The posture adopted by the weightlifter at the start of the Snatch underpins the success of the lift. Technically, a successful lift requires the bar to be raised off the ground to an overhead position in one continuous movement; failure to do so results in an unsuccessful attempt (18). The Snatch lift demands precise whole-body coordination performed by the weightlifter as explosively as possible making it a complex movement to learn and coach. For analytical purposes, the Snatch can be broken down into a sequence of phases characterized by the posture of the weightlifter relative to the bar position (13). The phases include the first pull, the transition, the second pull, and the catch (18). Although these phases are important components to master, it is the start position that is first and foremost fundamental to successful lifting.
The importance of the starting position to successful Snatch attempts is defensible because posture of the weightlifter at the beginning of the lift allows inertia of the bar to be overcome using the larger muscles that cross the hips (10,16). It is also at this early stage that the foundation is set to provide the weightlifter the best conditions to maximize time and clearance for the weightlifter to pull themselves under the bar that ultimately optimizes the chance of a successful lift (3,4,19). On the other hand, positional error of the joints and pelvis (i.e., poor posture) at the start position inevitably results in an unsuccessful lift as attempts to correct these errors during the lifting sequence are generally fruitless given the short total duration of time (<1.5 seconds) for the Snatch (5,10). Thus, consistency of body posture at the starting position has been a focus for coaches of novice weightlifters.
Despite the importance of the start position being highlighted in the literature, attention of most reports has been focused on the importance of peak vertical bar velocity and its relationship with success. This emphasis in the research has been justified based on the premise that a greater peak vertical bar velocity results in greater bar displacement at the catch phase, affording the weightlifter more time and room to get under the bar to accept the load (18). Peak vertical bar velocity occurs during the second pull phase after the knees reach full extension (1,6,7–9) and has been reported to range from 1.50 ± 0.00 to 1.98 ± 0.09 m·s−1 for elite-level weightlifters for successful Snatch lifts (1,2,7-11,16,17). It is, however, constrained by the weightlifter's physical size, shape, and the load being attempted. For example, women produced higher peak bar velocities (1.98 ± 0.09 m·s−1) using lighter loads when compared to their male counterparts (1.66 ± 0.06 m·s−1) who lifted heavier loads (7). Based on the previous research outlined above, measurement of bar velocity during lift attempts would be expected to provide a good source of feedback and coaching focus for an individual weightlifter.
Of those studies that do address technical issues related to the Snatch, descriptive analyses of joint angle displacement, extension, or flexion are reported for ideal or successful lifts (7,8). Only a few studies have provided technical descriptions of the “correct” starting position and coaching points in qualitative terms (i.e., focal point above standing eye level, back arched, shoulders ahead of the bar) (1,3,19). One of those aforementioned papers put forward a brief technical description explaining how to determine foot stance and body posture for the Snatch start position, yet no empirical evidence to support this contention was provided by comparing successful and unsuccessful lifts (3). Only one study has provided measurements based on digitizing video data for an optimal start position (1). Like bar velocity, it is widely accepted that starting positions will differ between weightlifters, and it is their anthropometric characteristics, which predominantly govern how they should best set up. For example, it was proposed that taller weightlifters require a greater knee flexion angle to get into an optimal start position (1). The generalization of an ‘optimal’ starting posture is therefore a contentious issue and should be considered on an individual basis. How coaches determine the optimal setup position for their athlete is currently based on subjective observations and experience taking into consideration the weightlifter's anthropometry and fitness parameters. We contend that from multiple observations of both successful and unsuccessful lifts, modern data-mining tools such as partition modeling can provide valuable feedback for the individual weightlifter. Moreover, coaching novice weightlifters can be challenging because high variability in performance and numerous technical errors can be expected. These modern analytical techniques can provide concise and manageable feedback for both weightlifter and coach by ordering parameters by level of importance (14). Such information is particularly important at the early learning stages for novice weightlifters where overcoaching by trying to correct all technical flaws can become overwhelming.
