Fighting Slavery through Statistics: A Discussion of Five Promising Methods to Estimate Prevalence in the United States

It is much easier to protect what one can count and identify. Modern slavery prevention and intervention efforts can become more effective and targeted with better information about how many victims exist and where they are likely to be found.

Measuring prevalence in modern slavery has historically presented several challenges. Definitional disagreements, lack of national coordination between all anti-slavery stakeholders and their data collection systems, and limited funding have played roles in the difficulties of obtaining a widely accepted and methodologically rigorous national prevalence estimate for human trafficking in the United States. Some global estimation efforts have included disaggregated national estimates for the U.S. alongside many other countries, but a more-focused and customized effort is required to estimate modern slavery in a country as populous, diverse, and critically nuanced as the U.S.

There are some concerns about the divisive repercussions of multiple, wide-ranging previous estimates in the U.S. and some of their particular foci on specific target populations, such as minors or foreign nationals, and industries such as commercial sexual exploitation of minors or agricultural labor.

As one of the first countries in the world to pass a comprehensive national anti-human trafficking law and one of the only countries in the world to publish its annual Trafficking in Persons Report through the Department of State, the U.S. has the opportunity to maintain its leadership and continue its progress on this critical human rights issue. One possible way forward is through more support for a national human trafficking estimate.

Modern slavery prevalence measurement in developed countries presents a consistent challenge to the existing survey-based methods of estimation, since the ineffective rule of law and generally low levels of extreme vulnerability among the populations in many of these countries make slavery harder to locate. However, other developed countries, such as the UK and the Netherlands, have managed to use alternative methods of obtaining national prevalence estimates. Unfortunately, if we only continue to study modern slavery in developing countries, where it may be traditionally easy to find through survey methods, we will mischaracterize the established and dominant role of stronger market economies and developed countries in this issue.

To better advocate for committed and sustained funding for anti-slavery efforts in the U.S., and to develop benchmarks and baselines for future research and progress measurement, it is critical to come together as a field and engage in collaborative efforts to determine national prevalence estimates. Success in the U.S. context would encourage many other developed countries facing similar constraints and contribute substantially to the information that exists about human trafficking in the United States, both domestically and from other countries.

There are three important reasons why better, and more-reproducible, slavery prevalence estimates are needed—to help:

1. Raise public and government awareness of the size of the problem.

2. Identify the complex resource needs of anti-slavery activities.

3. Assess the success or failure of specific anti-slavery activities.

High-level precision is particularly important for evaluating the effectiveness of anti-slavery activities that are targeted to a specific area before the activities are judged worthy of more general deployment. Such measures might include enhanced enforcement of existing laws; passage of specific new laws; vigorous information campaigns designed to educate people about how to identify, and then report, activities that could involve modern slavery. It is important to realize, however, that just because a measure “sounds good,” that does not prove its effectiveness. In fact, such measures might even prove to be counterproductive.

While there have been attempts to do this in the past, concerns have been raised that such studies focused only on specific populations, such as minors or foreign nationals, and on specific industries, such as commercial sexual exploitation of minors or agricultural labor. Some may argue that these industries are easier to identify or target, but that there are substantial issues related to networks such as domestic servitude and traveling sales crews that are more difficult to identify.

However, there are potential solutions to this. The following five methods are among the most promising opportunities to develop a national human trafficking prevalence estimate for the United States in the short- to medium-term future. Some of these efforts have been executed successfully in other developed countries, and other pilot or smaller studies indicate that national-level replication could be possible in the near future.

1) Multiple Systems Estimation (MSE) has been conducted successfully in the United Kingdom in 2014 and in the Netherlands in 2016. This method primarily applies a form of capture-recapture from administrative data from multiple concurrent lists of identifiable victims to estimate a victim population size. Basically, this method involves analyzing three to four concurrent lists of identifiable victims for apparent overlap—incidences where a specific victim was named on one or more lists in the same time period. This helps statisticians estimate how likely it is that the known list of victims collected by these groups is identifying the same victims and who they may be missing.

Several U.S. stakeholders are interested in applying MSE to administrative data in the United States, but are currently hindered by multiple jurisdictional, confidentiality, and systems-based challenges. However, this method is generally considered to be the most reliable and ideally suited for developed countries, where generally lower vulnerability conditions may result in overall lower prevalence rates, while stronger law enforcement and rule of law makes modern slavery more difficult to detect.

These conditions render nationally representative household surveys, such as those commissioned by the Walk Free Foundation in partnership with Gallup World Poll, less useful in the United States context. While developing countries may have survey respondents who are more likely to be personally affected by modern slavery, this is very difficult to detect with the standard sample sizes of 1,000 to 2,000 people in developed countries. Given these limitations and the fact that many developed countries maintain reliable and identifiable victim data, MSE becomes an attractive option in that context.

