The reliability and precision of measurement in sports medicine are of concern in both research and clinical practice. The validity of conclusions drawn from a research project and the rationale for decisions made about the care of an injured athlete are directly related to the precision of measurement. Through analysis of variance, estimates of reliability and precision of measurement can be quantified. The purpose of this manuscript is to introduce the concepts of intraclass correlation as an estimate of reliability and standard error of measurement as an estimate of precision. The need for a standardized set of formulas for intraclass correlation is demonstrated, and it is urged that the standard error of measurement be included when estimates of reliability are reported. In addition, three examples are provided to illustrate important concepts and familiarize the reader with the process of calculating these estimates of reliability and precision of measurement.
Craig R. Denegar and Donald W. Ball
Kristie-Lee Taylor, Will G. Hopkins, Dale W. Chapman and John B. Cronin
The purpose of this study was to calculate the coefficients of variation in jump performance for individual participants in multiple trials over time to determine the extent to which there are real differences in the error of measurement between participants. The effect of training phase on measurement error was also investigated. Six subjects participated in a resistance-training intervention for 12 wk with mean power from a countermovement jump measured 6 d/wk. Using a mixed-model meta-analysis, differences between subjects, within-subject changes between training phases, and the mean error values during different phases of training were examined. Small, substantial factor differences of 1.11 were observed between subjects; however, the finding was unclear based on the width of the confidence limits. The mean error was clearly higher during overload training than baseline training, by a factor of ×/÷ 1.3 (confidence limits 1.0–1.6). The random factor representing the interaction between subjects and training phases revealed further substantial differences of ×/÷ 1.2 (1.1–1.3), indicating that on average, the error of measurement in some subjects changes more than in others when overload training is introduced. The results from this study provide the first indication that within-subject variability in performance is substantially different between training phases and, possibly, different between individuals. The implications of these findings for monitoring individuals and estimating sample size are discussed.
Andrea J. Braakhuis, Kelly Meredith, Gregory R. Cox, William G. Hopkins and Louise M. Burke
A routine activity for a sports dietitian is to estimate energy and nutrient intake from an athlete’s self-reported food intake. Decisions made by the dietitian when coding a food record are a source of variability in the data. The aim of the present study was to determine the variability in estimation of the daily energy and key nutrient intakes of elite athletes, when experienced coders analyzed the same food record using the same database and software package. Seven-day food records from a dietary survey of athletes in the 1996 Australian Olympic team were randomly selected to provide 13 sets of records, each set representing the self-reported food intake of an endurance, team, weight restricted, and sprint/power athlete. Each set was coded by 3–5 members of Sports Dietitians Australia, making a total of 52 athletes, 53 dietitians, and 1456 athlete-days of data. We estimated within- and between- athlete and dietitian variances for each dietary nutrient using mixed modeling, and we combined the variances to express variability as a coefficient of variation (typical variation as a percent of the mean). Variability in the mean of 7-day estimates of a nutrient was 2- to 3-fold less than that of a single day. The variability contributed by the coder was less than the true athlete variability for a 1-day record but was of similar magnitude for a 7-day record. The most variable nutrients (e.g., vitamin C, vitamin A, cholesterol) had ~3-fold more variability than least variable nutrients (e.g., energy, carbohydrate, magnesium). These athlete and coder variabilities need to be taken into account in dietary assessment of athletes for counseling and research.
Antonio Dello Iacono, Stephanie Valentin, Mark Sanderson and Israel Halperin
estimates. Finally, sensitivity of the PF outputs obtained from the strain gauge and force plate was assessed by comparing the smallest worthwhile change (SWC) and standard error of measurement (SEM), and interpreted using the thresholds proposed by Liow and Hopkins. 15 Statistical significance was set at
Kai-Yu Ho, Brenda Benson Deaver, Tyrel Nelson and Catherine Turner
legs were combined for each task during analysis. Interrater and intrarater reliability were analyzed using intraclass correlation coefficients (ICC 3, k ) and standard error of measurement (SEM). ICC values were classified according to the following criteria 5 : poor (<.4), fair (.4–.7), good (.7
Salman Nazary-Moghadam, Mahyar Salavati, Ali Esteki, Behnam Akhbari, Sohrab Keyhani and Afsaneh Zeinalzadeh
to .56) 0.08 33.97 0.15 Abbreviations: ACLD, anterior cruciate ligament deficient; CV, coefficient of variation; HS, high speed; ICC, intraclass correlation coefficient; LS, low speed; MMDC, minimal metrically detectable change; SEM, standard error of measurement; SS, self-selected speed. Therefore
Lindsay B. Baker, Kelly A. Barnes, Bridget C. Sopeña, Ryan P. Nuccio, Adam J. Reimel and Corey T. Ungaro
error of measurement. * p < .001. ** p < .01. *** p < .05 (PRESTORAGE vs. POSTSTORAGE within each storage condition). Figure 2 —Regression of PRESTORAGE sweat [Na + ] on POSTSTORAGE sweat [Na + ] for the four storage conditions: (a) −20 °C (frozen); (b) 8 °C (refrigerated); (c) 23 °C (room); and (d
Kelsey Picha, Carolina Quintana, Amanda Glueck, Matt Hoch, Nicholas R. Heebner and John P. Abt
were violated prior to analysis. To detect RT differences across sessions, a repeated-measures analysis of variance was used. To determine reliability and variance, ICC 3,1 , standard error of measurement, and minimal detectable change at a 90% confidence interval were calculated. An ICC value of .50
Joel Garrett, Stuart R. Graham, Roger G. Eston, Darren J. Burgess, Lachlan J. Garrett, John Jakeman and Kevin Norton
“fatigued,” whereas the remaining samples considered to be “nonfatigued.” 3 , 7 Descriptive statistics are reported as mean (SD). Typical error of measurements (TE) were calculated using all 23 participants, expressed as a CV (±90% CI), were calculated to assess reliability for each variable. 19 The
Susana Cristina Araújo Póvoas, Peter Krustrup, Carlo Castagna, Pedro Miguel Ribeiro da Silva, Manuel J. Coelho-e-Silva, Rita Liliana Mendes Pereira and Malte Nejst Larsen
Shapiro–Wilk test. Absolute reliability for maximal distance, HR peak , and HR submax in each of the 3 Yo-Yo tests was assessed using the typical error of measurement (TEM), absolute and expressed as a percentage of the coefficient of variation (%CV). Random errors were interpreted as good, moderate, and