As previously noted, quantitative support for coaching recommendations for a ‘good’ setup position is currently limited. No study to date has addressed the differences in starting position between successful and unsuccessful lift attempts. The aims of this case study were to use high-speed 3D motion analysis of successful and unsuccessful lifts for a single weightlifter with the purposes to evaluate the importance of bar velocity to success of a lift and to explore key features of the Snatch starting position. Taking a traditional approach, it was hypothesized that peak bar velocity would be different for successful and unsuccessful lifts at any given load. In addition, it was hypothesized that the pelvis and joint angles at the starting position would differ between successful and unsuccessful lifts. We also took a more novel approach to explore the kinematic variables to determine if the starting position is important to the outcome of the attempt and to provide tailored feedback for the athlete based on the observations. It was anticipated that this case study would provide a foundation for further development of 3D kinematic evaluation of a particular athlete's weightlifting movements within the laboratory setting.
Experimental Approach to the Problem
A case study approach was adopted where a novice weightlifter completed a total of 133 Snatch attempts. A novice weightlifter was invited to participate as the method used was previously untested and the time commitment required for data collection at this stage would be disruptive to an elite class weightlifter's training schedule. Therefore, this case study was viewed as a means to evaluate the method and data analyses before engaging higher qualified weightlifters. All Snatch attempts were captured in a laboratory using VICON MX (500 Hz) and subsequently analyzed using VICON Nexus (version 1.4.116). The loads lifted ranged from 75 to 100% of 1 repetition maximum (1RM). Bar velocity, pelvis angle, hip, knee and ankle joint angles in the sagittal plane at the start position for successful and unsuccessful lifts were compared. Finally, partition modeling was also used to identify key features for the subject that would predict the outcome (success of the lift).
The subject (age = 25 years, stature = 171 cm, mass = 74.8 kg, inter anterior–superior iliac spine (ASIS) distance = 24.7 cm, leg length: left = 88.0 cm, right =86.5 cm, knee girth: left = 10.9 cm, right = 10.7 cm, ankle girth: left = 7.9 cm, right = 7.6 cm) was a novice male weightlifter (Snatch 1RM = 80 kg) who had <6 months of coaching. For all testing black Lycra™ shorts and weightlifting shoes (Adidas: AdiStar, 2004, China) were worn. The Institutional Review Board for Human Investigation approved all experimental procedures. Before the first visit to the laboratory, the subject was briefed regarding the purpose of the study and the data collection procedures. The subject volunteered to take part and provided informed consent before the commencement of data collection.
There were 5 training loads of single repetitions for data collection (i.e., 60, 65, 70, 75, and 80 kg).
Stature, mass, and limb lengths were recorded by a trained (level 1: ISAAC) Anthropometrist using a standardized protocol (15). A wall-mounted stadiometer (Seca: Hamburg, Germany) was used to measure the subject's standing and sitting height. Body mass was measured using digital scales (Wedderburn: Melbourne, Australia). Anthropometry tape (Tsutsumi: Japan) and sliding callipers (Tsutsumi) were used to measure all other lengths, including leg length, knee width, ankle width and ASIS distance. These measurements were later entered into the VICON Nexus program. Bony landmarks were identified for the placement of retroreflective ball markers for joint angle assessment using VICON Nexus software. Six VICON cameras were set up around the laboratory to capture 3D motion of the subject and the movement of the barbell (Figure 1). Two cameras were set in front and behind the weightlifter to allow for detection of the markers on the barbell and 4 set in a diagonal plane to the weightlifter for detection of the lower limb markers. All cameras were elevated approximately 2 m (6 ft) above ground level, and each lens was directed toward the middle of the platform to ensure that they were not facing each other to avoid the mistake of detecting another camera as a marker. A check was made to ensure that a minimum of 2 cameras could detect each of the markers used which was also a setting in the Nexus software. Once detection of all markers was confirmed, a calibration was performed using an L-wand with 5 reflective markers to ensure that the cameras were able to detect marker movement. The image count for each camera was set to a minimum count of 8,000 frames and to an accuracy of not >0.20% error. This process was repeated until all cameras were setup to the required accuracy. The camera positions were then calibrated using the L-wand again by placing it in the middle of the weightlifting platform and setting it as the point of origin to allow the cameras to position themselves in space.
