Sample size estimation in locomotion kinematics and electromyography for statistical parametric mapping

Abstract

In biomechanics, kinematic and electromyographic data can be represented as one-dimensional (1D) waveforms and compared by using 1D hypothesis tests. These statistical techniques are increasingly applied in the study of locomotion. However, although widely agreed as a key step to obtain reliable and replicable findings, no a priori sample size estimation is usually conducted. This can also be done in 1D tests by calculating the statistical power – i.e., the probability of rejecting the null hypothesis when it is false – by using statistical parametric mapping. With the present study we characterised the parameters needed to estimate sample size in locomotion, and how they impact on statistical power in 1D tests. First, noise and signal in kinematics and electromyography were defined using experimental data on locomotion in physiological and pathological participants. Then, 1D power analysis was performed in representative conditions, and a dataset of tabulated sample sizes was generated. Kinematic and electromyographic data showed a smooth Gaussian noise, with amplitude and full-width-at-half-maximum depending on the physiological or pathological condition, and the considered joint or muscle. Given a certain noise, statistical power increased i) with greater signal amplitude and signal full-width-at-half-maximum, ii) when setting a region of interest and iii) when using a paired (vs. unpaired) study design. The present work provides initial benchmarks for appropriate sampling in 1D hypothesis testing, meant to evaluate statistical power in 1D tests and assists sample size estimation in studies on locomotion.