Testing P-Technique Factor Analysis With Non-Normal Time Series.

Area of Science:

• Statistics
• Data Science
• Psychometrics

Background:

• Multivariate time series data are increasingly prevalent, driven by advancements in data collection tools like wearable sensors and brain imaging.
• Analyzing this data is complex due to common issues such as non-normality (e.g., in mood, heart rate, brain signals) and autocorrelation.
• P-technique factor analysis is a key method for developing measurement models for time series data.

Purpose of the Study:

• To propose and evaluate a novel bootstrap procedure for analyzing multivariate time series data.
• To address the challenges of non-normality and temporal dependencies inherent in such data.
• To compare the performance of the bootstrap procedure against existing analytic methods.

Main Methods:

• A bootstrap procedure was developed to handle non-normal and serially correlated time series data.
• Statistical properties were examined using simulated datasets.
• The proposed method was illustrated with two empirical datasets.

Main Results:

• The bootstrap procedure outperformed an existing analytic procedure for time series data exhibiting excessive kurtosis.
• For normally distributed and skewed time series data, the existing analytic procedure demonstrated superior performance compared to the bootstrap method.

Conclusions:

• The bootstrap procedure offers a viable alternative for analyzing time series data with high kurtosis.
• The choice of analytic procedure should consider the distributional properties of the time series data.
• Further research can explore extensions and applications of the bootstrap method in time series analysis.