Linear discriminant function analysis in neuropsychological research: some uses and abuses.

Area of Science:

• Neuropsychology
• Statistical analysis
• Research methodology

Background:

• Linear discriminant function analysis (LDFA) is frequently misinterpreted and misapplied in neuropsychological research.
• Methodological issues like shrinkage and stepwise selection significantly impact LDFA results, yet are often overlooked.
• These problems affect both the classification accuracy and the inferential validity of LDFA applications.

Purpose of the Study:

• To address the prevalent misinterpretation and misapplication of LDFA in neuropsychological studies.
• To elucidate the impact of methodological issues, specifically shrinkage and stepwise selection, on LDFA.
• To offer practical guidance for accurate interpretation of LDFA outcomes in research.

Main Methods:

• Review of existing literature on LDFA applications and limitations in neuropsychology.
• Illustrative examples demonstrating potential abuses of LDFA.
• Application of specific procedures to a familiar research problem to showcase accurate LDFA interpretation.

Main Results:

• Demonstration that shrinkage and stepwise selection procedures introduce significant biases in LDFA.
• Identification of common errors in classifying groups and drawing inferences using LDFA.
• Provision of a framework for more reliable and valid application of LDFA.

Conclusions:

• LDFA is a powerful statistical tool for exploring group differences when used correctly.
• Accurate interpretation of LDFA results is essential to avoid erroneous conclusions in neuropsychological research.
• Adherence to recommended methodological practices ensures the integrity and utility of LDFA in scientific inquiry.