Abstract:
|
Understanding the structure of face-to-face social contact networks can improve models for disease transmission. A social contact survey was administered in Niakhar, Senegal in conjunction with a cluster-randomized influenza vaccine trial. Participants reported numbers and locations of contacts for three days, and participants and household compounds are uniquely identified through Census data. Numbers of contacts were frequently rounded to multiples of five, so we implement a latent variable model to simultaneously estimate two rounding probabilities and parameters for a negative binomial model for the degree distribution. The model predicts degree based on symptom status, age, compound size, day relative to the survey day, and time of day. We also derive an estimator for the probability that a contact occurs to a member of one's own compound. We compare the network properties to those for other countries and make recommendations to refine survey design.
|