Criterion for the simultaneous selection of a working correlation structure and either generalized estimating...

Area of Science:

• Statistics
• Biostatistics
• Longitudinal Data Analysis

Background:

• Generalized estimating equations (GEE) are standard for marginal analysis of correlated data.
• Quadratic inference functions (QIF) offer an alternative, combining unbiased estimating equations for improved asymptotic efficiency.
• Accurate correlation structure modeling enhances efficiency for both GEE and QIF.

Purpose of the Study:

• To improve parameter estimation by simultaneously selecting working correlation structures and choosing between GEE and QIF methods.
• To introduce a novel criterion, the trace of the empirical covariance matrix (TECM), for method selection.
• To address finite-sample variability in QIF by incorporating a penalty for variance inflation.

Main Methods:

• Proposed a criterion based on the trace of the empirical covariance matrix (TECM).
• Developed a penalty within TECM to account for finite-sample variance inflation in QIF.
• Evaluated the proposed criterion using simulation studies and a real-world longitudinal study.

Main Results:

• Demonstrated the necessity of penalizing variance inflation for the QIF approach.
• Showed that the proposed TECM criterion performs effectively in selecting optimal methods and correlation structures.
• The TECM criterion improved parameter estimation compared to existing methods.

Conclusions:

• The proposed TECM criterion offers a robust method for selecting between GEE and QIF, and for choosing correlation structures.
• Penalizing variance inflation is crucial for the effective application of QIF in finite samples.
• The TECM-based approach enhances parameter estimation accuracy in correlated data analysis.