Testing model fit in longitudinal data analysis against alternatives with omitted covariates

Area of Science:

• Biostatistics
• Longitudinal Data Analysis
• Statistical Modeling

Background:

• Model misspecification, including omitted covariates, is a common issue in statistical analysis.
• Omitted covariates can arise from unmeasured confounders, measurement errors, or informative censoring.
• Longitudinal data offers unique opportunities to detect time-varying omitted covariates.

Purpose of the Study:

• To develop a method for testing the existence of omitted covariates in longitudinal data analysis.
• To address challenges in specifying alternative models when omitted covariates are present.
• To propose a quasi-score test statistic that simplifies the detection of omitted covariates.

Main Methods:

• Focus on longitudinal data analysis using generalized estimation equations (GEE).
• Development and application of a quasi-score test statistic.
• Asymptotic chi-square distribution under the null hypothesis of no omitted covariates.

Main Results:

• The quasi-score test effectively avoids complex numerical integration required by alternative models.
• The test's significance level and power were studied in linear and logistic regression models.
• The proposed test was applied to analyze excessive daytime sleepiness data.

Conclusions:

• The quasi-score test provides a robust method for detecting omitted covariates in longitudinal data.
• This approach simplifies model fitting and improves the reliability of statistical inferences.
• The test is a valuable tool for researchers dealing with potential model misspecifications in longitudinal studies.