Twelve young adults (9F/3 M, 20.53 (1.25) years, 1.69 (0.08) m, 62.3 (11.4) kg) participated. Inclusion criteria included participation in resistance training for at least six months at a frequency of ≥ 2 days/week and that the kettlebell swing was a novel movement. Exclusion criteria included neurological, muscular or cognitive disorders, and current or previous injuries, particularly of the lower back, which could be made worse by exercise. In addition, subjects were excluded if they had previously been coached on the kettlebell swing or attempted to learn the movement independently. This study was approved by the institutional review board at the hosting university. Informed consent was obtained from all individual participants included in the study.

We collected kinematic data using an 8-camera Vicon motion capture system (Centennial, CO, USA) at a sample frequency of 100 Hz. A full-body PSIS marker set was attached to each subject according to the Vicon Plug-In Gait model (Gutierrez-Farewik et al. 2006). An extra marker was placed on the side of the kettlebell to identify repetitions. Kinematic data was filtered using a fourth-order zero-lag Butterworth filter with a 6 Hz cutoff frequency.

Subjects performed kettlebell swings on five separate days over the course of a one-week period (Table 1). We collected data on the subjects’ kettlebell swing performance on the first and last day. During the five days in between, subjects practiced the kettlebell swing on three of those days with supervision from researchers. The subjects did not practice on the other two days. Subjects with a body mass equal to or greater than 165 lbs used a 25 lbs kettlebell and subjects with less than a body mass of 165 lbs used a 15 lb kettlebell. The selection of this body mass cut-off and kettlebell weight was informed by previous intervention studies focused on exercise (Lake and Lauder 2012; Fortner et al. 2014). Since our goal was to learn the skill and not exercise, we reduced the weight used in these studies by 5 lbs to minimize the chance of fatigue. In a previous study that used the same sets, repetitions, and load cut-offs, the maximum rating of perceived exertion was 4 out of 10 suggesting that fatigue was minimized (Beerse et al. 2020).

Kinematic data were collected at four time points during the study. On the first day, three sets of 20 repetitions were collected before practice (pre-practice) and three sets of 20 repetitions were collected after a practice session (first practice). All practice sessions comprised of five sets of 20 repetitions. One week later, we collected data on three sets of 20 repetitions (post-practice). As previously mentioned, within that week subjects completed three practice sessions on three separate days. An adaptation condition immediately followed the post-practice sets, where subjects performed three sets of 20 repetitions with a water-filled kettlebell. The water-filled kettlebell was equivalent in mass as their practiced kettlebell, but larger in size (diameter: practiced = 14 cm; water = 32 cm). Subjects rested for a minimum of three minutes in between each set of kettlebell swings to minimize fatigue.

Before the first pre-practice set of kettlebell swings, subjects watched a video of reconstructed motion capture data of a skilled demonstrator and listened to a list of verbal cues, such as “push the hips back, keeping the back straight”. These forms of instruction were available to the subjects before each set and followed a learner-regulated approach, where subjects selected when and how often they wanted to rewatch the video and hear the cues. Subjects opted to receive these instructions on average 3.9 times (range: 2–8), not including the required viewing at the beginning of the study. The minimum possible opportunities to receive instructions was 26, once before each new set. Subjects received no augmented feedback during the data collection or practice sessions.

Kettlebell swing repetitions were partitioned using the marker on the right side of the kettlebell. The first and last repetition of each set were removed from analysis, as these repetitions often differ from the rest of the repetitions even for skilled performers (Bullock et al. 2017). The task-level variable within the UCM analysis was a consistent vertical COM trajectory across repetitions and the local variables were segment angles. Our previous findings suggested that subjects stabilized vertical COM position with segmental angle variability when learning the kettlebell swing (Beerse et al. 2020). Seventeen sagittal plane segment angles were calculated: foot, shank, thigh, pelvis, upper arm, lower arm, and hand bilaterally, and the trunk, head, and kettlebell.

