Sensitivity of discrete symmetry metrics: Implications for metric choice
Allen Hill 
Julie Nantel
Gait asymmetry is present in several pathological populations, including those with Parkinson’s disease, Huntington’s disease, and stroke survivors. Previous studies suggest that commonly used discrete symmetry metrics, which compare single bilateral variables, may not be equally sensitive to underlying effects of asymmetry, and the use of a metric with low sensitivity could result in unnecessarily low statistical power. The purpose of this study was to provide a comprehensive assessment of the sensitivity of commonly used discrete symmetry metrics to better inform design of future studies. Monte Carlo simulations were used to estimate the statistical power of each symmetry metric at a range of asymmetry magnitudes, group/condition variabilities, and sample sizes. Power was estimated by repeated comparison of simulated symmetric and asymmetric data with a paired t-test, where the proportion of significant results is equivalent to the power. Simulation results confirmed that not all common discrete symmetry metrics are equally sensitive to reference effects of asymmetry. Multiple symmetry metrics exhibit equivalent sensitivities, but the most sensitive discrete symmetry metric in all cases is a bilateral difference (e.g. left—right). A ratio (e.g. left/right) has poor sensitivity when group/condition variability is not small, but a log-transformation produces increased sensitivity. Additionally, two metrics which included an absolute value in their definitions showed increased sensitivity when the absolute value was removed. Future studies should consider metric sensitivity when designing analyses to reduce the possibility of underpowered research.
gait asymmetry, Monte Carlo simulation
Reuse
Citation
@article{hill2022,
author = {Hill, Allen and Nantel, Julie},
publisher = {Public Library of Science},
title = {Sensitivity of Discrete Symmetry Metrics: {Implications} for
Metric Choice},
journal = {PLOS ONE},
volume = {17},
number = {5},
pages = {e0268581},
date = {2022-05-19},
url = {https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0268581},
doi = {10.1371/journal.pone.0268581},
langid = {en},
abstract = {Gait asymmetry is present in several pathological
populations, including those with Parkinson’s disease, Huntington’s
disease, and stroke survivors. Previous studies suggest that
commonly used discrete symmetry metrics, which compare single
bilateral variables, may not be equally sensitive to underlying
effects of asymmetry, and the use of a metric with low sensitivity
could result in unnecessarily low statistical power. The purpose of
this study was to provide a comprehensive assessment of the
sensitivity of commonly used discrete symmetry metrics to better
inform design of future studies. Monte Carlo simulations were used
to estimate the statistical power of each symmetry metric at a range
of asymmetry magnitudes, group/condition variabilities, and sample
sizes. Power was estimated by repeated comparison of simulated
symmetric and asymmetric data with a paired t-test, where the
proportion of significant results is equivalent to the power.
Simulation results confirmed that not all common discrete symmetry
metrics are equally sensitive to reference effects of asymmetry.
Multiple symmetry metrics exhibit equivalent sensitivities, but the
most sensitive discrete symmetry metric in all cases is a bilateral
difference (e.g. left—right). A ratio (e.g. left/right) has poor
sensitivity when group/condition variability is not small, but a
log-transformation produces increased sensitivity. Additionally, two
metrics which included an absolute value in their definitions showed
increased sensitivity when the absolute value was removed. Future
studies should consider metric sensitivity when designing analyses
to reduce the possibility of underpowered research.}
}