Process of contagion is driven not only by early infectious adopters but also by a critical mass of easily influenced individuals. An important research question is to statistically infer individuals' heterogeneous latent traits which can be used to rank and identify key entities and factors for precise contagion control. We propose a latent-traits modeling approach by extending IRT models to simultaneously incorporate and infer individual-level infectivity, susceptibility and baseline risk while controlling and estimating effects of environmental factors, from infection incidents and interaction data. And we offer Bayesian estimation with MCMC algorithms for the model.
The approach is illustrated with an application to fashion contagion where multilaterally connected customers' purchases across multiple products are potentially influenced by each other. The proposed model has better in-sample and out-sample fits than competing and benchmark models. The estimated results of the application show that the most infectious customers are not necessarily the most frequent buyers, nor the most connected ones; and that both the baseline adoption risks due to personal brand preferences and the sensitivities to marketing mix are highly heterogeneous among people.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.