamic advantage may itself be a practice effect; tools and objects are routinely reached for and picked up more often by the dominant arm. In addition, it is not plausible in our view to attribute dominance differences in skill with tool use to control of dynamics. For example, finger- and hand-based tasks such as writing with a pen, drawing with a pencil, and using chopsticks show pronounced skill differences between the dominant and nondominant sides—differences that are unlikely to be explained by anticipation of interaction torques, given the light weight of the effectors and objects involved. Both objections to the notion of a baseline dynamic advantage for the dominant side appear borne out in a recent study of chopstick use, in which after six weeks of practice, skill with the nondominant hand was as good as with the dominant hand (19). Likewise, left-handed writing can improve substantially with extended practice—whether over several months with weekly training (7), within 28 d of daily training (20), or in the case of “converted” left-handers (21).

So why can’t we write or use chopsticks just as well with our nondominant hand in the absence of practice? Most of us have, at some point, given nondominant handwriting a try and it is interesting to ask what exactly it is we get wrong. We certainly do not write the wrong letter entirely, but the attempted shape often deviates from the shape made by the dominant hand and there is more variability between attempts (see Fig. 1). It is as if we know the shape we want but can’t get it to come out right. In a previous study (22), we had subjects learn across days to rotate their wrist to quickly move a cursor through a U-shaped tube without touching the sides; in effect they were making an upside-down “U” shape. Participants converged on a consistent U-shaped trajectory across days of practice, a shape they could get right even on day 1 if they moved very slowly. Thus, it appears that the correct shape was known right away in a more abstract sense, but it required practice to execute it at normal tempo. Notably, in a follow-up study, this gain in motor acuity did not transfer to the other wrist (23); a lack of transfer of trajectory control that directly parallels what happens with handwriting (this task showed a baseline dominant–nondominant difference). We have argued that these are continuous control tasks that require de novo skill learning, which occurs via acquisition of a feedback control policy (24). Handwriting, in this framework, is a particular example of continuous control of trajectory shape (25). Thus, our core hypothesis is that dominance is just the sum of de novo control policies accrued over time task-by-task.

Fig. 1.

Examples of handedness for handwriting with distal and proximal joints. (A) Examples of writing the letter “A” and the number “8” with Right and Left hands on a piece of paper. The scale bars in both images represent 1 cm. (B) Same as (A), but written on a blackboard, requiring movements of proximal upper extremity joints. Scale bars represent 15 cm. In both panels, the subject was asked to write them as quickly as possible.

To test our dominance hypothesis, we designed a 3D center-out reaching task, using markerless tracking, where the subjects reached to five targets with their hands (regular reaches). In addition, we used two modified versions of this task. First, to test the DDH, we attached a 4 lb weight to the wrist of the subjects (weighted reaches). In the second, to mimic tool use, we attached a very light bamboo stick to the forearm of the subjects, effectively a single forearm chopstick (stick reaches). In all three conditions, the subjects were asked to perform point-to-point reaches with their dominant and nondominant arms. This design allowed us to directly pit the two hypotheses against each other. DDH predicts a dominant advantage for the weighted reach condition, since added inertial load is precisely the dynamic challenge for which the dominant hemisphere is claimed to be specialized. Our first prediction was that differences between the dominant and nondominant arm would be absent or minimal for the two reach tasks. Our second prediction was that with stick use we would see a large dominant/nondominant difference in trajectory quality because of life-long practice with objects and tools using the dominant side. Crucially, the stick is too light to impose meaningful dynamic complexity, so the DDH has no principled basis to predict dominance here over and above the weighted reach. The dissociation between these two conditions therefore directly distinguishes the accounts. In a second experiment, we tested the practice hypothesis’ most distinctive prediction: use of a novel effector with no asymmetric practice history should eliminate the dominant advantage. DDH predicts the opposite—its claimed advantage is hemispheric and task-general, and therefore should also manifest at the elbow. Our third prediction, therefore, was that elbow writing would show no dominant advantage. If the elbow experiment confirmed our prediction, we added an extra experiment to both check that the elbow has the intrinsic capacity for accurate handwriting and that dominance does not manifest in a second-order way as a learning advantage. Our fourth prediction was that equal practice would produce the same degree of improvement in either elbow, and that both would show superior performance to the nondominant hand. If our four predictions were borne out then it would suggest that there is no inherent baseline motor control advantage in the dominant arm—it simply accumulates more practice with trajectory-shaping actions required by objects and tools.

Results

A Novel Markerless 3D Center-Out Reaching Task for Probing Control of Inertial Dynamics and Novel Tool Use.

To capture movement trajectories in 3D, we designed a center-out reaching task. This task is comprised of five targets positioned at equal distances (17.05”) from a center point located at the midline of the table edge that is closest to the subject (Fig. 2A). All subjects were right-handed, neurologically healthy, young adults (SI Appendix, Tables S1-S3). They were instructed to make straight reaches starting from the center point to each target sequentially, returning to the center in between. The table was positioned such that the subjects’ elbow joints were fully extended when they reached the targets at the far end of the table (T2 and T4 in Fig. 2 A and C). Movement time for all reaches was set to ~350 ms, using a digital metronome. This timing was mostly achieved without significant differences between right and left reaches (SI Appendix, Fig. S1A). As the subjects performed the task, their movements were recorded using two high-speed (170 frames per second) cameras. Joint pose estimation was achieved through markerless deep learning methods (26, 27) (Fig. 2B). This markerless method was recently validated in healthy subjects performing similar reaches (28) to those used in this study. The resulting trajectories for each of the five targets were predominantly straight (Fig. 2 C and D) and exhibited bell-shaped velocity profiles (Fig. 2E), despite differences in joint angle changes (Fig. 2F). These findings replicate the results of Morasso’s 2D center-out reaching task (29) in 3D.

Fig. 2.

