Share this article on:

00005768-201004000-0001900005768_2010_42_762_connaboy_reliability_4miscellaneous-article< 117_0_21_8 >Medicine & Science in Sports & Exercise©2010The American College of Sports MedicineVolume 42(4)April 2010pp 762-770Measures of Reliability in the Kinematics of Maximal Undulatory Underwater Swimming[APPLIED SCIENCES]CONNABOY, CHRIS1,2; COLEMAN, SIMON3; MOIR, GAVIN4; SANDERS, ROSS2,31School of Life Sciences, Edinburgh Napier University, Edinburgh, Scotland, UNITED KINGDOM; 2Centre of Aquatics Research and Education, University of Edinburgh, Edinburgh, Scotland, UNITED KINGDOM; 3Department of Physical Education, Sport and Leisure Studies, University of Edinburgh, Edinburgh, Scotland, UNITED KINGDOM; and 4Exercise Science Department, East Stroudsburg University, East Stroudsburg, PAAddress for correspondence: Chris Connaboy, M.Sc., Sport & Exercise Science, School of Life Sciences, Edinburgh Napier University, Edinburgh, Scotland, United Kingdom; E-mail: for publication February 2009.Accepted for publication August 2009.ABSTRACTPurpose: The purposes of this article were to establish the reliability of the kinematics of maximal undulatory underwater swimming (UUS) in skilled swimmers, to determine any requirement for familiarization trials, to establish the within-subject (WS) variability of the kinematics, and to calculate the number of cycles required to accurately represent UUS performance.Methods: Fifteen male swimmers performed 20 maximal UUS trials (two cycles per trial) during four sessions. The magnitude of any systematic bias present within the kinematic variables was calculated between session, trial, and cycle. Random error calculations were calculated to determine the WS variation. An iterative intraclass correlation coefficient (ICC) process was used to determine the number of cycles required to achieve a stable representation of each kinematic variable.Results: Significant differences were found between session 1 and all other sessions for several variables, indicating the requirement for a familiarization session. Results indicated a wide range of WS variation (coefficient of variation [CV] = 1.21%-12.42%). Reductions in WS variation were observed for all variables when the number of cycles of data used to calculate WS variation was increased. Using six cycles of data, including additional cycles of data, provided diminishing returns regarding the reduction of WS variation. The ICC analysis indicated that an average of nine cycles (mean ± SD = 9.47 ± 5.63) was required to achieve the maximum ICC values attained, and an average of four cycles (mean ± SD = 3.57 ± 2.09) was required to achieve an ICC of 0.95.Conclusions: After determining the systematic bias and establishing the requirement for a familiarization session, six cycles of data were found to be sufficient to provide high levels of reliability (CVTE = 0.86-8.92; ICC = 0.811-0.996) for each of the UUS kinematic variables.The undulatory underwater swimming (UUS) performed during the start and turns of swimming races is becoming a popular topic for both performance analysis and empirical research (5,6,12-14). The underwater phases of the starts and turns are crucial sections of overall race performance because, with the exception of the dive, the underwater phases of the starts and turns represent the fastest parts of the race. With the recognition of the importance of UUS to the overall swimming race time (14), the component kinematic variables used to describe UUS are becoming more frequently assessed measures of performance.The increasing assessment of UUS performance coupled with the growing popularity of UUS as a topic for research necessitates that the key kinematic components of UUS performance should be evaluated in terms of their reliability. Information and research evaluating the reliability of the kinematic variables used to describe and assess UUS performance are sparse. In particular, there is little information regarding the relative contributions of systematic bias and/or within-subjects (WS) variation to the overall reliability of specific UUS kinematic variables. Furthermore, there is little information regarding the consistency or repeatability (test-retest reliability) of specific measures of UUS performance obtained from repeated trials by the same individuals.According to Hopkins (8) and Hunter et al. (11), reliability should be analyzed in terms of its component parts, namely, (i) systematic bias, (ii) WS variation or "random error," and (iii) test-retest correlation. Systematic bias refers to the occurrence of a systematic or nonrandom change in the group mean (for a specific variable) between two or more trials (8). Factors such as fatigue, motivation, and learning or practice effects can all contribute to the potential for the occurrence of a systematic bias. According to Hopkins et al. (9), once the magnitude and effects of the systematic bias are determined, researchers can instigate appropriate practices that minimize its occurrence (i.e., familiarization trials to reduce the leaning effect).The WS variation can include variations from several sources (biological, measurement errors, etc.). Hopkins (8) states that the WS variation is the most important type of reliability measure. WS