University of Notre Dame.
Multivariate Behav Res. 2022 Nov-Dec;57(6):895-915. doi: 10.1080/00273171.2021.1919047. Epub 2021 May 17.
There is an increasing need to analyze multivariate time series data due to the rapid development of data collection tools such as smartphone APPs, wearable sensors, and brain imaging techniques. P-technique factor analysis allows researchers to establish a measurement model for these time series. Analyzing such data is challenging because they are often non-normal (e.g., steps, heart rate, sleep, mood, and brain signals) and correlated at nearby time points. We propose using a bootstrap procedure to accommodate both the non-normality and the dependency of nearby time points. We explore the statistical properties with simulated data and illustrate the test with two empirical data sets. The results of the simulation study include (1) the bootstrap procedure performed better than an existing analytic procedure for time series data with excessive kurtosis (2) an existing analytic procedure performed better than the bootstrap procedure for normal time series and skewed time series.
由于智能手机应用程序、可穿戴传感器和脑成像技术等数据收集工具的快速发展，人们越来越需要分析多元时间序列数据。P 技术因子分析允许研究人员为这些时间序列建立测量模型。分析这类数据具有挑战性，因为它们通常是非正态的（例如，步长、心率、睡眠、情绪和脑信号），并且在附近的时间点上相关。我们建议使用自举程序来兼顾非正态性和附近时间点的相关性。我们使用模拟数据探索了统计性质，并使用两个经验数据集说明了该检验。模拟研究的结果包括：(1) 自举程序在处理具有过高峰度的时间序列数据方面比现有的分析程序表现更好；(2) 对于正态时间序列和偏态时间序列，现有的分析程序比自举程序表现更好。
