Pediatric Exercise Science, 2010, 22, 135-151 2010 Human Kinetics, Inc.

A Multivariate Analysis of Performance in Young Swimmers

Jos M. Saavedra and Yolanda Escalante
University of Extremadura

Ferran A. Rodrguez
University of Barcelona
This study develops multivariate models to predict swimming performance based on multidimensional assessment. 66 male (age 13.6 0.6 y) and 67 female (11.5 0.6 y) swimmers undertook a test battery including a sports background and training questionnaire, anthropometry, general and specific fitness tests, and technique. Competitive performance (LEN scores in three best events) was the predicted variable. A multiple linear regression model explained 82.4% of performance variability in males (based on age, sitting height, 30-min test, 6 50 m at 1:30, and swimming index) and 84.5% in females (age, 30-min test, 6 50 m at 1:30, and velocity at 50 m). Discriminant analysis using a four-group split-sample approach correctly classified 94.1% of the best male swimmers (based on age, 30-min test, 6 50 m at 1:30, shoulder extension, arm span, and height), and 71.0% of the best females swimmers (30-min test, horizontal floating, velocity at 50 m, and age). Chronological age was the main predictor of performance in this age category. Main predictive variables pertained to the anthropometric (particularly in males), specific fitness (aerobic speed and endurance), and technical domains (particularly in females). In these ages competitions should be organized according to year of birth and not by age categories.

Sports performance is the result of a complex process involving many factors. Performance capacity has been studied using a multidimensional assessment approach in different sports such as ice hockey (9), rowing (35), orienteering (8), tennis (39), rugby (28), Australian football (19), soccer (25,29), handball (10), triathlon (41), weightlifting (11), and volleyball (12,13). If performance was taken as a continuous variable, multiple linear regression (MLR) is the most common statistical technique (10,12,25,39). If, however, performance is considered a discrete variable, usually categorized in two groups (i.e., selected/nonselected, elite/nonelite), discriminant analysis (DA) seems to be the preferred technique (1,8,1012,19,28,29,35).

Saavedra and Escalante are with the University of Extremadura, Cceres, Spain. Rodrguez is with the University of Barcelona, Barcelona, Spain. 135

136 Saavedra et al.

Swimming performance is determined by the interaction of morphological, physiological, psychological and technical factors, based on individual genetic endowment, and continuously modulated by the training process (13). Some studies predicting swimming performance using MLR, both in adults (32,34) and young swimmers (1,4,15,17,18), have been able to explain between 20% and 93% of variability in performance. Discrepancies may be related to differences in the performance criterion: 25 yd (32,34), 100 freestyle (1,15) 100 breastroke (17) and 400 freestyle (4,18). One study has predicted performance in two groups (faster and slower) using DA, with prediction percentages over 82% (21). More recently, more novel statistical techniques have been used, such as neural network technology (23) and nonregression models (31) which have predicted performance on 99% and 60% of the cases, respectively. However, if these multivariate techniques are to be applied to talent selection in swimming, it is of equal interest to identify the best individual young swimmers (MLR) as it is to predict the best group of swimmers (DA), especially if we bear in mind that only about forty out of the one hundred best swimmers in the 1314 year old ranking will continue to rank among the one-hundred best at 1718 years (38). Therefore, the aims of this study were (a) to analyze swimming performance in young peripubertal athletes by developing multivariate predictive models (MLR and DA) based on a wide variety of assessments from a multidimensional perspective, (b) to compare their predictive power with that of models developed in nonswimming related studies; and (c) to determine which group of variables actually predicted performance level at young age.

Methods
Subjects
Subjects were 133 young swimmers, 66 males (13.6 0.6 years) and 67 females (11.5 0.6 years), selected as the best national swimmers of their age category in Spain. According to the National swimming federation (RFEN) the alevn category includes females of 1314 years and males of 1112 years of age. The study was approved by the Bioethics Committee of the University of Extremadura (Spain). The swimmers parents or legal tutors signed an informed written consent previously to their participation.

Assessment Procedures
All subjects undertook a comprehensive battery of tests, which included assessment in the following domains: (a) sports background and training status, (b) anthropometry, (c) general fitness tests, (d) specific fitness tests, (e) technical analysis, and (f) multidimensional evaluation. In accordance with the aims of the study (i.e., to develop multivariate models explaining swimming performance from a multidimensional perspective), a considerable number of assessments were included as predictive variables to ensure comprehensive evaluation. All tests are widely used in swimmers assessment and were performed one week after the National summer championships, the most important competition of the season.

Analysis of Performance in Young Swimmers

137

Sports Background and Training Status

This domain was assessed by an ad hoc questionnaire including 26 items: 7 related to social background, 6 on sports practice, and 13 items on swimming training and competition. This questionnaire assessed the relationship between swimming performance and variables such as previous sports participation (both for swimmers and parents), swimming practice (age of begin to swim and compete), number of weekly training sessions, training volume per session (pool and dryland), amount of specialized training, number of meetings per season, and training methods (circuit training, elastic bands, weights, etc.). After exploratory analysis (i.e., building specific MLR and DA models restricted to this specific domain), only training volume variables were used for further statistical analyses.

