Premium
The sensitivity of respondent‐driven sampling
Author(s) -
Lu Xin,
Bengtsson Linus,
Britton Tom,
Camitz Martin,
Kim Beom Jun,
Thorson Anna,
Liljeros Fredrik
Publication year - 2012
Publication title -
journal of the royal statistical society: series a (statistics in society)
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.103
H-Index - 84
eISSN - 1467-985X
pISSN - 0964-1998
DOI - 10.1111/j.1467-985x.2011.00711.x
Subject(s) - respondent , robustness (evolution) , sampling bias , statistics , variance (accounting) , sampling (signal processing) , simple random sample , selection bias , sample (material) , population , selection (genetic algorithm) , social network (sociolinguistics) , lesbian , econometrics , sample size determination , computer science , psychology , mathematics , demography , biology , machine learning , filter (signal processing) , psychoanalysis , law , business , chemistry , sociology , world wide web , social media , biochemistry , accounting , chromatography , political science , computer vision , gene
Summary. Researchers in many scientific fields make inferences from individuals to larger groups. For many groups, however, there is no list of members from which to draw a random sample. Respondent‐driven sampling (RDS) is a relatively new sampling methodology that circumvents this difficulty by using the social networks of the groups under study. The RDS method has been shown to provide unbiased estimates of population proportions given certain conditions. The method is now widely used in human immunodeficiency virus related studies among high risk populations globally. We test the RDS methodology by simulating RDS studies on the social networks of a large Lesbian, gay, bisexual and transgender Web community. The robustness of the RDS method is tested by violating, one by one, the conditions under which the method provides unbiased estimates. Simulations indicate that the bias is large if networks are directed or respondents choose to invite people on the basis of characteristics that are correlated with the study outcomes. The bias and variance increase if participants invite close as opposed to more distant friends whereas sampling in denser networks sharply reduces variance. However, the RDS method shows strong resistance to sampling without replacement, low response rates and certain errors in the participants’ reporting of their network sizes, as well as the selection criteria of seeds. The effects of network structure and the number of seeds and coupons are also discussed.