Generalized canonical correlation analysis of matrices with missing rows: a simulation study

Area of Science:

• Multivariate statistics
• Data analysis methodologies

Background:

• Canonical correlation analysis (CCA) is a statistical method to analyze relationships between two sets of variables.
• Handling missing data in multivariate analyses presents significant challenges.
• Existing methods like GENCOM have limitations in analyzing matrices with missing rows.

Purpose of the Study:

• To develop and evaluate a novel method for generalized canonical correlation analysis (GCCA) capable of handling matrices with missing rows.
• To compare the performance of the proposed GCCA method against the existing GENCOM algorithm.
• To assess improvements in model fit and the recovery of true data structures.

Main Methods:

• The proposed method integrates Carroll's (1968) CCA approach with the missing data handling strategy from the OVERALS technique (Van der Burg, 1988).
• A simulation study was conducted to rigorously assess the performance of the new method.
• The new method's performance was benchmarked against the GENCOM algorithm (Green and Carroll, 1988).

Main Results:

• The proposed method demonstrated superior performance compared to the GENCOM algorithm.
• Key improvements were observed in both model fit indices.
• Enhanced recovery of the underlying data structure was achieved by the new method.

Conclusions:

• The novel generalized canonical correlation analysis method effectively addresses matrices with missing rows.
• This new approach offers significant advantages over existing procedures like GENCOM for analyzing complex datasets.
• The findings suggest a more robust and accurate tool for multivariate data analysis in the presence of missing information.