A novel 3D center-out reach task with weight perturbation and stick-use variants. (A) Image of the experimental table and the targets (numbers with T indicate the targets). (B) Representative images and superimposed 2D detected keypoints of a subject performing reaches (Right and Middle panels), and a 3D skeletal model derived from the middle panel (Left panel). Scales indicate 100 mm. (C) Reach trajectories and the table. The black dot indicates the starting center point (origin), and the red points indicate the target positions. The scales for the x, y, and z directions represent 100 mm. (D) Reach trajectories for five targets shown for individual targets for a representative subject. The scales for x, y, and z directions indicate 100 mm. (E) Similar bell-shaped velocity profiles of reaches shown in (D). The scale is for all subpanels. NT indicates normalized time. (F) Reaching for each target results in different joint angle changes. The plots are for the same subject shown in panels (D and E). NT indicates normalized time. (G–I) Representative images for the regular, weight, and stick reach conditions, respectively.

We conducted three experiments using this setup. In the first experiment, the subjects used the palms of their hands to touch the start and end positions. We referred to these as “regular reaches” (Fig. 2G). In the second experiment, we tested the effects of an inertial challenge by attaching a 4 lb weight to the wrist. Subjects performed the same reaches with the weights on their wrists, referred to as “weight reaches” (Fig. 2H). In the third experiment, we examined the effects of a tool-use challenge while performing the same task. To achieve this, we “elongated” the subjects’ forearms by attaching a very lightweight (83 g) bamboo stick. The stick extended 20” distally from the mid-palm to its tip and 8” proximally toward the forearm for stability. Importantly, the 8” proximal extension effectively eliminated any wrist movement contribution. The subjects performed the reaches by touching the start and end positions with the stick tip, which was tracked via an attached LED light. The table was positioned so that the elbow angles were fully extended at the furthest targets (T2 and T4). We referred to these reaches as “stick reaches” (Fig. 2I). After processing the data, we extracted and analyzed the end-effector 3D trajectories (Fig. 2 C and D). In all three experiments, the subjects performed 10 to 12 reaches per target per side (right and left), all reaches within the same movement time. This resulted in a total of 2,863 (1,437 left, 1,426 right) regular reaches across 23 subjects, 1,075 (548 left, 527 right) weight reaches across 10 subjects, and 1,319 (671 left, 648 right) stick reaches across 11 subjects. No systematic differences were observed between the dominant and nondominant sides in basic kinematic metrics such as duration, maximum velocity, and endpoint errors (SI Appendix, Fig. S1) or in basic joint angles such as elbow, shoulder abduction, and shoulder flexion angles (SI Appendix, Fig. S2).

Statistical Shape Analysis for Comparison of Movement Trajectory Shapes.

To analyze the shapes of trajectories, we utilized statistical shape analysis (30), a set of algorithms designed to accurately compare shape differences. Briefly, a shape can be intuitively defined by landmarks—specific points outlining a figure, such as the vertices of a triangle (SI Appendix, Fig. S3A). Each shape configuration has two components: intrinsic shape elements (landmark arrangements, angles, etc.) and nonshape elements (size, position, rotation). Differences in size, position, or rotation alone do not alter the inherent shape (SI Appendix, Fig. S3A). To focus exclusively on true shape differences (31), nonshape components must first be removed through Procrustes analysis (32). This method eliminates variations in size, position, and rotation, allowing shapes to be mapped into a high-dimensional, mathematically curved “shape space.” To simplify analysis, shapes are projected onto a tangent space—a linear, flat approximation of shape space—typically centered at the mean shape, enabling straightforward linear analyses (33). To illustrate this concept, we consider a basic example using triangles defined by three landmarks. After applying generalized Procrustes analysis (SI Appendix, Fig. S3A), we find that triangle shapes can vary along two orthogonal axes: narrow versus wide angles, and right versus left tilting (SI Appendix, Fig. S3B). Any triangle shape can thus be represented as a combination of these two axes (SI Appendix, Fig. S3C). Kendall showed that the resulting shape space for planar triangles is a sphere (34) (SI Appendix, Fig. S3D), from which the tangent space for simplified analysis is derived (SI Appendix, Fig. S3E).

To practically demonstrate the analysis workflow, we generated 15 example triangles (SI Appendix, Fig. S3F) and projected them onto the tangent space using statistical shape analysis methods (SI Appendix, Fig. S3G). Next, we performed a principal component analysis (PCA) in this tangent space, which identified two orthogonal axes explaining different amounts of variability across these triangles (SI Appendix, Fig. S3H). This PCA (often called shape PCA) captures all relevant information present in this triangle set. While PCA provides valuable shape insights, linear discriminant analysis (LDA) offers further advantages by identifying axes that best separate custom-defined groups. For instance, when classifying triangles into two groups (right-tilted narrow triangles versus left-tilted wide triangles), LDA finds the axis (LD1) along which group differences are maximized (SI Appendix, Fig. S3I). Moving shapes along this axis demonstrates how they transition from one subgroup shape into the other (SI Appendix, Fig. S3I). Using this method, we compared the trajectory shapes of the reaches across all groups.

No Advantage for the Dominant Arm in Setting of Altered Dynamics.

Annett et al. found that movements of the nondominant hand tend to be noisier than those of the dominant hand (35). To determine whether this asymmetry also exists in 3D reaches, we quantified within-subject variability using the root mean squared deviation after Procrustes matching, calculated separately for each subject and each target. Since each subject completed 10-12 reaches per target, this provided a reliable estimate of individual movement noise.

In the regular reach condition, the nondominant side showed greater variability for four out of five targets (Fig. 3 A and B), demonstrating that our task and analysis approach were sensitive enough to detect this established asymmetry. When an inertial challenge was introduced during the weighted reaches, variability increased for both sides, but there was no significant difference between them (Fig. 3 A and B), suggesting that the dominant arm had no advantage dealing with the challenge of altered limb dynamics.

