Semiparametric estimation of covariance matrices for longitudinal data

Area of Science:

• Statistics
• Biostatistics
• Longitudinal Data Analysis

Background:

• Estimating covariance structure in longitudinal data is challenging due to irregular time points.
• Existing semiparametric models offer ways to estimate variance and correlation functions but lack asymptotic property analysis for their estimators.
• The quasi-maximum likelihood estimator (QMLE) for covariance models has unknown asymptotic properties, hindering robust application.

Purpose of the Study:

• To investigate the asymptotic properties of the quasi-maximum likelihood estimator (QMLE) for covariance models with longitudinal data.
• To develop more robust covariance function estimators applicable to a wider range of mean regression models, including rough functions.
• To establish theoretical guarantees for the QMLE in estimating correlation function parameters.

Main Methods:

• Extended semiparametric covariance models to accommodate general conditional mean functions (parametric, nonparametric, semi-parametric).
• Introduced a difference-based method to mitigate bias in varying-coefficient partially linear mean regression models with rough functions.
• Derived consistency and asymptotic normality for the QMLE under specified technical conditions.

Main Results:

• Developed a more robust estimator for the covariance function, suitable for diverse situations.
• Established the consistency and asymptotic normality of the quasi-maximum likelihood estimator (QMLE) for correlation function parameters.
• Demonstrated the proposed approach's effectiveness through simulation studies and a real data example.

Conclusions:

• The study provides theoretical justification for using QMLE in longitudinal data covariance estimation.
• The proposed methods enhance robustness and applicability across various mean regression scenarios.
• The findings contribute to more reliable statistical inference for longitudinal data analysis.