Assessment and Validation of Respondent Driven Sampling

Jing Li, University of Wisconsin at Madison
Giovanna Merli, University of Wisconsin at Madison
Erik Nordheim, University of Wisconsin at Madison
William Whipple Neely, University of Wisconsin at Madison

We assess the statistical validity of respondent-driven-sampling (RDS), an approach that is becoming increasingly popular for sampling hard to reach populations. RDS is based on certain assumptions about the social network connecting members of the population. It purports to allow valid inference from the sample to all potential members of the population of interest and to yield statistically unbiased estimates of behaviors of hidden populations. We evaluate the validity of a number of the key assumptions of RDS with a simulation model that is able to describe potential populations of FSWs and includes a realistic structure of their social networks. By modifying the model parameters, we evaluate what effects the possible violation(s) of the assumptions have on the inference that can be drawn. Variation in model parameters are empirically informed by exploratory research conducted among female sex workers in Shanghai, China. We also explore a wider range of input parameters to allow our results to be generalizable to a broader range of hard-to reach populations.