variation influences the accuracy of an assessment of change in a specific variable within an experimental study. Smaller values of WS variation enable a more precise assessment of meaningful changes in a variable (8).Except where a single episode (trial or cycle) of performance is the principal focus of a study, it is considered that the use of multiple trials provides a more stable and representative account of biomechanical variation (2,3,10). The number of trials used to assess baseline levels of performance and/or a subsequent change in performance (i.e., postintervention) is an important methodological consideration (10). The stability of the mean value of multiple trials is greatly influenced by the stability of the variation across trials. If insufficient trials are used to ascertain a representative mean value for a specific variable, then the reported mean value of the collected trials will not accurately represent the performance. Consequently, the validity of the specific variable to assess performance would be limited. Both Bates et al. (2) and Salo et al. (20) highlight the requirement for empirical work to be undertaken to ascertain the number of trials necessary to provide stable, representative data before analyzing performance. According to Portney and Watkins (17), test-retest reliability methods are commonly used to evaluate the stability or repeatability of a specific variable across repeated trials.Without such a priori investigations, empirical research investigating changes in variables to assess overall performance could produce results that falsely support a null hypothesis as a result of insufficient statistical power (3). By clarifying the extent of any learning, practice, and/or fatigue effects associated with repeated collection of the specified variables, by examining the WS variation apparent in the specific sample, and by ascertaining the requisite number of trials required to accurately determine WS variation, the factors influencing statistical power can be adequately addressed.The aims of the present study are threefold: (i) to determine the magnitude of systematic bias among session, trial, and/or cycle to provide guidelines for familiarization in future studies of UUS; (ii) to establish the WS variation of the key biomechanical measures commonly used to assess maximal UUS to determine the accuracy with which changes in performance can be measured; and (iii) to analyze the test-retest reliability to indicate the number of cycles/trials required to study changes in the kinematics of UUS.METHODSParticipantsFifteen skilled male swimmers (mean ± SD; age = 19 ± 3.3 yr, height = 1.82 ± 0.05 m, weight = 74.8 ± 8.6 kg, arm length = 0.52 ± 0.03 m, trunk length = 0.66 ± 0.03 m, thigh length = 0.37 ± 0.02 m, shank length = 0.45 ± 0.03 m, foot length = 0.16 ± 0.01 m, competitive swimming experience = 9.4 ± 3.2 yr) from the university swimming team participated in the study. Ethical approval was gained from the local ethics committee. Written informed consent was obtained from each participant. Written parental consent was obtained from each of the five participants younger than 18 yr of age.Study DesignA single group, repeated-measures study design was used to assess the requirement for familiarization trials, to determine the WS variability, and to establish the test-retest reliability. During each session, participants completed five trials of UUS to maximize swimming speed. All participants attended four testing sessions, with each testing session separated by 7 d. All four sessions were conducted at the same time of day to minimize the influence of diurnal biological variation on performance (18). Participants were also asked not to practice the task throughout the 4-wk testing period and asked to refrain from strenuous exercise 24 h before each session.Experimental ProtocolBefore entering the water, participants were marked with circles of black oil-based body paint (3 cm in diameter) at the joint centers of the wrist, shoulder, hip, knee, ankle, and fifth metatarsal phalangeal joint (5th MPJ) of the foot on the right side of the body. Before undertaking the maximal UUS trials, each swimmer performed a standardized 20-min warm-up at the beginning of each testing session.Each performance trial consisted of the participant swimming 15 m under water using a UUS technique. Each trial started with the participant in the water; a push start off the wall was performed to reach a designated depth (approximately 0.60-1.00 m) below the surface of the water to negate the effects of wave drag (23). The push off from the wall was only used to achieve the correct depth and orientation (horizontal with respect to the camera) and not as a means to maximize swimming velocity. Once the required depth and orientation were achieved, swimmers were then required to accelerate toward a marker on the pool floor 10 m away, representing the start of the filming area. Participants were instructed to maximize swimming velocity as they passed over the first marker and maintain maximal UUS velocity throughout the designated filming area until they passed over a second marker, 5 m from the first.A two-dimensional cinematographic technique was used to collect the video data. Participants were recorded with