Determining optimal trial size using sequential analysis

Abstract

When characterising typical human movement profiles, the optimal number of trials analysed for each participant should ensure a stable mean. Sequential analysis is one method able to establish the number of trials to stability by assessing a moving point mean against a set bandwidth. As the total trial number determining this bandwidth is selected arbitrarily, the effect of applying different total trial numbers on the results of sequential analysis was investigated. Twenty participants performed 30 trials of overarm throwing, and sequential analyses were applied to three dimensional (3-D) kinematic data over 10, 20 and 30 trial numbers. We found a total of 20 to be the preferred trial number for sequential analyses. Erroneous results were produced consistently by 10 trial number groups, while moving point means were statistically unchanged after the 10th trial. Subsequently, sequential analyses were applied to 20 trials to establish trials to stability in discrete and time series elements of the 3-D kinematic data. The results suggest that a trial size between 13 and 17 provides stable means for overarm throwing kinematics.

Selecting the number of trials in experimental biomechanics studies

“…Typically, ICC analysis has resulted in an estimated four trials for stability, suggesting it to be less conservative than the SET. Based on the t-statistic (Figure 1), this value is in the region where the confidence intervals on the mean are changing quite rapidly with number of trials and a similar observation has been made based on experimental data (Taylor et al 2015). Hence, the ICC method holds the risk of under-estimating the number of trials required to achieve stability.…”

“…Thus, the number of trials to achieve stability for a given reference number of trials and standard deviation threshold can be obtained directly from this simulation model for any movement without the need to invest the time and resource in collecting experimental data. The currently used criteria of 20 reference trials and a threshold of 0.25 resulted in 9 ± 8 trials (mean ± 95% confidence intervals; Figure 3) to achieve a stable estimate of the mean which encompassed all the results from the experimental studies (Bates et al 1983;Hamill & McNiven 1990;James et al 2007;Racic et al 2009;Taylor et al 2015). Interestingly, some of the experimental SET results tended to lie above the mean simulation line particularly as the reference number of trials increased.…”

“…The simulations were carried out for N REF between 3 and 50 and THRES between 0.1 and 0.5. Each sample involved 100,000 computer generated data-sets from which the mean and 95% confidence intervals for N STABLE were evaluated (Taylor et al 2015). The simulation model was validated by using threshold values of 0 (expect N STABLE = N REF ) and 2 (expect N STABLE → 1).…”

“…In subplots (a)-(c) the thin blue line with filled circles represent the SET results from the experimental breast range of motion data; averaged over all subjects, bras and co-ordinate directions. In subplot (b), the thin red lines and unfilled squares represent the results from previous experimental applications of the SET analysis (Bates et al 1983;Hamill & McNiven 1990;Rodano & Squadrone 2002;James et al 2007;Racic et al 2009;Taylor et al 2015). (Table 2).…”

“…These have covered a range of movements including running (Bates et al 1983), walking (Hamill & McNiven 1990), vertical jumping (Rodano & Squadrone 2002), landing (James et al 2007), continuous jumping (Racic et al 2009) and overarm throwing (Taylor et al 2015). In brief, SET requires a reference number of trials and standard deviation threshold to be selected.…”