Anthropometry
Anthropometric measurements were taken according to standardized procedures (30) by an ISAK (International Society for the Advancement of Kinanthropometry) certified anthropometrist. Measures included body dimensions (height, sitting height, arm span, and weight), lengths and widths (hand and foot), skinfolds (triceps, subescapular, biceps, supraspinale, abdominal, front thigh, and medial calf), breadths (biacromial, biiliac, bitrochanteric, knee, elbow, and wrist), girths (chest, arm flexed, gluteal, thigh, and leg). The following proportionality indices were calculated: body mass index, arm span/height index, bitrochanterial/biacromial breadth index, biacromial breadth/height index, bitrochanterial breadth/height index, chest girth/height index, and gluteal girth/height index. Body composition was assessed using a two-compartment model (24). Sum of six skinfolds was used as main adiposity index. In addition, although there is no wide agreement as to which regression equations should be used to estimate fat mass from skinfolds measures in athletic populations of children and adolescents, we chose to use the linear regression equations based on the sum of six skinfolds used by Carter in Olympic athletes of any age and for each gender category (2). Somatotype was determined using the anthropometric method (16), and the three components (endomorphy, mesomorphy, and ectomorphy) were analyzed separately. Sexual maturation was assessed from the development of secondary sex characteristics according to Tanner (40). Stages of genital (males), breast (females), and pubic hair (males and females) development were rated in five stages by a medical expert by inspection. The age of menarche (females) was assessed by recall.

General Fitness Tests

General fitness was assessed using the Eurofit test battery (7): shuttle run test assessed general aerobic endurance, flamingo balance assessed general balance, plate tapping assessed segment velocity of the upper limbs, sit and reach assessed flexibility of the trunk and lower limbs, horizontal jump assessed explosive strength of the lower limbs, hand dynamometry assessed grip, abdominals in 30 s assessed trunk power, flexed arm hang assessed muscular resistance of the arms and shoulders, and shuttle run test 10 5 m assessed agility-velocity.

138 Saavedra et al.

Specific Fitness Testing

Specific fitness was assessed by means of flexibility, sprint swimming, endurance swimming, and hydrodynamic tests. Flexibility testing included the measurement of the maximum range of movement of the shoulder, and ankle flexion and extension, from video records on standard positions. Swimming tests were all performed at maximal effort. A 10-m test was used to assess maximum speed; starting from an immersed position, the swimmer accelerated during the first 10 m (start) and maintained maximum speed during 10 m (finish). The 30-min test (27) was used to assess aerobic endurance. The 6 50 m with 1:30-min:s start at personal stroke was used to assess speed endurance (3). Hydrodynamic evaluation included glide and floatability tests (3): (a) to assess gliding hydrodynamic position and explosive strength of the lower limbs we measured by visual inspection the distance gained by gliding in prone position after turning, and after starting; (b) to assess horizontal floating we measured the time lapsed between horizontal floating (initial position) and vertical floating in steady position; and (c) to assess vertical floating we rated the part of the body remaining out of the water from a zero score if the head was completely immersed up to seven if the water surface was at the level of the neck.

Technical Analysis
Video analysis of swimming technique, both qualitative and semiquantitative was also undertaken. Qualitative analysis consisted of technical assessment by an experienced coach with more than ten years of experience in underwater technical evaluation. Swimmers were filmed from sagital and frontal planes while swimming in their fastest stroke, and their technique compared with a standard technical pattern (23) to identify the most common technical errors (36). Filming was carried out in a 25-m pool, on a separate trial from every plane, and starting from the water. Subsequently, a technical swimming index was calculated, as the total number of errors made by each swimmer divided by the total number of possible errors (37). Semiquantitative kinematic analysis was performed during a 50-m all-out swim for the calculation of: start velocity (010 m), approach velocity (17.525 m), pull-out velocity (2532.5 m), turn velocity (17.532.5 m), finish velocity (4050 m), stroke rate, stroke length, and swimming index (6).

Multidimensional Evaluation
Combined analysis of variables from the different domains (sports background and training status, anthropometry, general fitness tests, specific fitness tests, and technical analysis) was made by developing multivariate models (see Statistical analysis for details).

Performance Evaluation
LEN table (22) was used to assess competitive performance level. LEN scores measure how close a certain personal best time is from the World record in each competitive event, allowing also comparing two time records, within or at different events. Individual performance level was quantified as the sum of LEN scores in the three best personal events during the season. 50-m personal best times were not

Analysis of Performance in Young Swimmers

139

included assuming that the swimmers were not specialized in sprint swimming at their young age; in fact, 50 m events are not included at the National championships program. Thus, performance was evaluated as the result in fastest three competitive events swum in one of the fours strokes (i.e., freestyle, backstroke, butterfly, or breastroke) at any of four different race distances (i.e., 100, 200, 400, 800 females, and 1,500 m males). These criteria were chosen to obtain an overall assessment of competitive performance, not restricted to a single event (stroke and/or distance).