a stationary underwater camera (KY32 CCD; JVC Corporation, Yokohama, Japan) at 50 fields per second. The camera was located 1 m below the surface of the water and 12 m from the line of action. The optical axis of the camera was perpendicular to the plane of motion of the swimmer. A capture window of 4 m, in line with the horizontal motion of the swimmer, ensured that a minimum of two complete kick cycles could be recorded. Two consecutive kick cycles were recorded per trial to facilitate the calculation of the within-trial (between-cycle) variability in UUS kinematics. Participants repeated the procedure for a total of five times per session, with a 5-min rest interval between trials to minimize the effects of fatigue.Bilateral symmetry was assumed, and only the side of the body facing the camera was digitized to define a five-segment model of the swimmer's body comprising the arm, trunk, thigh, shank, and foot. The segment end point landmarks were digitized using the Ariel Performance Analysis System (APAS-2000 Arial Dynamics, 2000, San Diego, CA). Two kick cycles plus 15 additional frames on either side of the start and end of the kick cycles were digitized to enable the accurate identification of the start/end points of each cycle, to avoid distortion in the calculation of time derivates of position data (22) and to minimize end point errors in the filtering process (24). Before determining the start/end points of each kick cycle, the data were interpolated (by a factor of 4), decreasing the time interval between successive data points from 0.02 to 0.005 s. The start of a kick cycle was defined as the frame corresponding to the initiation of an upward movement (in the y-plane) of the marker placed on the fifth MPJ, through a complete kick cycle, to the frame immediately preceding the frame corresponding to the initiation of an upward movement of the marker on the fifth MPJ of the next kick cycle. The raw position-time data from the APAS system were then transformed to produce the displacement data using a subject-derived two-dimensional linear scale on the basis of the known length of the swimmer's thigh (7). The raw data were demeaned and detrended before a Fourier transform and Fourier reconstruction in which the harmonics up to 7 Hz were retained.Data AnalysisA total of 19 individual kinematic variables were calculated for each kick cycle: 1) average swimming velocity, 2) kicking frequency, and 3) cycle length; joint range of motion of 4) shoulder, 5) hip, 6) knee, and 7) ankle; maximum angular velocity of 8) shoulder, 9) hip, 10) knee, and 11) ankle; joint center amplitudes of 12) wrist, 13) shoulder, 14) hip, 15) knee, 16) ankle, and 17) 5th MPJ; 18) maximum angle of attack; and 19) mean absolute angle of attack.Average swimming velocity.The average horizontal hip velocity (U) was used as a representation of the average swimming velocity (U) for each cycle. The average horizontal velocity of the hip (m·s−1) was obtained for each kick cycle and calculated as the difference in the horizontal displacement of the hip throughout a kick cycle divided by the time taken to complete the cycle, with displacement values obtained from the APAS output.where U is the average swimming velocity, d2 and d1 are the final and initial horizontal displacements of the hip, respectively, and t is the time taken to complete one kick cycle.Equation (Uncited)Kicking frequency.Kicking frequency (f) was calculated as the inverse of the time taken to complete a kick cycle.where f is the cycle frequency and t is the time taken to complete one kick cycle.Equation (Uncited)Cycle length.Cycle length (CL) was calculated as the horizontal displacement of the hip marker during one complete kick cycle.Joint range of movement and maximum angular velocity.The minimum and maximum angular displacements of each of the relevant joints angles were calculated from the smoothed angular displacement data, and the range of movement (ROM) was calculated as the difference between these two values for each of the respective joints. The joint angle was defined as the angle formed at the joint by the movement of the component limbs (Fig. 1). Maximum angular velocity was derived from the smoothed angular displacement data.FIGURE 1-The swimmer's joint center markings and joint angle definitions.Joint center amplitude.The amplitudes of the oscillations of the joint centers at the wrist, shoulder, hip, knee, ankle, and fifth MPJ were calculated. The amplitudes were calculated as the difference between the maximum and the minimum values from the reconstructed vertical displacements of the segment end point data.Angle of attack.The angle of attack (AoA) was determined from the reconstructed segment end point data of the ankle joint and the fifth MPJ. The AoA was calculated as the arc cosine of the dot product of the vectors of the line of the foot and the tangent of the path of the foot at every instance throughout the kick cycle. The maximum AoA achieved and the absolute mean AoA value were calculated for each kick cycle.Statistical AnalysisBefore calculating the systematic bias, the WS variation, and the test-retest reliability, each of the kinematic variables was individually assessed for heteroscedasticity using the methods outlined by Bland and