2) Respondent Driven Sampling (RDS) was originally introduced by Douglas Heckathorn (2002) to provide sufficient statistical parameters for chain-referral methods to determine statistically valid indicators and to counteract issues of initial sample selection, volunteerism, referral methods, homophily or non-random referral, and potential overrepresentation by those with larger personal network biases.

Originally, RDS—a process where initial survey respondents are identified and then asked to identify other respondents who share similar characteristics, in an attempt to locate difficult-to-find populations—was challenged by a lack of statistical parameters that would allow statisticians to infer the total population from this nonrandom sample. However, Heckathorn’s contributions allowed researchers to start using this effective method of identifying respondents, as well as to statistically infer information about the total population.

Sheldon Zhang has championed this work in the anti-slavery field among migrant workers in San Diego and in North Carolina, and improved upon the referral methods to further reduce the biases that have traditionally reduced the statistical validity of such approaches. While promising, this type of chain-referral approach also requires incentives for participation and may be difficult and costly to replicate on a national scale.

3) Taking online-based surveys of U.S. stakeholders is an important and relatively low-cost possible method of obtaining more information on human trafficking prevalence data and related issues, but would be difficult to use for statistical inference. There have been several promising efforts, such as the pilot online survey of United Against Slavery. The organization plans to begin integrating questions into its survey design related to human trafficking data and prevalence.

It is also worthwhile to consider methods for influencing the general population’s understanding of, interest in, and willingness to help eliminate contemporary slavery.

This could be done almost for free, using some form of social media, and at a relatively low cost with local advertising. Evaluating the impact of this type of information campaign would be relatively inexpensive, since traditional survey methods work with the general public.

4) National aggregation of statewide prevalence estimates is another possible method to determine human trafficking prevalence in the United States. A recent study from the University of Texas in Austin used available administrative data and targeted qualitative interviews in Houston, Texas, to estimate that 313,000 people are enslaved in Texas. This is a promising step for a large and populous state, but for this method to gain traction, other states will have to follow suit with their own statewide estimates.

It was noted earlier that prevalence estimates tend to be imprecise. As the following numbers show, “imprecise” is an understatement. Scaling that number of 313,000 modern slaves up to the world’s population, there would have to be about 86 million modern-day slaves on Earth. That number compares with the International Labor Organization’s global estimate of about 20 million and the Walk Free Foundation’s of about 46 million, of whom only 60,000 are in the United States.

Moreover, it is reasonable to assume that Texas has a below-world-average slavery rate. A recent study by Martin and Smith (2015) estimated that there were about 148,000 modern slaves in Texas, less than a half of the 313,000 estimate. These estimates were derived using different definitions of slavery, different data sources, and different analytic methods, and resulted in an extraordinarily wide range.

5) Gallup public opinion surveys on human trafficking awareness and willingness to report to hotlines and/or law enforcement is a potential method that has already yielded some compelling preliminary results. In early 2016, the Gallup World Poll included a set of items on the Gallup Daily Poll for the United States related to human trafficking awareness and the general public’s willingness to report human trafficking. This information is useful to the field due to the valuable, publicly accessible, and anonymous case and call information that is published online by the National Human Trafficking Hotline, as well as the importance of having a victim-centered and non-law-enforcement hotline known to survivors.

While the survey in 2016 indicated that many respondents would report a crime of human trafficking to law enforcement directly, an encouraging number of respondents were both aware of the National Human Trafficking Hotline and indicated that they would report crimes there. This method can be further developed alongside a more robust vulnerability model to try to determine the statistical validity of using the publicly available information on calls to the National Human Trafficking Hotline as an approximation of the incidence of human trafficking in the United States.

These efforts should also reflect the best practices in applied research and replication standards, including transparent and reproducible methods, as well as the development of theoretically grounded models. The development of a theoretically grounded model, in particular, is necessary to advance our understanding of vulnerabilities to human trafficking in the United States and around the world. To date, there has been moderate success to this effect by employing the United Nations-derived human security framework.

It is important to identify a large set of potential survey prevalence predictors for which local data are typically available—a challenge for slavery experts. A related, but different, version of this approach had been implemented by the Walk Free Foundation when it developed its global vulnerability model and prevalence estimates for more than 160 countries using the global slavery index, and also by Martin and Smith (2015), when they used regression analysis to estimate slavery statistics for the 50 U.S. states.

In general, showing the effectiveness of any intervention is complicated. It requires:

1. Conducting pre-intervention surveys in two or more reasonably similar areas.

2. Implementing the intervention in some of the area(s), but not in the “comparison” area(s).

3. Conducting post-intervention surveys in the same areas.

4. Comparing (average) pre-post change between intervention and comparison areas.

Surveys involved in such demonstration projects have to be well-designed. As noted earlier, RDSs, if they are carefully implemented, can produce unbiased estimates even for hard-to-reach hidden populations. In designing such studies, it is especially important to obtain final sample sizes that guarantee adequate power for detecting any meaningful change in slavery prevalence.

