Multiple imputation of missing data in large studies with many variables: A fully conditional specification approach...

Area of Science:

• Psychological research methodology
• Statistical modeling
• Data analysis techniques

Background:

• Multiple imputation (MI) is widely used for missing data in psychology.
• Conventional MI methods struggle with datasets containing numerous variables, often requiring simplification.
• Simplification can lead to unstable imputation models and loss of information.

Purpose of the Study:

• To propose an advanced MI method integrating dimension reduction techniques with fully conditional specification.
• To address the limitations of conventional MI in high-dimensional psychological datasets.
• To offer a more robust approach for handling missing data in complex research.

Main Methods:

• Developed a novel approach extending fully conditional specification (FCS) with dimension reduction (e.g., partial least squares).
• Conducted simulation studies to compare the proposed method against variable selection, composite scores, and principal component analysis-based MI.
• Applied the method to real-world psychological data.

Main Results:

• The proposed MI method demonstrated accurate results in challenging scenarios where other methods failed.
• Partial least squares-enhanced FCS proved more stable and effective for high-dimensional data.
• The method successfully handled missing data without significant information loss.

Conclusions:

• The novel MI approach offers a powerful solution for missing data problems in psychological research with many variables.
• Integrating dimension reduction with FCS enhances imputation accuracy and model stability.
• This method provides practical benefits for researchers dealing with complex datasets.