Measuring Vulnerability and Estimating Prevalence of Modern Slavery

Addressing modern slavery is a significant challenge for governments and non-government organizations (NGOs), whose efforts are hampered by a range of factors, not least a lack of understanding what increases the risk of enslavement and the extent of the crime at a national level. Slavery is a hidden and diverse crime, and understanding vulnerability to slavery remains a work in progress, although we have learned a great deal about the factors that help modern slavery flourish from practitioners and qualitative researchers.

Empirical evidence suggests a connection between slavery and problems such as corruption, conflict, poverty, discrimination, and the impact of a weak rule of law, poor or declining economic conditions, and adverse environmental change. While this is an important starting point, without measurement, we cannot know where to focus interventions.

To provide a reliable evidence base upon which governments, civil society groups, and businesses can build more-effective responses, the Global Slavery Index (GSI) has applied a statistical approach to identifying the factors that are correlated with increased risk of enslavement. Central to the Index are an assessment of factors that explain or predict the prevalence of modern slavery, and an estimation of the prevalence of modern slavery in 167 countries. This article discusses the development of the vulnerability model and the methods applied in the estimation of modern slavery for the GSI.

Image courtesy of Walk Free Foundation, Global Slavery Index.

Understanding Vulnerability

Vulnerability to modern slavery is affected by a complex interaction of factors related to the presence or absence of protection; respect for rights; physical safety and security; access to the necessities of life such as food, water, and health care; and patterns of migration, displacement, and conflict. Understanding vulnerability is a critical component to estimating prevalence of human trafficking. Vulnerability assessments allow us to determine countries with similar risk profiles more objectively by looking at a wide array of risk factors that help us better understand the drivers causing human trafficking to flourish.

Theorizing vulnerability to slavery

The GSI vulnerability model is guided by human security and crime prevention theories. Human security as a developing security subfield has many overlapping and diverging definitions without any clear “consensual definition” (Tadjbakhsh and Chenoy, 2007) among scholars. The most basic shared concept of human security involves a focus on the safety and well-being of individuals, regardless of their citizenship status or relationship to a nation-state.

Much of the literature focuses on human security’s rise to prominence after the Cold War, reflecting a new type of security studies that no longer privileges state security. As Graham and Poku (2000) state, traditional security is a focus on “individuals qua citizens,” while human security is a focus on “individuals qua persons.” Importantly, the field of human security represents a shift away from state-centered or traditional security to a focus on individuals as the referents of security.

This method of using human security theory to conceptualize vulnerability in relationship to global human trafficking prevalence is an attempt to connect to a larger discourse on vulnerability and to force us to think comprehensively about the potential risks that we are evaluating in the Index. There is also value in theoretically deriving and guiding our thinking on vulnerability as a more-defensible position. We will also be contributing to a larger academic literature on how these models are constructed in general. Now, the consideration of human security theory and literature lets variable be included, excluded, and guided in part by a theory, while remaining an empirically exploratory approach.

The Current Vulnerability Model

The 2016 Index includes an assessment of vulnerability used to measure the factors linked to the probability of enslavement in each country and to cluster countries that are similarly vulnerable to slavery. This provides a starting point for estimating slavery in countries for which no reliable, national-level data exists. Underpinning the building of this model is the need to quantify the risk of enslavement in ways that can be compared across countries and regions. Durgana and Brown discuss the challenges of global data reporting and the limits it imposes on the work described on page 36 of this issue.

Image courtesy of Walk Free Foundation, Global Slavery Index.

Creating and testing the vulnerability model

The model has undergone several rounds of review and iterative improvements since 2014. When preparing for the 2016 and 2018 GSIs, the existing model was considered to be a useful starting point, since results generated by this model tested well through extrapolation against newer survey results (Joudo Larsen, Datta, and Bales, 2015). These results were fairly close (within one percentage point when considered as a proportion of the population in 2016), which validated the decision to use the 2014 model as the starting point. When comparing the 2014 model against a comprehensive model of human security, it was clear that there were important gaps in the list of variables (for example, there were no available data on environmental security and gender-based insecurity across all countries).

