Predicting Bobsled Pushing Ability From Various Combine Testing Events.

SAVE OFFLINE

PREDICTING BOBSLED PUSHING ABILITY FROM VARIOUS COMBINE TESTING EVENTS CURTIS L. TOMASEVICZ,1,2 JACK W. RANSONE,2

AND

CHRISTOPHER W. BACH2

1

Department of Biological Systems Engineering, University of Nebraska, Lincoln, Nebraska; and 2Nebraska Athletic Performance Laboratory, University of Nebraska, Lincoln, Nebraska ABSTRACT

KEY WORDS bobsleigh, strength and conditioning, winter

Tomasevicz, CL, Ransone, JW, and Bach, CW. Predicting bobsled pushing ability from various combine testing events. J Strength Cond Res 34(9): 2618–2626, 2020—The requisite combination of speed, power, and strength necessary for a bobsled push athlete coupled with the difficulty in directly measuring pushing ability makes selecting effective push crews challenging. Current practices by USA Bobsled and Skeleton use field combine testing to assess and identify specifically selected performance variables in an attempt to best predict push performance abilities. Combine data consisting of 11 physical performance variables were collected from 75 subjects across 2 winter Olympic qualification years (2009 and 2013). These variables were sprints of 15, 30, and 60 m, a flying 30-m sprint, a standing broad jump, a shot toss, squat, power clean, body mass, and dry-land brake and side bobsled pushes. Discriminant analysis (DA) in addition to principle component analysis (PCA) was used to investigate 2 cases (case 1: Olympians vs. non-Olympians; case 2: National Team vs. non-National Team). Using these 11 variables, DA led to a classification rule that proved capable of identifying Olympians from non-Olympians and National Team members from nonNational Team members with 9.33 and 14.67% misclassification rates, respectively. The PCA was used to find similar test variables within the combine that provided redundant or useless data. After eliminating the unnecessary variables, DA on the new combinations showed that 8 (case 1) and 20 (case 2) other combinations with fewer performance variables yielded misclassification rates as low as 6.67 and 13.33%, respectively. Using fewer performance variables can allow governing bodies in many other sports to create more appropriate combine testing that maximize accuracy while minimizing irrelevant and redundant strategies.

Address correspondence to Curtis L. Tomasevicz, ctomasevicz2@unl. edu. 34(9)/2618–2626 Journal of Strength and Conditioning Research Ó 2018 National Strength and Conditioning Association

2618

the

sports, performance, Olympics

INTRODUCTION

P

rofessional sport organizations and practitioners have historically attempted to identify physical and physiological characteristics predictive of sporting success via combine-like settings. At these combine events, performance tests typically include a series of sport and position specific drills, exercises, interviews, aptitude tests, and physical tests with the intent of identifying aspects of an athlete that will predict future sporting success (4,9,13). However, using sport-specific tests as predictive measures of performance has elicited varying degrees of success (4,9,12–14). The sport of bobsled (4-man) requires multiple positionspecific athletes with distinct athletic skills that include strength, speed, power, and biomechanical pushing efficiency, which are primarily skills focused on maximizing push start performance. The importance of push start times on bobsled performance has been well-documented (3,10,11,14). In the 2010 and 2014 Olympics, the average start time was strongly correlated with the down time in each heat of the top 15 finishing teams (r = 0.69) (8). Furthermore, 90% of those teams had an average start time rank within 4 places of their final ranking further illustrating the importance of the push at the start of a race (8). Bobsled creates a unique challenge with the translation of off-ice individual performance tests to on-ice team performance. In an attempt to best predict pushing ability, a variety of physical tests have been incorporated into various bobsled combines by top bobsled and skeleton programs worldwide, including the United States, Germany, and Australia (2,15,17). Osbeck et al. previously investigated the validity of the field testing performed by the United States Olympic Committee in 1996 to bobsled start performance. The success variables in the study were individual on-ice brakeman and side pushes (14). Although these objective success variables may seem to be an obvious performance indicator, an individual on-ice push lacks clear assessment of an athlete’s ability to push with a team. Therefore, this investigation uses a binary logic classification success variable of Olympian/

TM

Journal of Strength and Conditioning Research

Copyright © 2018 National Strength and Conditioning Association. Unauthorized reproduction of this article is prohibited.

the

TM

Journal of Strength and Conditioning Research

| www.nsca.com

aim of the investigation could be applied to several athlete assessment combines to ensure the most efficient and accurate testing method possible.