Fig. 3.

Trajectories under the three reaching conditions—regular, weighted, and stick. (A) Within-subject variance measured as RMSD of the full Procrustes distances between the reach trajectories. The colors of bar graphs are indicated in (B). Data are shown as mean ± SEM and the individual paired data points (each light gray line indicates a subject). *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001; Paired t test or Wilcoxon matched-pairs signed rank test. (B) Representative end-effector movement trajectories for all conditions for a single subject (each trajectory is a single reach). T1-T5 in the regular left reach condition indicate the five targets; the same target labeling applies across all other conditions. The x, y, z scale bars indicate 100 mm. (C) The tangential velocity profiles for Right and Left sides for all conditions for Target-2. Data are shown as mean (the line) ± SEM (shaded area). Note that the break point in deceleration phase exists only in the stick reaches. NT indicates normalized time. (D) Mean trajectory shapes for target-2 across all subjects and conditions. Each trajectory is the mean trajectory shape for a single subject. The x, y, z scale bars indicate 100 mm.

Tool Use Unmasked a Clear Dominant Arm Advantage.

In the stick reach condition, a marked dominant vs. nondominant difference became apparent across targets (Fig. 3 A and B). Qualitative inspection revealed that reach trajectories in stick reaches were distinctly altered, particularly showing an increased curvature toward the end of the reaches. This was noticeable both at the individual reach level (Fig. 3B) and at the mean trajectory shape level (Fig. 3D and SI Appendix, Fig. S4A).

To evaluate whether this reflected a shift in motor control strategy, we examined tangential velocity profiles. Only stick reaches exhibited a clear inflection (“break point”) during the deceleration phase (Fig. 3C and SI Appendix, Fig. S4B), suggesting a two-phased movement: an initial ballistic component followed by an “adjustment” phase, similar to that in reach-and-grasp (36). This pattern appeared consistently for both dominant and nondominant arms, indicating a systematic effect of tool use.

Shape PCA Revealed Distinct Trajectory Shapes Induced by Tool Use.

To quantify trajectory-shape changes, we calculated mean trajectory shapes for each subject and target, eliminating within-subject variability effects. A shape PCA (37) on these mean shapes revealed that the first principal component (PC1)—accounting for 57 to 76% of total variance—consistently separated stick reaches from both regular and weighted reaches (SI Appendix, Fig. S5 A–C). PC1 axis reflected increased curvature in the distal phase of the trajectory (SI Appendix, Fig. S5A). This indicates that the stick reaches were distinct in their mean shape compared to regular and weighted reaches. Weight reaches did not differ from regular reaches along any principal axis (SI Appendix, Fig. S5 B–E; the axes beyond the third were not shown, as the top three PCs consistently explained a cumulative 93-96% of the variance).

To assess whether these shape changes were present at the single-reach trajectory level, we trained a binary support vector machine (SVM) classifier to predict group (Regular vs Stick) of a given single reach. Using an 80-20 train-test split (and equal number of samples from each group), and five-fold cross validation, the model achieved high classification accuracy: 93% for right and 91% for left reaches (SI Appendix, Fig. S5F). This demonstrates that shape differences were also evident at the single-reach level.

LDA Identified Distinct Tool-Use and Dominance Axes.

While shape PCA found axes of maximal variance without regard to group labels (regular vs weighted vs stick or dominant vs nondominant), a key question remains: Can specific shape changes linked to tool-use or dominance be identified? Identifying tool-use or dominance axes could provide specific insights into how trajectory shapes are altered, potentially involving combinations of multiple PC axes. To identify the linear combinations of PCs that best separate the groups, we used canonical variate analysis (32), which is equivalent to Fisher’s LDA. This supervised method incorporates group labels as input, enabling the algorithm to find axes that maximize separation between the groups. Using this approach, we compared the mean reaches of subjects for each group and target. Using the top 20 PCs (capturing >99% of variance), the first linear discriminant (LD1) consistently distinguished stick reaches across all targets (Fig. 4 A and B and SI Appendix, Fig. S6A). We termed this axis the “Tool-use Axis”. No axis consistently separated regular from weighted reaches (Fig. 4 A–C and SI Appendix, Fig. S6 A–D), reinforcing that the inertial challenge did not significantly alter trajectory shape.

Fig. 4.

Tool-use induces a unique trajectory shape and unmasks a dominance advantage. (A) Comparisons of linear discriminant (LD) scores for target-2 reaches. (B) Quantification of LD1 scores capturing the trajectory difference observed in stick reaches (“Tool-use axis”). (C) Quantification of the LD2 or LD3 capturing the Right- and Left-sided differences observed only in stick reaches (“Dominance axis”). Data are shown as mean ± SEM along with individual paired data points. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001; two-way ANOVA with Tukey’s post hoc multiple comparisons test. (D) Manipulating Right stick (Target-2) trajectories along the tool-use and dominance axes transforms their shapes to resemble those of Right regular and Left stick reaches, respectively. The x, y, z scale bars indicate 200 mm. (E) Schematic of hypothetical tool-use and dominance axes in the shape space (formed by the top 20 PCs). (F) Quantification of the cosine similarity between the tool-use and dominance axes (each point is for a target). (G) The tool-use axis captures higher variance than the dominance axis. **P < 0.01, paired t test after log transformation. (H) Differential contribution of each PC to tool-use and dominance axes. Thin lines represent each target, and thick lines represent the mean values.

Interestingly, we observed that the second or third LD axis revealed significant differences between the right and left stick reaches (Fig. 4 A and C and SI Appendix, Fig. S6A). We termed this axis the “Dominance Axis”, which was specific to stick reaches. Notably, no dominance axis was detectable in regular or weighted reaches (Fig. 4 A and C and SI Appendix, Fig. S6 B–D). This indicates that only tool-use induced a shape difference between the dominant and nondominant sides.

