Exploiting test structure: case series, case-control comparison, and dissociation.

Area of Science:

• Neuropsychology
• Biostatistics
• Psychometrics

Background:

• Neuropsychological tests typically comprise multiple items, with scores aggregated at the subject level.
• Standard analysis often treats subject scores as single data points, overlooking the inherent hierarchical structure (item within subject).
• This hierarchical structure presents an opportunity for more sophisticated statistical modeling.

Purpose of the Study:

• To present statistical methods that exploit the multi-level structure of neuropsychological test data.
• To demonstrate how accounting for item-level variance can enhance statistical power and analytical nuance.
• To provide a framework for analyzing case series, case-control comparisons, and dissociation using binomial general linear models.

Main Methods:

• Utilizing the binomial general linear model to analyze data with a two-level structure (item nested within subject).
• Developing methods for case series analysis, case-control comparisons, and dissociation testing.
• Incorporating multiple predictors (categorical and continuous) into the models.

Main Results:

• The proposed methods allow for the incorporation of test length and item dispersion effects on score variance.
• Exploiting test structure can lead to increased statistical power compared to traditional single-score analyses.
• The framework accommodates complex hypothesis testing, including the investigation of explanatory factors beyond group status.

Conclusions:

• Analyzing neuropsychological test data at the item level offers significant advantages over aggregated scores.
• The binomial general linear model provides a flexible tool for exploiting test structure in various research designs.
• These methods enhance researchers' ability to conduct nuanced analyses and test complex hypotheses in neuropsychological research.