METHODS Experimental Approach to the Problem

This investigation evaluates current USABS combine practices for their accuracy in predicting future Olympic and National Team bobsled success. In preparation for the 2010 and the 2014 Winter Olympics, bobsled combine tests were held in the Figure 1. An athlete performs a left side push on a dry-land sled at the Lake Placid, NY push track. www. qualifying years of 2009 and nathancrumpton.com. 2013. Combine testing consisted of body mass collection in addition to 10 performance test variables: 15-m sprint, 30-m non-Olympian for case 1 and National Team/non-National sprint, 60-m sprint, 30-m fly, standing broad jump, underhand Team for case 2. The USA Bobsled and Skeleton (USABS) shot toss, power clean, back squat, dry-land brakeman push, has established a set of anthropometric and performance and dry-land side push. The testing was done over a 3-day tests using a combine of 11 physical performance variables period. The sprints, broad jump, and shot toss were measured in an attempt to gain insight into the pushing capabilities of on the first day. The squat, power clean, and body mass were athletes when transferred to a team setting on an ice track collected on the second day, leaving the 2 dry-land bobsled (18). Admittedly, this success variable carries a partial subpushes for the third day. Adequate self-determined rest time jective component that is described in the team selection was given to each subject between attempts and between criteria (17). However, it is believed that an improved objecevents, eliminating any potential fatigue factor. Using subtive combine will reduce the need for subjective contribusequent selection to Olympic and National Teams as a binary tion. To date, no investigation has assessed the ability of success measure, DA along with PCA was used to assess the current USABS combine practices to determine future bobability of these tests to predict future success, and furthermore, sled performance success. As such, the primary aim of this the effectiveness of each performance test. article was to determine the ability of current USABS combine practices to differentiate between success level (OlymSubjects pian vs. non-Olympian; National Team vs. non-National Seventy-five male athletes (mean 6 SD, age: 26.5 years 6 5.2) were invited by USABS to participate in combine testing Team), and subsequently, determine the most efficient and at the United States Olympic Training Center (Lake Placid, effective combination of performance tests that may help NY, USA) in 2 Olympic qualification years (33 athletes in create a predictive model of athletic potential for on-ice 2009; 42 athletes in 2013). Athletes were selected for comperformance via off-ice combine testing. This is critical bine participation via a recruitment process outlined by because time and velocity lost at the start could potentially USABS, which includes top performance at regional bobsled magnify threefold by the finish (10). It was hypothesized that push championships and active recruitment from experithe discriminant analysis (DA) and principle component enced bobsled coaches (18). The performance results from analysis (PCA) methods applied to the USABS combine the combines were acquired from freely available online data would lead to different combinations of testing variables sources with further detail regarding testing protocol prothat will be a reduction in the number of tests and an vided with further approval by USABS officials (16). Instituimproved misclassification rate. tional Review Board approval was considered a nonissue Additionally, through this example with the United States because this investigation was retrospective in nature, and bobsled team, the secondary aim of this investigation was to the data were obtained from freely available online sources; bring attention to potential evaluation errors commonly therefore, the study was exempt from review board approval. committed through combine testing in a variety of sports either through redundancy in combine tests or through Procedures undertesting of key performance athletic attributes. The Combine Performance Variables. Body mass was assessed using a digital scale (Model BWB-800a; Tanita Corporation of statistical approach using DA and PCA presented in the first VOLUME 34 | NUMBER 9 | SEPTEMBER 2020 |

2619

Copyright © 2018 National Strength and Conditioning Association. Unauthorized reproduction of this article is prohibited.

Predicting Bobsled Pushing Ability

2620

the

0.20 0.13 0.19 0.16 0.18 6 6 6 6 6 6 6 6 6 6

17.5 13.7 17.8 13.6 18.1

97.2 101.2 96.3 99.6 96.2

6 6 6 6 6

6.7 2.7 7.0 4.1 7.3

4.77 4.58 4.82 4.61 4.84

6 6 6 6 6

0.20 0.12 0.18 0.13 0.18

4.84 4.64 4.88 4.69 4.91

6 6 6 6 6

Sprints (15, 30, and 60 m) and Fly Runs (30 m). Sprint performances were captured via electronic timing system (TC Timing System; Brower Timing Systems, Draper, UT, USA) in one movement following a 1-m “rock-in” start. The rock-in start allows an athlete to shift his weight posterior but must keep 1 foot on the ground within 1 m of the initial timing eye or photocell gate. A rock-in start can also be defined as a 1-m fly-in. Photocell gates were set up so that timing started as the athlete broke the initial gate (set at a height of 20 cm). Gates were set at 15, 30, and 60 m from the initial eye and were set at a height of 1 m. Athletes were allowed to perform 3 separate trials of the sprint test, with their best performance for each testing variable being taken regardless of trial. This means that the 15, 30, and 60 m may be taken from separate trials. However, the 30-m fly time was recorded as the best difference between corresponding 30- and 60-m splits within 1 trial.