Tool-Use and Dominance Axes Captured Distinct Shape Features.

What do the tool-use and dominance axes represent geometrically and are they composed of different underlying shape dimensions? To answer these questions, we first morphed reach trajectory shapes along these axes and investigated the resulting shape changes. Moving the right stick reaches along the tool-use axis resulted in reach shapes similar to right regular reaches, while moving them along the dominance axis resulted in shapes similar to left stick reaches (Fig. 4D and SI Appendix, Fig. S7 A–D). These differences were characterized by increasingly sharp curvature in the distal portion of the trajectories in stick reaches and straighter trajectories in regular reaches for the tool-use axis. Dominance axis changes were characterized by further increased midline curvatures without substantially affecting the endpoint direction. The conversions in the opposite directions along these two axes of right regular and left stick reaches were also done which gave consistent results.

To compare axes directly and quantify these differences, we computed the cosine similarity between them and found only 20 ± 18% overlap (Fig. 4 E and F), indicating that tool-use and dominance modulate different shape features. Specifically, tool-use induced a qualitative shape-change compared to regular reaches, whereas dominance induced variation around the stick reach shape. Next, we quantified the extent of information contained in each axis. We quantified how much variance each axis captured in the larger shape space and found that the tool-use axis explained 6.45 times more variance than the dominance axis (Fig. 4G). To demonstrate the composition of each axis, we calculated the percent contribution of each PC axis to the tool-use and dominance axes and found that the compositions were markedly different (Fig. 4H): While the tool-use axis was primarily composed of PC1 (~70%), the dominance axis was dominated by PC2 (~40%), with smaller contributions from PC1, PC3, PC4, and others. These findings confirm that tool-use (versus regular reach) and arm dominance elicited qualitatively different changes in reach trajectory shape.

Using a De Novo Effector Abolished the Dominance Effect for Handwriting.

Our findings indicate that dominance becomes dramatic and obvious in the setting of tool use, when control of a markedly nonstraight trajectory is required. This is notable, given that tool use is a key component of many skilled motor behaviors in everyday life. These results are consistent with our practice-based hypothesis. To further test this idea, we also examined the reverse scenario: if hand dominance reflects long-term practice effects, then using a de novo end effector—one with little to no prior use in daily life—should eliminate any dominance-related differences during skilled task performance.

To test this scenario, we asked 11 healthy subjects (SI Appendix, Table S4) to perform a standard writing task using their hands (Fig. 5A). We then asked the same subjects to perform the task using their elbows—a body part that, while involved in most upper extremity movements, is almost never used as an end effector in daily life, especially tool use. To accomplish this, we securely attached a pen to the medial side of the distal arm, positioning the tip just beyond the olecranon and ensuring that the pen remained stable. This configuration enabled visual guidance during writing and prevented the elbow from touching or dragging on the table (Fig. 5B). The ink trace produced during writing constitutes a direct physical record of the pen-tip trajectory—the end-effector path—and therefore the shape analysis described below operates at the same level of description as the kinematic trajectory analyses in Experiment 1. No inference from letter shape to joint-level kinematics is required or intended. The subjects were asked to write two characters (“A” and “8”) on a piece of paper (SI Appendix, Fig. S8A) in prespecified configurations to prevent differences due to variations in writing preferences (38) (SI Appendix, Fig. S8B). The time it took to write these characters eight times was recorded. The nondominant side was slightly slower than the corresponding dominant side for both handwriting and elbow-writing. Elbow-writing was also significantly slower than the handwriting (SI Appendix, Fig. S8C). To compare shapes, we first preprocessed the data after scanning the experimental sheets. After cleaning and extracting individual characters (SI Appendix, Fig. S8D), we rescaled the size of the characters (SI Appendix, Fig. S8E) and normalized line thickness (SI Appendix, Fig. S8F). The resulting final images for all characters showed clear visual differences (Fig. 5 C–F and SI Appendix, Figs. S9 and S10). To extract the features from the individual character images, we used a convolutional neural network (ResNet-50) trained on the ImageNet dataset and took the last fully connected layer (1000FC) of this network (Fig. 5G and SI Appendix, Fig. S8G). This resulted in good clustering of “good” and “bad” shapes visually (Fig. 5H and SI Appendix, Fig. S8H), with a good overlap of the group labels onto these clusters (Fig. 5I and SI Appendix, Fig. S8I). To quantify these differences, we performed an LD analysis on the four groups for each character. For both characters, we found good separation between the dominant handwriting and the nondominant hand- and elbow-writing (Fig. 5 J–M). Importantly, there were no significant differences or trends between the dominant and nondominant elbow characters (Fig. 5 J–M).

Fig. 5.

Writing with a de novo effector (elbow) abolished the dominant/nondominant difference. (A and B) Experimental setups of writing with hand (A) and elbow (B). For elbow writing, the pen was securely attached to the elbow to prevent wobbling. (C–F) Examples of written characters, “A” and “8”, produced with the dominant and nondominant sides using the hand (C and D) and elbow (E and F), respectively. D: dominant and ND: nondominant. (G) Schematic of feature extraction of the characters. Each character image was run-through the ImageNet-trained ResNet-50 network, and features were selected from the last fully connected layer (1000FC). (H) Visualization of features for all “A” characters using t-SNE plot. Note that all “good-shaped” “A” characters cluster together. (I) Color-coded groups of characters in the t-SNE plot shown in (H). (J) Comparisons of different LD axes using LDA of “A” characters. Note that LD1 emerges as the “dominance” axis. (K) Quantification of mean LD1 scores of each subject. (L and M) Same as (J and K) but for character “8” writing. Data are shown as both mean ± SEM and as individual paired data points. ***P < 0.001, ****P < 0.0001; repeated measures ANOVA with Šidák’s multiple comparisons post hoc test.

