Estimators for Respondent Driven Sampling

Abstract

When random sampling is not possible because of lack of a sampling frame, a different approach is needed to aquire the statistical information that is desired. Respondent driven sampling is a series of methods to sample hard to reach populations that don't provide a sampling frame. It might possible to find reasonable estimators that allow for statistical inference from the data that can be collected. Bias is one of the main problems that can occur and to find and correct for a these biases can be challenging. In this thesis we derive two asymptotically correct estimators. One for the population prevalence and one for the homophily of the population. The difference in the degree between two parts of the population causes a bias that affects the estimates. In this thesis the coverage of the 95% confidence interval is usually below 95% for finite samples. However, this thesis does give a not too complicated example of an RDS-simulation that can be understood and adjusted.