Altman (4). If heteroscedasticity was not present, the raw data were used in the reliability calculations. If the data were found to be heteroscedastic, then the data were log transformed using 100× natural logarithm of the observed value (8).Systematic BiasSystematic bias was determined using repeated-measures ANOVA (RM ANOVA). One three-way RM ANOVA was completed for each independent variable to calculate whether the magnitude of difference among the mean values for each session (n = 4), trial (n = 5), or cycle (n = 2) was statistically significant. The α value was set at 0.05. Any significant intersession, trial, or cycle differences were assessed from the a priori planned comparisons using the Bonferroni procedure. Where any statistically significant differences occurred, those sessions, trials, and cycles were removed from further calculations of reliability (WS variation and test-retest correlations).WS VariationWS variation was calculated for 3-30 cycles. For the sake of brevity, a reduced number of cycles were reported (i) to highlight the initial changes in WS variation with the inclusion of additional single cycles and (ii) to highlight the extent of changes in WS variation calculated from greater numbers of cycles. WS variation was reported for 3, 4, 5, 6, 12, 18, 24, and 30 cycles as both typical error (TE) and coefficient of variation (CVTE). TE was calculated aswhere MSEn is the mean square error value from the RM ANOVA from n repeated cycles.Equation (Uncited)The coefficient of variation was determined aswhere TEn is the TE of n number of cycles and Mn is the mean value from the same n repeated cycles. Confidence intervals (95%) for CVTE were calculated using the methods outlined by Tate and Klett (21).Equation (Uncited)Test-Retest ReliabilityThe test-retest reliability for the all cycles (n = 30) was evaluated using a mixed-model (1,3) intraclass correlation coefficient (ICC) (16). The stability of the variation in each kinematic variable was assessed using the methods proposed by James et al. (10). The initial ICC was determined for two cycles. An iterative process was then conducted, whereby repeated ICC values were performed including an additional cycle with each iteration up to a maximum of 30 cycles. The maximum ICC value for all 30 cycles and the 95% confidence intervals (upper and lower limits) were calculated. The 95% confidence intervals for the ICC were calculated using the methods outlined by McGraw and Wong (15). To assess the stability of each variable, we calculated the minimum number of cycles required to achieve the maximum ICC value. To determine the minimum number of cycles required to achieve a stable representation of the variation within each of the kinematic variables, we also calculated the number of cycles required to achieve ICC values of 0.85, 0.90, and 0.95.RESULTSStatistically significant differences between testing sessions were found for U (P = 0.049), f (P = 0.045), and CL (P = 0.044). Results from the Bonferroni planned comparisons indicated systematic bias between the first and the remaining three testing sessions for U, f, and CL. Figure 2 highlights the differences in mean values between session 1 and the following three sessions for U, f, and CL. No further significant differences were found in mean values across the four testing sessions for any of the remaining kinematic variables. No significant differences in mean values were found across trial or cycle for any of the kinematic variables. To isolate the effects of the systematic bias between session 1 and the remaining three sessions, we included only the data from the final three testing sessions (30 cycles) in the subsequent reliability analyses (WS variation/test-retest correlations).FIGURE 2-Reliability (systematic bias) of maximal UUS kinematic variables. Mean values for session (1-4) for kicking frequency (A), average swimming velocity (B), and cycle length (C). *Statistically significant difference (P > 0.05). Values are session means; bars are SD.Table 1 shows the results for the WS variation. Depending on whether heteroscedasticity was present, random error was expressed in either absolute or ratio form.TABLE 1. Reliability (random error) of maximal UUS kinematic variables.The results presented in Table 1 illustrate that the extent of random error within the respective kinematic variables varies markedly (i.e., knee ROM for three cycles, CVTE = 1.21% compared with wrist joint center amplitude [JCA] for three cycles, CVTE = 12.85%). Irrespective of the relative differences in random error for each of the variables, the reliability was found to improve as the number of trials used to determine the average score increased (from n = 3 to n = 30). The greatest increases in reliability (reduction in %CVTE) were apparent within the initial changes in the number of trials used to calculate the mean value, with an average of 0.59% reduction in %CVTE by using four cycles compared with three cycles, a further 0.40 average reduction by using five cycles compared with four cycles, and a further 0.37% average reduction in %CVTE by using six cycles compared with five cycles. Beyond six cycles, the use of additional cycles of data to calculate the mean value resulted in diminishing returns; for every