Hidden populations are, by definition, hard-to-reach, but are often of great interest in scientific studies; examples of such populations include the homeless, injection drug users, and men who have sex with men. Respondent- driven sampling (RDS) involves peers recruiting their peers (usually friends and acquaintances) through the use of a small number of uniquely-numbered coupons per participant. RDS employs a 'double incentive' system, where individuals are compensated both for participating in the study and for successfully recruiting other eligible participants. Not only does RDS allow members of hidden populations to be recruited efficiently in this manner, it also provides information on mixing between different subpopulations (otherwise known as 'homophily') and on network sizes of individuals, both important factors underlying the structure of social networks. Although RDS is currently being used in a large number of studies around the world, methodology to analyze the data being generated by these studies needs to be developed and validated. Our application has three overarching aims; to develop statistical models to analyze homophily and network size, two important features of the structure of social networks; to develop weighting schemes which correct for the biased nature of RDS samples, to allow standard statistical approaches to be applied; and to develop a simulation environment to investigate the statistical properties of our approaches and the robustness of parameter estimates to model assumptions. Our application melds state-of-the-art statistical inference and computational modeling in order to address long-standing questions in sociology. Respondent-driven sampling (RDS) is a technique to sample individuals from 'hidden' populations, such as the homeless, injection drug users, and men who have sex with men, which involves peers recruiting their peers. Not only does RDS allow members of hidden populations to be recruited efficiently in this manner, it also provides information on mixing between different subpopulations (otherwise known as 'homophily') and on network sizes of individuals, both important factors underlying the structure of social networks. Our application has three overarching aims; to develop statistical models to analyze homophily and network size, two important features of the structure of social networks; to develop weighting schemes which correct for the biased nature of RDS samples, to allow standard statistical approaches to be applied; and to develop a simulation environment to investigate the statistical properties of our approaches and the robustness of parameter estimates to model assumptions. ? ? ?
Showing the most recent 10 out of 12 publications
