MIXED MODEL AND ESTIMATING EQUATION APPROACHES FOR ZERO INFLATION IN CLUSTERED BINARY RESPONSE DATA WITH APPLICATION...

Area of Science:

• Adolescent health research
• Biostatistics
• Public health

Background:

• Adolescent dating violence is a significant public health concern.
• Existing survey data may contain excessive zeros due to non-relationship responses, complicating analysis.
• Accurate statistical methods are crucial for understanding and addressing dating violence.

Purpose of the Study:

• To propose and evaluate statistical approaches for analyzing zero-inflated clustered binary response data common in adolescent health studies.
• To address the issue of excessive zeros in dating violence surveys.
• To compare the performance of different analytical methods in handling this data characteristic.

Main Methods:

• Development of likelihood-based (ML) and generalized estimating equations (GEE) approaches.
• Utilizing mixed models to account for cluster effects.
• Employing Gaussian-Hermite quadrature (GHQ) approximation for ML estimation.
• Conducting simulation studies to assess bias, efficiency, and robustness of ML and GEE methods.

Main Results:

• Simulation studies evaluated the performance of ML and GEE methods.
• The study examined the bias, efficiency, and robustness of the proposed statistical techniques.
• Reanalysis of the NEXT Generation Health study data demonstrated the impact of accounting for zero inflation.

Conclusions:

• Properly accounting for zero inflation is critical for accurate analysis of adolescent dating violence data.
• The proposed ML and GEE methods offer robust solutions for zero-inflated clustered binary data.
• These advanced statistical techniques enhance the reliability of findings in adolescent health research.