A robust likelihood approach to inference about the kappa coefficient for correlated binary data.

Area of Science:

• Statistics
• Biostatistics
• Data Analysis

Background:

• The agreement kappa coefficient is crucial for assessing inter-rater reliability, especially with correlated data.
• Traditional methods often require specific distributional assumptions or detailed modeling of all correlation levels, limiting their applicability.
• A need exists for flexible statistical tools that can handle clustered and correlated data without stringent assumptions.

Purpose of the Study:

• To develop a robust likelihood function for the agreement kappa coefficient applicable to correlated data.
• To enable standard statistical inference tools, such as likelihood ratio and score tests, without requiring knowledge of underlying data distributions.
• To provide a parametric robust likelihood approach suitable for general clustered data scenarios.

Main Methods:

• Construction of a legitimate likelihood function for the agreement kappa coefficient.
• Development of a parametric robust likelihood approach.
• Application to general clustered data scenarios, bypassing the need to model all correlation levels.

Main Results:

• The proposed method provides a valid likelihood function for agreement kappa with correlated data.
• It facilitates the use of likelihood ratio tests and score tests, irrespective of the underlying data distributions.
• Simulations and real data analyses confirm the advantages of this robust procedure for clustered data.

Conclusions:

• The parametric robust likelihood approach offers a powerful and flexible tool for analyzing agreement in correlated and clustered data.
• This method enhances statistical inference by removing the dependency on specific distributional assumptions.
• The demonstrated advantages in simulations and real-world applications highlight its practical utility in various scientific fields.