Respondent-Driven Sampling and Spatial Autocorrelation

Abstract

Respondent-driven sampling (RDS) is a type of sampling method used to survey rare and hard-to-reach populations. RDS was developed to address the issue of bias associated with snowball sampling in qualitative research. Although, RDS has evolved by addressing major issues involved with the snowball sampling method, the issue of how the presence of spatial autocorrelation (SA) affects RDS had not been studied. SA refers to the clustering of similar attribute values in geographic space. Quantitative studies show that the presence of positive SA leads to an underestimation of the appropriate sample size. If RDS is not affected by SA, then the samples are expected to be dispersed in geographic space and not clustered around a sampling seed that initiates a sequence of respondents. This chapter presents impacts of SA on RDS when a social network displays a geographic pattern. The geographic distribution of the samples and associated socioeconomic and demographic variables are analyzed with respect to sequences of respondents. Social network RDS data for Rio de Janeiro, Brazil, are analyzed. Previous research indicates that, in these social network RDS data, samples are clustered around their initial seeds and do not spread out in geographic space as the sequence of respondents progresses. This tendency may result in increased sampling variance, which raises a concern about appropriate sample size determination in RDS.