We next addressed whether the observed dominant/nondominant shape differences could be explained by a speed–accuracy tradeoff. The nondominant side was slightly slower than the dominant side for both effectors (SI Appendix, Fig. S8C). To formally test whether execution time accounted for LD1 differences, we fit linear mixed-effects models with LD1 as the dependent variable, Side, Effector, and their interaction as fixed effects, execution time as an additional covariate (centered at the grand mean), and a random intercept for Subject: LD1 ~ Side × Effector + Time + (1|SubjectID). For character “A,” execution time was not a significant predictor of LD1 after accounting for Side and Effector (P = 0.186), indicating that the dominance effect on shape quality is entirely independent of movement speed. For character “8,” execution time was a significant positive predictor of LD1 (β = +0.170, P = 0.046), meaning that slower execution was associated with marginally better shape quality—a speed–accuracy trade-off. However, the nondominant side, despite being slower and thus receiving this marginal benefit, still produced substantially worse characters. Accordingly, the Side × Effector interaction remained highly significant (P = 1.55 × 10⁻⁷), confirming that the nondominant disadvantage persisted after statistically controlling for execution time. To further test whether the dominant and nondominant sides operated under different speed–accuracy regimes, we compared the above model to an expanded model including a Side × Time interaction [LD1 ~ Side × Effector + Side × Time + (1 | SubjectID)] using likelihood ratio tests. This interaction did not improve model fit for either “A” [χ²(1) = 0.066, P = 0.797) or “8” (χ²(1) = 0.916, P = 0.339], confirming that the time–quality slope did not differ between dominant and nondominant sides. Together, these analyses rule out a speed–accuracy tradeoff as an explanation for the observed dominance effects.

These results support the idea that hand-dominance effects on shape quality are practice-dependent and do not generalize to de novo effectors like the elbow, reinforcing the view that dominance arises through experience rather than a baseline asymmetry.

Practice Improved Writing Quality to the Same Degree in Both Elbows.

The finding that elbow-writing abolished the dominance effect is consistent with a practice-based account: because neither elbow has been used as an end effector in daily life, neither side holds a meaningful advantage at baseline. However, this result invites two related alternative interpretations that must be addressed. First, one might argue that the elbow is simply an imprecise effector incapable of producing high-quality shapes regardless of practice—in which case the absence of a dominance effect would reflect a ceiling effect rather than anything meaningful about practice asymmetry. Second, perhaps the dominance effect in a previously unused effector expresses itself in superior learning with practice. A strong prediction of the practice hypothesis addresses both concerns simultaneously: if practice—not intrinsic effector capacity—determines both the level of performance and the asymmetry between sides, then when trained, both elbows should be able to perform better than the nondominant hand (ruling out a low motor control ceiling) and the improvement in performance should be of the same magnitude in each elbow.

To test this, we trained 12 subjects to write with either their dominant (six subjects) or nondominant (six subjects) elbows and reassessed character quality using the same LD1-based framework (SI Appendix, Table S5). Consistent with the de novo effector results, both dominant and nondominant elbows began at similarly poor shape quality prior to training, comparable to nondominant handwriting (Fig. 6 A and B) and at similar duration patterns (SI Appendix, Fig. S11). Following training, elbow-writing quality improved substantially for both sides (Fig. 6 A and B). Critically, posttraining elbow performance exceeded nondominant handwriting quality (Fig. 6B), directly ruling out a ceiling effect and demonstrating that the elbow is capable of generating high-quality shapes when given adequate practice. The improvement was indistinguishable between sides: dominant and nondominant elbows reached equivalent posttraining quality (Fig. 6B), and the learning effect size—defined as the difference in LD1 between post- and pretraining—did not differ between sides (Fig. 6C). These results make two points. First, the elbow is not an inherently imprecise effector: its baseline poor performance reflects inexperience, not a fixed control incapacity, just as foot-artists who lack arms demonstrate that precision is practice-dependent rather than body-part-specific (39). Second, when practice is equated across sides, the dominant advantage does not reemerge. Together with the baseline elbow-writing results, this provides direct experimental support for the practice-based hypothesis: limb dominance in skilled motor tasks reflects the accumulated asymmetry of lifelong practice rather than an intrinsic control difference between the two sides.

Fig. 6.

Practice of elbow-writing with either the dominant or nondominant elbow led to symmetric improvements in skill that surpassed that of the nondominant hand. (A). Examples of written character “8”, produced with the dominant and nondominant sides using the hand at baseline and elbow pre- and posttraining. D: dominant and ND: nondominant. (B) Quantification of resulting “8” characters as in Fig. 5. **P < 0.005 for elbow posttraining vs pretraining, and P < 0.05 for elbow posttraining vs nondominant hand, Two-way repeated measures ANOVA with Fisher’s LSD post hoc test. (C) Learning effect sizes, defined as the difference between before and after the training, do not differ between dominant and nondominant sides. n.s. nonsignificant, Two-tailed t test.

