Pairwise likelihood estimation of latent autoregressive count models

Area of Science:

• Epidemiology
• Biostatistics
• Computational Statistics

Background:

• Latent autoregressive models are crucial for infectious disease time series analysis.
• Evaluating the likelihood function of these models is computationally challenging.
• Current simulation-based methods are intensive and difficult to validate.

Purpose of the Study:

• To propose a computationally efficient and robust alternative to simulation-based methods for latent autoregressive models.
• To explore the computational and methodological aspects of a weighted pairwise likelihood approach.
• To apply the proposed method to real-world infectious disease data.

Main Methods:

• Weighted pairwise likelihood estimation.
• Estimation of robust standard errors.
• Investigation of numerical integration techniques.
• Application to monthly invasive meningococcal disease data from Greece and Italy.

Main Results:

• The weighted pairwise likelihood approach provides a feasible alternative for likelihood evaluation.
• The method allows for robust standard error estimation.
• Numerical integration plays a key role in the computational efficiency and accuracy.

Conclusions:

• The weighted pairwise likelihood approach is a viable and computationally advantageous method for latent autoregressive models in infectious disease analysis.
• This approach offers improved assessment of model approximations.
• The method demonstrated effectiveness on invasive meningococcal disease data.