15.7 10.8 14.9 12.2 15.4

183.8 192.5 181.8 191.5 180.4

Squat (kg) Power clean (kg)

6 6 6 6 6 6 6 6 6 6

1.6 1.0 1.4 0.9 1.4

137.3 151.4 134.1 146.8 133.1

Standing Broad Jump. The jump was performed from a hard surface into a sand pit. From a standing position with both feet touching a starting line, athletes were instructed to jump as far as possible (horizontal direction). A successful jump consisted of both take-off and landing phases being completed with both feet simultaneously. Distance between starting line and heel of the closest foot at landing was recorded. Athletes were allowed 3 separate jumps, and the best distance was used for analysis.

3.09 3.23 3.06 3.22 3.03

0.19 0.17 0.18 0.15 0.18

15.4 17.0 15.0 16.8 14.8

Shot toss (m) Broad jump (m)

6 6 6 6 6 6 6 6 6 6

0.16 0.10 0.16 0.09 0.15

Underhand Shot Toss. For shot toss testing, athletes stood on a 10-cm raised block (to simulate the start block on a bobsled track) and were instructed to throw an official Olympic-sized shot (16 pounds; 7.26 kg) underhand using both hands as far as possible. Distance from the front of the block to the initial landing site was recorded for analysis. The best toss of 3 attempts was recorded.

6.82 6.59 6.88 6.6 6.92

0.31 0.21 0.30 0.18 0.30

3.16 3.06 3.18 3.05 3.21

30-m fly (s) 60 m (s)

Power Clean. A 1 repetition maximum (1RM) effort power clean was performed from the floor with a shoulder-width grip on the bar. Athletes were allowed self-determined warm-up and rest time between attempts. They were instructed to clean the bar as explosively as possible to maximize bar height before the catch phase. Athletes could choose their own attempted weight and were allowed up to 3 misses at each weight. Only their best completed attempt was recorded.

0.08 0.06 0.07 0.06 0.08 2.11 2.03 2.13 2.05 2.13 Overall Olympians Non-Olympians National Team Non-National Team

*Data presented as mean 6 SD.

0.15 0.09 0.14 0.08 0.15 6 6 6 6 6 6 6 6 6 6

3.72 3.59 3.75 3.61 3.77

30 m (s) 15 m (s) Mean values (m)

TABLE 1. Mean values of each performance variable.*

Body mass (kg)

Brake push (s)

Side push (s)

America, Inc., Arlington Heights, IL, USA) with athletes wearing “minimal athletic clothing” and no shoes.

Back Squat. For squat testing, athletes self-selected their weight resistance for an absolute 3RM effort regardless of body weight. Adequate squat depth was determined as below femur parallel and visually judged by a certified strength and condition specialist. Warm-up time and rest time were self-determined to eliminate fatigue influence. For safety concerns, the current point system used by the TM

Journal of Strength and Conditioning Research

Copyright © 2018 National Strength and Conditioning Association. Unauthorized reproduction of this article is prohibited.

the

TM

Journal of Strength and Conditioning Research

| www.nsca.com

TABLE 2. Loadings and standardized loadings for Olympian/non-Olympian (case 1) and National Team/non-National Team (case 2).* Loadings

15 m

30 m

60 m

30-m Broad fly jump

Shot toss

Case 1: Olympian/nonOlympian L 219.67 224.54 12.66 20.19 20.41 0.53 L-star 29.22 211.50 5.94 20.09 20.19 0.25 Case 2: National Team/nonNational Team L 5.72 23.38 21.41 21.10 1.12 1.00 L-star 3.03 21.79 20.75 20.58 0.59 0.53

Power Body clean Squat mass

0.07 0.03

20.03 0.07 20.01 0.03

0.01 0.01

20.02 0.03 20.01 0.02

Brake push

Side push

2.28 21.92 1.07 20.90 26.39 23.38

4.20 2.22

*L = loading; L-star = standardized loading.