Statistical Analysis
Unless specified, data are expressed as means SD (SD). The normality and equal variance of the distributions were tested using the Kolmogorov-Smirnov and the Levene tests, respectively. Some typified variables (10-m all-out test, 6 50 m at 1:30 start, glide distance after turn and start, and all parameters from technical quantitative analysis) were expressed as z values. Before MLR analysis, Pearsons simple correlation coefficient (R), and simple and multiple linear coefficients of determination (R2 and Rm2) were calculated for all parameters in relation to performance level. As a secondary analysis, partial correlation coefficients (Rp) with chronological are as a control variables were also computed. For each type of assessment, MLR models were developed using the stepwise selection procedure. Performance was the predicted variable, and the number of candidate predictive variables in the model was limited to nv n / 5 (i.e., 13 variables) to prevent obtaining spurious relations among variables (26). Stepwise selection procedure consists in removing the variable with the largest probability of F if a certain preestablished value is exceeded (i.e., p .10). The equation is recomputed without the variable and the process is repeated until no more independent variables can be removed. Then, the independent variable not in the equation with the smallest probability of F is entered if the value is smaller than a certain preestablished value (i.e., p .05). All variables in the equation are again examined for removal. This process continues until no variables in the equation can be removed and no variables not in the equation are eligible for entry, or until the maximum number of steps has been reached. In DA, subjects were classified by the sample-splitting method in four groups according to their performance level (i.e., A the best group of swimmers, D the worst) using a stepwise selection procedure. The criterion used to determine whether a variable entered the model (i.e., discriminant function) was Wilkss Lambda, which measures the deviations within each group with respect to the total deviations. The sample-splitting method included initially the variable that most minimized the value of Wilkss Lambda, provided the value of F was greater than a certain critical value (i.e., F = 3.84 to enter). The next step was pairwise combination of the variables with one of them being the variable included in the first step. Successive steps were performed in the same way, always with the condition that the F-value corresponding to the Wilkss Lambda of the variable to select has to be greater than the aforementioned entry threshold. If this condition was not satisfied, the process was halted, and no further variables were selected in the process. Before including a new variable, an attempt was made to eliminate some of those already selected if the increase in the value of Wilkss Lambda was minimal, and the corresponding F-value was below a critical value (i.e., F = 2.71 to remove). Wilkss Lambda, canonical correlation index, and percentage of subjects correctly classified

140 Saavedra et al.

for the whole sample and for each category group were computed as indicators of performance predictive capacity. A P-value <.05 was considered to be statistically significant. Having set a two-tailed alpha value of .05, beta was set at four times alpha (i.e.,.20), according to Cohen (5). Since power is 1-beta (1.0 - .2 = .8), power was set at .80. Using a moderate effect size of .5, the calculated statistical power for the studied sample (i.e., n = 66 and 67 for males and females, respectively) was .82. The Statistical Package for Social Sciences (SPSS for Windows, version 15.0) was used for all analyses.

Results
A summary of chronological age, body size, and sexual maturation status of the swimmers is presented in Table 1. Swimming performance level, assessed as the mean of the three best scores on the LEN table, was 508.4 70.7 points (males) and 476.0 76.0 points (females) on average. Thus, swimmers reached, on average, about 50.8% (males) and 47.6% (females) of absolute World records in their three best events. Table 2 shows the main results from the different assessments and variable correlates with competitive swimming performance. Linear simple correlation (R) and partial correlations coefficients (Rp) calculated with age as control variable are shown for variables significantly correlated with performance (p .05). Multiple determination coefficients (Rm2) of MLR models for each sex are also shown. The selected variables included in the MLR models explained 82.4% and 84.5% of the variance in competitive performance in males and females, respectively. The variables selected to enter the models were five, both in males (age, sitting height, 30-min test, 6 50 m at 1:30 start, and swimming index), and in females (age, 30-min test, specific endurance index, 6 50 m at 1:30 start, and mean velocity at 50-m all-out test). Table 3 shows the DA of each assessment. In the multidimensional evaluation, predictive models correctly classified 72.7% and 68.3% of the sample, with only six variables for males (age, 30-min test, 6 50 m at 1:30 personal stroke, shoulder extension, arm span, and height), and four variables for females (30-min test, horizontal floating, velocity at 50 m personal stroke, and age). Table 1 Gender, Age, Body Size, and Sexual Maturation Status of the Swimmers
Age (years) Height (cm) Weight (kg) Pubic hair (stage)* Genital development (stage)* Breast development (stage)* Age of menarche (years) #
Note. Values are mean SD. *According to Tanner (1962; 40) # n = 19

Males (n = 66) 13.60 .56 171.12 7.50 57.95 8.18 3.61 1.158 3.47 1.57 -

Females (n = 67) 11.51 .55 154.75 7.47 43.96 7.17 2.99 1.19 2.63 1.01 10.92 .68

Table 2 Main Results From the Different Assessments and Correlates With Competitive Swimming Performance
Males (n = 66) Females (n = 67) Rp .330 .396 .263 N/A N/A .401 .507 .243 .386 8.12 1.34 4052 929 Rm2 Mean SD R Rp Rm2 Mean SD 9.80 2.89 4410 1136 .767* .634 .579* .587 .500 .409 .465 R

Test domain Test variables

.301

.246

.741* .504 .398 .459 .414 .401 .256 .326

.441 .375 .350 .309 .392 .287 .314 .345 .273 .263 .354 .294 .318 .265 .380 .304 .263

.281

Training status Training volume (h/wk) Training volume per session (m) Somatic evaluation Chronological age (years) Height (cm) Sitting height (cm) Arm span (cm) Weight (kg) Hand length (cm) Hand width (cm) Foot lenght (cm) Foot width (cm) Biacromial breadth (cm) Biiliac breadth (cm) Bitrochanteric breadth (cm) Knee breadth (cm) Elbow breadth (cm) Wrist breadth (cm) Chest girth (cm) Arm flexed girth (cm) Gluteal girth (cm) Thigh girth (cm) 13.60 0.56 171.12 7.50 80.46 4.14 177.48 8.77 57.95 8.18 17.55 2.15 9.43 0.62 24.89 1.42 8.40 0.61 37.77 1.93 25.50 1.75 29.36 2.12 9.74 0.47 6.87 0.42 5.45 0.30 90.19 5.05 28.46 2.27 84.92 7.75 51.20 5.95 11.51 0.55 154.75 7.47 77.24 4.10 158.86 8.72 43.96 7.17 15.46 0.99 8.08 0.57 22.20 1.62 7.22 0.50 33.79 1.62 23.03 1.82 27.06 2.30 8.66 0.41 5.92 0.32 4.82 0.25 82.78 6.34 24.85 2.12 80.34 7.23 47.07 7.27 .527 .252 .430 .287