Sixteen reflective markers (14 × 12.5 mm in diameter and 2 × 40 mm in diameter) were placed on the subject according to the VICON Plug-in Gait model (12,20). Larger reflective markers (40 mm in diameter) were used at the ASIS. A pilot study to determine camera setup and feasibility was performed, and one of the main barriers was marker detection. As the start position involved the weightlifter being crouched over, smaller ASIS markers were obscured by the weightlifter's torso. This problem was rectified using larger markers. Two additional large (40-mm diameter) reflective markers were also attached to the center point at each end of the barbell.
A static Plug-in Gait model of the weightlifter and a 3D representation of the barbell were created by the software for calibration purposes. Single repetitions of the Snatch were then performed and each lift was recorded individually. For each lift, data were processed to build a dynamic Plug-in Gait model, which allowed joint angles to be extracted from the start position of the lifts. Figure 2 shows the convention used to describe joint angles. Using this convention, a positive direction (increase in angle) equates to anterior tilt at the pelvis, hip flexion, knee flexion, and ankle dorsiflexion.
The vertical displacement of the barbell was used as a reference to determine the frame in which the start position was adopted by the weightlifter. A baseline height of the barbell was taken from the data collected and an increase of 1 mm in the vertical direction indicated that the bar had lifted off the ground. The frame before this occurred was selected as the frame from which joint angle data would be taken as the start position of the Snatch.
In total, 133 lifts were captured during 11 separate sessions (all at 9 am in the morning) spanning a period of 28 days. Between each session, a minimum of 3 and a maximum of 6 days of recovery were allowed. After a progressive warm-up, the order of lifts (attempts) was blocked and progressed by intensity starting with the lightest load (75% 1RM) and finishing with the heaviest load for that session. The randomization of lifts per load was anticipated to reduce the learning effect. The number of lifts within sessions ranged between 8 and 17. Rest between each attempt (repetition) was a minimum of 3 minutes, and the weightlifter did not make an attempt until he felt ready and not fatigued. Data collection was completed within a 4-week period to reduce the effect of training adaptation, and each session of data collection was limited to a maximum of 1.5 hours to reflect a regular training session. Successful lifts were determined by the guidelines for international competition (10). Bar departure was determined when either marker on the bar achieved positive displacement from the baseline and continued in an upward direction.
All data were analyzed using JMP 8.0 (SAS Institute, Inc) software. The distribution of successful and unsuccessful lifts by load was presented in the form of a contingency table. Only lifts at a load of 70 kg (84% 1RM) and above were further analyzed, because these loads represent lifts that were technically demanding. Variables included right side peak bar velocity and right side joint angles (pelvis, hip, knee, and ankle) at the start position. In addition, the difference between left and right (left–right) sides for these variables was reported. The left side variables were not analyzed, given the strong relationship between left and right sides.
Using the fit model platform, standard least squares 2 × 3 (success × load) factorial analyses of variance (analyses of variance [ANOVA]) were used to test differences. When significant (p ≤ 0.05) main effects for success were found, least square means difference with 95% confidence intervals (95% CI) was reported. When significant main effects for load were observed, Tukey's Honestly Significant Difference (HSD) post hoc was subsequently employed for each pairwise comparison, and the significant least square means difference with 95% CIs were reported.
In a separate analysis, all 133 lift attempts between 70 and 80 kg were randomly divided into training (70%) and holdback data sets (30%). Using the training data set, all joint angle variables (right side and left vs. right) at the starting position were entered into a recursive partitioning model. This form of analysis finds a set of ‘cut-offs’ for independent variables that best predict the categorical response (i.e., successful or unsuccessful). The purpose of the holdback data set was to enter new data into the partition model to gauge the appropriate number of splits used and evaluate the accuracy of the model. The decision of how many splits used for the partitioning was based on maximizing the R 2 of both the training and holdback data sets. When the R 2 of the holdback data set moved in a negative direction the splitting process was terminated and the final split was pruned back. Accuracy of the model was also assessed based on the area under the curve of the receiver operator characteristic (ROC) curves of both the training and holdback data sets.
The contingency table shows an increase in the proportion of failed attempts compared to successful lifts with greater loads (Table 1). Loads of 60 and 65 kg included 61 (56% of total) successful lifts with only 3 failed attempts (2% of total).