Discussion

Arm dominance is often considered as evidence for an intrinsic advantage of the dominant hemisphere in motor control, yet everyday superiority might instead reflect asymmetric practice with tools over long periods of time. Here we introduced an analysis of trajectory shape to determine the type of motor control challenge that best reveals the difference in ability between the dominant and nondominant arm. One possibility is that there is a fundamental baseline difference in capacity, perhaps based on hemispheric asymmetry, that would be unmasked even during the performance of basic 3D reaching movements. In this case, however, we only observed a small difference in trajectory variance in favor of the dominant arm. When we introduced a dynamic stressor to the reach movements in the form of a 4 lb weight attached to either wrist, there was an increase in trajectory variance to an equal extent in both arms. Notably, there was no qualitative difference in trajectory shape between either of the two reach conditions or the two arms. In contrast, when the targets had to be touched with a light stick attached to the forearm, the dominant/nondominant difference became obvious. The shape of the trajectory required with the stick was qualitatively different from baseline reaches, and the nondominant side showed markedly increased variability, obviously failing to match the performance of the dominant side. The difference was directly reminiscent of that seen when someone tries to write, for example, an “A” with the nondominant arm—the shapes of the “A’s” deviate considerably from the dominant side’s and are inconsistent. We concluded from these results that the dominant advantage is the ability to precisely control the curved trajectory of the tip of the tool, and that this is most likely a practice effect rather than an inherent task-independent advantage. To test this idea further, we had individuals attempt to write letters with a marker attached to their elbows. Notably, in this case, there was no dominant advantage; both elbows were equally clumsy, with their poor performance comparable to nondominant handwriting. Taken together, these results suggest that dominance is a practice effect: we have had life-long practice holding a variety of tools with our dominant hand and controlling their trajectories. In the absence of such practice, as in the elbow writing task, the advantage vanishes.

An important consideration in the elbow-writing results is whether the absence of a dominant-side benefit is attributable to a ceiling effect, i.e., writing with the elbow is just inherently so difficult that it swamps any dominant advantage. Several observations argue strongly against this possibility. First, the quality of elbow-generated characters was not worse than those of the nondominant hand; both produced recognizable shapes with comparable variability, thus controlling the shoulder to write with the elbow is not obviously harder than using the nondominant hand. Second, elbow writing does not constitute a qualitatively different control problem. Many everyday writing contexts—such as writing on a blackboard, signing one’s name on a large vertical surface, or producing giant works of graffiti—are executed primarily with proximal joints, yet still show robust dominance differences. Thus, proximal control demands themselves do not abolish lateralization. What distinguishes elbow-writing is not some kind of unique biomechanical challenge but the near-total absence of prior tool-use experience with either elbow as an end effector. In this sense, the elbow task is a de novo control problem, analogous to first-time use of an unfamiliar tool: participants know the intended shape but lack an effector-specific control policy to realize it.

The elbow training experiment provided the most direct empirical test of our practice hypothesis. If the absence of a dominance effect at baseline reflected an inherent incapacity of the elbows for precision—rather than simply the absence of practice—then training should fail to meaningfully improve their performance. This is not what we found, instead, the elbows’ baseline performance was symmetric and not worse than that of the nondominant hand, and then markedly surpassed it after training. It is interesting that both elbows improved to the same degree—training did not unmask a latent control advantage for the dominant limb’s elbow. This training experiment reinforces the view that “capacity for precision” is not a fixed control property of a body part but an acquired one, determined by practice history. This principle extends well beyond the elbow: individuals born without arms who learn to paint with their feet develop robust foot-dominance effects, with the preferred foot producing substantially better writing than the nonpreferred foot—a body part not conventionally considered a precision effector, yet one that acquires precision and lateralized skill through asymmetric practice (39). Taken together, the baseline elbow result, the training experiment, and the foot-artist literature converge on the same conclusion: what a body part can do is determined largely by what it has done. Future studies examining populations with informative dissociations between preference and practice—including left-handers who have been forced to use tools with their nonpreferred hand, stroke survivors with acquired hand preference changes, and amputees who develop skilled use of unconventional effectors—will be valuable in further testing the boundaries of the practice-based account proposed here.

Based on studies conducted over more than 20 y, Sainburg and colleagues have suggested that the dominant arm has a general trajectory control advantage, which is attributable to superiority in anticipating acceleration-dependent intersegmental dynamics (11). Conversely, the nondominant side is better at stabilization due to superior impedance control (13). This group has gone on to suggest that this asymmetry is because each of these two fundamental forms of limb control is lateralized to a different hemisphere (40). The core underlying assumption is that the motor control differences in the two arms transcend any specific task. This is a tricky issue, however, because any assay used to demonstrate this difference could itself be construed as a task. For example, Sainburg and colleagues have most often used a planar reaching apparatus with the arm supported by an air sled and with targets arrayed on a vertical screen (11–13). Why is this not a task that uses an air sled as a kind of object or a tool? One way to address this possibility would be to train the nondominant arm on this task and see if better anticipation of intersegmental dynamics ensued. If the nondominant arm indeed became kinematically indistinguishable from the dominant arm, then a case could be made that the original difference is just a practice effect. A learning experiment of this kind by this group, or any other, has not, to the best of our knowledge, been performed. Here, when a 4 lb weight was attached to the wrist of either arm, which increased the variance in the trajectories compared to baseline, there was no dominant advantage. This dissociation—no dominance for a dynamically challenging weighted reach, but robust dominance for stick use where asymmetric practice history distinguishes the conditions—is precisely what the practice hypothesis predicts and what DDH, as a task-general account, cannot readily explain, especially as it could be argued that free 3D reaches are a more basic motor control challenge than planar reaches with the arm strapped to an apparatus with targets on a screen. A potential objection might be that we did not examine EMG activations, which can show larger differences than the kinematics (41, 42). Although this might be true, a learning effect could explain EMG differences also. It was not, however, a dynamic stressor during baseline reaches that brought out a stark dominant advantage but use of a forearm-strapped tool; a tool that was very light and so would not significantly alter arm dynamics. It appears instead that, just as in handwriting, the nondominant arm is unable to consistently control the trajectory of the tip of the tool, which is consistent with lack of a tool-specific control policy. This was confirmed when both the dominant and nondominant elbows were equally impaired in writing letters; neither had practiced elbow-controlled tool use. The DDH cannot explain this set of results. We of course agree that there are tools that will require acquired control policies that take inertial dynamics into account for precise trajectory control, for example, using a tennis racket or a sledgehammer, but it does not follow from this that the dominant arm has a generic dynamic advantage. We note that DDH’s proposed double dissociation has two halves: a dominant-side trajectory control advantage and a nondominant stabilization advantage. The second half addresses a different control phenomenon and is not relevant to our question, which is ‘why does the dominant side outperform the nondominant side at skilled trajectory-shaping tasks?’ The DDH’s answer to this question is entirely contained in its dominant-side trajectory control claim, which is what our experiments directly test and challenge. The nondominant stabilization advantage, by contrast, is DDH’s account of why the nondominant side excels at a different class of tasks altogether, and testing it would not bear on the conclusions we draw here. It should be noted that the nondominant limb’s stabilization superiority could also just be a practice effect, which could be tested in future experiments.