141

(continued)

Table 2 (continued)
Males (n = 66) Females (n = 67) Rp .268 .448 .269 Rm2 R .310 Rp Rm2 R

142

Test domain Test variables

.278

.259 .244

Mean SD 34.00 2.49 3.72 2.04 20.96 0.95 52.72 2.26 49.68 4.49 19.66 1.87 61.76 20.93 9.08 2.20 2.57 0.92 4.28 1.08 3.92 1.03 .242 .369 .275 .571 .365 .277

Mean SD 30.57 2.03 2.64 2.39 22.09 0.99 53.49 3.18 51.89 3.59 18.32 1.91 78.74 22.38 12.69 2.21 3.58 1.01 3.67 0.89 3.68 1.04

Leg girth (cm) Arm span/height index Biacromial breadth/height index Chest girth/height index Gluteal girth/height index Body mass index (kg/m2) Sum of six skinfolds Fat mass (%) Endomorphy Mesomorphy Ectomorphy General fitness testing Shuttle run endurance (stage) Flamingo balance (attempts) Plate tapping (s) Sit-and-reach (cm) Horizontal jump (cm) Hand dynamometry (kg) Abdominals in 30 s (reps) Flexed arms hang (s) Shuttle run 10 5 m (s) Specific fitness testing Shoulder flexion (grades) Shoulder extension (grades) 9.83 1.36 6.14 2.91 11.91 1.46 9.48 6.71 196.0 16.33 28.65 6.84 26.09 3.67 43.65 16.52 20.44 1.07 .385 .279 .312 .508 .346 .351 .550 .351 .379 3.53 8.52 62.68 13.39 7.51 1.45 5.12 .86 13.40 1.48 10.46 6.27 177.6 14.26 15.92 4.21 21.84 2.44 24.55 14.72 22.78 1.19 .360 .540 .291 .362 .287 .294 8.37 11.03 65.60 13.70 .238 .302

(continued)

Table 2 (continued)
Males (n = 66) Females (n = 67) Rp Rm2 R Rp Rm2 R

Test domain Test variables

Mean SD 163.76 6.50 108.9 12.46 0.01 0.97 1.24 0.11 0.00 0.91 7.29 1.16 0.02 0.97 0.01 0.96 .476 .700* .685* .528 .526 .488 0.01 0.97 .459 .286 .539 .501 .308 .365 .674 .840* .775* .425 .458 .551 .613

Mean SD 165.61 6.58 112.27 9.11 0.10 0.98 1.11 0.07 0.01 0.98 7.36 1.19

Ankle flexion (grades) Ankle extension (grades) 10-m test maximal velocity (z value) 30-min test (ms-1) 650 m at 1:30 start (z value) Gliding distance (prone; m) Gliding after turn (z value) Gliding after start (z value) Technical analysis Start velocity (z value) Approach velocity (z value) Pull-out velocity (z value) Turn velocity (z value) Finish velocity (z value) Mean velocity (z value) Stroke rate (z value) Stroke length (z value) Swimming index (z value) MLR models 0.00 0.97 0.00 0.97 0.00 0.97 0.00 0.97 0.00 0.93 0.03 0.93 0.02 0.97 0.00 0.97 0.01 0.94 .251 .250 .317 .293 .248 .824# .471 .442 .415 .505 .466 .548 .523 .448 .576* 0.00 0.97 0.01 0.93 0.10 0.94 0.00 0.97 0.02 0.96 0.00 0.97 0.02 0.96 0.00 0.97 0.00 0.97 .624 .411 .630 .654 .616 .676* .332 .488 .543 .365 .484 .552 .490 .599 .404 .509

.845#

143

Note. Pearsons linear simple correlation (R) and partial correlations coefficients (Rp) calculated with age as control variable are indicated for variables significantly correlated with performance (P.05). Multiple determination coefficients (Rm2) of multiple linear regression (MLR) models for males and females are also shown at the bottom. N/A: Not applicable (chronological age is the control variable for partial correlations) * Variables entered in MLR models # Multiple regression determination coefficients (Rm2) for all variables in MLR models

Table 3 Discriminant Analysis Models for the Different Assessments and the Overall Multidimensional Evaluation
Evaluations Sports background and training status A Group (best swimmers) B Group C Group D Group (not-so-good swimmers) Whole sample Wilkss L Canonical correlation index Variables in the model Somatic evaluation A Group (best swimmers) B Group C Group D Group (not-so-good swimmers) Whole sample Wilkss L Canonical correlation index Variables in the model General fitness testing A Group (best swimmers) B Group C Group D Group (not-so-good swimmers) Whole sample Wilkss L Canonical correlation index Variables in the model Specific fitness testing A Group (best swimmers) B Group C Group D Group (not-so-good swimmers) Whole sample Wilkss L Canonical correlation index Variables in the model Males (n = 66) 70.6% 56.3% 35.3% 31.3% 34.8% .797 .450 Training volume per session 88.2% 43.2% 58.8% 56.3% 62.1% .397 .822 Height, arm span, age 37.9% 0% 47.1% 68.8% 35.3% .724 .638 Hand dynamometry, abdominals in 30 s 70.6% 75.0% 64.7% 62.5% 68.2% .544 .807 6x50 m at 1:30 start, 30-min test, shoulder extension Females (n = 67) None* 70.6% 52.9% 31.3% 29.4% 46.3% .452 .851 Age, sexual maturation (pubic hair) 58.8% 35.3% 37.5% 68.8% 46.3% .597 .722 Hand dynamometry, shuttle run endurance 76.5% 41.2% 62.5% 82.4% 65.7% .390 .789 30-min test, horizontal floating (continued)
144

