Statistical models for social networks have a long history in related public-health and the social and behavioral sciences. They can be used to provide precise stochastic representations of complex social structure, to compare theory to data and to simulate virtual networked populations that retain the essential properties of a theory or of data. This project will address fundamental issues in the statistical modeling of social networks and expands the existing capabilities. These are directly applicable to the epidemiological aspects of HIV/AIDS and STI both in the U.S. and internationally. Exponential-family random graph models are capable of representing the complex dependencies in social phenomena, and have been well studied in SNA. However, they do not represent the social endogeneity of nodal characteristics but only that of the relations. This project will address this deficiency by jointly stochastically modeling both the relational and individual variables via a novel class of exponential-family random network models. The majority of network data collection relies on sampling of the social network or is subject to missing data issues when a census is attempted. This project will develop new forms of network link- tracing designs that more efficiently collects information from the network while preserving the privacy of the networked population. Valid statistical inference from link-traced data is difficult because of th strong and often unknown dependencies in it. This project will develop a new framework for likelihood-based inference for social network models based on link-traced data when the covariates and outcome variables measured on the nodes are social endogenous. Many questions in health-related SNA are multivariate and can be stated as hypotheses about regressions of individual outcome variables on other covariates and their relational information. This project will extend network regression models to the more realistic situation where the outcomes, covariates and social relations are socially endogenous. The conceptual and methodological innovations will be applied to inferring HIV / STI prevalence among the IDU population in Los Angeles County and to HIV / STI among MSM in EU counties via the SIALON II project. The IDU data arise from an innovative link-tracing design to sample this hard-to- reach population. The social network structure of IDU will be inferred, and network regression will be used to analyze their HIV / STI prevalence. Privatized network sampling will be used in the MSN study.
Social interactions and processes are important determinants of human health, and especially the epidemiological aspects of HIV/AIDS and STI. This project addresses fundamental issues in the statistical modeling of social networks and expands the existing capabilities. These include better models for social processes, better ways to collect data, and better ways to identify factors important for health outcomes.