Respondent driven sampling: determinants of recruitment and a method to improve point estimation

Area of Science:

• Social Sciences
• Epidemiology
• Statistics

Background:

• Respondent-driven sampling (RDS) is a link-tracing design used to estimate characteristics of hidden populations.
• Potential biases in RDS arise from non-random coupon distribution or non-random participant presentation for interviews.
• This study investigates the sources of bias in an RDS study and proposes a weighting method to mitigate them.

Purpose of the Study:

• To explore biases in Respondent-driven sampling (RDS) studies stemming from non-random coupon distribution and interview presentation.
• To propose and apply novel weighting methods to reduce bias caused by non-random presentation for interviews in RDS.
• To evaluate the effectiveness of the proposed weighting method on population proportion estimates.

Main Methods:

• Analyzed data on coupon offers and interview presentations in relation to age and socioeconomic status.
• Estimated population proportions using inverse probability weighting for both coupon offers and interview presentations.
• Compared weighted estimates against unweighted RDS estimates to assess bias reduction.

Main Results:

• Younger men and men of higher socioeconomic status were under-recruited due to lower coupon offer rates and interview presentation rates, respectively.
• Weighting for non-random interview presentation by age and socioeconomic status significantly improved estimates for the lowest socioeconomic group (38% RMSE reduction).
• The weighting method also improved estimates for tribe and religion (19-29% RMSE reduction) but had minimal impact on age, sexual activity, or HIV status estimates.

Conclusions:

• Data on the characteristics of individuals to whom recruiters offer coupons can be leveraged to reduce bias in RDS studies.
• The proposed weighting method shows promise for improving the accuracy of RDS estimates, particularly for sensitive socioeconomic variables.
• Further research and validation are necessary to fully evaluate this novel bias-reduction technique in RDS.