additional cycle used in the calculation of the mean values, the smaller the reduction in the %CVTE.The data presented in Table 2 show the levels of performance stability achieved for each of the kinematic variable measured. The maximum ICC value recorded (0.996) was for U, with the maximum ICC values ranging from 0.952 to 0.996. The number of cycles required to reach the maximum ICC values ranged from 3 to 23 (mean ± SD = 9.74 ± 5.63 cycles). With the exception of the knee JCA, all variables achieved an ICC values of 0.90 (or greater) after three cycles, and all except knee JCA attained an ICC value of 0.95 (or greater) after six cycles. The number of cycles required for all kinematic variables to achieve an ICC value of 0.95 ranged from 2 to 11 (mean ± SD = 3.57 ± 2.09 cycles).TABLE 2. Reliability (retest reliability) of maximal UUS kinematic variables.DISCUSSIONThe aims of the present study were threefold. The first of those aims was to determine the existence of any systematic bias between session, trial, and/or cycle to establish the extent of any learning, motivation, and/or fatigue effects. The results of the study clearly indicate the requirement of a single familiarization session before collecting sufficiently reliable kinematic data. The systematic and statistically significant (P > 0.05) changes in the mean between session 1 and the remaining three sessions for U, f, and CL suggest that for this specific experimental protocol, familiarization is required before reliable values of these three key kinematic variables are obtained.The requirement for further familiarization and practice of a skill commonly undertaken by swimmers may at first appear counterintuitive. However, the protocol used to acquire the UUS kinematic data is relatively novel compared with the normal UUS practices within training and/or racing. Although skilled swimmers may be accustomed to turning off the wall and performing maximal UUS, the protocol used within the present study required that the swimmers did not use the push off from the wall to maximize swimming velocity rather simply to attain the required depth and orientation. The present protocol enabled the UUS performance to be isolated from any variation in the execution of the push off from the wall. This difference/novel aspect of the protocol may have resulted in the values recorded. As it can be seen from Figure 2, U, f, and CL were significantly greater in session 1 compared with the remaining three sessions, representing the learning required before the swimmers could successfully adhere to experimental protocol and not use the wall to maximize swimming velocity.No significant differences were found within each session for any of the kinematic variables. This suggests that the five trials conducted per session with 5 min of recovery between each trial did not cause any fatigue (or at least not enough to alter the kinematics of the UUS performance significantly). Likewise, the results also indicate that there was no learning effect present between trials and that the motivation to produce maximal UUS performance during the five trials was maintained. Furthermore, the between-cycle (within-trial) variation was not statistically significant. This indicates that the swimmers were able to achieve and maintain maximal UUS velocity within the allotted 5-m filming area and that the variation in the kinematics was stable across both cycles of data collected.The second aim of the present study was to establish the extent/magnitude of the WS variation apparent in the kinematic variables commonly used to describe and evaluate maximal UUS performance in skilled swimmers. Because there are no other studies that have reported reliability data for UUS performance, no direct comparisons can be made with the data produced from the protocol used within the present study. However, the levels of %CVTE reported for three cycles (1.21%-12.85%) are similar in magnitude to the research of Hunter et al. (11), which examined the reliability of kinematic data in sprint running (three-cycle %CVTE range = 0.6%-19.9%). Similar to the results of Hunter et al. (11), the %CVTE of all the 19 kinematic variables examined in the present study were found to reduce when the number of cycles used to calculate an average score was increased. Nevertheless, care must be taken when increasing the number of trials/cycles used to calculate a mean value to represent the data. An increase in the number of trials recorded per session (beyond the five trials used within the present experimental protocol) may introduce systematic bias because of fatigue affecting performance. The effects of fatigue should always be considered when deciding on the exact number of trial/cycles to observe within each session. In addition, Rodano and Squadrone (19) suggest that there may be an optimal number of trials required to achieve stability in joint kinematic data, beyond which no increase in stability will be observed.The largest values for WS variability within the kinematic data were found for the shoulder ROM (three-cycle %CVTE = 9.17%), wrist amplitude (three-cycle %CVTE = 12.85%), shoulder amplitude (three-cycle %CVTE = 8.66%), and shoulder angular velocity (three-cycle %CVTE = 9.68%). The higher levels of variability