If it were possible to obtain the requisite number and type of victim lists from law enforcement agencies and/or from other organizations throughout the nation, MSE methods could also be used. The advantage of MSE over RDS is that MSE is simpler and costs less. The disadvantage is that MSE may not be able to provide all relevant details about the nature of local enslavement. In contrast, protecting respondent privacy by adequate methods of data anonymization presents an added challenge for RDSs.

Traditional survey methods can be employed to assess the general population’s understanding of, interest in, and willingness to help eliminate contemporary slavery. However, given the rapid changes in public awareness of human trafficking and diversity of thought on vulnerability modeling to human trafficking, this approach will require much more academic and practitioner collaboration and discussion, as well as possibly an updated survey.

The U.S. federal government is making significant progress in evaluating the feasibility of MSE for the U.S. context and human trafficking data. Many stakeholder groups and nonprofits, such as the McCain Institute and United Against Slavery (UAS), continue to convene workshops and efforts to address these issues.

Unfortunately, statistically valid direct methods, such as MSE and RDS, are unlikely to be practical for estimating prevalence at the national level. However, we think that a promising estimation strategy could be constructed by regressing statistically valid prevalence estimates that were obtained for a representative sample of local areas on a set of potential predictors of slavery prevalence that are available at appropriate local levels, possibly including markers inspired by the human security theory.

This approach involves major challenges for statisticians, as well as for experts on modern slavery. Given that obtaining survey-based unbiased prevalence data is likely to be the most time-consuming and expensive component of the proposed approach, it is important to design a minimally adequate set of local areas for survey-based prevalence estimation. This is the statistician’s challenge.

Note: This article builds upon Dr. Durgana’s original post, published by DEVEX, and includes input from Dr. Zador after their joint technical discussion at the American Association for the Advancement of Science (AAAS) in March 2017.

Further Reading

Aggarwal, C.C., and Philip, S.Y. (eds.) 2008. A general survey of privacy-preserving data mining models and algorithms (PDF download). Privacy-preserving data mining: Models and algorithms. New York, NY: Springer US, pp. 11–52.

Bales, K., Hesketh, O., and Silverman, B. 2015. Modern Slavery in the UK: How Many Victims? Significance, pp. 16–21.

Barrick, K., Lattimore, P.K., Pitts, W.J., and Zhang, S.X. 2014. When farmworkers and advocates see trafficking but law enforcement does not: Challenges in identifying labor trafficking in North Carolina. Crime, Law and Social Change 61(2), pp. 205–214.

Busch-Armendariz, N.B., Nale, N.L, Kammer-Kerwick, M., Kellison, J.B., Torres, M.I.M., Cook-Heffron, L., and Nehme, J. 2016. Human Trafficking by the Numbers: Initial Benchmarks of Prevalence & Economic Impact in Texas (PDF download). Austin, TX: Institute on Domestic Violence & Sexual Assault, University of Texas at Austin.

Durgana, D. 2017. Fighting Slavery Through Statistics: Five Promising Methods to Estimate Prevalence. DEVEX.

Heckathorn, D.D. 2002. Respondent-Driven Sampling II: Deriving Valid Population Estimates from Chain-Referral Samples of Hidden Populations (PDF download). Social Problems.

Martin, H.M., and Smith, L.M. 2015. Historical overview and demographic analysis of human trafficking in the USA. International Journal of Public Law and Policy 5(3).

van Dijk, J.J.M., and van der Heijden, P.G.M. 2016. Multiple Systems Estimation for estimating the number of victims of human trafficking across the world, Research Brief. Vienna, Austria: UNODC.

Waldron, S. 2017. Experts Leverage Statistical Methods to Investigate Human Trafficking. American Association for the Advancement of Science (AAAS).

United Against Slavery (UAS).

Zhang, S.X. 2012. Looking for a hidden population: Trafficking of migrant laborers in San Diego County (PDF download). Final Report submitted to the National Institute of Justice, Office of Justice Programs, U.S. Department of Justice.

About the Authors

Davina P. Durgana, PhD is senior statistician and report co-author of the Global Slavery Index of the Walk Free Foundation. She was recently named the 2016 Statistical Advocate of the Year by the American Statistical Association and as a Forbes Top 30 Under 30 in Science in 2017 for her work on statistical modeling, human security theory, and human trafficking. She applies analytical models to understanding vulnerability, risk, and prevalence of human trafficking domestically and internationally.

Paul L. Zador, PhD is a senior statistician at Westat, Inc., an employee-owned research organization. His research interests have covered a range of topics, including statistical communication theory, public health, transportation safety, policy evaluation, research design, and data analysis. In 2006, working with other statisticians, Zador developed a sampling plan and analyzed data that documented a multi-decade-long period of terror in Guatemala based on the recently discovered archive of the Guatemalan police. More recently, Zador helped design a survey of bonded laborers in Tamil Nadu, India, and analyzed the data that was collected.

Back to Top

Tagged as: , , , ,

Leave a Response

Please note: comment moderation is enabled and may delay your comment. There is no need to resubmit your comment.