Dynamic factor analysis with multivariate time series of multiple individuals: An error-corrected estimation method.

Area of Science:

• Social and Behavioral Sciences
• Statistics
• Psychometrics

Background:

• Intensive longitudinal data, often multivariate time series, are increasingly used in social and behavioral research.
• Dynamic factor analysis (DFA) models individual processes from time series but struggles to integrate them across individuals due to estimation errors.

Purpose of the Study:

• To develop a computationally efficient and robust method for integrating individual-specific dynamic processes from multivariate time series.
• To accommodate estimation errors inherent in individual-specific parameter estimates.
• To improve the accuracy of random effects estimation in multi-individual time series analysis.

Main Methods:

• Proposed a novel statistical method combining dynamic factor analysis with techniques to handle estimation errors.
• The method is robust to model misspecification and non-normal data.
• Compared the proposed method against a Naive approach (ignoring estimation errors) using empirical and simulated data.

Main Results:

• Both methods yielded similar fixed effect parameter estimates.
• The proposed method demonstrated superior performance in estimating random effects compared to the Naive approach.
• The proposed method's advantage was more pronounced for shorter time series (T = 56-200).

Conclusions:

• The proposed method effectively integrates individual-specific processes in multivariate time series analysis while accounting for estimation errors.
• This approach offers improved random effects estimation, particularly beneficial for shorter longitudinal datasets.
• The method's computational efficiency and robustness enhance its applicability in social and behavioral sciences.