USABS is maximized at 200 kg; therefore, the values recorded may not indicate a true 3RM effort. Brakeman and Side Pushes. Push testing was performed using a modified, dry-land bobsled with wheels running on a rail track (Figure 1). Testing consisted of a 5-m “fly-in” start, from which point the electronic timing system was initiated and timing was collected from 5 m through 40 m down the track. The brakeman push was done from the rear of the sled with 2 hands placed on the 2 rear push handles. With the wheels of the sled on the parallel rails as shown in Figure 1, the athlete ran between the rails and loaded (jumped) into the sled when he felt he could not further accelerate the sled. The side push was done on either left or right side as chosen by the athlete with both hands on the side push bar as shown in Figure 1. The athlete loaded into the sled by first stepping on the side bunk of the sled and then stepping over

TABLE 3. Error matrix for classifying Olympians/ non-Olympians and National Team/non-National Team. Actual

Classified Olympians (National Team) Non-Olympians (non-National Team)

Olympians (National Team)

Non-Olympians (non-National Team)

13 (23)

6 (11)

1 (0)

55 (41)

the side and into the sled when the athlete determined that the sled was not further accelerating. Athletes were allowed up to 3 trials for both the brakeman and the side push testing. Statistical Analyses

All analyses were done using MATLAB (version 2014a; Mathworks, Inc., Natick, MA, USA). Before running the DA and the PCA, the overall mean values of the 11 variables were found for all 75 combine participants and then stratified into 2 cases for subsequent analysis (case 1: Olympians vs. non-Olympians; case 2: National Team vs. non-National Team). An analysis of variance (ANOVA) was run to ensure that significant differences existed between the variables of the 2 groups for both cases. That is, the mean values of every variable for the Olympians were tested to ensure a significant difference from the non-Olympians. The same statistical distinction was made by an ANOVA for case 2. The correlation found for each of the variables indicated strong enough relationships (r . 0.5) to perform a PCA without eliminating any test variables. This was confirmed by the Bartlett’s test of sphericity, which found the number of dimensions needed to explain the nonrandom variance of the data. The Kaiser-Meyer-Olkin index was also found to ensure the suitability of the data. First, to evaluate the present combine method, a DA was performed on the correlation matrix to determine loadings to classify athletes into their respective groups for both cases (Olympian or non-Olympian, National Team or nonNational Team). Previous probabilities were considered even, and cost of misclassification was zero across groups. The loadings were standardized by the covariance of each test variable to eliminate large variance within test variables and standardize variability in units (i.e., seconds, meters, kilograms, etc.). From the DA, a classification rule was established that categorized the athletes into appropriate groups. VOLUME 34 | NUMBER 9 | SEPTEMBER 2020 |

2621

Copyright © 2018 National Strength and Conditioning Association. Unauthorized reproduction of this article is prohibited.

Predicting Bobsled Pushing Ability deemed to have the greatest association with each other and similar effects on the total influence on the data set. These associations could be used to give an initial indication of which test variables could be merged for fewer test variables. A DA was then run for every possible combination of test variables to investigate the influence of each test variable on the percent error of misclassification. With 11 test variables, there were 211 or 2,048 Figure 2. Scree plot of the eigenvalues of PCA with percent of total variance. PCA = principle component possible combinations. Howanalysis. ever, the combination that excluded all the variables was not practical. This left 2,047 possible combinations. The combinations of test variables To test the classification rule, cross-validation was used, in with the lowest percent error were considered the best which each participant was individually assessed using the combinations. hold-one-out method. One participant was removed from the original data at a time, and the DA was run with a data RESULTS set consisting of 74 participants and 1 new participant. This Correlation of the test variables show that the tests most cross-validation method eliminates the tested participant closely related are the 30-m fly and the 60-m sprint at 0.98. from having an influence on the loadings and the classificaThe second and third most correlated variables are the brake tion rules. Then, the new participant’s group was compared and side pushes and the 30- and 60-m sprints at 0.97. Among with the actual group to test for misclassification. Misclassimultiple variables, it can be seen that the 4 sprint-related fication for the 2 cases included both type I (false-positive) variables are closely related with all correlations at over 0.84 and type II (false-negative) errors. The rate of misclassificawith each other. No other variables are correlated more than tion, or percent error, was calculated as the sum of all 0.71, but every test had at least 1 correlation with another misclassifications. test greater than 0.50. Mean values for all the participants are Next, a PCA was performed on the correlation matrix to presented in Table 1 along with the mean values for the determine which variables contributed the most to the Olympians, the non-Olympians, the National Team, and variance observed in the data set. The variables that the non-National Team. described the most variance in each component were

TABLE 4. Principle components with variance and cumulative variance. PCA Eigenvectors (variance; cumulative variance)

15 m

30 m

60 m

Broad 30-m fly jump

Shot toss

Power clean

Squat

Body mass

Brake push

Side push

PC1 (56.9%; 0.354* 0.359* 0.351* 0.336* 20.287 20.299 20.231 20.202 20.110 0.348* 0.332* 56.9%) PC2 (19.4%; 0.227 0.271 0.284 0.301 20.090 0.230 0.441* 0.354 0.523* 20.125 20.183 76.3%) PC3 (7.2%; 83.4%) 20.005 20.055 20.136 20.186 20.223 0.123 0.220 0.673* 20.295 0.364 0.397 *PCA = principle component analysis.