seen in those variables associated with the movements of the hands and the arms may be accounted for by the swimmers attempting to balance the forces (inertial recoil) produced when kicking. According to previous research (5), the hands and arms can act as an inertial damper, helping to maintain an efficient streamline position in the water while also enabling the effective production of a propulsive waveform down the remainder of the body. However, as with all the other variables, the %CVTE was found to be reduced when calculated from an increasing number of cycles of data.The key UUS performance variables of U, CL, and f were found to be very reliable with %CVTE values of 1.84%, 1.41%, and 3.89%, respectively (for three cycles). These reduced to 1.30%, 1.00%, and 2.75%, respectively, when calculated from the average of six cycles. The ROM at the knee was found to have the lowest %CVTE with 1.21% for three cycles improving to 0.86% for six cycles, signifying that, for skilled swimmers, the ROM used at the knee during maximal UUS performance is maintained within very narrow limits of variability. This is made more interesting when it is considered that the WS variability for the knee amplitude is more than twice the magnitude (three-cycle %CVTE = 3.02%, six-cycle %CVTE = 2.13%) of the knee ROM, suggesting that the swimmers may be actively trying to maintain the levels of knee ROM by manipulating the coordination of the oscillations of the various joint centers. However, much more research is required to fully explain the factors that interact to produce the coordination observed in the UUS performance of skilled swimmers.The third and final aim of the present study was to determine the test-retest reliability of the kinematic data to establish the number of cycles/trials necessary to achieve stable levels of performance and to accurately track changes in UUS kinematics over time. Although it should be noted that the ICC value at which test-retest reliability is deemed to be good (ICC > 0.75) is an arbitrary value (17), James et al. (10) suggested that the ICC is a more objective means of assessing the number of trials necessary to determine the stability of performance than other measures (i.e., sequential averaging) because it involves fewer arbitrary decisions to assess performance stability.Maximum ICC values were high for all the kinematic measures of UUS performance, ranging from 0.952 for knee JCA up to 0.996 for U for the total 30 cycles recorded. The initial interpretation of the results of the ICC analysis suggests that multiple cycles (mean ± SD = 9.47 ± 5.63) of kinematic data are required before the maximum ICC values are achieved for all 19 kinematic variables. The ICC analyses demonstrate that the test-retest reliability (stability) of the maximal UUS performance is strong for the majority of the kinematic variables measured. All except knee JCA achieved an ICC value of 0.95 within six cycles and with the majority (68%) achieving an ICC of 0.95 within three cycles. With further iterations of the repeated ICC calculations, diminishing returns were observed in the increases in the ICC value; hence, the relatively high number of cycles required to achieve the maximum ICC values (for 30 cycles) compared with the numbers required to achieve the 0.95 level (mean ± SD = 3.57 ± 2.09 cycles). Nevertheless, the ICC data should not be considered in isolation. WS variability data should also be taken into account when making decisions regarding the minimum number of cycles required to accurately represent the kinematics of UUS. By considering the magnitude of WS variation in the selection of the number of cycles required to ensure a reliable assessment of each of the kinematic variables, it provides a measure of the accuracy with which any future changes in the kinematics of UUS performance can be monitored. Furthermore, the results of ICC analysis should not be considered in isolation when determining the reliability of kinematic variables because they can be adversely influenced by the homogeneity of the sample tested, greatly affecting any interpretation of reliability (1,8).A pragmatic approach to the selection of the number of cycles used to represent the kinematic data needs to be adopted, balancing the need for ensuring high test-retest reliability and acceptable levels of WS variation with the practical, economic, and logistical concerns of collecting repeated cycles of data. As demonstrated in the present study, increasing the number of cycles used to calculate a mean value to represent a subject's UUS performance shows an increase in the levels of test-retest reliability and provides increasingly more reliable representations of the swimmers' performance (lower values of CVTE). Therefore, given the diminishing returns witnessed in the reduction of the %CVTE data beyond six cycles and the achievement of a 0.95 ICC value for all but the knee JCA at six cycles, it would seem reasonable to conclude that six cycles can be used to accurately represent the kinematics of skilled swimmers' UUS performance. Using six cycles, all the kinematic variables demonstrated good levels of reliability as signified by the low WS variation (six-cycle %CVTE range = 0.86%-8.92%) and high test-retest correlations (six-cycle ICC range = 0.811-0.996). However, as Atkinson and Nevill (1) noted, the extent of a variable's reliability is dependent on its intended use, and subsequently a researcher must determine whether it is sufficiently reliable to measure the smallest worthwhile change in an athlete's performance.CONCLUSIONSThe findings of the present study clearly indicate the requirement for a familiarization session before undertaking an assessment of UUS performance to ensure reliable data. In addition, the determination of the number of cycles required to provide accurate and reliable data has shown that an average of 10 cycles is required to achieve maximum ICC values (from all 30 cycles), whereas an average of only four cycles is required to achieve an ICC value of 0.95. However, it should be recognized that when making the decision of how many cycles of data should be used to accurately represent a kinematic variable, the data from the ICC analysis should not be viewed in isolation. Consideration of the WS variation is also necessary as having smaller and stable TE and %CVTE values and enables a more precise assessment of worthwhile changes in a variable. The changes in %CVTE data with the increase in the number of cycles used in the calculation of %CVTE clearly show improved levels of reliability with a greater number of cycles of data. However, the reduction in %CVTE beyond the values provided by six cycles of data led to diminishing returns, with less of a reduction in %CVTE seen with each additional cycle of data included in the calculation. To conclude, individual researchers and applied sports scientists need to consider the practical aspects of collecting many cycles of data balanced against the requirement to ensure a sufficiently reliable and accurate measurement of the kinematic variables with respect to the magnitude of the smallest worthwhile changes acceptable for their own testing.The results of the present study do not constitute endorsement by the American College of Sports Medicine.The authors received funding from the Edinburgh University Development Trust to purchase equipment.The authors thank the support of the Edinburgh University Development Trust that helped fund this research.REFERENCES1. Atkinson G, Nevill A. Statistical methods for assessing measurement error (reliability) in variables relevant to sports medicine. Sports Med. 1998;26:217-38. [Context Link]2. Bates BT, Osternig LR, Sawhill JA, James SL. An assessment of subject variability, subject-shoe interaction, and the evaluation of running shoes using ground reaction force data. J Biomech. 1983;16:181-91. [CrossRef] [Medline Link] [Context Link]3. Bates BT, Dufek JS, Davis HP. The effect of trial size on statistical power. Med Sci Sports Exerc. 1992;24(9):1059-68. [CrossRef] [Full Text] [Medline Link] [Context Link]4. Bland J, Altman D. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet. 1986;1(8476):307-10. [Medline Link] [Context Link]5. Connaboy C, Coleman S, McCabe C, Naemi R, Psycharakis S, Sanders RH. Tadpole, trout or tuna: the equivalence of animal and human aquatic undulatory locomotion. In: Proceedings of the XXVth International Symposium on Biomechanics in Sports. Ouro Preto (Brazil); 2007 August 23-27: Department of Sport, Federal University of Minas Gerais, Brazil; 2007. p. 75-8. [Context Link]6. Connaboy C, Coleman S, Sanders RH. Coordinating undulatory underwater swimming: a comparison of skilled versus unskilled swimmers. Motor Control. 2007;11:S189-90. [Context Link]7. Clothier PJ, Payne WR, Harvey JT, Blanksby BA, Benjanuvatra N. Technical note: the use of subject derived scale factors for one-camera 2D analysis in underwater swimming. J Hum Mov Stud. 2004;46:333-45. [Context Link]8. Hopkins WG. Measures of reliability in sports medicine and science. Sports Med. 2000;30(1):1-15. [CrossRef] [Full Text] [Medline Link] [Context Link]9. Hopkins WG, Schabort EJ, Hawley JA. Reliability of power in physical performance tests. Sports Med. 2001;31(3):211-34. [CrossRef] [Full Text] [Medline Link] [Context Link]10. James RC, Herman JA, Dufek JS, Bates BT. Number of trials necessary to achieve performance stability of selected ground force variables during landing. J Sports Sci Med. 2007;6:126-34. [Context Link]11. Hunter JP, Marshall RN, McNair P. Reliability of biomechanical variables of sprint running. Med Sci Sports Exerc. 2004;36(5):850-61. [Context Link]12. Lyttle AD, Blanksby BA. A look at gliding and underwater kicking in the swim turn. In: Proceedings of XVIII Symposium on Biomechanics in Sports: Applied Program: Application of Biomechanical Study in Swimming. Hong Kong (China): 2000 June 25-30; Department of Sports Science and Physical Education, The Chinese University of Hong Kong; 2000. p. 56-63. [Context Link]13. Lyttle AD, Keys M. The use of computational fluids dynamics to optimise underwater kicking performance. In: Proceedings XXIInd International Symposium of Biomechanics in Sport. Ottawa (Canada): University of Ottawa; 2004. p. 438-41. [Context Link]14. Mason B, Cossor J. What can we learn from competition analysis at the 1999 Pan Pacific Swimming Championships? In: Proceedings of XVIII Symposium on Biomechanics in Sports: Applied Program: Application of Biomechanical Study in Swimming. Hong Kong (China): Department of Sports Science and Physical Education, The Chinese University of Hong Kong; 2000. p. 75-82. [Context Link]15. McGraw KO, Wong SP. Forming inferences about some intra-class correlation coefficients. Psychol Methods. 