Peak Bar Velocity
Peak bar velocity and differences between left and right sides by load and success are shown in Figure 3. For peak bar velocity, no main effects were found for load (p = 0.2252), success (p = 0.4510), or their interactions (p = 0.3901). Similarly, no main effects for load (p = 0.8749), success (p = 0.4929), or their interactions (p = 0.2778) were found for the difference between left and right bar velocities. Although not included in the analysis, differences in peak bar velocity at the lighter loads were evident.
Joint Angles at Starting Position
Descriptive statistics for (right side) joint angles at start position by success and load are presented in Table 2. A 1.9° (95% CI = 0.3–3.4°, p = 0.0228) greater (more dorsiflexed) ankle joint angle for unsuccessful lifts was the only difference in joint angle found when compared to successful lifts. At the pelvis (3.7°, 95% CI = −1.0 to 8.5°, p = 0.1192), hip (0.0°, 95% CI = −2.6 to 2.6°, p = 0.9879), and knee (−3.9°, 95% CI = −8.8 to 0.9°, p = 0.1079) no differences between successful and unsuccessful lifts were found.
Across load, main effects for hip angle (p = 0.0038) and knee angle (p = 0.0319) were observed. After post hoc comparisons, hip angle was 2.5° (95% CI = 0.6–4.3°, p = 0.0061) more flexed in the 70 kg (88% 1RM) lifts compared to the 75 kg (94% 1RM) lifts. Similarly, the knee angle was 3.8° (95% CI = 0.4–7.2°, p = 0.0273) more flexed in the 70 kg (88% 1RM) lifts compared to the 75 kg (94% 1RM) lifts.
Descriptive statistics for the joint angles difference between left and right sides at the starting position by load and success are shown in Table 3. The only difference observed was at the knee where a main effect for load (p = 0.0024) was found. Post hoc comparisons revealed that greater flexion (−1.8°, 95% CI = −3.1 to −0.5°, p = 0.0051) occurred at the left knee relative to the right knee for the lifts with a load of 75 kg compared to the 70-kg lifts.
The decision tree shown in Figure 4 below identifies the pelvis and hip angles as key features to the outcome (successful or unsuccessful) of a Snatch lift. From this, a pelvis angle of >17.6° returned a successful lift (100%). When the pelvis angle is below this threshold (17.6°), a hip angle of <89.6° maintains a high (75%) chance of being successful. Alternately, when pelvis angle is <17.6° and hip angle is >89.6°, the chance of making a successful lift was only 27%. The key features (coaching points) from the decision tree are therefore pelvic anterior tilt and hip flexion at the starting position. Both R 2 (0.29 and 0.15) and ROC curves (Area Under the Curve [AUC] = 0.81 and 0.69) of the included (training data set) and the excluded (holdback data set), respectively, support the validation of the model.
Using 3D motion analysis of the bar and lower body, the primary goals for this case study were to evaluate the importance of selected variables to the success of Snatch lifts for a novice weightlifter. In addition, from the kinematics of the subject's starting posture analytical approaches were explored with the view to determine key technical features for feedback to the subject. Previously, only general descriptions providing guidelines of how best to setup for the Snatch have been presented throughout the literature (3,19). Thus, our current approach to this question we believe is novel.
Our findings further question the high importance placed by some coaches on peak bar velocity to success of Snatch lifts. Only at lighter loads (60 and 65 kg), where unsuccessful lifts were sparse (n = 3), was there any evidence to suggest that slower peak bar velocity led to unsuccessful lifts. At the higher and more challenging loads of 70 kg and above (88–100% 1RM), no differences in vertical bar velocity were found. Moreover, post hoc to the original hypothesis analysis showed that the success of the lift at 70 kg and above could not be predicted by using binary logistic regression (χ2[1, N = 69] = 1.8778, p = 0.1706). This further strengthens the contention that other factors, apart from bar velocity, are more important to the outcome of a snatch attempt. These other factors may include speed of the weightlifter to get under the bar (17), the position of the body during the catch, the stability of the weightlifter and the bar trajectory (1,10,13).
