Hui Yi

How can emerging statistical methodologies improve the global response to human trafficking?

Abstract:

Having accurate prevalence data is critical for enabling informed decisions about how to allocate scarce resources to strengthen human trafficking (HT) response. Traditional prevalence estimation strategies, namely those utilizing probability sampling designs that are based on combinations of stratified and multistage sampling, have long been used to study hidden populations. Such traditional survey strategies may give estimates that are biased and/or have low precision, typically due to issues relating to under coverage and small sample sizes that correspond to observations made on the hidden population. While all estimation methods have unmeasured errors, existing biases in measurement have motivated the development of more robust statistical methods for prevalence estimation of hidden populations, such as those constructed from network-based and secondary data collection sources. These include network scale-up method (NSUM), respondent-driven sampling, Vincent link tracing sampling, time–location sampling, all of which directly engage with members of the hidden population, such as those being trafficked, and multiple systems estimation and mark-recapture methods, which rely on administrative data. Challenges were frequently encountered when implementing the methods, such as data deficiencies and violations of assumptions which can impact estimation accuracy, but scant literature provided solutions to these challenges.

Drawing from our ongoing projects in West Africa where the traditional survey method involved a probability-based, stratified, and clustered multistage sampling design in which adult respondents in 3,070 households were interviewed about trafficking of children who reside in their household in three selected districts, I present the first attempt to estimate the prevalence of child trafficking using NSUM, which entailed questioning the same adult respondents about the trafficking-related activities of children in their personal networks. I also present novel advancements to improving the accuracy of prevalence estimation, including the development of the new top-code strategies for addressing unreliable data reported by the respondents in the survey interviews, and the application of Small Area Estimation methods to 40 chiefdoms, the lower administrative-level units within the districts, and compared to district-level prevalence estimates.

The presentation is expected to spark deep discourse in the effectiveness of statistical methods towards prevalence estimation, drawing greater attention to the potential innovations in the methodologies contributing to the reduction in prevalence of HT in West Africa and other corners in the world.

Bio:

Hui Yi is an Assistant Research Scientist/Statistician at the University of Georgia School of Social Work Center on Human Trafficking Research & Outreach since 2020. As a researcher/co-Investor working on $30+ million projects funded by the U.S. Department of State, including a newly granted $10 million project based in southern African countries running from 2022 to 2028, her daily work involves grant proposal development and executing quantitative research activities, including sampling design, survey design, supervising data collection in collaboration with local research partners, with a cumulation of over 12,500 survey data being originally collected from 13 hotspot regions of five African countries, i.e., Sierra Leone, Guinea, Senegal, Zambia, and Malawi. With these, she performs large-scale prevalence estimation of various forms of labor/sex exploitation using traditional survey methods and advanced statistical models such as Basic and Bayesian Network Scale-Up Methods, Respondent Driven Sampling Method, Longitudinal Analysis, Machine Learning Methods, etc.

She used to be a tenured faculty in a university in China for five years and served as a Principal Investigator on several intramural and extramural grants in optimizing highway infrastructure management using temporal spatial models and machine learning techniques. She taught six undergraduate/graduate courses in English and Mandarin and supervised more than 5 graduate and doctoral students. Her work has been published on the top journals in the application field of criminology, violence, genetics, and traffic safety, such as Trauma, Violence, & Abuse (Impact Factor: 8.29), Journal of Quantitative Criminology (5.08), Crime & Delinquency (2.85), Genetics (4.40), Traffic Injury Prevention (2.20).