Comparison of variance estimation approaches in a two-state Markov model for longitudinal data with misclassification

Area of Science:

• Statistics
• Biostatistics
• Machine Learning

Background:

• Alternating binary Markov models are used to analyze unobservable processes.
• Misclassification in observable data can affect the accuracy of transition probability estimation.
• Understanding the behavior of parameter estimates under misclassification is crucial for reliable analysis.

Purpose of the Study:

• To examine the variance-covariance parameter estimates in an alternating binary Markov model with misclassification.
• To compare the performance of different estimation procedures under varying degrees of misclassification.
• To identify the most robust estimation method for different sample sizes and misclassification probabilities.

Main Methods:

• The study employed three estimation procedures: observed information, jackknife, and bootstrap techniques.
• Variance components of estimated transition parameters were calculated using these methods.
• Simulation studies were conducted to compare variance estimates and assess the impact of misclassification.

Main Results:

• Observed information, jackknife, and bootstrap methods yielded similar variance estimates for large samples and moderate misclassification.
• Resampling methods (jackknife and bootstrap) are viable alternatives to programming partial derivatives.
• The bootstrap method demonstrated superior performance with smaller sample sizes or higher misclassification probabilities.

Conclusions:

• The choice of estimation procedure impacts the reliability of parameter estimates in misclassified Markov models.
• Bootstrap resampling is a robust and recommended method for handling misclassification, especially in challenging data scenarios.
• Accurate estimation of transition probabilities is essential for understanding the dynamics of unobservable processes.