In contrast to peak bar velocity, 3D kinematics of the weightlifter's start position were found to be important for successful lifts. In particular, the partition modeling highlighted the pelvis and hip joint angles as key features to the success of the attempt for the subject. From the traditional statistical testing (ANOVAs) used, only a difference at the ankle joint between successful and unsuccessful lifts was observed. A lack of difference at the pelvis and other joint angles was unexpected. Ankle dorsiflexion coupled with an extended hip in the start position would cause the weightlifter to be in a more erect position or sitting back into the lift, causing the shoulders to be in line with the bar or even behind the bar when a reference line is drawn. Based on previous studies describing the optimal start position (3,4,19), increased ankle dorsiflexion is contrary to the ‘good’ starting position. As a result, increased dorsiflexion as found would bring the shoulders behind the bar to compensate for the extended hips, causing weight distribution to go over the heels and not the balls of the feet.
Most notably, partition modeling revealed that pelvic and hip joint angles, and not the ankle angles, were key features to successful lifts, which was an important finding missed through the use of a series of univariate analysis (ANOVAs). Both the pelvis and hips are interlinked (Figure 2) and are unconstrained (free to move). The ankle on the other hand is constrained and fixed through contact with the ground. Therefore, complex interactions of the more proximal joints can occur that will result in more noticeable changes at the ankle. This is a likely explanation of why ANOVA testing found differences between successful and unsuccessful lifts at the ankle joint. The importance of the partitioning model lies in its provision of clear and verified feedback for the subject. It should be noted that it was anticipated to find a low R 2 value (<0.30) for the model, given that only variables relevant to the start position were included in the model and not during the movement phases of the lift.
Finally, it was observed that with an increase in load, significant changes occurred in the various lower limb angles in the starting position. Interestingly, the partition model highlighted that for the hip movement toward a better position was found (i.e., mean <89°) at the heavier load. Also, the number of lifts at the highest load (80 kg or 100% 1RM) consisted of only 3 lifts in total (1 successful and 2 unsuccessful). As this was the load which was considered a personal best for the subject, the number of lifts performed was limited. Hence, statistical power for the highest percentage lift was limited and had the lowest ratio of successful to unsuccessful lifts as compared to the other loads lifted. Future studies involving weightlifting may benefit from the investigation of other variables such as the weightlifter's stance width and bar location in relation to the feet, and assessing joint angles in other phases of the Snatch (i.e., the first pull, transition, and second pull).
As a preliminary study investigating the start position of the Snatch, with only one subject, a limitation was that a generalization could not be made for other weightlifters. Given the diversity of body sizes and shapes found in weightlifting, a more complete model could be built with a larger data set similar to the present case study but including the additional anthropometric measures used to construct the Plug-In Gait model. It should also be noted that the interpretation is exclusive to the subject and the partition model outlined is specific to the individual tested, but also within the specific ranges of motion. In addition, the subject involved in this study was a novice weightlifter. Novice weightlifters are inherently more variable in their performance and this is shown as measured by previous studies investigating elite weightlifters whom had a less variable bar velocity (1,2,7,9-11,16,17). Further, this study only used a lower body model; the incomplete model did not include the upper body. The presence of a shoulder marker may be useful as most coaches would use the coaching practice of having the shoulders over the bar for a good start position. However, it would be very difficult to create a full-body model of a weightlifter using 3D kinematics because of the numerous additional markers required (especially for the shoulder girdle) which could affect the movement of the weightlifter.
This case study describes a previously untested method for data collection and analyses to provide the subject with valuable coaching feedback. The feedback is derived from current technology (VICON Nexus Motion Analysis) and modern analytical tools (i.e., partition modeling) using multiple observations of both successful and unsuccessful Snatch attempts. The hierarchal outcomes of the analyses can assist the coach to determine which specific features from the starting position should be made a coaching priority. We consider this a good starting point for future research development in the area of Olympic weightlifting. With the ability to develop a model for an individual weightlifter, an automated system collecting and analyzing data regarding an individual's Snatch lifts could be developed in the future to provide athletes and coaches tailored feedback. Finally, for analytical purposes, an important observation from this case study was that with univariate analysis, essential information is lost as the interactions between variables are overlooked. Therefore, other analytical tools should be considered.
The authors wish to thank the Australian Weightlifting Federation and Phoenix Weightlifting Club for their support and participation in the study. We would also like to express our utmost gratitude particularly to Mr. Robert Kabbas and Mr. Eric Rosario for their support of this investigation.