1996;1:30-46. [CrossRef] [Full Text] [Medline Link] [Context Link]16. Morrow JR, Jackson AW. How "significant" is your reliability? Res Q Exerc Sport. 1993;64:352-5. [CrossRef] [Medline Link] [Context Link]17. Portney LG, Watkins MP. Foundations of Clinical Research: Applications to Practice. 2nd ed. Upper Saddle River (NJ): Prentice-Hall; 2000. p. 742. [Context Link]18. Reilly T, Robinson G, Minors DS. Some circulatory responses to exercise at different times of day. Med Sci Sports Exerc. 1984;16(5):477-82. [CrossRef] [Full Text] [Medline Link] [Context Link]19. Rodano R, Squadrone R. Lower limb kinetic variability in vertical jump exercises. J Appl Biomech. 2002;18(1):75-83. [Context Link]20. Salo A, Grimshaw PN, Viitasalo JT. Reliability of variables in the kinematic analysis of sprint hurdles. Med Sci Sports Exerc. 1997;29(3):383-9. [CrossRef] [Full Text] [Medline Link] [Context Link]21. Tate RF, Klett GW. Optimal confidence limits for the variance of a normal distribution. J Am Stat Assoc. 1959;54:674-82. [Context Link]22. Vaughan CL. Smoothing and differentiation of displacement-time data: an application of splines and digital filtering. Intl J Biomed Comput. 1982;13:375-86. [Context Link]23. Vennell R, Pease D, Wilson B. Wave drag on human swimmers. J Biomech. 2006;39:664-71. [CrossRef] [Medline Link] [Context Link]24. Vint PF, Hinrichs RN. Endpoint error in smoothing and differentiating raw kinematic data: an evaluation of four popular methods. J Biomech. 1996;29(12):1637-42. [CrossRef] [Medline Link] [Context Link] FAMILIARIZATION; RANDOM ERROR; VARIABILITY; RETEST RELIABILITY; TRIAL|00005768-201004000-00019#xpointer(id(R2-19))|11065213||ovftdb|SL0000461719831618111065213P84[CrossRef]|00005768-201004000-00019#xpointer(id(R2-19))|11065405||ovftdb|SL0000461719831618111065405P84[Medline Link]|00005768-201004000-00019#xpointer(id(R3-19))|11065213||ovftdb|00005768-199209000-00017SL00005768199224105911065213P85[CrossRef]|00005768-201004000-00019#xpointer(id(R3-19))|11065404||ovftdb|00005768-199209000-00017SL00005768199224105911065404P85[Full Text]|00005768-201004000-00019#xpointer(id(R3-19))|11065405||ovftdb|00005768-199209000-00017SL00005768199224105911065405P85[Medline Link]|00005768-201004000-00019#xpointer(id(R4-19))|11065405||ovftdb|SL000055311986130711065405P86[Medline Link]|00005768-201004000-00019#xpointer(id(R8-19))|11065213||ovftdb|00007256-200030010-00001SL00007256200030111065213P90[CrossRef]|00005768-201004000-00019#xpointer(id(R8-19))|11065404||ovftdb|00007256-200030010-00001SL00007256200030111065404P90[Full Text]|00005768-201004000-00019#xpointer(id(R8-19))|11065405||ovftdb|00007256-200030010-00001SL00007256200030111065405P90[Medline Link]|00005768-201004000-00019#xpointer(id(R9-19))|11065213||ovftdb|00007256-200131030-00005SL0000725620013121111065213P91[CrossRef]|00005768-201004000-00019#xpointer(id(R9-19))|11065404||ovftdb|00007256-200131030-00005SL0000725620013121111065404P91[Full Text]|00005768-201004000-00019#xpointer(id(R9-19))|11065405||ovftdb|00007256-200131030-00005SL0000725620013121111065405P91[Medline Link]|00005768-201004000-00019#xpointer(id(R15-19))|11065213||ovftdb|00060744-199603000-00004SL00060744199613011065213P97[CrossRef]|00005768-201004000-00019#xpointer(id(R15-19))|11065404||ovftdb|00060744-199603000-00004SL00060744199613011065404P97[Full Text]|00005768-201004000-00019#xpointer(id(R15-19))|11065405||ovftdb|00060744-199603000-00004SL00060744199613011065405P97[Medline Link]|00005768-201004000-00019#xpointer(id(R16-19))|11065213||ovftdb|SL0000696619936435211065213P98[CrossRef]|00005768-201004000-00019#xpointer(id(R16-19))|11065405||ovftdb|SL0000696619936435211065405P98[Medline Link]|00005768-201004000-00019#xpointer(id(R18-19))|11065213||ovftdb|00005768-198410000-00010SL0000576819841647711065213P100[CrossRef]|00005768-201004000-00019#xpointer(id(R18-19))|11065404||ovftdb|00005768-198410000-00010SL0000576819841647711065404P100[Full Text]|00005768-201004000-00019#xpointer(id(R18-19))|11065405||ovftdb|00005768-198410000-00010SL0000576819841647711065405P100[Medline Link]|00005768-201004000-00019#xpointer(id(R20-19))|11065213||ovftdb|00005768-199703000-00014SL0000576819972938311065213P102[CrossRef]|00005768-201004000-00019#xpointer(id(R20-19))|11065404||ovftdb|00005768-199703000-00014SL0000576819972938311065404P102[Full Text]|00005768-201004000-00019#xpointer(id(R20-19))|11065405||ovftdb|00005768-199703000-00014SL0000576819972938311065405P102[Medline Link]|00005768-201004000-00019#xpointer(id(R23-19))|11065213||ovftdb|SL0000461720063966411065213P105[CrossRef]|00005768-201004000-00019#xpointer(id(R23-19))|11065405||ovftdb|SL0000461720063966411065405P105[Medline Link]|00005768-201004000-00019#xpointer(id(R24-19))|11065213||ovftdb|SL00004617199629163711065213P106[CrossRef]|00005768-201004000-00019#xpointer(id(R24-19))|11065405||ovftdb|SL00004617199629163711065405P106[Medline Link]8945665Measures of Reliability in the Kinematics of Maximal Undulatory Underwater SwimmingCONNABOY, CHRIS; COLEMAN, SIMON; MOIR, GAVIN; SANDERS, ROSSApplied Sciences442