Abstract:
Panel or longitudinal analysis of dynamic systems using scientifically motivated mechanistic models is becoming more widespread thanks to the increasing availability of data and to recent advances in statistical inference methodology that aim at dispensing with linearity and Gaussianity assumptions. Examples of such inference tools are iterated filtering algorithms. However, development of these algorithms has focused on a multivariate time series framework---without explicitly considering panel settings---and has relied on sequential Monte Carlo, which needs to be made scalable for large panels. Panel data can be useful both to disentangle within- and between-individual features and to deal through partial pooling of data with parameter weak identifiability and biases. In this talk, I will present a novel panel iterated filtering algorithm, which extends existing methodology to nonlinear non-Gaussian panel models, and I will illustrate how it can be applied to study infectious disease dynamics.