Analysis of Performance in Young Swimmers

145

Table 3 (continued)
Evaluations Technical analysis A Group (best swimmers) B Group C Group D Group (not-so-good swimmers) Whole sample Wilkss L Canonical correlation index Variables in the model Multidimensional evaluation A Group (best swimmers) B Group C Group D Group (not-so-good swimmers) Whole sample Wilkss L Canonical correlation index Variables in the model Males (n = 66) 82.4% 20.0% 76.5% 66.7% 62.5% .605 .672 Turn velocity, stroke rate 94.1% 62.5% 64.7% 68.8% 72.7% .405 .893 Age, 30-min test, 6 50 m at 1:30 personal stroke, shoulder extension, arm span, height Females (n = 67) 65.7% 43.8% 62.5% 70.6% 76.5% .523 .691 Velocity 50 m individual stroke 71.4% 50.0% 62.5% 88.2% 68.3% .427 .845 30-min test, horizontal floating, velocity at 50 m personal stroke, age

Note. Data are percentage of subjects correctly classified in each of the four-category groups (A group, best swimmers; D group, not-so-good swimmers) and within the whole sample. Wilkss Lambda, canonical correlation index, and variables entered in the model are indicated. * No discriminant predictive models were obtained.

Discussion
This study contributes with MLR models predicting performance as a continuous variable in young swimmers at the top national level with a high degree of predictive power (explaining over 82% of the variance in performance in both sexes), as well as with DA models which categorized performance at four levels reaching predictive values in the fastest swimmers group of 94% in males and 71% in females.

Comparison With Other Sports

The results show a swimming performance prediction by MLR models of 82% (males) and 85% (females) based on only five predictive variables. These values are far higher than those reached in other sports: 46% in tennis (39), 29% in soccer (25), or 21% in volleyball (12); only in triathlon a higher prediction was achieved (98%), although it is worthy to note that that study was performed with recreational athletes (41). Three of the variables selected (i.e., age, aerobic endurance, and

146 Saavedra et al.

speed endurance) entered the predictive models in both sexes. This is consistent with studies on young athletes (12,25,39), in which aerobic endurance was shown to be a strong predictive variable. With regard to DA, variables discriminated among performance groups on 73% (males) and 68% (females) of all swimmers with six and four variables, respectively. These results are lower than those obtained in studies on other sports: 83% in rowing (35), 94% in rugby (28), 80% in Australian football (19), 8284% in male handball (10), and 84% in weightlifters (11). There are some studies, however, which showed similar or lower prediction values: 55% in ice hockey (9), 64% in orienteering (8), 6970% in female handball (10), and 79% in volleyball (13). However, it is worthy to note that our study is to our knowledge the only one in which subjects were categorized in four sample-splitted groups according to performance level. Despite the obvious difficulties this creates for categorization (26), prediction of the best performance group was achieved at 94% in males and 71% in females. With regard to the variables selected for our study, age and aerobic endurance (30-min test) entered the discriminant models in both sexes. In other studies the selected variables belonged to: physical fitness and anthropometric domain in ice hockey (9), rugby (28), Australian football (19), handball (10), rowing (35), and weightlifting (11); technical domain in volleyball (13); psychological domain in soccer (29); and anthropometric domain (age) in orienteering (8).

Multiple Lineal Regression Models in Swimming

As we have seen, the results show high prediction performance values with a large percentage of variance explained by the MLR models developed (males 82%, females 85%). These values are higher than those obtained in performance prediction in 400 m freestyle in French swimmers (males and females, 74%; 4), 100 m freestyle in Greek swimmers (males, 59%; female 17%; 15), 100 m breaststroke in Estonian swimmers (females, 28%; 17), or 100 m butterfly (males, 20%) and 100 m freestyle (males, 20%) in Australian swimmers of similar age and competitive level (1). However, prediction rates in sprint swimming (25 yd) are close to (males 75%, female 72%; 34) or higher than (males, 89%; 18,32) in our results, though in the first study the competitive level of the swimmers was lower, and in the second the swimmers were adults. The variables selected in the model in both genders were age (males, R2 = .59; females, R2 = .55), aerobic endurance (males, R2 = .49; females, R2 = .71), and speed endurance (males, R2 = .47; females, R2 = .60). The models did not select anthropometric variables, except sitting height in males (R2 = .34), which differs from other studies in which these were the most selected variables (1,4,15,18,32,34); it should be remembered, however, that most of these studies did not assess specific fitness. Age was the strongest predictor in males and the third in females, in agreement with a previous study (19), which suggests that to minimize the effect of age on performance in the formative years, swimming competitions should be organized by year of birth rather than by age category (two years of age together). The 30-min test, an indicator of aerobic endurance which was validated with velocity at 4 mmolL1 lactate as a reference criterion (27), was the strongest predictor of performance in females, and the second strongest in males. This finding points at aerobic endurance as an essential factor for performance at this age, and supports

