Selecting the number of trials in experimental biomechanics studies

Abstract

Experimental biomechanics studies often involve the comparison of mean values from individuals across two or more experimental conditions. The purpose of this study was to evaluate two existing methods for determining the number of trials necessary to estimate these means. The sequential estimation technique (SET) was investigated in terms of the influence of input data distribution on the outcome. Paired samples t-tests were investigated in terms of the interaction between the number of subjects and number of trials necessary to achieve an acceptable level of statistical power. Simulation models were developed to perform SET and paired samples t-tests on representative synthetic input data. The SET results confirmed that the number of trials to achieve a stable estimate of the mean is independent of the input distribution provided the mean and standard deviation are fixed. For the commonly used 20 reference trials and 0.25 standard deviation threshold 9 ± 8 trials were needed to achieve stability. The paired t-test results confirmed that both number of subjects and number of trials can have a marked effect on the statistical power, e.g. a power of 0.80 can be achieved for effect size of 0.80 using 15 subjects and at least 19 trials or 20+ subjects and only 3 trials. The SET method suffers from arbitrary convergence criteria and neglecting intra-subject variance and, thus, should be applied with extreme caution. In contrast, statistical power can provide a more objective and conclusive means for determining the number of trials required for a given experimental situation.