Online Program

Exploring the Effectiveness of Logistic Regression with Respondent-Driven Sampling Data (309595)

Keywords: respondent driven sampling, logistic regression

Respondent-driven sampling (RDS) is a snowball type of sampling method primarily utilized for reaching hidden populations whose sampling frame is unknown. RDS consists of several waves of incentivized peer-to-peer recruitment that finishes when the desired sample size is met. With this sampling method, there is a tendency to oversample highly connected people who share similar characteristics (i.e., homophily). Much of the current research into RDS sampling methodology has focused on estimator properties for population means or proportions, while less work has been done on model performance using RDS data. In this poster, we will discuss results from a simulation study designed to analyze the estimation performance of a basic logistic regression model with RDS data. We found that estimator performance varied depending on how the population level homophily was related to the response and explanatory variables. We also explored estimator properties of a random effects logistic model to account for clustering in a RDS sample, but this model performed worse than a basic logistic model in our study.