In 2016 and 2018, where data was available, existing variables were updated with more regularly updated, reliable, and/or current sources, and new variables not previously available were included. The steps taken to create the vulnerability model were:

1. Identification of relevant variables after a review of human security and crime prevention theories

2. Assessment of variables using (i) data updated and published regularly that (ii) possess an explicit and transparent methodology, (iii) are available for the majority of the countries examined, and (iv) are based on reliable data;

3. Compiling an initial list of variables for statistical testing, using logarithmic transformation of variables where the data was not normally distributed;

4. Normalization of the data to a linear scale (1 to 100) where 100 represented greatest vulnerability to slavery;

5. Applying multi-collinearity checks to the full list of variables, using a threshold of 0.8 as high collinearity;

6. Conducting Principal Factor Analysis to identify which of the variables group together into distinct dimensions;

7. Organizing variables into relevant dimensions;

8. Using imputation methods to address missing data at the dimension level in 2016 and the variable level by region for 2018;

9. Calculating dimension-level vulnerability scores by averaging the relevant data for each country;

10. Calculating an overall score for each country by averaging the sum of the dimension scores for each country.

The resulting 2016 vulnerability model comprises 24 variables that fall into four dimensions through factor analysis: Political Rights and Safety; Financial and Health Protections; Protection for the Most Vulnerable; and Conflict.

Dimension 1: Political Rights and Safety

The Political Rights and Safety dimension encompasses measures of: confidence in judicial system, political instability, access to weapons, same same-sex rights, displaced persons, government responses to slavery, and political rights and civil liberties.

Dimension 2: Financial and Health Protections

The Financial and Health Protections dimension encompasses measures of: financial inclusion (the ability of an individual to borrow and to receive wages), access to cell phones, social security provisions available to citizens, undernourishment, incidence of tuberculosis, and access to improved water sources.

Dimension 3: Protection for the Most Vulnerable

The Protection for the Most Vulnerable dimension encompasses measures of financial inclusion (an individual’s ability to get access to emergency funds); violent crime; women’s physical security; income inequality; and discrimination toward immigrants, minority groups, and people who are intellectually disabled.

Dimension 4: Conflict

The Conflict dimension encompasses measures of the impact of terrorism, the presence of conflicts within a country’s legal boundaries, and the total number of refugees.

Table 1—Retained Variables and Factor Loadings from Principal Factor Analysis

Future Improvements

Following intensive consultations with the GSI Expert Working Group, several areas for improvement were considered for the 2018 vulnerability model. These include, but are not limited to:

a) Including an environmental degradation variable to cover an existing gap in environmental insecurity information;

b) Including the United Nations Statistical Capacity Indicator, which can help make empirically sound determinations on post-estimation adjustments, based on data reliability;

c) Normalizing all variables to be tested, using an alternative to the current standard approach (where minimum values are set to 0 and maximum values are set to 100), in which the means of all variables are set to 0 and standard deviations to 1;

d) Possibly weighting the resulting dimensions or factors by eigenvalues to distinguish the most-powerful dimensions from the less-influential ones on vulnerability to trafficking.

Ensuring the vulnerability model contains the groundwork for year-to-year comparability is a growing area of interest, since the model continually improves and develops. Ultimately, strong development of the GSI vulnerability model will assist in the long-term aim of developing a predictive model to understand global trends in slavery prevalence.

Estimating Prevalence

Measuring the number of people in modern slavery is difficult due to the hidden nature of this crime and low levels of victim identification. In 2016, the prevalence estimates from the Index were based on new data from nationally representative, random sample surveys conducted through the Gallup World Poll in Bangladesh, Bolivia, Brazil, Cambodia, Chile, Dominican Republic, Ethiopia, Ghana, Guatemala, Hungary, India, Indonesia, Mauritania, Mexico, Myanmar, Nepal, Pakistan, Philippines, Poland, Russia, South Africa, Sri Lanka, Tunisia, and Vietnam. These surveys involved face-to-face interviews with more than 28,000 respondents.

We conducted an additional survey in Thailand, but excluded it from this analysis because it was administrated only in Thai and had substantially limited coverage of the population of interest. Using a more-appropriate sampling method for these cases will be considered in the future.

The prevalence estimates from the 2016 Global Slavery Index reflect one of three methodologies:

1. Direct estimation after a nationally representative random sample survey (25 countries)

2. Multiple systems estimation (2 countries)

3. Extrapolation based on statistical modeling of risk to which relevant survey data has been applied (140 countries)

1. Direct estimation through surveys

As part of ongoing improvements to the Index methodology, WFF tested the use of surveys about modern slavery in seven countries in 2014, conducted as part of the Gallup World Poll, which represents 95 percent of the world’s population. The target sample is the entire civilian non-institutionalized population, age 15 and older.

