Panel Data Analysis via Mechanistic Models.

Area of Science:

• Statistics
• Computational Biology
• Epidemiology

Background:

• Panel data, or longitudinal data, comprises multiple time series from distinct units.
• Mechanistic modeling describes dynamic systems generating observations, assuming negligible inter-unit interactions.
• Existing models often lack flexibility for arbitrary nonlinear, partially observed panel data.

Purpose of the Study:

• To develop a flexible framework for statistical inference on panel data.
• To accommodate arbitrary nonlinear and partially observed mechanistic models.
• To enable likelihood-based inference for complex panel data structures.

Main Methods:

• Utilizes iterated filtering techniques for likelihood-based inference.
• Builds upon nonlinear partially observed Markov process models.
• Employs a simulation-based approach for latent Markov processes, ensuring model class applicability.

Main Results:

• Demonstrates a 'plug-and-play' methodology applicable to a wide range of panel models.
• Successfully applies the framework to a toy example and two epidemiological case studies.
• Addresses inferential and computational challenges associated with complex models and large datasets.

Conclusions:

• The proposed framework offers significant flexibility for mechanistic modeling of panel data.
• Iterated filtering provides a robust approach for inference in complex, partially observed systems.
• The methodology is well-suited for analyzing epidemiological data and other scientific domains.