Symposium: An Introduction to Modelling the Risk of Modern Slavery
While there are more people living in slavery now than at any point in human history, its measurement has become more difficult than ever as its illegal status has led to the crime becoming more hidden. The emergence of new and diverse forms of the crime have further increased the difficulty of arriving at a reliable estimate. Given the experience of measuring similarly “hidden” crimes, such as sexual assault and domestic violence through random sample population surveys, the Walk Free Foundation adopted this approach in an effort to develop reliable national estimates of modern slavery. Since 2014, a total of 54 nationally representative household surveys have been implemented through the Gallup World Poll across 48 countries. These surveys underpin the first global effort to model and predict the risk of modern slavery.
Although there is more data available on modern slavery since Walk Free first adopted national surveys to measure modern slavery, the challenge remains that surveys cannot be conducted in every country. In order to produce the most reliable estimates possible for the 167 countries covered in the 2018 Global Slavery Index, an extrapolation methodology using hierarchical Bayes models was developed. This statistical approach takes into account respondent-level survey data and country-level predictors in order to estimate country averages. These estimates are obtained using Bayes theorem, which helps completing the necessary computations, and can be used to incorporate prior knowledge about the prevalence of modern slavery. The analysis summarized here builds upon previous extrapolation-based approaches, making it possible to estimate the risk of modern slavery at the individual and country-level and inform prevalence estimates beyond the sample of 48 countries.
Predicting the risk of modern slavery
The analysis was based on survey data collected through the Gallup World Poll and country-level data from Walk Free Foundation’s vulnerability model. Data from the modern slavery surveys was used to estimate the risk model and a broader set of surveys was used for extrapolation purposes.
The process of estimating the prevalence of modern slavery in 167 countries began with identifying individual and country-level variables that have a significant relationship with forced labour and forced marriage at the individual level. On the individual level, demographic factors such as age, gender and employment status, as well as socio-economic and psychographic risk factors, such as feelings about household income, life evaluation scores and negative experienced affect, help predict risk, as well as country-level vulnerability factors.
Several models, each with a larger number of predictor variables, were tested before the “base” model was identified as the model achieving the best balance between predictive accuracy and geographic coverage. Multi-level models were then fitted in order to extrapolate results beyond the sample of 48 countries. Based on the individual-level risk factors identified, as well as country-level vulnerability scores, a hierarchical Bayes modelling approach was used to accurately predict the forced labour and forced marriage status of individuals. Average weighted predicted probabilities were then calculated using the best-fitting predictive model to estimate the average prevalence of modern slavery at the country level.
Implications for policy and next steps
Second, we have shown that a hierarchical Bayes modelling approach can be used to accurately predict the forced labour and forced marriage status of individuals and the average prevalence of modern slavery at the country-level. This is an important finding, but increased identification of victims in the surveys would allow for the expansion of our predictive models and further enhance the accuracy of our predictions.
Our analysis is not without limitations inherent to any cross-sectional research endeavour. We cannot ascertain the direction of causality, and it is quite possible, for example, that forced labour engenders lower life evaluation scores, rather than life evaluations being a protective factor. Another important consideration is that the worldwide coverage of the prediction data is not matched in the model estimation data. The latter includes a subset of countries that were selected based on criteria that leads to the exclusion of countries in Western Europe, Northern America and developed Asia. Essentially, this means that we cannot test whether the risk factors identified in our sample behave the same way in these regions.
Having said this, the current model is modest in scope and the risk factors unlikely to vary greatly across regions, that is, being female will remain a risk factor for forced marriage; being in a situation of poverty will remain a risk factor for forced labour. Data in developed countries would refine our understanding of risk factors in both low- and high-risk countries, building out our understanding of modern slavery and how best to tackle it.
This piece has been prepared as part of the Delta 8.7 Modelling the Risk of Modern Slavery symposium. Read all the responses here.
Jacqueline Joudo Larsen is Head of Research for the Walk Free Foundation.
Pablo Diego-Rosell is a Senior Consultant at Gallup.
This article has been prepared by Jacqueline Joudo Larsen and Pablo Diego-Rosell as contributors to Delta 8.7. As provided for in the Terms and Conditions of Use of Delta 8.7, the opinions expressed in this article are those of the authors and do not necessarily reflect those of UNU or its partners.
 Bayes hierarchical linear model with weak priors, 7 demographic predictors, 6 “Base” variables from the World Poll, one country-level predictor (Weighted Vulnerability Score), country-level random intercepts, and a cross-level interaction between currently owning a business and region (South Asia vs rest).
 The modern slavery module was only deemed suitable for face-to-face interviewing. Among these countries, those with high expected prevalence and/or large populations were prioritized, and lastly, countries were selected to provide a sufficient sample within each of the strata used for global estimation (see International Labour Office & Walk Free Foundation (2017). Methodology of the global estimates of modern slavery: Forced labour and forced marriage. ILO: Geneva.