Analysis of Performance in Young Swimmers

147

the notion that aerobic training should be one of the main training goals in young swimmers, more even so in the females. The third common predictor for both sexes was the 6 50 m test swum at 1:30 at personal stroke. This test is considered an index of speed endurance (3) and suggests that mixed, aerobic-anaerobic specific endurance also plays a role in determining performance. However, we need to take into account that both aerobic and speed endurance are also strongly correlated with age; partial correlations with chronological age as control variable (Table 2) showed that both aerobic endurance (males, Rp=.54; females Rp=.55), and speed endurance (males, Rp=.50; females Rp=.61) were in fact highly correlated with swimming performance, independently from age, although at a somewhat lower level. From the technical point of view, the swimming index was in males not as good as a predictor of performance (R2=.33) as in a previous report (R2=.898; 18). In females though, most technical variables showed high partial correlations with performance (Rp=.36.60). From an overall perspective, our results indicate that age and specific fitness, particularly those related with aerobic and speed endurance, are main predictors of performance as a continuous variable, and that general fitness, anthropometric (including sexual maturation) and technical variables are less relevant, except for females, in which technical variables reach high levels of correlation with performance even when controlling for age.

Discriminant Analysis in Swimming

The results show high performance prediction of the DA models both in males (73% of the whole sample and 91% of the best swimmers correctly classified) and in females (68% of the whole sample and 71% of the best swimmers correctly classified). These values are similar than those obtained in young U.S. swimmers of similar age (83%) and adult swimmers (88%; 21), although in those studies performance was categorized in only two groups (i.e., faster and slower swimmers). On analyzing each of the assessments performed, we can see that in the sports background and training status domain, training volume per session (71% of best swimmers correctly classified) was the only variable entering the model, whereas no variables were selected in females. This may indicate that that training volume is only a key factor for performance in males; this difference may be due to the better performance level of males in the sample as compared with the females (males: 508 LEN points, females: 476 LEN points, p = .12). This argument has already been suggested as a possible explanation for different predictive models in males and females at this age (15). However, this has to be interpreted with caution since other factors, such as training intensity, were not included in the model due to methodological reasons. With regard to the anthropometric assessment domain, the DA variables in males were height, arm span, and age (62% of best swimmers correctly classified). The first two variables were also closely related in the current study (R = .918, p < .01), since taller swimmers usually show a larger arm span, which benefits swimming efficiency (i.e., larger stroke length). Arm span has also been identified as a good predictor of performance in young swimmers using a MLR model (R2 = .454 p < .001; 18). In females, the discriminant variables were age and sexual maturation (pubic hair; 46% of best swimmers correctly classified). This gives

148 Saavedra et al.

an indication of the relevance of sexual maturation in swimming performance at young, peripubertal age (24). In general fitness testing in males, discriminant variables were hand dynamometry and abdominals in 30 s, although with low discriminant power (35% of all swimmers correctly classified). In females, also hand dynamometry and shuttle run endurance were selected (performance prediction: 46% of all swimmers correctly classified). This last variable indicates general aerobic endurance which, in our study, was also closely related to specific aerobic endurance (R = .55, p < .001). Thus the relevance of this capacity in the formative years is to be acknowledged. As to specific fitness testing, the discriminant variables in males were 6 50 m at 1:30 start at personal stroke, 30-min test, and shoulder extension (68% of all swimmers correctly classified). The first variable informs on speed endurance, a determining capacity in swimming performance in 100 and 200 m (14). 30-min performance also highlights the importance of aerobic endurance in swimming (14) at the formative stages, and is in complete agreement with MLR results. In females, 30-min performance were once more selected, together with horizontal floating (performance prediction: 66% of all swimmers correctly classified), which informs on good body position while swimming, given that drag is directly proportional to the cross sectional area of body (23). In the technical analysis domain, the discriminant variables in males were turn velocity and stroke rate (63% of all swimmers correctly classified). The first of these variables underlines the importance of efficient turns, only second to swimming speed in 50 m; since quantitative analysis was made in a 50-m swim, the relevance of turns at longer distances may be even higher. Stroke rate points out the importance of swimming with the lowest cycle frequency to guarantee efficient swimming (20). A study carried out on swimmers of similar ages also selected this variable to predict performance (21). In females, only one variable (i.e., mean velocity in 50 m) was selected in the semiquantitative analysis (77% of best swimmers correctly classified), which suggests that the fastest females in 50 m are those who obtain the best performance also in longer distances at this age. Finally, in the multidimensional evaluation (i.e., all variables considered) the discriminant variables for males were age, 30-min test, 6 50 m at 1:30 start at personal stroke, shoulder extension, arm span, and height (73% of all swimmers, and 94% of best swimmers correctly classified). It is worthy to note that, the three first variables were also selected by the MLR model; this is a strong indication that (a) chronological age, as discussed before, is the main predictor of performance in this age category, and (b) aerobic endurance and speed endurance are key factors for performance among young swimmers of this age. Also body linear dimensions and shoulder flexibility seem to discriminate among best and not-do-good male swimmers. In females the discriminat variables were 30-min test, horizontal floating, 50-m mean velocity, and age (68% of all swimmers, and 71% of best swimmers correctly classified). Again, aerobic endurance and age were selected in the DA model, together with swimming speed and horizontal floatability.