Stick use in our experiment required a two-phase trajectory strategy—an initial ballistic transport phase followed by a late-phase curvature associated with a deceleration “break point”. These two phases defined the stick trajectory shape, which was qualitatively different from the shape of regular reaches (Tool-use Axis). It was the shape of the stick trajectory that was modulated along our Dominance Axis, indicating that the nondominant arm showed increased variability in execution of the same shape made by the dominant arm. Similarly, it was the letter shape made by dominant arm that was poorly duplicated by the nondominant arm, and by both dominant and nondominant elbows. We note that the forearm attachment of the stick, while differing from conventional hand-held tool use, mirrors a well-recognized feature of skilled technique in several everyday and sporting contexts. Writing on a blackboard or signing one’s name on a large vertical surface—as illustrated in Fig. 1B—is executed primarily with proximal joints with minimal wrist involvement, yet shows robust dominant-side advantages. Similarly, in tennis, minimizing wrist involvement and treating the racket as an extended forearm is considered a hallmark of proficient performance—and dominant-side advantages are well established in both contexts (43). The forearm-attached stick in our task captures this same mechanical relationship, suggesting that the dominant advantage we observe is directly relevant to naturalistic trajectory-shaping actions rather than being an artifact of the attachment method. It should be noted also that the DDH was formulated entirely with tasks that did not involve the hand. What is it that makes control of trajectory shape difficult when controlling a cursor or holding a stick or a pen? The desired shape is not in question in the case of writing, it just does not come out the way we want it. In a previous study we showed that it takes many days of practice to guide a cursor through an arc-shaped channel using wrist movements (22). This is an example of a task where the shape of the trajectory is the control challenge; get the shape wrong and cursor will hit the sides. Notably, the dominant side was initially better than the nondominant side on this task. Subsequent training was then done with the nondominant wrist, which then became more skilled than the dominant wrist (23). Interestingly, the acquired skill did not transfer to the dominant wrist; initial trajectories were clumsy in the same way that we saw here for the nondominant stick reaches and elbow letter writing. Two conclusions can be drawn from these arc-task papers. First, there is an initial tool-use advantage for the dominant hand; presumably from a lifetime of using tools. Second, it takes practice to map the goal shape onto the right motor commands. These conclusions are consistent with studies that have shown that practice with the nondominant arm can lead to the same proficiency as on the dominant side at a variety of tasks (7, 19–21).

Overall, these results are consistent with a view we have put forward previously that trajectory shape, when it is an overt goal of the task (draw an “8”) and critical to successful performance, is planned before execution (25). We have previously demonstrated this distinction between planning and execution of curved trajectories in an obstacle-avoidance task using the dominant arm (25). We have also shown that left premotor lesions impair trajectory-shape imitation while leaving basic reach execution intact (44)—evidence for the effector-independence of premotor abstract shape representations. These results have led us to posit uncoupling of specification of trajectory-shape from quality of its execution (45). In the case of handwriting, it is usually possible to discern what shape is being attempted by the nondominant arm, for example the letter “A”, even though its execution is poor. In a recent study of handwriting, there was transfer of the A shape but the way in which the shape was drawn, in this case the direction of the middle horizontal stroke, often did not transfer (38). A similar distinction between planning and execution has been made when it comes to discrete sequence tasks. Specifically, there is a difference in knowing what element comes next in a sequence versus the quality of execution of that individual element (46–48). Of direct relevance is work showing that the effector-independent sequence order transfers from one hand to the other, but there is a component that does not and instead depends on practicing with that specific effector (49). Thus, it appears that in the cases of either letter shape or sequence order, there is a more abstract effector-independent representation that then can be turned into effector-specific motor commands. This distinction parallels the structure versus parameterization framework from the generalized motor program tradition (50), which partitions learning into structural features (relative timing and force patterns) that transfer between effectors, and parameterization (absolute scaling of those features) that is effector-specific and requires practice. On this account, the recognizable but poorly executed nondominant “A” reflects intact structure with impaired parameterization: the abstract shape representation transfers, but its precise motor realization does not. Consistent with this view, studies of professional athletes show that reductions in interlimb performance asymmetry are skill-specific rather than domain-general—asymmetries diminish only for the specific practiced skill and do not generalize across tasks (51, 52). This is precisely the pattern our practice hypothesis predicts and is difficult to reconcile with a task-general hemispheric account. In the case of stick-use, an overt representation of trajectory shape need not be invoked as it is not the goal. The common factor then is that complex nonstraight trajectories, and the production of more elaborate shapes like letters, must be controlled, and that such trajectory control is unfamiliar to the nondominant arm, unlike, for example, simple point-to-point reach movements. In addition, the quality of trajectory control improves by practicing with the specific effector. In the case where the trajectory shape is the goal, such as for letters, planning with an intermediate shape representation can serve as a scaffold to initialize practice. The availability of such a representation is apparent in the effector-independent component of letters. In the case of the stick no such representation is available and so practice alone must suffice.