2622

the

TM

Journal of Strength and Conditioning Research

Copyright © 2018 National Strength and Conditioning Association. Unauthorized reproduction of this article is prohibited.

the

TM

Journal of Strength and Conditioning Research

| www.nsca.com

TABLE 5. Optimal test combinations.* Case 1: Olympian vs. non-Olympian Combinations 15 m 30 m 60 m 30-m fly Broad jump Shot toss Power clean Squat Body mass Brake push Side push 1 2 3 4 5 6 7 8

X X X X X X X X

X X X X X X X X

X X X X X X X

X X

X X

X

Case 2: National Team vs. non-National Team Combinations 15 m 30 m 60 m 30-m fly Broad jump Shot toss Power clean Squat Body mass Brake push Side push 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

X X X X X X X X X X X X X X X X X X X X

X X

X X X X X X X X X X X X X

X X X X X

X X

X X X X X X X X X X X X X X X X X X

X

X

X X

X

X

X

X X X

*Eliminated test variables are marked with “X.”

All variables were significantly different (p # 0.05). The Kaiser-Meyer-Olkin index of sampling adequacy measured to be 0.83. This classified the data as “meritorious,” indicating that the data are well suited for the analysis. The Bartlett’s test of sphericity indicated that all 11 dimensions are needed to explain the nonrandom variation of the data. Loadings and standardized loadings for the 11 variables from the DA are presented in Table 2, both for case 1 and case 2. When the DA was performed for case 1, the error matrix (using the hold-one-out method with the data) resulted in Table 3. Seven of the total 75 athletes were misclassified compared with the actual classification of the athletes, 1 as false negative and 6 as false positive. This gives a percent error of 9.33% (7 of 75). The error matrix for case 2 is also in Table 3, where the number of misclassifications is 11 of 75 (all are false positives), giving a percent error of 14.67%.

A PCA was then performed on the correlation matrix to obtain the most significant principle components, their eigenvalues, and their corresponding contribution to the total variance. By using the correlation matrix instead of the covariance matrix, the total variance of all the principle components (PCs) equaled 11, the number of variables. The eigenvalue of each PC divided by the total variance then gave the percent variance contribution from that PC. The PCA gave eigenvalues, with corresponding cumulative values, that showed the top 3 PCs accounted for more than 83% of the total variance as seen in Figure 2. The critical eigenvalue was set at 0.79 because subsequent PCs had a percent change of less than 50% of the preceding PC and were therefore deemed less significant. PC4 through PC11 each showed less than 5% of the total variance (16.6% cumulative), deeming the association of those variables less VOLUME 34 | NUMBER 9 | SEPTEMBER 2020 |

2623

Copyright © 2018 National Strength and Conditioning Association. Unauthorized reproduction of this article is prohibited.

Predicting Bobsled Pushing Ability significant than PC1 through PC3. The top 3 PCs are shown in Table 4 along with the cumulative variance of each PC. The top related variables within each PC are denoted with an asterisk. A DA was run again on the correlation matrix 2,047 times, in which each possible combination of the test events was evaluated. Considering just case 1, of the 2,047 tests, 136 combinations of tests resulted in a rate of misclassification (percent error) as good as or better than the original inclusion of all 11 tests (9.33%). Twenty of those 136 tests showed an improved percent error over using all the tests. Furthermore, 8 combinations equally had the best percent error of 6.67% (Table 5). For case 2, using the hold-one-out method, 107 combinations had a percent error of 14.67% or lower, which is the percent error of using all 11 available tests. Of those 107 combinations, 20 had a percent error of only 13.33% (Table 5).