Limitations
Firstly, in spite of the relative homogeneity of the samples, several variables are correlated with age, which appears to be the main single predictor of swimming

Analysis of Performance in Young Swimmers

149

performance at this age category in both sexes. The computation of partial correlations coefficients, to a certain extent, allowed weighing this effect. Secondly, the physical and technical assessments made are, in most cases, indirect estimations of physical and technical capacities and abilities; however, we must consider that the large number of subjects evaluated (n = 133) and the large number of tests performed (ninety), would have made more sophisticated measurements (i.e., ergospirometric measurements, biomechanical analysis) impractical. Thirdly, while most swimmers were of peripubertal age, 24 (17 males, 7 females) were already at the adult sexual stage; although we initially believed that this fact would influence performance, the results show that this was not so. Finally, performance was assessed as the best score in each swimmers three fastest events, and not just in one. Perhaps using a different criterion would have changed our results; however, we took into consideration that the swimmers were in their formative years, in which full specialization is not normally attained. Therefore, considering the best three events would have given a better picture of the overall performance capacity of the swimmers.

Conclusions
This study has developed multivariate models to predict swimming performance which, in overall comparison with other sports, reached similar or higher predictive power. Likewise, the models have shown to be efficient in predicting performance, both by detecting the swimmers individual potential (MLR: 82% of males, and 84% females of variability in performance was explained by the model), and by identifying the best group of swimmers (DA: 94% males, 71% females correctly classified). Most selected variables pertain to the anthropometric (particularly in males), specific fitness, and technical domains (particularly in females), outstanding chronological age and aerobic and speed endurance. Altogether, these results support the notion of the multidimensional nature of this sport. Two practical implications may be derived: (a) in these ages competitions should be organized according to year of birth and not by age categories, and (b) aerobic and speed endurance should be considered not only as capacities which guarantee long-term athletic development, but also as predictors of performance itself. Acknowledgments
This study was partially granted by the Spanish National Swimming Federation (Real Federacin Espaola de Natacin) and the Spanish Higher Sports Council (Consejo Superior de Deportes). The authors wish to thank James McCue for revising the English text and each and every one of the participants in the study. During completion of this paper J.M. Saavedra was a visiting researcher at the University of Ulster with a grant awarded by the Ministry of Education and Science of Spain (Ministerio de Educacin y Ciencia; JC2007-00316). Y. Escalante was also a visiting researcher at the same University with partial grants awarded by the Autonomic Government of Extremadura (Junta de Extremadura) and Social European Funds (GRU07109-GRU08118). We would like to acknowledge the work of the two anonymous reviewers who made the quality of the work being improved.

150 Saavedra et al.

References
1. Blanksby, B.A., J. Bloomfield, M. Ponchard, and T.R. Ackland. The relationship between anatomical characteristics and swimming performance in state age-groups championship competitors. J. Swimming Res. 2:3036, 1986. 2. Carter, J.E.L. Body composition of Montreal Olympic athletes. In: Physical structure of Olympic athletes (part I), J.E.L. Carter (Ed.). San Diego: Karger, 1982, pp. 107116. 3. Cazorla, G. Tests spcifiques dvaluation du nageur. Cestas: Fdration Franaise de Natation, 1993. 4. Chatard, J.C., S. Padilla, G. Cazorla, and J.R. Lacour. Influence de la morphologie et de lentranement sur la performance en natation. Sci. Tech. Act. Physiques Sport. 8:2328, 1987. 5. Cohen, L. Statistical power analysis for the behavioural sciences. Hillsdale, NJ: Lawrence Erlbaum Associates, 1988. 6. Costill, D.L., J. Kovaleski, D. Porter, J. Kirwan, R. Fielding, and D. King. Energy expenditure during front crawl swimming: predicting success in middle-distance events. Int. J. Sports Med. 6:266270, 1985. 7. Council of Europe. Handbook for the Eurofit Test of Physical Fitness. Roma: CONI, 1988. 8. Creagh, U., and T. Reilly. A multivariate analysis of kinanthropometric profiles of elite female orienteers. J Sports Med Phys Fitness. 35:5966, 1985. 9. Deshaies, P., D. Pargman, and C. Thiffault. A psycho-biological profile of individual performance in junior hockey players. In: Psychology of motor behavior and sport 1978, G.C. Roberts and K.N. Newell (Eds.). Champaign, IL: Human Kinetics, 1979, pp. 3650. 10. Fernndez, J.J., M.H. Vila, and F.A. Rodrguez. F.A. Modelo de estudio de la estructura condicional a travs de un anlisis multivariante enfocado a la deteccin de talentos en jugadores de balonmano. Eur. J. Hum. Mov. 12:175191, 2004. 11. Fry, A.C., D. Ciroslan, and M.D. Fry. C.D., LeRoux, B.K. Schilling and L.Z. Chiu. Anthropometric and performance variables discriminating elite American junior men weightlifters. J. Strength Cond. Res. 20:861866, 2006. 12. Gabbett, T., and B. Georgieff. Physiological and anthropometric characteristics of Australian junior national, state, and novice volleyball players. J. Strength Cond. Res. 21:902908, 2007. 13. Gabbett, T., B. Georgieff, and N. Domrow. The use of physiological, anthropometric, and skill data to predict selection in a talent-identified junior volleyball squad. J Sports Sci. 25:13371144, 2007. 14. Gastin, B. Energy system interaction and relative contribution during maximal exercise. Sports Med. 31:725741, 2001. 15. Geladas, N.D., G.P. Nassis, and S. Pavlicevic. Anthropometric and physical traits affecting sprint swimming performance in young swimmers. Int. J. Sports Med. 26:139144, 2005. 16. Heath, B.H., and J.E.L. Carter. A modified somatotype method. Am. J. Phys. Anthropol. 27:5774, 1967. 17. Jagomgi, G., and T. Jrime. The influence of anthropometrical and flexibility parameters on the results of breaststroke swimming. Anthropol. Anz. 63:213219, 2005. 18. Jrime, J., K. Haljaste, A. Cicchella, et al. Analysis of swimming performance from physical, physiological, and biomechanical parameters in young swimmers. Pediatr. Exerc. Sci. 19:7081, 2007. 19. Keogh, J. The use of physical fitness scores and anthropometric data to predict selection in an elite under 18 Australian rules football team. J. Sci. Med. Sport. 2:125133, 1999.

