CMStatistics 2015: Start Registration
View Submission - CMStatistics
B0192
Title: Closing the loop: Computational vs. analytical results in continuous-time modeling of infectious diseases Authors:  Carles Breto - The University of Michigan (United States) [presenting]
Abstract: The advent of cheap computer power promises to continue to catalyze empirical findings of scientific interest. Such findings sometimes come from data analyses based on models for which analytical results may not be readily available. However, such analytical results complete the circle: empirical findings are more likely to be valued by other scientists once they understand the model used to obtain them and, for such understanding, deriving model analytical properties has proved fundamental. An example of such a loop will be presented, focusing on plug-and-play computational statistical algorithms applied to the analysis of infectious disease data based on continuous-time stochastic processes of the SIR family, commonplace in ecology and epidemiology. Plug-and-play approaches allow the modeler to analyze data by simply writing computer code to simulate models. Empirical results will be presented based novel Markov chain models where the transition rates of the chain are subject to continuous-time white noise. These empirical findings led us to studying the properties of those Markov chains in the unconventional white-noise random environment. Finally, some of these properties will be presented, which reveal an unexpected change in the fundamental nature of the initial Markov chains. Such a fundamental change shows that with great computational power comes great analytical responsibility.