DISCUSSION The primary aim of this investigation was to evaluate the current practices of USABS combine testing for its accuracy in predicting future sporting success. This was done using a unique combination of statistical methods including DA and PCA. The current combine assessment method uses 11 test variables. However, the need for each variable is unknown as it helps determine bobsled-specific success. In this investigation, success was determined by qualification for the Olympic Team (case 1) and the National Team (case 2). Using the statistical methods, evidence was gathered to both reduce the number of test variables needed and improve the misclassification rate of the current system. Using DA to investigate the ability to differentiate between 2 separate groups of athletes in 2 different cases, it was revealed that current practices of USABS are capable of identifying an Olympian from a non-Olympian with a 9.33% error rate and a National Team member from non-National Team member with a 14.67% error. Pooling previous and future combine data together (given identical protocols) would decrease the error rates and theoretically provide USABS with a more precise determination of which variable combinations best predict future success. Nonetheless, the relatively low misclassification rates in both cases illustrate that the current combine testing is indeed a useful tool to classify athlete potential for an Olympic Team, a National Team, or neither. From a practical standpoint, however, it is important that the large number of testing variables accurately reflect the combination of performance characteristics/attributes desired. Some of the tests can be redundant and cloud the importance of certain variables. To estimate which testing variables could be eliminated without voiding key athletic attributes, the variables were initially broken into principle components by performing a PCA on the correlation matrix. The PCA revealed that 83.4% of the variance was determined by the first 3 PCs. The first PC (56.9% of the

2624

the

total variance) emphasizes the importance of the 4 sprint tests and 2 dry-land pushes. Therefore, those 6 variables have a proven association, so redundancy in those tests could exist. Further examination through the DA shows which, if any, of the 6 tests could be safely ignored. The second PC (19.4%) has the highest values in the body mass and power clean variables, which stresses the importance of both body size and lower-body power, whereas the third PC (7.1%) demonstrates the importance of strength of a successful bobsledder by the high squat variable value (Table 4). Some of the test variables that comprise the majority of the PCs can then be eliminated to rerun the DA to see if the misclassification rate is affected. For example, the close relationship of the 4 sprint variables and the 2 push variables in the first PC would indicate that 1 or more of the 6 variables redundantly measure the speed aspect of an athlete. To specifically determine which test variables can be eliminated, DA was run for every possible combination of combine tests. Although more than 90% of the test combinations of the possible 2,047 did not produce an improved percent error, several combinations did lead to an improved test method. For case 1, 20 combinations had a better percent error and 8 of those resulted in the lowest percent error of 6.67% (Table 5). By eliminating up to a maximum of 5 test variables (30-m sprint, 30-m fly, the broad jump, and the 2 dry-land pushes), the same conclusions could be drawn with even better accuracy, as these tests did not aid in selecting the 14 Olympians from the original 75 that tested. The results of this investigation suggest that when seeking Olympic potential, the only tests necessary are the 15- and 60-m sprints, the shot toss, the power clean, and the squat, along with the athlete’s body mass. When considering case 2, there were 20 combinations of tests that led to a lower percent error (13.33%) than the inclusion of all the test variables (14.67%) (Table 5). In a similar manner to case 1, it was observed that some tests could be eliminated to improve the testing method efficiency. Up to 6 tests could be excluded, leaving only the 30-m sprint, shot toss, power clean, squat, and brakeman push. It should be noted that one combination was observed that improved the percent error in both case 1 and case 2. The test combination in which the 30-m sprint, 30-m fly, and the broad jump are eliminated leads to the minimum percent error for both cases. Therefore, to aid coaches with the most efficient testing method and lowest error rate, a combine of 8 tests would be advised. These 8 tests include the 15-m sprint, 60-m sprint, shot toss, power clean, squat, body mass, and both dry-land bobsled pushes. In a related study, Colyer et al. (2017) used only a PCA to reduce the number of events for an efficient combine in the sport of skeleton with no DA. After the PCA, Colyer et al. then factored the top variables from each PC together using a multiple regression analysis to predict on-ice start velocity, but not overall sport-specific success as several subjective

TM

Journal of Strength and Conditioning Research

Copyright © 2018 National Strength and Conditioning Association. Unauthorized reproduction of this article is prohibited.

