Respondent-Driven Sampling: An Overview in the Context of Human Trafficking

Jessica P. Kunke, Adam Visokay, Tyler H. McCormick

TL;DR

Studying human trafficking is notoriously difficult because the population is "hidden" by stigma, illegality, and restricted movement. This paper provides a comprehensive overview of Respondent-Driven Sampling (RDS)—a chain-referral technique that leverages social networks to reach these marginalized groups—while highlighting the unique statistical and ethical hurdles encountered when applying it to the trafficking context.

What Problem Does This Paper Address?

Traditional survey methods rely on a "sampling frame"—essentially a master list of the population from which researchers can randomly pull participants. For human trafficking, no such list exists. Victims are often isolated, distrustful of authority, and may not have autonomy over their own movements.

This creates a massive data gap: current global estimates of trafficking victims range wildly from 12.3 million to 45.8 million. Without reliable data, policymakers cannot effectively allocate resources or evaluate interventions. The authors explore how RDS attempts to bridge this gap by turning the "hard-to-reach" nature of the population into a sampling advantage through peer-to-peer recruitment.

Methods

The paper details the mechanics of Respondent-Driven Sampling (RDS), which functions as a hybrid between a sampling strategy and a statistical estimation method:

  1. The Seed Phase: Researchers start with a small "convenience sample" of known individuals from the target population, called "seeds."
  2. The Recruitment Chain: Seeds are given a limited number of coupons to recruit peers. When a recruit completes the study, they receive a primary incentive, and the person who recruited them receives a secondary incentive.
  3. Waves: This process repeats in "waves." Mathematically, if the chain continues long enough, the sample "forgets" the initial bias of the non-random seeds and begins to reflect the actual diversity of the underlying social network (similar to a Markov Chain Monte Carlo process).
  4. Statistical Correction: To account for the fact that "popular" people with many contacts are more likely to be recruited, researchers use Horvitz-Thompson estimators. These re-weight the data by the inverse of a participant's "degree" (number of social connections), giving more weight to the responses of more isolated individuals.

The paper also discusses newer variants like Network Sampling with Memory (NSM), where researchers strategically direct recruitment toward "bridge nodes" to explore the network more efficiently, and Randomized RDS (RRDS), which uses phone-based randomization to reduce recruitment bias.

Datasets

As a methodological overview and review paper, this work does not introduce a new primary dataset. Instead, it synthesizes findings and challenges from several landmark RDS studies, including:

  • Studies of sex workers in Serbia, Montenegro, and Russia (Simic et al., 2006).
  • Trafficking prevalence among unauthorized migrant laborers in San Diego (Zhang et al., 2014).
  • Risk factors for sex trafficking in Ohio (Chohaney, 2016).
  • Surveys of women exchanging sex in New York City (Carrillo et al., 2020).

Key Insights and Ideas

  • The "Reciprocity" Requirement: RDS assumes that if Person A can recruit Person B, Person B is equally likely to recruit Person A. In trafficking, this often fails. Power dynamics or "stranger" referrals (where a recruiter barely knows the recruit) can violate this assumption, leading to biased estimates.
  • The Mobility Trap: A major assumption of RDS is that participants have the mobility to visit a study site or redeem a coupon. In many trafficking scenarios (like controlled brothels), victims are physically prevented from leaving, meaning RDS may systematically exclude the most severely exploited individuals.
  • Incentive Goldilocks Zone: Incentives must be high enough to motivate participation but not so high that they encourage "non-targets" (people not in the group) to try to talk their way into the study.
  • Trust as Infrastructure: The success of RDS isn't just about the math; it’s about the "social capital" of the seeds. If the initial seeds don't trust the researchers, the recruitment chain will "die" before it reaches the necessary depth to overcome bias.

Key Results

The paper highlights that while RDS is gaining traction—used in roughly 16% of human trafficking measurement studies reviewed by Barrick and Pfeffer (2021)—it is not a "silver bullet."

Quantitatively, the authors note that violations of RDS assumptions (like non-random recruitment or lack of network connectivity) can lead to "substantial issues with statistical inference." For example, in the San Diego study of migrant laborers, the "seeds" were all Mexican, which resulted in a sample that failed to represent non-Mexican unauthorized migrants due to "ethnic homophily" (the tendency to associate with people like oneself).

Limitations and Future Work

The authors identify several critical limitations and areas for growth:

  • Network Bottlenecks: If a social network is fragmented into "pockets" (e.g., street-based workers vs. organized brothel workers), the recruitment process can get "stuck" in one pocket and fail to see the rest of the population.
  • Hidden Degrees: Accurately estimating a participant's "degree" (how many people they know in the target population) is difficult. People may underreport or simply not know the status of their peers.
  • Hybrid Approaches: The most promising future direction involves combining RDS with other methods, such as Mark-Recapture (comparing overlaps between different sampling efforts) or Venue-Based Sampling, to create more robust population estimates.
  • Technology: Moving toward phone-based or digital RDS may help reach those with restricted physical mobility, though it introduces new risks regarding digital surveillance and safety.

Summary updated

Abstract

Respondent-driven sampling (RDS) is both a sampling strategy and an estimation method. It is commonly used to study individuals that are difficult to access with standard sampling techniques. As with any sampling strategy, RDS has advantages and challenges. This article examines recent work using RDS in the context of human trafficking. We begin with an overview of the RDS process and methodology, then discuss RDS in the particular context of trafficking. We end with a description of recent work and potential future directions.
BibTeX
@inproceedings{kunke2023respondent,
  author = {Jessica P. Kunke and Adam Visokay and Tyler H. McCormick},
  title = {{Respondent-Driven Sampling: An Overview in the Context of Human Trafficking}},
  booktitle = {CHANCE},
  year = {2023},
  doi = {10.1080/09332480.2023.2290949},
}
Figure