TY - JOUR
T1 - Statistical properties of Multiscale Regression Analysis
T2 - Simulation and application to human postural control
AU - Likens, Aaron D.
AU - Amazeen, Polemnia G.
AU - West, Stephen G.
AU - Gibbons, Cameron T.
N1 - Funding Information:
This work was supported in part by the Robert Cialdini Leap Forward Fund within the ASU Department of Psychology and the National Science Foundation [ BCS 1255922 ] (P. Amazeen). Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author and do not necessarily reflect the views of the National Science Foundation.
Publisher Copyright:
© 2019 Elsevier B.V.
PY - 2019/10/15
Y1 - 2019/10/15
N2 - Multiscale Regression Analysis (MRA) is a promising new tool for the analysis of bivariate time series that is based on Detrended Fluctuation Analysis (DFA) and Ordinary Least Squares (OLS) regression. The method was developed within the economics and environmental science literatures (Kristoufek, 2015, 2018; Kristoufek and Ferreira, 2018) and is beginning to be applied in other scientific domains. To date, however, no systematic studies have investigated the behavior of the estimator with respect to short time series. This paper fills that gap by assessing the performance of the MRA estimator using time series with varying length, distribution, and structure (e.g., autocorrelation, stationarity). Simulations show that MRA performs well under many circumstances with as few as 512 observations. Linear and quadratic time trends contribute considerable systematic bias; however, using a detrending polynomial of order ≥ 2 effectively attenuates time trend associated deviations from expected values. We apply MRA to a previously published dataset in order to explore the relationship that emerges between body segments during an act of quiet standing. Results suggest that the velocity of the hip asymptotically depends on velocity of the ankle. In contrast, ankle velocity was a much weaker predictor of shoulder velocity. We conclude by providing suggestions for best practice and future model development.
AB - Multiscale Regression Analysis (MRA) is a promising new tool for the analysis of bivariate time series that is based on Detrended Fluctuation Analysis (DFA) and Ordinary Least Squares (OLS) regression. The method was developed within the economics and environmental science literatures (Kristoufek, 2015, 2018; Kristoufek and Ferreira, 2018) and is beginning to be applied in other scientific domains. To date, however, no systematic studies have investigated the behavior of the estimator with respect to short time series. This paper fills that gap by assessing the performance of the MRA estimator using time series with varying length, distribution, and structure (e.g., autocorrelation, stationarity). Simulations show that MRA performs well under many circumstances with as few as 512 observations. Linear and quadratic time trends contribute considerable systematic bias; however, using a detrending polynomial of order ≥ 2 effectively attenuates time trend associated deviations from expected values. We apply MRA to a previously published dataset in order to explore the relationship that emerges between body segments during an act of quiet standing. Results suggest that the velocity of the hip asymptotically depends on velocity of the ankle. In contrast, ankle velocity was a much weaker predictor of shoulder velocity. We conclude by providing suggestions for best practice and future model development.
KW - Detrended fluctuation analysis
KW - Dynamics
KW - Fractal regression
KW - Multiscale regression analysis
KW - Posture
UR - http://www.scopus.com/inward/record.url?scp=85067789996&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85067789996&partnerID=8YFLogxK
U2 - 10.1016/j.physa.2019.121580
DO - 10.1016/j.physa.2019.121580
M3 - Article
AN - SCOPUS:85067789996
SN - 0378-4371
VL - 532
JO - Physica A: Statistical Mechanics and its Applications
JF - Physica A: Statistical Mechanics and its Applications
M1 - 121580
ER -