Exploratory factor analysis with structured residuals for brain network data.

Area of Science:

• Neuroscience
• Network Analysis
• Statistical Modeling

Background:

• Dimension reduction is crucial for analyzing complex network data, particularly in neuroscience.
• Exploratory Factor Analysis (EFA) is a common technique, but often overlooks known data structures like brain symmetry.
• Ignoring such a priori information can negatively impact the accuracy and interpretability of analyses.

Purpose of the Study:

• To demonstrate the detrimental effects of ignoring a priori structure in factor analysis.
• To introduce a novel technique, Exploratory Factor Analysis with Structured Residuals (EFAST), that incorporates structural information.
• To validate EFAST's effectiveness using large-scale brain-imaging network datasets.

Main Methods:

• Developing the EFAST technique to integrate structural constraints into the factor analysis framework.
• Utilizing structured residuals to account for known relationships within the data.
• Applying EFAST to three diverse brain-imaging network datasets.

Main Results:

• EFAST demonstrated a superior model fit compared to standard EFA across all datasets.
• The factors derived using EFAST showed enhanced interpretability, reflecting underlying brain structure.
• The adverse consequences of ignoring symmetry in factor analysis were clearly illustrated.

Conclusions:

• Incorporating a priori structural knowledge, such as brain symmetry, significantly improves factor analysis.
• EFAST offers a robust method for dimension reduction in network analysis, particularly for neuroimaging data.
• An R package is available to facilitate the application of EFAST by researchers.