Lesions and functional imaging experiments in humans suggest that effector-independent representations, both for trajectory shape and sequence order and for tool use skills, localize to premotor areas (44, 53–56). This is of great interest as premotor areas very often have bilateral representations (53, 57), consistent with an effector-independent capacity (58, 59). Lesions in the left hemisphere cause deficits in tool use and trajectory-shape imitation in either arm (44)—a pattern consistent with left-lateralized effector-independent representations in the premotor cortex, though we note that ideomotor apraxia encompasses a broader constellation of deficits beyond trajectory shape specifically. More broadly, we acknowledge that tool-relevant action representations involve multiple interacting components—semantic knowledge, haptic prediction, and sensorimotor expectations about action outcomes—and that the inferior parietal lobule, a critical lesion site for ideational apraxia, plays an important role in this multimodal architecture (60). Our account focuses specifically on the motor planning and trajectory-shaping component of tool use, and we do not claim to offer a comprehensive account of tool representation. Lesions of the premotor cortex and the supplementary motor area (SMA) in humans cause severe impairments in reproducing sequential movements at a particular tempo from memory, despite intact manual dexterity (61). Relatedly, lesions of the premotor cortex cause limb-kinetic apraxia, in which movements that require a certain temporal sequence (e.g., catching a ball) are most affected (62). Additionally, temporary cooling of the dorsal premotor cortex (PMd) in monkeys impaired both spatial accuracy and the speed of corrective movements, consistent with “deactivation” of the control policy (63). In a functional imaging study of the arc task, mentioned above, we found learning-related changes in the contralateral PMd and SMA (64). In primates, a similar division between sequence order and individual element execution has been found for premotor and primary motor cortex, respectively (65, 66). Finally, a study of novel tool use in nonhuman primates showed that the direction of tool movement (goal) was coded in the ventral premotor cortex and uncoupled from the direction of hand movement needed to control it, which was coded in the primary motor cortex (59). This view of tools as engaging distinct motor planning processes sits comfortably with recent neuroimaging evidence challenging the body schema extension hypothesis. Contrary to the influential idea—originating with Iriki et al. in nonhuman primates (67)—that tools are extensions of limb neural correlates during use, expert tool users show less tool-hand neural correlation than novices; successful prosthetic use is predicted by a more categorically distinct neural correlates of the artificial limb (68, 69). Based on these findings, we speculate that the dominant and nondominant arms differ with respect to the granularity of the mapping between the premotor cortex and primary motor cortex, with increased information processing capacity between these cortical areas contralateral to the dominant side (see also Deng and Haith for a related argument (70)). This greater information processing capacity increases the fidelity with which the primary motor cortex is able to reliably and repeatedly execute an appropriate movement to realize the effector-independent plan held in premotor regions.

In conclusion, of the 10 items in the Edinburgh handedness inventory, eight involve tool-use, a ninth throwing a ball, and the tenth opening the lid of a box (71). Throwing a ball shares the same operative feature as the tool-use items: the dominant hand takes the primary trajectory-shaping role in directing an object toward a goal. On our account, what unifies these items is not a philosophically precise definition of tool use but the consistent asymmetry of practice in dominant-hand trajectory-shaping actions across a lifetime. Notably, to the degree that hand preference has been seen in chimpanzees, it is for tool-use and not for simple reaching (72). The challenge with tool use and object-mediated actions more broadly, we would argue, is control of the unfamiliar trajectory shapes they require, whether the shape needs to be overtly represented (written letters), or not (stick-use). In all cases, trajectory-shape control needs to be mastered through practice. Thus, dominance is contextual and learned. It emerges through the accumulation of effector-specific control policies that can piggy-back on a more abstract trajectory representation-capacity, possibly localized to premotor cortical areas. We speculate that even when skilled trajectory control is achieved with a given tool, there is continued interplay between premotor and motor cortices on the contralateral side. This could perhaps be contrasted with finger individuation/dexterity, which is primarily localized to the primary motor cortex (73–78). An interesting fact in support of a distinction between arm trajectory control and finger individuation, is that cellists and violinists use their dominant arm for bowing and their nondominant hand for the finger board (79, 80). Indeed, it has been argued that the motor control challenge of bowing is greater than that of fingering (79, 81, 82). We conclude that lateralization, as captured by the Edinburgh handedness inventory, is due to task-by-task practice of the trajectory-shaping challenges that arise when the dominant hand takes its primary causal role in object-mediated actions—whether with a conventional tool, a thrown object, or any implement whose tip or end must be guided to a goal. We suggest that this form of skilled shape control might be particularly dependent on task-specific changes in communication and connectivity between the premotor and motor cortex. We end with a provocative speculation: The vast repertoire of hand-held tools is indicative of human inventiveness, thus strong population-level dominance could be seen as the indirect motor control consequence of human thought.

Materials and Methods

Neurologically healthy adults performed a 3D center-out reaching task to five targets under three conditions: regular reaches, weighted reaches (4 lb wrist weight), and stick reaches (a lightweight bamboo stick attached to the forearm), recorded with a dual-camera markerless motion capture system at 170 fps. All procedures were approved by the University of California, Los Angeles Institutional Review Board, and were performed in accordance with the Declaration of Helsinki. Reach trajectories were reconstructed in 3D using DeepBehavior, then analyzed using a statistical shape analysis pipeline—using the Shapes package in R—that quantifies trajectory geometry, with group differences assessed via Canonical Variate Analysis (LDA). In a separate experiment, participants wrote characters (“A” and “8”) with their dominant and nondominant hands and elbows; written characters were analyzed using a ResNet-50 convolutional neural network to extract shape features, followed by PCA and LDA to compare character quality. A group of participants completed an elbow-writing learning experiment involving 2,400 practice characters with each elbow, with pre- and posttraining recordings compared using the same pipeline. More detailed Materials and Methods are described in the Supplemental Materials and Methods (SI Appendix).

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

Data used in the study are included in the manuscript and/or SI Appendix. The basic code regarding the shape analysis is available at https://github.com/aarac/ArmDominance (83).