the

TM

Journal of Strength and Conditioning Research variables could not be factored in Ref. 5. Additionally, by simply keeping 1 variable from each of the best PCs, the potential to over-reduce exists. If this method was applied to the data presented in this study for bobsled, it would have been concluded the best combine would consist of the 30-m sprint, body mass, and squat tests, a combine that would lead to an error rate of 14.67% for case 1 and 17.33% for case 2. This is a less accurate combine than the current combine used by USABS and a much less accurate method than the method presented here that includes a DA in addition to the PCA. Osbeck et al. (14) indicated that the 6-item combine administered in 1996 to bobsled athletes would be improved if the 60-m sprint was one of the eliminated variables along with the 100-m sprint and the 5-hop test. Interestingly, the results of this investigation show that in some combinations, the 60-m sprint does contribute to the prediction of both Olympians and National Team members. The results most likely differ because of the inclusion and exclusion of several other test variables. Furthermore, Osbeck et al. used a single on-ice brake push as the variable to measure success, where a successful athlete in this investigation is one who is selected to the Olympic Team for case 1 or the National Team for case 2. Additionally, despite the elimination of body mass based on investigative findings, coaches may still need to collect an athlete’s body mass as it contributes to the total team mass and may play a role in the selection that is dependent on the athlete’s teammates. Similarly, the brake and side pushes can be used as a technique assessment for subjective inclusion in the selection process by coaching staff despite the fact that the pushes are not needed in some of the most efficient combine assessments. This study has some limitations, including the aforementioned maximum point value for a 200-kg 3RM squat test used in the current USABS protocol. Determination of a true 3RM by allowing an athlete to self-select a resistance greater than 200 kg may have altered the results of the DA and PCA. Furthermore, the authors acknowledge that this investigation was limited to the current combine testing procedures of USABS specifically. Thus, it is possible that the statistical methodology used in this investigation may not apply directly as well to other combine-like settings in other sports because of the variance in number of measures obtained, similarity of events performed, and time course of events. Additionally, because of changes in USABS combine practices, this study spans only 2 Olympic-qualifying combines and their subsequent selection. Further analysis of additional combine performance may improve the accuracy; however, combine testing procedure would need to remain constant. Knowing this same problematic situation can be prevalent with other combine testing methods in many amateur and professional sports (National Basketball Association, National Football League, etc.) that use similar combine testing methods, the secondary aim of the investigation was

| www.nsca.com

to use the results of the first objective as a demonstration that combine assessments in other sports could improve in efficiency and accuracy. Whether it be in a combine setting or a continuous monitoring approach, advancing technology now allows coaches and trainers to gather more data on athlete performance than ever before, creating an increasing importance on the ability to interpret the data correctly (1,6,9,12,20). More specifically, collecting data from athlete assessments, such as combines, can be useful (7,9,12,14,19). However, the ability of coaches and sporting organizations to decipher the results is more important as that is what ultimately determines performance implications for the future success of a sporting organization. Therefore, coaches and trainers need to administer an appropriate test that will give information that will not only be beneficial but also maximize the ability of the combine testing to do as designed—predict future sporting success. In that sense, the results of this investigation indicate that collecting more data may not always be better in aiding organizational performance decisions. In some cases, the end results can be improved with the elimination of certain data, enabling organizations to build a more efficient and effective model to determine athlete selection for national, world, and Olympic championship competition. This investigation shows that current USABS combine practices are capable of differentiating between Olympians and non-Olympians, as well as National Team and nonNational Team members at a success rate of 90.67 and 85.33%, respectively. However, this rate can be improved upon by altering current combine testing via the elimination of certain testing variables through multivariate statistical methods. These objective results, taken together with the subjective eye of experienced bobsled professionals, can provide clear practical suggestions regarding the improvement of combine testing accuracy and efficiency. The conclusions drawn from this investigation provided empirical evidence to apply to other amateur and professional sports in which combine performance testing methods are used to assess athletic ability and predict sport-specific competitive success.

PRACTICAL APPLICATIONS The knowledge gained in this investigation can first be useful for the specific selection of elite bobsled push athletes from a competitive pool to select future Olympic and National Teams. The statistical analysis shows that the efficiency of the current combine assessment for USABS could be improved with not only fewer tests but also with a better misclassification rate. Applying the statistical methods could help coaches recruit the right athletes, evaluate current athletes, and select the most successful teams of push athletes for on-ice success. Second, the statistical methods used in this investigation can be used in many sports in the same manner. Several sports that use a version of a combine made up of several objective athletic VOLUME 34 | NUMBER 9 | SEPTEMBER 2020 |

2625

Copyright © 2018 National Strength and Conditioning Association. Unauthorized reproduction of this article is prohibited.

Predicting Bobsled Pushing Ability assessments can apply the statistical methods from this study to ensure that they are establishing a fair and successful combine. Where different sports may have different definitions of a classification rule, the general application of the DA and PCA would show what tests could be eliminated to avoid unuseful and redundant data collection.