We conducted a Motor Equivalence Analysis to generate error and solution deviations for each kettlebell swing repetition (Scholz et al. 2003; Mattos et al. 2011; Selgrade and Chang 2015). The segment angles were time normalized to every 1% of the kettlebell swing cycle. The following UCM analyses were conducted at each 1% and summed across these time points to generate single values for each repetition. For the evaluation of the pre-practice, first practice, and post-practice conditions we employed three different mean configurations. The traditional mean configuration was the average segment angles across all repetitions for that condition, making the assumption that subjects were attempting to stabilize an average posture throughout the sets (Scholz and Schöner 1999). The initial mean configuration was the average segment angles of the first ten repetitions of the pre-practice condition, making the assumption that subjects were attempting to stabilize their initial posture. The practiced mean configuration was the average segment angles of the last ten repetitions of the post-practice condition, making the assumption that subjects were attempting to stabilize what would eventually become their preferred posture.

Geometric models were constructed to relate changes of sagittal plane segment angles to vertical COM position and then linearized around each mean configuration to create the Jacobian (J). The null space of the Jacobian defines the uncontrolled space, or solution space, as defined in Eq. 1.

$$ J\left( } \right) \cdot \varepsilon = 0 $$

(1)

Deviations of the segment angles (θ) away from the mean configuration (θ°) were calculated and were partitioned (Eqs. 2 and 3) into two subspaces, solutions (ΘUCM) or errors (θORT).

$$ \theta_ = \mathop \sum \limits_^ \varepsilon_^ \cdot \left( } \right) \cdot \varepsilon_ $$

(2)

$$ \theta_ = \left( } \right) - \theta_ $$

(3)

At this point, rather than summing these deviations to variances, we normalized the deviations by the square root of their degrees of freedom (Eqs. 4 and 5).

$$ motor \; equivalent \; deviations = \frac }} }} $$

(4)

$$ nonmotor \; equivalent \; deviations = \frac }} $$

(5)

The motor equivalent deviations (ME) contained the solution space deviations, while the non-motor equivalent deviations (nME) contained the error space deviations. A ratio of the ME/nME deviations was calculated as a synergy index (Scholz et al. 2007; Mattos et al. 2011) and log transformed to account for a non-normal distribution.

We implemented a similar approach to evaluate the adaptation condition to the water-filled kettlebell. For this evaluation, we employed only two mean configurations, the traditional and the practiced. The same definitions for these mean configurations applied. Therefore, the traditional mean configuration assumed that subjects modified their deviations towards the average posture of the adaptation condition with the water-filled kettlebell. This approach expected the modified task dynamics of the water-filled kettlebell to demand the emergence of a new mean configuration that subjects sought to stabilize. The practiced mean configuration assumed that subjects modified their deviations back towards the posture they preferred with the metal kettlebell during the post-practice condition.

To compare how subjects modified their deviations when learning the kettlebell swing, we calculated the mean ME, nME, and SCIDS for the first ten and last ten repetitions (labeled Start and End, respectively) for each condition (pre-practice, first practice, and post-practice) and mean configuration (traditional, initial, and practiced). For ME and nME deviations, differences between the mean configurations were not compared as it was expected that deviations were larger when using a mean configuration calculated from outside of the current performance. In other words, it was reasonable to expect that the traditional mean configuration would present with smaller deviations compared to the initial and practiced, except for the start of the pre-practice and end of the post-practice conditions, respectively. Therefore, we conducted two-way ANOVAs (3 condition × 2 time) with repeated measures on ME and nME deviations to compare differences from the start to the end of a condition (time factor) and between conditions (condition factor). Shapiro–Wilk tests were conducted to assess normality and a log transformation was applied to non-normal variables. Tukey post-hoc tests were conducted in the case of statistically significant main effects or interactions. For the ME/nME deviation ratio, we conducted a three-way ANOVA (3 condition × 2 time × 3 configuration) with repeated measures. The mean configuration was included as a factor since the ratio removes the inherent magnitude effect described earlier making it a more meaningful comparison.

To assess how subjects adapted to the water-filled kettlebell during the transfer task, we conducted paired t-tests on the traditional and practiced mean configurations, separately. Again, we chose not to include the mean configuration as a factor due to the expectation that the traditional mean configuration will result in smaller deviations. Whether subjects reduced their nME deviations during the adaptation condition was the more pertinent question. Eta squared and Cohen’s d were calculated as effect sizes. Statistical analysis was conducted in R Statistical Software (v4.3.2; R Core Team 2023). Significance level was set at alpha = 0.05.