The initial set of questions was developed in consultation with the GSI Expert Working Group and Gallup. The test questions focused on identifying situations experienced by a respondent or by a respondent’s immediate family members, and fell into two broad categories: unfree labor, and forced marriage.

These questions were then refined in early 2014 through testing for each language. Larsen and Diego-Rossell describe the development of the survey instrument and the methodology in detail in this issue.

2. Multiple systems estimation

Multiple Systems Estimation (MSE) has been used by human rights data analysts in recent years to estimate hidden populations in conflict situations, such as casualty counts in the Syrian civil war (Human Rights Data Analysis Group, 2013). MSE applies a form of capture-recapture methods to estimate a population size for multiple lists.

MSE can be applied in countries where nationally representative random sample surveys will not necessarily work. This is particularly the case in more “developed” countries, where low levels of vulnerability mean that there are few cases to report; law enforcement is strong and organized crime is more hidden; and the resulting numbers are so small that even if they were not hidden, they would be highly unlikely to be found and selected for interview in a random sample survey (Bales, Hesketh, and Silverman, 2015).

The United Kingdom government used MSE in 2014 to estimate the number of people in modern slavery. Drawing on data from the UK National Crime Agency Strategic Assessment, six lists were compared to derive the estimation of between 10,000 to 13,000 people in modern slavery. The midpoint of this range was used to calculate the percent of the population estimated to be in modern slavery in 2015 (the year the estimate was calculated). That is, 11,500/64,097,085 x 100 = 0.02 percent. This figure is used in the Index for the UK and to extrapolate for countries with a similar risk profile, predominantly those in Western Europe. The Royal Statistical Society recently published a paper summarizing the process (Bales, Hesketh, and Silverman, 2015).

MSE was also tested in the Netherlands in 2016 (van Dijk and van der Heijden, 2016). This work resulted in an estimate of 17,500 for the Netherlands, which equates to 0.104 percent of the population. Cruyff, van Dijk, and van der Heijden describe the application of this method in detail in this issue.

3. Calculating prevalence for non-survey countries

Where no hard data point was available from the methods described above, an estimate was extrapolated from the data available. The extrapolation process takes the GSI vulnerability measures as a starting point. The process followed these steps:

1. Group and order countries based on vulnerability measures, from low to high. This aggregated countries that were more alike compared to those that were less alike.

2. Within each group, identify existing survey data points. The average proportion of enslavement from these surveys was calculated for each group and applied to all countries within the group for which there were no survey data.

3. Make a limited set of adjustments to better account for conflict; geopolitical concerns; state-sanctioned forced labor; and a final, downward adjustment for Small Island Developing States (SIDS). (There were three diversions from this general rule; a similar adjustment was made for Madagascar, since it has a socio-economic environment similar to SIDS; while the UN lists Singapore as a SIDS, it is markedly different from other nations in this list due largely to stronger economic conditions, which is a strong pull factor; and no adjustment was made for Haiti—although it is recognized as one of the SIDS, there was high confidence in the existing data point for Haiti, which was based on random sample survey data.)

Data Points Collected through Direct Surveys or MSE

In 2014, survey data made it possible to calculate minimum estimates of migrant workers enslaved in Qatar, Saudi Arabia, and Malaysia (the data for these countries were not applied to other countries in their grouping) with reference to the relevant labor force in each country. These data points, and those for Brazil, Ethiopia, Indonesia, Nepal, Nigeria, Pakistan, Russia, and DRC, were treated slightly differently within their cluster averages. In addition, new data points for the United Kingdom and the Netherlands, derived by applying MSE, have also been incorporated.

In total, 28 data points (25 from the WFF surveys, two derived from the application of multiple systems estimation in the UK and the Netherlands, and one survey in DRC; Johnson, Scott, Rughita, et al., 2010) were then used as the foundation from which to extrapolate to the remainder of the 167 countries (see Table 2).