ACKNOWLEDGMENTS The authors would like to thank the USABS coaches and staff. Coach Mike Kohn displayed great cooperation and assistance in the collection process and sharing the combine testing data. Coach Brian Shimer continues to have a willingness to think outside the box and explore new approaches to finding the best athletes to represent the United States in the sport of bobsled. The authors also thank Dr. Kent Eskridge for his assistance and inspiration to apply the statistical methods described in this investigation. The authors have no conflicts of interest to disclose.

REFERENCES 1. Buchheit, M. Want to see my report, coach? Sport science reporting in the real world. Aspetar Sport Med J 6: 36–43, 2013. 2. Bullock, N, Gulbin, JP, Martin, DT, Ross, A, Holland, T, and Marino, F. Talent identification and deliberate programming in skeleton: Ice novice to winter Olympian in 14 months. J Sports Sci 27: 397–404, 2009. 3. Bullock, N, Martin, DT, Ross, A, Rosemond, D, Holland, T, and Marino, FE. Characteristics of the start in women’s World Cup skeleton. Sports Biomech 7: 351–360, 2008. 4. Burr, JF, Jamnik, RK, Baker, J, Macpherson, A, Gledhill, N, and McGuire, EJ. Relationship of physical fitness test results and hockey playing potential in elite-level ice hockey players. J Strength Cond Res 22: 1535–1543, 2008. 5. Colyer, SL, Stokes, KA, Bilzon, JLJ, Cardinale, M, and Salo, AIT. Physical predictors of elite skeleton start performance. Int J Sports Physiol Perform 12: 81–89, 2017. 6. Evenson, KR, Goto, MM, and Furberg, RD. Systematic review of the validity and reliability of consumer-wearable activity trackers. Int J Behav Nutr Phys Act 12: 159, 2015. 7. Foster, C, Thompson, NN, and Snyder, AC. Ergometric studies with speed skaters: Evolution of laboratory methods. J Strength Cond Res 7: 193–200, 1993.

2626

the

8. IBSF. IBSF j International Bobsleigh and Skeleton Federation. Available at: http://www.ibsf.org/en/, 2017. Accessed October 22, 2017. 9. Kuzmits, FE and Adams, AJ. The NFL combine: Does it predict performance in the national football league? J Strength Cond Res 22: 1721–1727, 2008. 10. Leonardi, LM, Komor, A, and Dal Monte, A. An interactive computer simulation of bovsled pushoff phase with a multimember crew. In: Biomechanics X-B. Champaign, IL: Human Kinetics, 1987. pp. 761–766. 11. Lopes, AD and Alouche, SR. Two-man bobsled push start analysis. J Hum Kinet 50: 63–70, 2016. 12. McGee, KJ and Burkett, LN. The National Football League combine: A reliable predictor of draft status? J Strength Cond Res 17: 6–116, 2003. 13. McGill, SM, Andersen, JT, and Horne, AD. Predicting performance and injury resilience from movement quality and fitness scores in a basketball team over 2 years. J Strength Cond Res 26: 1731–1739, 2012. 14. Osbeck, JS, Maiorca, SN, and Rundell, KW. Validity of field testing to bobsled start performance. J Strength Cond Res 10: 239–245, 1996. 15. Sanno, M, Goldmann, JP, Braunstein, B, Heinrich, K, and Bru¨ggemann, GP. Joint specific mechanical power during vertical jumps of elite bobsleigh athletes. ISBS-Conf Proc Achieve 1: 1–2, 2013. 16. USABS. Combine test results. Available at: https://www.teamusa. org/USA-Bobsled-Skeleton-Federation/Results/Combine-Test, 2017. Accessed October 22, 2017. 17. USABS. Criteria. Available at: http://www.teamusa.org/USABobsled-Skeleton-Federation/Resources/For-Athletes/Criteria, 2017. Accessed October 22, 2017. 18. USABS. Features, events, results j team USA. Available at: https:// www.teamusa.org/USA-Bobsled-Skeleton-Federation, 2017. Accessed October 22, 2017. 19. Vandewalle, H, Peres, G, Heller, J, Panel, J, and Monod, H. Forcevelocity relationship and maximal power on a cycle ergometer. Correlation with the height of a vertical jump. Eur J Appl Physiol Occup Physiol 56: 650–656, 1987. 20. Wright, SP, Hall Brown, TS, Collier, SR, and Sandberg, K. How consumer physical activity monitors could transform human physiology research. Am J Physiol Regul Integr Comp Physiol 312: R358–R367, 2017.

TM

Journal of Strength and Conditioning Research

Copyright © 2018 National Strength and Conditioning Association. Unauthorized reproduction of this article is prohibited.