Analysis of Performance in Young Swimmers

151

20. Keskinen, K.L., L.J. Tilli, and P.V. Komi. Maximum velocity swimming: Interrelationship of stroking characteristics, force production and anthropometric variables. Scand J Med Sci Sports. 11:8792, 1989. 21. Kilka, R.J., and W.G. Thorland. Physiological determinants of sprint swimming performance in children and young adults. Pediatr. Exerc. Sci. 6:5968, 1994. 22. LEN. Comparative performance tables for swimming (1997-2000). Munich: Ligue Europene de Natation and German Swimming Federation, 1996. 23. Maglischo, E.W. Swimming fastest: a comprehensive guide to the science of swimming. Champaign, IL: Human Kinetics, 2003. 24. Malina, R.M., and C. Bouchard. Growth, maturation, and physical activity. Champaign, IL: Human Kinetics, 1991. 25. Malina, R.M., B. Ribeiro, J. Aroso, and S.P. Cumming. Characteristics of youth soccer players aged 13-15 years classified by skill level. Br. J. Sports Med. 41:290295, 2007. 26. Norman, G.R., and D.L. Streiner. Biostatistics. Ontario: BC-Decker, 2000. 27. Olbrecht, J., O. Madsen, A. Mader, H. Liesen, and W. Hollmann. Relationship between swimming velocity and lactic concentration during continuous and intermittent training exercises. Int. J. Sports Med. 6:7477, 1985. 28. Piennar, A.E., M.J. Spamer, and H.S. Steyn. Identifying and developing rugby ta among 10-year-old boys: a practical model. J Sports Sci. 16:691699, 1998. 29. Reilly, T., A.M. Williams, A. Nevill, and A. Franks. A multidisciplinary approach to talent identification in soccer. J Sports Sci. 18:695702, 2000. 30. Ross, W.D., and M.J. Marfell-Jones. Kinanthropometry. In: Physiological testing of the elite athlete, J.D. MacDougall, H.A. Wenger, and H.J. Green (Eds.). Ottawa: Canadian Association of Sports Sciences, 1982, pp. 75115. 31. Sekuli c, D., N. Zeni c, and N.G. Zubcevi c. Non linear anthropometric predictors in swimming. Coll. Antropol. 31:803809, 2007. 32. Sharp, R.L., J.P. Troup, and D.L. Costill. Relationship between power and sprint freestyle swimming. Med. Sci. Sports Exerc. 14:5356, 1982. 33. Silva, A.J., A.M. Costa, P.B. Oliveira, et al. The use of neural network technology to model swimming performance. J Sports Sci Med. 6:117125, 2007. 34. Smith, L. Anthropometric measurements, and arm and leg speed performance of male and female swimmers as predictors of swim speed. J Sports Med Phys Fitness. 18:153168, 1978. 35. Smith, R.M., and W.L. Spinks. Discriminant analysis of biomechanical differences between novice, good and elite rowers. J Sports Sci. 13:377385, 1995. 36. Soares, S.M., R.J. Fernandes, C.M. Carmo, J.V. Santos-Silva, and J.P. Vilas-Boas. Avaliao qualitative da tcnica em Natao. Apreciao de consistncia dos resultados produzidos por avaliadores com experincia e formao similiares. Port. J. Sports Sci. 1:2232, 2001. 37. Soares, S.M., R.J. Fernandes, J. Silva, and J.P. Vilas-Boas. The importance of technical evaluation on the training control in swimming Subjective analysis of the swimming technique of a juvenile group of swimmers from the Association of the North of Portugal. In Proceedings of the IV Congress of Notational Analysis of Sports, Hughes, M. and F. Tavares (Eds.). Porto, Portugal: Center for Team Sports Studies (Faculty of Sports Sciences and Physical Education), 1998, pp.261-267. 38. Sokolovas, G. Analysis of USA swimmings all time top times. Port. J. Sports Sci. 6:315317, 2006. 39. Solanellas, F., and F.A. Rodrguez. Multidisciplinary evaluation and performance prediction of tennis players of different age and sex categories. In Book of Abstracts. First Annual Congress of the European College of Sport Science. Nice, France: ECSS, 1996, pp. 345-349. 40. Tanner, J.M. Growth and adolescence. Oxford: Blackwell, 1962. 41. Van Schuylenbergh, R., B.V. Eynde, and P. Hespel. Prediction of sprint triathlon performance from laboratory tests. Eur. J. Appl. Physiol. 91:9499, 2004.