Table 2—Data Points for Extrapolation

To make sound decisions regarding the applicability of these 28 data points to the countries for which no reliable, national-level data exist, the WFF team grouped countries based on the measures of vulnerability. Certain countries have been retained in our model despite missing data points on Dimensions 3 and 4 of the vulnerability measures. Barbados, Brunei, Cape Verde, Equatorial Guinea, North Korea, and Suriname are all in this group, so their results should be interpreted with caution. These countries have less than 50 percent of data available on Dimensions 3 and 4; that is, data available for 4 or fewer variables in the 10 variables in both Dimensions 3 and 4.

As was the case in 2014, a K-means cluster analysis was run to group the 167 countries into distinct groups. In simple terms, K-means is a statistical method that groups similar items into clusters, ensuring that items not in the same cluster are as different as possible. This is achieved by allocating an item to the cluster with the nearest centroid—the mean of the cluster. The cluster’s mean is then recalculated and the process of allocating items to clusters begins again until no items change groups, or those changes do not make a substantial difference in the definition of clusters.

Tests were run to determine the ideal group size between 10 and 15 groups. We found that 12 groups had the highest Pseudo-F scores. These groups were sufficiently distinct on overall mean values, although the minimum and maximum values did indicate some overlap among countries at the bottom of one list and top of the next. The final grouping of countries is set out in Table 3.

Table 3—Allocation of Countries to 12 Groups as a Result of K-means Clustering

The research team reviewed where the data points fell across these groups (see Table 4) and used this information to calculate the average proportion of a population in modern slavery for each cluster. Where no survey data points were available in a cluster, the average of the two surrounding clusters was applied. This average became the starting point for the extrapolation process and was applied to each country in a cluster for which a survey was not available.

Table 4—Survey Data by Group

A limited set of adjustments were made to better account for state-sanctioned forced labor, conflict, and geopolitical concerns. A final, downward adjustment for Small Island Developing States ensured the extent of enslavement was not over-estimated in these nations. A list of all countries affected by the four types of adjustment is presented in Table 5.

Table 5—Summary of Countries Affected by Adjustments

Conclusion

Since its inception in 2011, the Global Slavery Index has remained committed to the iterative science and constant improvement of its models and techniques. Equally committed to transparency in our methods and efforts, we invite everyone to provide constructive feedback and criticism in our joint effort to better understand the conditions of exploitation and slavery that persist around the world and within our countries.

Our intent for the impact of this work is global and far-reaching. Since the Global Slavery Index was first published, we have seen a substantial increase in media coverage, and significant government and civil society initiatives to tackle global slavery. Findings from the Index have been quoted in parliamentary discussions, used as material in government and business workshops, and disseminated in numerous publications around the world. More recently, GSI data has been used in the commercial sector to identify and rank risks in supply chains—information that is critical to taking steps to prevent and mitigate the risk of slavery in business operations.

In facing a challenge this large, the entire anti-trafficking field must pool its best thinking and resources to leverage our counter-attack. While measurement may not seem the most obvious method to fight this crime, it is essential in setting a baseline against which progress can be measured and effective interventions identified.

Further Reading

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

Graham, D.T., and Poku, N.K (eds.). 2000. Migration, Globalization and Human Security. London: Routledge.

Human Rights Data Analysis Group.

Johnson, K., Scott, J., Rughita, B., Kisielewski, M., Asher, J., Ong, R. 2010. Association of Sexual Violence and Human Rights Violations with Physical and Mental Health in Territories of the Eastern Democratic Republic of the Congo. Journal of the American Medical Association 304(5).

Joudo Larsen, J., Datta, M.N., and Bales, K. Modern Slavery: A global reckoning. Significance 12, (5).

Tadjbakhsh, S., and Chenoy, A.M. 2007. Human Security: Concepts and Implications. London: Routledge.

UK Human Trafficking Centre. 2014. NCA Strategic Assessment: The Nature and Scale of Human Trafficking in 2013, Ref. 0093-UKHTC. National Crime Agency.

Van Dijk, J., and van der Heijden, P. G.M. 2016. On the potential of Multiple Systems Estimation for estimating the number of victims of human trafficking across the world. Paper presented at the UN Crime Commission, Vienna, Austria.

About the Authors

Jacqueline Joudo Larsen is a criminologist and senior research manager at the Walk Free Foundation and author of the Global Slavery Index. She oversees the survey research program and previously led research on human trafficking, international students, and violent extremism at the Australian Institute of Criminology.

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.

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.