1. Bartonietz, KE. Biomechanics of the snatch: Toward a higher training efficiency. Strength Cond J
18: 24–31, 1996.
2. Campos, J, Poletaev, P, Cuesta, A, Pablos, C, and Carratala, V. Kinematic analysis of the snatch in elite male junior weightlifters of different weight categories. J Strength Cond Res
20: 845–850, 2006.
3. Derwin, BP. The snatch: Technical description and periodization program. Natl Strength Cond Assoc J
12: 6–14, 1990.
4. Garhammer, J. The power clean: A kinesiological evaluation. Natl Strength Cond Assoc J
1: 40, 61–63, 1984.
5. Garhammer, J. A comparison of maximal power outputs between elite male and female weightlifters in competition. Int J Sports Biomech
7: 3–11, 1991.
6. Garhammer, J. Weightlifting performance and techniques of men and women. In: International Conference on Weightlifting and Strength Training
. Lahti, Finland: Gummerus Printing, 1998. pp. 89–94.
7. Gourgoulis, V, Aggelousis, N, Antoniou, P, Christoforidis, C, Mavromatis, G, and Garas, A. Comparative 3-dimensional kinematic analysis of the snatch technique in elite male and female Greek weightlifters. J Strength Cond Res
16: 359–366, 2002.
8. Gourgoulis, V, Aggelousis, N, Garas, A, and Mavromatis, G. Unsuccessful vs. successful performance in snatch lifts: A kinematic approach. J Strength Cond Res
23: 486–494, 2009.
9. Gourgoulis, V, Aggelousis, N, Kalivas, V, Antoniou, P, and Mavromatis, G. Snatch lift kinematics and bar energetics in male adolescent and adult weightlifters. J Sports Med Phys Fitness
44: 126–131, 2004.
10. Gourgoulis, V, Aggelousis, N, Mavromatis, G, and Garas, A. Three-dimensional kinematic analysis of the snatch of elite greek weightlifters. J Sports Sci
18: 643–652, 2000.
11. Hoover, DL, Carlson, KM, Christensen, BK, and Zebas, CJ. Biomechanical analysis of women weightlifters during the snatch. J Strength Cond Res
20: 627–633, 2006.
12. Hughes, G, Watkins, J, and Owens, N. Gender differences in lower limb frontal plane kinematics during landing. Sports Biomech
7: 333–341, 2008.
13. Isaka, T, Okada, J, and Funato, K. Kinematic analysis of the barbell during the snatch technique of elite Asian weighlifters. J Appl Biomech
12: 508–516, 1996.
14. Morgan, S, Williams, MD, and Barnes, CJ. Modeling of player movement interactions in game play for team invasion sports. In: Proceedings of 7th International Symposium on Computer Science in Sport
. Canberra, Australia: 2009.
15. Norton, K, Whittingham, N, Carter, L, Kerr, D, Gore, C, and Marfell-Jones, M. Measurement techniques in anthropometry. In: Anthropometrica
. Norton, K and Olds, T, eds. Sydney, Australia: UNSW Press, 1996. pp. 25–76.
16. Okada, J, Iijima, K, Fukunaga, T, Kikuchi, T, and Kato, K. Kinematic analysis of the snatch technique by Japanese and international female weightlifters at the 2006 junior world championship. Int J Sport Health Sci
6: 194–202, 2008.
17. Stone, MH, O'Bryant, HS, Williams, FE, Johnson, RL, and Pierce, KC. Analysis of bar paths during the snatch in elite male weightlifters. Strength Cond J
20: 30–38, 1998.
18. Stone, MH, Pierce, KC, Sands, WA, and Stone, ME. Weightlifting: A brief overview. Strength Cond J
28: 50–65, 2006.
19. Takano, B. Coaching optimal technique in the snatch and clean and jerk–part 1. Natl Strength Cond Assoc J
15: 33–39, 1993.
20. Telhan, G, Franz, JR, Dicharry, J, Wilder, RP, Riley, PO, and Kerrigan, DC. Lower limb joint kinetics during moderately sloped running. J Athl Train
45: 16–21, 2010.